Author SHA1 Message Date
mesh-tools cacf537004 Merge pull request 'The checker's failure is the build step's failure, and a verdict that hides its failing test is refused (issue 1901)' (#84) from fix/1901-failure-contract into main 2026-10-26 23:44:21 +00:00
jschoubben f4a86f5884 Merge main into fix/1901-failure-contract: the checker pins mesh-sdk a2759ec, the group's sdk head (issue 1901)
mesh/delivery merged: source.main moved to novox/mesh-delivery@cacf537004c9
mesh/delivery-group group fix/1901-failure-contract merged: every member is merged
mesh/check checking f4a86f58843d on cacf537004c9 | novox/mesh-delivery#84 in group fix/1901-failure-contract
2026-10-27 00:39:35 +01:00
mesh-tools 9f1b227fb8 Merge pull request 'The indexer fills each repository's taken key from merges it applied or a drained stream's heads, so a merge heard while the state was missing reads on the machines (issue 1280)' (#83) from fix/1280-indexer-fills-taken into main 2026-10-26 23:22:01 +00:00
jschoubben 6c79e456a1 The indexer confirms a forge message's ack with the server before it fills taken keys (issue 1280)
mesh/check checking 6c79e456a100 on 9f1b227fb8c0
mesh/delivery merged: source.main moved to novox/mesh-delivery@9f1b227fb8c0
2026-10-27 00:20:16 +01:00
jschoubben 0b980203fe Taken keys are filled only from merges the indexer applied or a drained forge stream's heads (issue 1280) 2026-10-27 00:18:23 +01:00
jschoubben 73bf290f58 The indexer fills every repository's taken key from the source's head at start and when the state appears (issue 1280) 2026-10-27 00:14:41 +01:00
mesh-tools 475bb4bb0b Merge pull request 'The view's tools read their own reach by its bare name, and a test holds every state the commands read to what is issued (issue 1280)' (#82) from fix/1280-view-reads-its-reach into main 2026-10-26 22:48:25 +00:00
jschoubben efadc691e7 The view's tools read their own reach by its bare name, as the mesh issues it, and a test holds every state the commands read to what is issued (issue 1280)
mesh/check being carried to efadc691e7ea on 475bb4bb0b8f | from 4af36c69a761
mesh/delivery merged: source.main moved to novox/mesh-delivery@475bb4bb0b8f
view_reach asked for mesh-delivery-view.reach, a name the mesh never issued, so it failed on the machines;
the same line was not gofmt'd, which failed the try of main. Quick fix: the name. Thorough for this module: the
manifest test now refuses any stdio.State literal in cmd/ that is neither an own state nor listed in reads.
2026-10-26 23:27:05 +01:00
mesh-tools 4af36c69a7 Merge pull request 'Delivery records are true at the source: the indexer records what it took in, the view derives what reached the machines, a settle verb for held deliveries (issue 1280)' (#80) from fix/delivery-records-true into main 2026-10-26 21:48:50 +00:00
mesh-tools bd107d2015 Merge pull request 'A version that fails on one machine while another runs it healthy is that machine's failure, and goes no further meanwhile (issue 1496, ADR 0557 §1-§3)' (#81) from fix/1496-start-health into main 2026-10-26 21:37:58 +00:00
jschoubben f007ca4227 A node that runs a version alone-failed elsewhere still gets its want at it (ADR 0557 §2, issue 1496)
mesh/delivery merged: source.main moved to novox/mesh-delivery@bd107d20154c
mesh/delivery-group group fix/1496-start-health merged: every member is merged
mesh/check checking f007ca42279a on 4b36cc29a6ae | novox/mesh-delivery#81 in group fix/1496-start-health
Review of 50c4674: the bar on new wants naming F while a node fails it alone
also caught a node that runs F with no live want, the issue-1945 recovery path:
it went on neither list and its node-engine gave the module up. The bar now
holds only a node that neither wants nor runs F.
2026-10-26 22:34:25 +01:00
jschoubben 50c4674dfa While a node fails a version alone, no further node gets a want for it (ADR 0557 §2, issue 1496)
ADR 0557 version 9: a failure that is one node's holds nothing and sets nothing
back, but the version reaches no node that does not want it yet until it is
over. A node whose want does not name F gets no new want naming it (rule 2, in
words naming the failing nodes); wants already written stay.
2026-10-26 22:30:00 +01:00
jschoubben c12b870104 The indexer binds its taken state when it appears, as issue 1790 asks of every state a delivery adds: an indexer started before the bus holds it still takes merges in and says the key it did not write (issue 1280, ADR 0556 clause 1)
mesh/check checking c12b870104e8 on bd107d20154c | novox/mesh-delivery#80 in group fix/delivery-records-true
mesh/delivery merged: source.main moved to novox/mesh-delivery@4af36c69a761
mesh/delivery-group group fix/delivery-records-true merged: every member is merged
2026-10-26 22:25:39 +01:00
jschoubben 92aedb9365 The first pass after a start clears failed-alone for every module and known node (ADR 0557, issue 1496)
Review round 2 of 16e4e61: gating the first-pass clear on a failed have left a
condition an earlier process raised standing for good once its node recovered
while no assigner watched. The first pass now clears it for every module and
every node the presence bucket knows, a quiet clear at the router for one that
is not open.
2026-10-26 22:21:00 +01:00
jschoubben 368204115c Pin the teller at mesh-sdk 68d65e6, the group's sdk head merged with sdk main (issue 1280) 2026-10-26 22:20:45 +01:00
jschoubben 971cdb44e2 Merge remote-tracking branch 'origin/main' into fix/delivery-records-true
# Conflicts:
#	modules/mesh-delivery-indexer/cmd/mesh-delivery-indexer/main.go
#	modules/mesh-delivery-indexer/go.mod
#	modules/mesh-delivery-indexer/go.sum
#	modules/mesh-delivery-indexer/manifest_test.go
#	modules/mesh-delivery-indexer/module.json
#	modules/mesh-delivery-view/cmd/mesh-delivery-view-tools/main.go
#	modules/mesh-delivery-view/go.mod
#	modules/mesh-delivery-view/go.sum
#	modules/mesh-delivery-view/internal/view/read.go
#	modules/mesh-delivery-view/manifest_test.go
#	modules/mesh-delivery-view/module.json
2026-10-26 22:20:45 +01:00
jschoubben 16e4e61a0f Only an online healthy witness, and only a first failure, keep a version from being held; failed-alone clears when a node leaves (ADR 0557, issue 1496)
Review of 2ee7589: a node that sleeps after running F healthy kept its stale
have as a witness, so F stayed unheld while it failed on the online nodes; and a
failed-alone condition was never cleared once its node left the module.

A witness must now be online by the watchman's presence. F failing on a second
node is no longer one node's failure: the hold and the set-back stand (ADR 0557
version 6, §2). A remembered failed-alone key whose node is no longer one of the
module's nodes is cleared once; the first pass after a start clears it for every
node the presence bucket knows, for a module with a failed have.
2026-10-26 22:16:38 +01:00
mesh-tools 4b36cc29a6 Merge pull request 'Placement from the catalogue: the bridge carries it, the indexer writes it, the assigner compares it (issue 1126)' (#76) from feat/1126-nodes-compose into main 2026-10-26 21:11:26 +00:00
mesh-tools 31ddd09087 Merge pull request 'The checker ends a check whose heads no candidate holds, and starts a group's check once its heads settle (issue 1890)' (#79) from fix/1890-end-superseded-checks into main 2026-10-26 21:10:56 +00:00
jschoubben 68173de8bd Pin the indexer, the teller and the view at mesh-sdk b4310df: the delivery-records work merged with sdk main (issue 1280) 2026-10-26 22:08:40 +01:00
jschoubben 960e3309bc Merge remote-tracking branch 'origin/main' into fix/delivery-records-true
# Conflicts:
#	modules/mesh-delivery-view/go.mod
#	modules/mesh-delivery-view/go.sum
2026-10-26 22:07:28 +01:00
jschoubben 2ee7589474 A version healthy on another node is not held for one node's failed health check (ADR 0557, issue 1496)
On ace, 2026-10-26 21:19, a newly assigned mesh-build-node failed its first
health check at 8a98386f (later 31be2f24) while shanks, g14 and novox had run it
healthy for an hour. Being first by name, ace was the canary, and the assigner
held the version and moved the three healthy nodes back.

A failed health check of the newest fingerprint now holds it only when no other
node's have names it healthy (ADR 0557 §2 keeps ADR 0429 §6's hold and set-back
for that case). Otherwise the failure is that node's (ADR 0557 §1): nothing is
held or set back, the node keeps its want, and the assigner raises the node's
condition once (failed-alone.<node>.<module>), naming the node, the module, the
fingerprint and the have's words, and clears it once when it ends; the first
pass after a start clears what a previous process may have raised.
2026-10-26 22:06:32 +01:00
jschoubben 297992a4ae Pin mesh-sdk 9796444 (l2-next: feat/1126-nodes-compose merged with sdk main #101, the standing-test race fixes) (issue 1126)
mesh/delivery merged: source.main moved to novox/mesh-delivery@4b36cc29a6ae
mesh/delivery-group group feat/1126-nodes-compose merged: every member is merged
mesh/check checking 297992a4ae55 on 4b36cc29a6ae | novox/mesh-delivery#76 in group feat/1126-nodes-compose
2026-10-26 22:05:41 +01:00
jschoubben 3f85230417 Pin mesh-sdk 5f45175 (l2-next: feat/1126-nodes-compose merged with sdk main) after merging main (issue 1126) 2026-10-26 21:59:21 +01:00
jschoubben 43c3158bc6 Merge remote-tracking branch 'origin/main' into feat/1126-nodes-compose
# Conflicts:
#	modules/mesh-delivery-view/go.mod
#	modules/mesh-delivery-view/go.sum
2026-10-26 21:57:03 +01:00
jschoubben d95e9f3229 End a check whose heads no candidate holds, and start a group's check once its heads settle (issue 1890)
mesh/delivery merged: source.main moved to novox/mesh-delivery@31ddd090876f
mesh/check pass d95e9f322913 on d581f9659a78 | unit 191 run, 525 kept | integration 201 run, 591 kept | composition passed on novox
2026-10-26 21:44:12 +01:00
mesh-tools d581f9659a Merge pull request 'Tries record their asks; the try limit counts tries on a build node; build_usage shows capacity, hours and steps (issue 1890)' (#78) from fix/capacity-follows-cpu-share into main 2026-10-26 20:33:34 +00:00
jschoubben ea94f9234c Merge remote-tracking branch 'origin/main' into fix/capacity-follows-cpu-share
mesh/delivery merged: source.main moved to novox/mesh-delivery@d581f9659a78
mesh/delivery-group group fix/capacity-follows-cpu-share merged: every member is merged
mesh/check checking ea94f9234c97 on 3dd94f714d6e | novox/mesh-delivery#78 in group fix/capacity-follows-cpu-share
2026-10-26 21:29:44 +01:00
jschoubben c19468b04e Pin the checker and the view at mesh-sdk 805b4d9: the capacity work merged with sdk main 2026-10-26 21:29:44 +01:00
jschoubben c47afd8f4b build_usage says a seat's build hours, and when it is outside them
mesh/delivery closed: a newer head of the same pull request
mesh/delivery-group group fix/capacity-follows-cpu-share checking: 1 of 3 member(s) ready
mesh/check checking c47afd8f4b88 on 3dd94f714d6e | novox/mesh-delivery#78 in group fix/capacity-follows-cpu-share
A build module's capacity.<node> now carries its node's build window (the operator, 2026-10-26: g14
builds only from 18:00 to 06:00); the seat shows it, outside it says it takes nothing new, and a
value the module cannot read shows as any hour with the value and why. The checker's and the view's
SDK pin moves to mesh-sdk d00dac1.
2026-10-26 21:22:36 +01:00
jschoubben afd5a3ba07 A module a node wants and runs nothing countable of holds no head back; the reached head's event names its merges as not yet on the machines; pin mesh-sdk e44cbaf (review of issue 1280, ADR 0556 clause 2) 2026-10-26 21:15:31 +01:00
jschoubben 7dbad4c6d2 Pin mesh-sdk 33144781 (taken, reach, settle layouts) in the indexer, the view and the teller (issue 1280) 2026-10-26 21:05:29 +01:00
jschoubben 083a5042f6 The view reads the wants by module once per derivation of the reach 2026-10-26 21:03:56 +01:00
jschoubben 351ca7c290 The teller checks a settle ask's warrant against mesh-delivery.settle, the act its options bind (issue 1280, ADR 0556 clause 5) 2026-10-26 20:58:20 +01:00
jschoubben f907be2c14 The view derives per repository up to which merge every node runs it, keeps it as its state reach and says each reached head, so a fix on the machines reads delivered (issue 1280, ADR 0556 clauses 2 and 3) 2026-10-26 20:56:12 +01:00
jschoubben 6201f58c20 The indexer says per repository the last merge it took in, so the view knows its modules speak for that merge (issue 1280, ADR 0556 clause 1) 2026-10-26 20:49:00 +01:00
jschoubben 8fee764c73 Pin mesh-sdk feat/1126-nodes-compose 03dce12, the group's head: the checker refuses a pin that is neither trunk nor the group's head (issue 1126)
mesh/delivery-group group feat/1126-nodes-compose checking: 4 of 6 member(s) ready
mesh/delivery closed: a newer head of the same pull request
mesh/check checking 8fee764c734e on d581f9659a78 | novox/mesh-delivery#76 in group feat/1126-nodes-compose
2026-10-26 20:35:45 +01:00
mesh-tools 3dd94f714d Merge pull request 'The checker fails any repository's change that adds a binary, in any commit (issue 1577)' (#77) from fix/1577-checker-refuses-binaries into main 2026-10-26 19:32:27 +00:00
jschoubben c716670be0 The checker pins the SDK of review round 1, and its hid-the-failure cases join the older build module's test (issue 1901) 2026-10-26 20:19:07 +01:00
jschoubben 352066dde6 Review of L2: the assigner binds the indexer's placement state when it appears, with no restart (D1); the three new unit tests folded into existing ones, so no count rises (D2); pin mesh-sdk 3968849 (issue 1126)
mesh/delivery closed: a newer head of the same pull request
mesh/check checking 352066dde611 on 3dd94f714d6e | novox/mesh-delivery#76 in group feat/1126-nodes-compose
2026-10-26 20:12:00 +01:00
jschoubben d6bd31a166 Say the try limits as what they are: a check when a try is asked
Tries let in while waiting may later all run at once; each node's build capacity bounds what runs
on it (review of issue 1890's try cap).
2026-10-26 20:11:08 +01:00
jschoubben c0a4b139a2 A try queued behind the build seat's checks no longer blocks the next try
Only tries a build node holds count toward the 4 at once; tries waiting for a node are bounded
apart at 8, and the refusal names both counts (issue 1890: at 19:53 four tries queued behind 44
checks for up to 47 minutes refused every further try).
2026-10-26 20:05:51 +01:00
jschoubben 6d60953f98 A running try names its asks and where each stands; build_usage says a seat's CPU-bound capacity
The checker wrote a try's asks only at its end, so a 31-minute try could not be told queued from
running (issue 1890). The try's record now names each ask as it is put, then the node that took
it, its state and the phase its step says, written on the step key's own changes. build_usage
names a running try's asks and takes a build module's capacity.<node> (configured bounded by its
node's build CPU, issue 1806's follow-up) when it is current with mesh-config.
2026-10-26 19:58:39 +01:00
jschoubben 17ed06ee4d Pin mesh-sdk's fix/1901-failure-contract head 81ec7bd7 (issue 1901) 2026-10-26 19:50:47 +01:00
jschoubben 9427dac823 The checker says loudly when a build module hid a failure's cause (issue 1901)
A failed test with no cause, or a passing line as its cause, was worded as if it were the failure: a try of mesh-catalog
failed with a passing '✔' line of the first test file. Under mesh-sdk's failure contract the verdict's words and its
status line now start with 'the build module hid the failure (<module> at <version>)', and the check still fails.
2026-10-26 19:50:05 +01:00
jschoubben d74e082155 Checker refuses a program a commit of the branch added and a later one removed: it stays in history (issue 1577)
mesh/delivery merged: source.main moved to novox/mesh-delivery@3dd94f714d6e
mesh/delivery-group group fix/1577-checker-refuses-binaries merged: every member is merged
mesh/check checking d74e0821552c on df8b9ac3b1b5 | novox/mesh-delivery#77 in group fix/1577-checker-refuses-binaries
2026-10-26 19:47:53 +01:00
jschoubben 2e0013ec25 The assigner's wants test goes on to placement: a placed node's want names the build alone and its own want says it composes its own (issue 1126, ADR 0547 §2)
mesh/delivery closed: a newer head of the same pull request
mesh/check failed: ace: novox/mesh-catalog#505 does not merge into the trunk: files in conflict: .gitignore | 2e0013ec2500 on df8b9ac3b1b5 | unit 1245…
2026-10-26 19:44:55 +01:00
jschoubben 2493aa6529 Pin mesh-sdk feat/1126-nodes-compose 9713c30: placement carries a settings-form module's values (issue 1126)
mesh/delivery closed: a newer head of the same pull request
mesh/check failed: ace: novox/mesh-catalog#505 does not merge into the trunk: files in conflict: .gitignore | 2493aa65299e on df8b9ac3b1b5 | unit 1245…
2026-10-26 19:40:38 +01:00
jschoubben 85c89b16ef The grants part composes what the catalogue's placement places beside the assignments, and holds a seat held once for the mesh where placement says (issue 1126, ADR 0549 §3, G5)
mesh/delivery closed: a newer head of the same pull request
mesh/check failed: unit: git.novox.be/novox/mesh-catalog/checks/nobinaries.TestTheCatalogueHoldsNoCompiledProgram: modules/mesh-config/cmd/mesh-config…
2026-10-26 19:33:27 +01:00
jschoubben 4f75925410 Nodes compose: placement from the catalogue, node records, self wants, own and pair credentials, core boxes and the authority from the vault (issue 1126)
mesh/delivery closed: a newer head of the same pull request
mesh/check failed: unit: git.novox.be/novox/mesh-catalog/checks/nobinaries.TestTheCatalogueHoldsNoCompiledProgram: modules/mesh-config/cmd/mesh-config…
One commit for the branch feat/1126-nodes-compose of mesh-delivery: its 16 commits squashed so its history does not carry the 17.8 MB compiled mesh-delivery-assigner a commit added by mistake and a later one removed (issue 1577). The tree equals the former head ff6c038d. The operator allowed this rewrite of the feature branch on 2026-10-26.
2026-10-26 19:30:18 +01:00
jschoubben 820c458f4c The captured clean fetches carry the programs a current build module answers, so lines asserted whole do not warn (issue 1577) 2026-10-26 19:16:25 +01:00
jschoubben dc9d65524d A check judged by a build module that reports no programs says so in its status line, still passing: the refusal did not run (issue 1577) 2026-10-26 19:08:00 +01:00
jschoubben 1a4c788ed7 Checker fails a change whose fetch reports a compiled program, and warns when a build module reports none: build output kept getting committed (issue 1577) 2026-10-26 19:01:40 +01:00
mesh-tools df8b9ac3b1 Merge pull request 'Checker, builder and indexer bind states created after they start (issue 1790)' (#75) from fix/1790-state-appears into main 2026-10-26 14:47:40 +00:00
jschoubben fcb1e40582 Pin mesh-sdk fix/1790-state-appears at 65ee43f: a refused hearing of a bucket's creation is said and heard again at a reconnect (issue 1790, review)
mesh/delivery merged: source.main moved to novox/mesh-delivery@df8b9ac3b1b5
mesh/delivery-group group fix/1790-state-appears merged: every member is merged
mesh/check being carried to fcb1e4058262 on df8b9ac3b1b5
2026-10-26 15:25:53 +01:00
jschoubben 4973372612 Merge remote-tracking branch 'origin/main' into fix/1790-state-appears 2026-10-26 15:25:06 +01:00
jschoubben 3f6c085e8a Pin mesh-sdk fix/1790-state-appears at 3ac644b (issue 1790) 2026-10-26 15:14:00 +01:00
jschoubben 040e3097b5 checker, builder, indexer: a state of their own not on the bus at start is bound when it appears; mesh-grants composes each principal's hearing of its buckets' creation (issue 1790)
A delivery restarted the checker and the builder before the grants part created their new state hearing;
OptionalBucket answered nil and the hearing was never kept. They, and the builder's reasks and queues and the
indexer's takeins, now open their own states through mesh-sdk process.AwaitBucket (mesh-sdk
fix/1790-state-appears d515c71), which binds on the bus's STREAM.CREATED advisory and writes the latest kept
value. mesh-grants takes the same SDK, whose composer grants that advisory.
2026-10-26 15:13:00 +01:00
mesh-tools feef96ed16 Merge pull request 'build_usage: how much each build node is used and whether it may build, in one call' (#74) from feat/build-usage-tool into main 2026-10-26 14:09:32 +00:00
jschoubben 4f722899df View: build_usage does not answer eligibility it cannot read; a resolved capacity key with no readable value is the puller's 0
mesh/delivery merged: source.main moved to novox/mesh-delivery@feef96ed160e
mesh/check checking 4f722899df25 on feef96ed160e
A capacity this tool cannot read is unknown to it, not 0: the seat says eligible_known false and the node not known.
A key mesh-config holds without a readable value is what the puller reads as 0, and was dropped silently.
2026-10-26 14:58:47 +01:00
jschoubben 166a918b43 View: build_usage counts keys a build module ended for being left running apart, lists every node a capacity names, and says capacity unknown when mesh-config cannot be read
Since mesh-sdk #90 a build module ends a left-running key as failed at the moment it notices; counted as finished,
such a step put hours into the medians and the time at capacity. The view takes mesh-sdk's step bound and restart
reason from the SDK it now pins.
2026-10-26 14:58:47 +01:00
jschoubben 1b4c860f63 Checker: its announcement test expects tries readable, as the delivery view now reads them 2026-10-26 14:58:47 +01:00
jschoubben 3b6945b2db View: build_usage answers how the build nodes are used in one call; the checker's tries name their asks
The operator's question (how much are the build nodes used, is ace eligible) took six calls over four tools.
build_usage reads it from the bus's states: seats held (have), capacity (mesh-config resolved), pauses, the
build modules' steps by kind with left-running keys, the last hour and day with total and median time and time
at capacity, the running tries, and the backlog the view program reads from the seats' streams. A try now names
the asks it put, and the checker announces its tries readable, so the view tells a try's asks from a check's.
2026-10-26 14:58:41 +01:00
mesh-tools 4174e23c04 Merge pull request 'The checker and the builder end or re-ask what the build seat gave up, and a try can be ended by hand (issue 734)' (#73) from fix/734-lost-ask into main 2026-10-26 13:54:17 +00:00
jschoubben 3993175e45 Checker, builder and grants part: pin mesh-sdk main, which holds both the dead-letter hearing (#92) and the placement (#91) (issue 734)
mesh/delivery merged: source.main moved to novox/mesh-delivery@4174e23c0492
mesh/delivery-group group fix/734-lost-ask merged: every member is merged
mesh/check checking 3993175e457f on 4174e23c0492
2026-10-26 14:45:33 +01:00
jschoubben 60c7ffc981 Checker, builder and grants part: hear DEAD_LETTERS again until heard, say when they do not, and compose the same grant (issue 734, review round 1)
mesh-grants moves to the same mesh-sdk (3203358c), so its shadow user list grants the checker and the builder the
DEAD_LETTERS reads the controller's list will, and the take-over is not blocked by a difference. The checker and the
builder hear DEAD_LETTERS through buildrun.DeadLetterHearing: in the background, again at every reconnect and every ask
waited for while deaf, and keep whether they hear it in their new state hearing, which checker_status and builder_stuck
say as dead_letters. checker_try says that a try ended by the not-taken bound leaves its ask in the stream.

Not done as asked: a module health tool. The catalogue refuses a tool check on a module with no runtime of its own and no
other check beside it (ADR 0227 rule 8, ADR 0366 §1), so the deaf state is said in the status tools and the log.
2026-10-26 14:44:53 +01:00
jschoubben 33de9dde4f Checker and builder: wait on the build seats through mesh-sdk's LostAskWaiter, so an ask the seat gave up on ends every asker's wait (issue 734)
The waiter moved to mesh-sdk buildrun (fix/734-lost-ask, d96b9fd6), where every asker of the build seats shares it.
The builder hears DEAD_LETTERS for the go, node and image seats and asks a given-up part of a build again at once, with
the seat's words as why, instead of waiting for the check bound. The checker keeps what it does with a lost ask (a try
ends as error, a check is asked again) and its verdict words.
2026-10-26 14:44:52 +01:00
jschoubben 6b85220fa4 Checker: an ask the build seat gives up on ends its try as error and asks a check again; a try no node takes ends at the step bound; checker_try_end ends one by hand (issue 734)
A check ask handed out five times during a bus reload was given up into DEAD_LETTERS and no build node wrote a step for it,
so try-1793017340003993788 stayed running for ever, held one of the four try places, and checker_try of its head kept
answering its id. The checker now hears the copies DEAD_LETTERS keeps of the build seats' given-up asks (an ordered
consumer from its start; the grant is mesh-sdk/mesh-controller DeadLetterReads, branch fix/734-lost-ask) and ends the
step it waits for: a try ends as error saying why, a check is asked again under its claim, up to three times, then fails
as not heard. A try whose ask no node takes within the step bound ends the same way (a check's backlog stays no fault,
ADR 0525 §9). checker_try_end ends a try by hand, freeing its place and stopping its judging.
2026-10-26 14:44:40 +01:00
mesh-tools 1cbae7d41c Merge pull request 'The checker asks the packages a node could not test again, on a node that has what they need (issue 890, ADR 0552)' (#72) from feat/890-placement into main 2026-10-26 13:40:45 +00:00
mesh-tools b3a4b09f5b Merge pull request 'The indexer takes in a registered module whose repository became readable, and serves indexer_take_in (issue 1945)' (#71) from fix/indexer-takes-in into main 2026-10-26 13:39:13 +00:00
mesh-tools c8b26735c8 Merge pull request 'The assigner says when it holds the bus back instead of skipping it silently (issue 1945)' (#70) from fix/bus-bar-loud into main 2026-10-26 13:13:22 +00:00
jschoubben 907e092abc mesh-delivery-indexer reads unread registered modules again in one background run, so a forge timeout holds no merged message or source revision past its ack wait (issue 1945, review round 1)
mesh/delivery merged: source.main moved to novox/mesh-delivery@b3a4b09f5b6f
mesh/check being carried to 907e092abc45 on b3a4b09f5b6f
The take-in tool's bound now says the take-in is still running and where its answer will be.
2026-10-26 14:11:12 +01:00
jschoubben 3073495a06 checker: pin mesh-sdk 33395f68, feat/890-placement merged with main (issue 890)
mesh/delivery merged: source.main moved to novox/mesh-delivery@1cbae7d41cd0
mesh/delivery-group group feat/890-placement merged: every member is merged
mesh/check checking 3073495a0670 on 1cbae7d41cd0
2026-10-26 14:06:10 +01:00
jschoubben 16c411e436 Merge remote-tracking branch 'origin/main' into feat/890-placement 2026-10-26 14:05:40 +01:00
jschoubben 2d841870a8 Assigner: only a put still on its steps is named when a node leaves neither list (issue 1945)
mesh/delivery merged: source.main moved to novox/mesh-delivery@c8b26735c8cf
mesh/check pass 2d841870a873 on c8b26735c8cf | carried from 1d416827157d | unit 185 run, 518 kept | integration 195 run, 584 kept | composition passed…
Review round 2: an ended put stays in the bucket and owns no want.
2026-10-26 14:03:48 +01:00
jschoubben 7f8ce7f5c0 checker: the placed ask names the unmet packages, so a package without a kept verdict is not run or counted twice (issue 890, ADR 0552 §4; review round 1). Pins mesh-sdk 45564e69. 2026-10-26 14:02:16 +01:00
jschoubben 07b5f13474 mesh-delivery-indexer reads a registered module again once its repository becomes readable, and takes one in on ask (issue 1945)
A registered module whose repository the forge answered 404 (car-hunter, snake, before the forge's account could read
them) was read again only when the indexer started again or the controller wrote a new build record of it, so the grants
part's composed user list stayed short of their subjects and the take-over was refused. A failed read now keeps the
module as unread, read again on the next source revision and the next merged message (no timer of its own), and the new
tool indexer_take_in (the tools bundle mesh-delivery-indexer-tools, through the indexer's takeins state, as the
builder's builder_ask_again) takes one in now and answers what it stored or why not.
2026-10-26 14:01:34 +01:00
jschoubben f4aac599ca Assigner: name a running put when a node leaves neither list, and say the bus bar in plain words (issue 1945)
Review round 1: while an operator's put runs through its steps it owns the want, so the clear said
'the node does not run it' of a node that runs the bus.
2026-10-26 14:00:32 +01:00
jschoubben 2d63f0ba62 Assigner: say the bus bar instead of skipping it silently, and clear a neither condition only in true words (issue 1945)
A bus whose move only the operator's put may make (ADR 0546 §8) got no want and nothing in the outcome:
nats on novox, running an old-form build, was on no list while the pass cleared its condition and logged
that it had a want again. The bar now names its holding, puts a node without a live want on neither list
(its condition needs the operator: assigner_put), and a neither condition is cleared as 'has a want again'
only when a live want exists.
2026-10-26 13:55:42 +01:00
mesh-tools 1d41682715 Merge pull request 'mesh-grants says loudly when it holds the user list back, and clears it after a restart (issue 1945)' (#69) from fix/grants-hold-loud into main 2026-10-26 12:44:36 +00:00
jschoubben 705765d113 checker: a placed ask that every node declines for longer than its bound (30 min) fails the check naming the need (issue 890, ADR 0552 review) 2026-10-26 13:42:46 +01:00
jschoubben 16a97ed65d checker: ask the packages a node left unmet again of a node that has what their tests need, and put a declined ask again (issue 890)
A check that landed on ace passed with mesh-build-node's integration tests skipped. The build module now leaves such a
package unmet; the checker keeps the passes and asks the levels from the first unmet one with the need named, putting the
ask again under a new id when a node lacking it declines it. Re-pins mesh-sdk feat/890-placement.
2026-10-26 13:40:00 +01:00
mesh-tools ecf532af0e Merge pull request 'checker_status reads the node-engine at its running commit, as the checker decides (issue 568)' (#68) from fix/status-and-holds into main 2026-10-26 12:39:56 +00:00
jschoubben ce600df98a Say the held-list state on a part's first pass, so a warning raised before a restart is cleared
mesh/delivery merged: source.main moved to novox/mesh-delivery@1d416827157d
mesh/check being carried to ce600df98afc on 1d416827157d
2026-10-26 13:37:31 +01:00
jschoubben b4a6a34705 mesh-grants: a user list held back after take-over raises the warning mesh-grants.user-list-held and clears it when the list advances, so the hold is not only in a tool's output (issue 1945) 2026-10-26 13:36:39 +01:00
jschoubben a7e51b416b checker_status reads the node-engine at its running commit, as the checker's decision does, so it stops listing every candidate as waiting for a trunk move (issue 568)
mesh/delivery merged: source.main moved to novox/mesh-delivery@ecf532af0e85
mesh/check being carried to a7e51b416bc1 on ecf532af0e85 | from 71f786f14662
2026-10-26 13:25:18 +01:00
mesh-tools 71f786f146 Merge pull request 'mesh-grants: take-over and hand-back by the operator's Telegram Approve, the sealed list, and every grant the loaded list holds (issue 1945)' (#67) from feat/1945-user-list-switch into main 2026-10-26 12:21:50 +00:00
mesh-tools 848c6fdc56 Merge pull request 'Checker: one check per set of heads, a trunk or sibling move carries the verdict, a merge turn holds a claim (ADR 0548, issue 568)' (#65) from feat/548-one-check into main 2026-10-26 12:01:32 +00:00
mesh-tools ead9ee2d13 Merge pull request 'Indexer: index a module registered from any branch, by its registered ref, never a newer pull request's (issue 1945)' (#66) from fix/indexer-registered-modules-2 into main 2026-10-26 12:01:24 +00:00
mesh-tools 4861d1f78d Merge pull request 'Checker: keep no package whose tests skipped, so a skipped pass is never reused as proof (issue 890)' (#64) from fix/checker-skips-not-kept into main 2026-10-26 12:01:19 +00:00
jschoubben 5c76c9d3ac mesh-grants: a list in force with no users kept is checked against the users the bus loaded, so the first list after take-over keeps every user (review of the switch, issue 1945)
mesh/delivery merged: source.main moved to novox/mesh-delivery@71f786f14662
mesh/delivery-group group feat/1945-user-list-switch merged: every member is merged
mesh/check being carried to 5c76c9d3ac84 on 71f786f14662 | novox/mesh-delivery#67 in group feat/1945-user-list-switch
2026-10-26 12:54:55 +01:00
jschoubben 699cbfb7c1 Indexer: a registered module takes its registered ref, not a newer branch's record; no read at an older commit after a failure
mesh/delivery merged: source.main moved to novox/mesh-delivery@ead9ee2d138d
mesh/check checking 699cbfb7c131 on ead9ee2d138d
2026-10-26 12:52:29 +01:00
jschoubben 319d56be7e Indexer: a registered module is read from every repository it was built from, whatever the branch, and at the registered commit when the head is gone 2026-10-26 12:45:35 +01:00
mesh-tools 11d3678984 Merge pull request 'Indexer: index a module registered from its own repository, so the grants part composes its grants (issue 1945)' (#63) from fix/indexer-registered-modules into main 2026-10-26 11:33:41 +00:00
jschoubben f55f150896 mesh-grants: after take-over the list in force is not replaced by one composed past a failed read or one dropping its users for want of a hash; the switch's effect says when the bus node does not watch the key (review of the switch, issue 1945). Pins mesh-sdk c2a410bb. 2026-10-26 12:31:31 +01:00
jschoubben e1f1b5d18b Checker: a package with skipped tests is withheld from the kept verdicts, not the whole result (issue 890)
mesh/delivery merged: source.main moved to novox/mesh-delivery@4861d1f78d28
mesh/delivery-group group fix/checker-skips-not-kept merged: every member is merged
mesh/check failed: ace: the check of novox/mesh-catalog#492 refused: the installed go version could not be read: `go version` with GOTOOLCHAIN=local f…
2026-10-26 12:29:14 +01:00
jschoubben f0fe454145 Checker: a result with skipped tests keeps no verdict, so a skipped pass is run again (issue 890) 2026-10-26 12:23:57 +01:00
jschoubben 4026f6c188 checker: a check is never held by a turn its own heads wait in, the node-engine read at its running commit is named and carried when it moves, and a gate bucket made after the start is opened before the next claim (ADR 0548 §3, §6; issue 568, review round 1)
mesh/delivery merged: source.main moved to novox/mesh-delivery@848c6fdc5648
mesh/delivery-group group feat/548-one-check merged: every member is merged
mesh/check failed: ace: the check of novox/mesh-delivery#65 refused: the installed go version could not be read: `go version` with GOTOOLCHAIN=local f…
2026-10-26 12:14:50 +01:00
jschoubben 29cfb8c177 Pin mesh-sdk at the group's head 83ee7769 (issue 1945) 2026-10-26 12:06:58 +01:00
jschoubben a092ee9ff9 mesh-grants writes the user list sealed to the bus node and names it in user-list.sealed; the switch turns that key; take-over state says switched only once the bus node loaded it (ADR 0547 clauses 7, 8; issue 1945)
The bus node can only take over a list that exists: each pass composes the accounts block with the vault's hashes, seals
it to the key the bus node says in its have, puts it in the part's own object bucket by digest and writes the key with
digest, revision, user count, users left out and which list to follow. take-over is refused, naming why, while no list is
sealed to the bus node's current key or the loaded list does not grant the bus node its read.

Review of fix/1945-takeover-blockers: no responders on the router's record is an error, not 'no record'; an answer whose
record cannot be reached is not closed; an ask that failed to publish is taken back at the router; an error acting on an
answer is logged. Pins mesh-sdk 175a5567.
2026-10-26 11:59:41 +01:00
jschoubben 6c3cbee35c checker: pin mesh-sdk at the group's head f5c233a9288c, which holds the merge turns and the carried verdict (ADR 0548 §10; issue 568) 2026-10-26 11:45:13 +01:00
jschoubben 646f83642a Indexer: read a registered module at the source's head when it names one, else at the registered commit
mesh/delivery merged: source.main moved to novox/mesh-delivery@11d3678984af
mesh/check pass 646f83642a01 on 11d3678984af | unit 171 run, 502 kept | integration 181 run, 568 kept | composition passed in 39ms on novox
2026-10-26 11:41:46 +01:00
jschoubben d9bfba1e62 Indexer: read one directory of a registered module's repository, leave a repository the source names to its merges, accept master as a trunk 2026-10-26 11:36:22 +01:00
jschoubben f1bb7747ae checker: one integration test through the real bus proves a merge turn holds a check off until it goes and a carry of unchanged digests runs nothing; it extends the equal-inputs test, so the count stays (ADR 0548 §3, §5; issue 568 D7) 2026-10-26 11:36:15 +01:00
jschoubben d2eb3e5ecf checker: the line of a carried verdict names the head on the new trunk, then where it was carried from; a verdict being carried is pending and says so (ADR 0548 §6, §8; issue 568 D6) 2026-10-26 11:35:16 +01:00
jschoubben d91ea41f64 checker: only a head git refused gets a whole check after a trunk move; a conflicting carry fails naming the files and asks nothing (ADR 0548 §7; issue 568 D5) 2026-10-26 11:34:08 +01:00
jschoubben 5fed429827 checker: a verdict names the siblings it read at their trunk, and a sibling's move is carried as a trunk move is (ADR 0548 §6; issue 568 D4) 2026-10-26 11:33:35 +01:00
jschoubben e2c41ffa5a checker: a carry keeps the composition it passed while no composed fingerprint changed, and composes again otherwise (ADR 0548 §5; issue 568 D3) 2026-10-26 11:32:31 +01:00
jschoubben 13e4ed8515 checker: a trunk move never ends a check; once it ended, its pass is carried and names the commits it was carried from, and a failed carry names the move (ADR 0548 §1, §5; issue 568 D2) 2026-10-26 11:30:46 +01:00
jschoubben c66fe11e16 checker: claim no check of a repository an approved merge holds a turn on, and claim it when the turn goes (ADR 0548 §3; issue 568 D1) 2026-10-26 11:29:25 +01:00
jschoubben c44b9f91cd Indexer: gofmt 2026-10-26 11:23:27 +01:00
jschoubben 82da252cf0 Indexer: index the manifests of modules registered from their own repositories, so the grants part has no gap for them 2026-10-26 11:23:21 +01:00
mesh-tools 32bc4f3c48 Merge pull request 'Checker: read every test result of a level and name every failure in the verdict, the try and the status line (issue 1921)' (#62) from fix/1921-every-failure-r into main 2026-10-26 10:18:11 +00:00
mesh-tools ed1125b66d Merge pull request 'Assigner: take a stale ask away only once its held condition's clear was said, so a failed clear is retried (issue 1906)' (#61) from fix/1906-clear-retried into main 2026-10-26 10:09:42 +00:00
jschoubben 366bc750da Assigner: clear a stale hold's condition before taking its ask away, so a failed clear is retried at the next pass (issue 1906)
mesh/delivery merged: source.main moved to novox/mesh-delivery@ed1125b66d91
mesh/check pass 366bc750da87 on ed1125b66d91 | unit 168 run, 489 kept | integration 21 run, 712 kept | composition passed on ace
2026-10-26 11:03:09 +01:00
jschoubben 04a56b7aff Name every failed test in a check's verdict and a try's failed list (issue 1921)
mesh/delivery merged: source.main moved to novox/mesh-delivery@32bc4f3c48b2
mesh/delivery-group group fix/1921-every-failure-r merged: every member is merged
mesh/check checking 04a56b7aff66 on 32bc4f3c48b2
The checker stopped reading a level's test results at the first that did
not pass and kept only its first failure, so a check named one fault per
round. It now reads and counts every result and writes every failed test
with its first line into the verdict's Failures and the try's failed
list (at most 50, then how many more); Failed keeps the first one's
words whole (issue 1901). Pins mesh-sdk fix/1921-every-failure.
2026-10-26 10:58:13 +01:00
130 changed files with 8248 additions and 438 deletions
+3
View File
@@ -1 +1,4 @@
vendor/
# A compiled program never enters a commit (issue 1577): `go build .` in a command directory leaves one there.
modules/*/cmd/*/mesh-*
!modules/*/cmd/*/mesh-*.go
@@ -172,6 +172,9 @@ func TestTheAssignerStartsAndSaysWhatItKeeps(t *testing.T) {
// 1297/B2b: a bucket it reads that the bus does not hold stops the start, naming the bucket.
func TestTheAssignerDoesNotStartWithoutABucketItReads(t *testing.T) {
for _, missing := range readBuckets {
if missing == delivery.PlacementBucket {
continue // it starts without it, below
}
b := startBed(t, missing)
select {
case err := <-b.start(t):
@@ -218,6 +221,34 @@ func TestTheAssignerDoesNotStartWithoutABucketItReads(t *testing.T) {
t.Fatal("the put was not taken further once its bucket was made")
}
}
// Review of L2 (D1): without the indexer's placement state it starts all the same, and once the bucket is made (the
// indexer and the assigner come in one delivery) it is read with no restart: a module only the placement names is
// said as placed by one source only.
pb := startBed(t, delivery.PlacementBucket)
pdone := pb.start(t)
select {
case err := <-pdone:
t.Fatalf("without %s the assigner stopped: %v", delivery.PlacementBucket, err)
case <-time.After(2 * time.Second):
}
placement, err := pb.js.CreateKeyValue(pb.ctx, jetstream.KeyValueConfig{Bucket: delivery.PlacementBucket, History: 5})
if err != nil {
t.Fatal(err)
}
pl, _ := delivery.EncodePlacement(delivery.Placement{Node: "n1", Modules: []delivery.PlacedModule{{Module: "web"}}})
plkey, _ := delivery.PlacementKey("n1")
if _, err := placement.Put(pb.ctx, plkey, pl); err != nil {
t.Fatal(err)
}
for deadline := time.Now().Add(15 * time.Second); ; time.Sleep(200 * time.Millisecond) {
if c, ok := pb.conditionsSaid(t)[delivery.AssignerModule+".placement.n1.web"]; ok && c.State == asks.StateRaised {
break
}
if time.Now().After(deadline) {
t.Fatal("the placement state made after the start was not read")
}
}
}
// putJSON writes a value into one of the bed's buckets.
@@ -406,6 +437,32 @@ func TestWantsAreWrittenAndNoDifferenceIs(t *testing.T) {
if !cleared {
t.Errorf("the condition of web on neither list on g14 was not cleared once g14 got its want")
}
// Issue 1126 (ADR 0547 §2): the catalogue's placement reaches the indexer's state naming web on every node the
// assignments name it on. Each node then composes its own: its want names the build alone (Self, no composition), and
// its own want says it composes its own, which the node agent reads (mesh-host's TestNats...ComposesAndAppliesItsOwn).
for _, node := range []string{"ace", "g14", "shanks"} {
pl, err := delivery.EncodePlacement(delivery.Placement{Node: node, Modules: []delivery.PlacedModule{{Module: "web"}}})
if err != nil {
t.Fatal(err)
}
pkey, _ := delivery.PlacementKey(node)
b.putJSON(t, delivery.PlacementBucket, pkey, pl)
}
if w := b.wantNow(t, "ace", "web", "ace's want once placed: the build alone, composed by the node", func(w delivery.Want) bool { return w.Self }); w.Composed != "" || w.FP != fNew {
t.Errorf("ace's want once placed = %+v", w)
}
nkey, _ := delivery.WantNodeKey("ace")
var self bool
for deadline := time.Now().Add(10 * time.Second); time.Now().Before(deadline) && !self; time.Sleep(100 * time.Millisecond) {
if e, err := b.buckets[delivery.AssignerWantsBucket].Get(b.ctx, nkey); err == nil {
nw, err := delivery.DecodeNodeWant(e.Value())
self = err == nil && nw.Self
}
}
if !self {
t.Error("ace's own want does not say it composes its own once placed")
}
}
// 1297/B9 (the assigner's half, ADR 0528 clause 6): started whole, the assigner holds a fingerprint whose health check
@@ -59,6 +59,7 @@ var readBuckets = []string{
delivery.CandidatesBucket,
delivery.ManifestsBucket,
delivery.ModulesBucket,
delivery.PlacementBucket,
delivery.PresenceBucket,
delivery.HaveBucket,
delivery.StepsBucket(goBuildModule),
@@ -98,6 +99,12 @@ func open(ctx context.Context, js process.BucketOpener, logf func(string, ...any
logf("the bucket %s cannot be opened (%v): no put is taken further until the assigner starts again", name, err)
continue
}
if name == delivery.PlacementBucket && err != nil {
// The indexer's placement state comes with the second milestone (ADR 0549 §2): without it the controller's
// assignments alone name where modules run, as before any placement was read (ADR 0549 §5).
logf("the bucket %s cannot be opened (%v): the controller's assignments alone are read until the assigner starts again", name, err)
continue
}
if err != nil {
return nil, process.EssentialErr("assigner", name, "write a want from the whole mesh", err)
}
@@ -151,7 +158,7 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
logger.Printf("what the assigner keeps could not be said on the bus, so its readers are not granted until it is: %v", err)
}
logger.Printf("started on the account %s, reading %d buckets", account.User, len(kvs))
return follow(ctx, js, kvs, sayVia(ctx, js), emitting{said}, logger)
return follow(ctx, js, conn, kvs, sayVia(ctx, js), emitting{said}, logger)
}
// emitting is the process connection as the puts' caller: its calls, and its events with no options.
@@ -169,6 +176,7 @@ var watched = []string{
delivery.BuilderBuildsBucket,
delivery.CandidatesBucket,
delivery.ModulesBucket,
delivery.PlacementBucket,
delivery.PresenceBucket,
delivery.HaveBucket,
delivery.AssignerAsksBucket,
@@ -179,7 +187,7 @@ var watched = []string{
// follow watches every bucket a pass reads and makes a pass at start, once every watch handed what stands, and after
// each change; changes that come during a pass make one more. It returns when ctx ends or a watch ends.
func follow(ctx context.Context, js jetstream.JetStream, kvs buckets, say func(asks.Condition) error, bus caller, logger *log.Logger) error {
func follow(ctx context.Context, js jetstream.JetStream, nc process.Subscriber, kvs buckets, say func(asks.Condition) error, bus caller, logger *log.Logger) error {
w := newWorld()
dirty := make(chan struct{}, 1)
wake := func() {
@@ -263,7 +271,12 @@ func follow(ctx context.Context, js jetstream.JetStream, kvs buckets, say func(a
}
logger.Printf("the bucket %s is open: puts are taken further", delivery.AssignerPutsBucket)
}
started := 0
for _, name := range watched {
if kvs[name] == nil && name == delivery.PlacementBucket {
continue // not on the bus at this start (open said so)
}
started++
var opts []jetstream.WatchOpt
if name == delivery.HaveBucket {
opts = append(opts, jetstream.IncludeHistory()) // what each node ran healthy before, for a set-back
@@ -275,7 +288,20 @@ func follow(ctx context.Context, js jetstream.JetStream, kvs buckets, say func(a
if err := watch(delivery.ManifestsBucket, jetstream.MetaOnly()); err != nil {
return err
}
for range len(watched) + 1 {
// The indexer's placement state arrives with the same delivery as this assigner (ADR 0549 §2): when it is not on the
// bus at this start it is bound once the bus says it was created, and watched from then on, with no restart (issue
// 1790's kind). Until then the controller's assignments alone name where modules run (ADR 0549 §5).
placementAppeared := make(chan jetstream.KeyValue, 1)
if kvs[delivery.PlacementBucket] == nil && js != nil {
process.AwaitBucket(ctx, js, nc, delivery.PlacementBucket, "the controller's assignments alone are read",
logger.Printf).Ready(func(kv jetstream.KeyValue) { placementAppeared <- kv })
}
takePlacement := func(kv jetstream.KeyValue) error {
kvs[delivery.PlacementBucket] = kv
logger.Printf("the bucket %s is open: placement is read beside the controller's assignments", delivery.PlacementBucket)
return watch(delivery.PlacementBucket)
}
for range started + 1 {
select {
case <-ctx.Done():
return nil
@@ -297,6 +323,10 @@ func follow(ctx context.Context, js jetstream.JetStream, kvs buckets, say func(a
case err := <-errs:
return err
case <-settled:
case kv := <-placementAppeared:
if err := takePlacement(kv); err != nil {
return err
}
case <-dirty:
reopenPuts(p)
p.pass(ctx)
@@ -94,7 +94,7 @@ func TestOpensWhatItsManifestNames(t *testing.T) {
kvs[delivery.ControllerComposedBucket].(*watchKV).entries = []kvEntry{{key: ckey, value: cd}}
kvs[delivery.AssignerPutsBucket].(*watchKV).closed = true
kvs[delivery.AssignerWantsBucket].(*watchKV).wrote = wrote
go follow(ctx, nil, kvs, func(asks.Condition) error { return nil }, nil, log.New(io.Discard, "", 0))
go follow(ctx, nil, nil, kvs, func(asks.Condition) error { return nil }, nil, log.New(io.Discard, "", 0))
select {
case key := <-wrote:
if key != "want.ace.node" {
@@ -3,7 +3,7 @@ package main
import (
"context"
"errors"
"fmt"
"maps"
"slices"
"strings"
"time"
@@ -40,6 +40,9 @@ type passer struct {
// noWant is, by node and module, the reason last said why the node gets no want for the module (issue 1945): a reason is
// said in the journal when it appears or changes, and its clearing when the node has a want again; one line per change.
noWant map[string]string
// nodeFailed is, by node and module, the words of the condition last raised for a failure that is the node's own (ADR 0557
// §1, issue 1496): raised once per words, cleared once when it ends.
nodeFailed map[string]string
}
func (p *passer) clock() time.Time {
@@ -61,7 +64,23 @@ func (p *passer) pass(ctx context.Context) {
// controller has composed it and again whenever that digest changes (ADR 0539 §6).
func (p *passer) nodeWants(ctx context.Context) {
written := p.w.read(delivery.AssignerWantsBucket)
placed := p.w.placedNodes()
// A node the catalogue places composes its own composition (ADR 0547 §2): its own want names none.
for node := range placed {
key, err := delivery.WantNodeKey(node)
if err != nil {
continue
}
if before, err := delivery.DecodeNodeWant(written[key]); err == nil && before.Self {
continue
}
want := delivery.NodeWant{Self: true, Words: "the node composes its own composition (ADR 0547 §2)"}
p.put(ctx, delivery.AssignerWantsBucket, key, func() ([]byte, error) { return delivery.EncodeNodeWant(want) })
}
for node, d := range p.w.nodeComposed() {
if placed[node] {
continue
}
key, err := delivery.WantNodeKey(node)
if err != nil {
continue
@@ -98,6 +117,8 @@ func (p *passer) wants(ctx context.Context) {
}
}
wanted := p.w.wantsOf()
frozen := p.sayPlacement(p.w.placementDiffers())
placed := p.w.placedNodes()
puts := p.putsByModule()
p.staleAsks(ctx, trunk)
names := map[string]bool{}
@@ -114,12 +135,20 @@ func (p *passer) wants(ctx context.Context) {
// quiet at the router.
first := p.neither == nil
stale := map[string]neitherOne{}
defer func() { p.sayNeither(neither, stale) }()
left := map[string]string{}
defer func() { p.sayNeither(neither, stale, left) }()
for name := range names {
fp := trunk[name].FP2
name := name
m := assign.Module{Name: name, FP: fp, Manifest: manifests[name], Nodes: nodesOf[name], Now: time.Now(), Puts: puts[name],
Composed: func(node, fp string) (delivery.ComposedDigest, bool) { return p.w.composed(node, name, fp) },
Composed: func(node, fp string) (delivery.ComposedDigest, bool) {
if placed[node] {
// The node composes its own (ADR 0547 §2, §4): no composition of the controller's is waited for, the want
// names the build alone, and the node judges the move itself.
return delivery.ComposedDigest{Judgement: delivery.JudgedNeither, Words: "the node composes its own"}, true
}
return p.w.composed(node, name, fp)
},
GrantedAt: func(fp string) bool {
key, err := delivery.ControllerGrantKey(name, fp)
return err == nil && p.w.has(delivery.ControllerGrantsBucket, key)
@@ -153,8 +182,11 @@ func (p *passer) wants(ctx context.Context) {
last[node] = p.w.lastHealthy(node, name, fp)
}
out := assign.Wants(m, nodes, wanted[name], last)
p.yeses(ctx, name, m, placed, wanted[name], &out)
gatherNeither(neither, name, out)
leftNeither(left, m, wanted[name], out)
p.sayNoWant(name, m.Nodes, wanted[name], out)
p.sayNodeFailed(name, fp, m.Nodes, nodes, out, first)
if first {
startClears(stale, m, out)
if fp != "" && !m.Held && out.Held == nil && !out.Lifted {
@@ -182,6 +214,9 @@ func (p *passer) wants(ctx context.Context) {
p.recordAsk(ctx, name, fp, need)
}
for node, want := range out.Wants {
if frozen[node+"."+name] {
continue // one source alone names it (ADR 0549 §5): no want, and the one written before stays
}
key, err := delivery.WantKey(node, name)
if err != nil {
p.logf("no want of %s on %s is written: %v", name, node, err)
@@ -210,7 +245,7 @@ func (p *passer) sayNoWant(module string, nodes []string, before map[string]deli
}
reason := ""
if h, held := out.Rules[node]; held && (!written || w.Removed) {
reason = fmt.Sprintf("rule %d: %s", h.Rule, h.Words)
reason = h.Reason()
}
last, said := p.noWant[key]
switch {
@@ -229,6 +264,60 @@ func (p *passer) sayNoWant(module string, nodes []string, before map[string]deli
}
}
// sayNodeFailed raises, once per node and words, the condition of a failed health check of a module's newest fingerprint fp
// that is the node's own, since fp runs healthy elsewhere and is not held (ADR 0557 §1, issue 1496), and says it in the journal;
// it clears the condition once when the node no longer fails alone, or is no longer one of the module's nodes. The first pass
// after a start remembers nothing a previous process raised: it clears, for every module, the condition of every node the
// presence bucket knows that does not fail alone, since a node that recovered while no process watched leaves no failed have to
// find it by (a clear of a condition not open is quiet at the router). A condition the router did not take is said again at the next pass.
func (p *passer) sayNodeFailed(module, fp string, nodes []string, known assign.Nodes, out assign.Outcome, first bool) {
if p.nodeFailed == nil {
p.nodeFailed = map[string]string{}
}
look := map[string]bool{}
for _, node := range nodes {
look[node] = true
}
for key := range p.nodeFailed {
if node, m, _ := strings.Cut(key, "."); m == module {
look[node] = true
}
}
if first {
for node := range known.Online {
look[node] = true
}
}
for _, node := range slices.Sorted(maps.Keys(look)) {
key := node + "." + module
f, failing := out.NodeFailed[node]
last, said := p.nodeFailed[key]
switch {
case failing && f.Words+"|"+strings.Join(f.Others, ",") != last:
p.logf("%s on %s: the health check at %s failed here (%s), while it runs healthy on %s: the failure is %s's, so nothing "+
"is held or set back", module, node, fp, f.Words, strings.Join(f.Others, ", "), node)
if p.say != nil {
if err := p.say(assign.NodeFailedCondition(node, module, fp, f, true)); err != nil {
p.logf("the condition of %s failing on %s alone could not be said, and is said at the next pass: %v", module, node, err)
continue
}
}
p.nodeFailed[key] = f.Words + "|" + strings.Join(f.Others, ",")
case !failing && (said || first):
if said {
p.logf("%s on %s no longer fails its health check alone", module, node)
}
if p.say != nil {
if err := p.say(assign.NodeFailedCondition(node, module, fp, assign.NodeFailure{}, false)); err != nil && said {
p.logf("the condition of %s failing on %s alone could not be cleared, and is cleared at the next pass: %v", module, node, err)
continue
}
}
delete(p.nodeFailed, key)
}
}
}
// neitherOne is a module assigned to a node that holds no want for it after a pass, and the rule holding its first want back.
type neitherOne struct {
node, module string
@@ -242,6 +331,48 @@ func gatherNeither(into map[string]neitherOne, module string, out assign.Outcome
}
}
// leftNeither says, by node and module, for every node of m not on neither list after the pass, why: "" when the node has a live
// want for it (written before and kept, or written in this pass), else why it has none. sayNeither clears a neither condition in
// these words, so a clear never says a want the node does not hold (issue 1945: "nats on novox has a want again" while the bus bar
// left it with none).
func leftNeither(into map[string]string, m assign.Module, before map[string]delivery.Want, out assign.Outcome) {
for _, node := range m.Nodes {
if _, on := out.Neither[node]; on {
continue
}
w, written := out.Wants[node]
if !written {
w, written = before[node]
}
why := ""
h, ruled := out.Rules[node]
switch {
case written && !w.Removed:
case (m.Held && !out.Lifted) || out.Held != nil:
why = "it is held, and its held condition says so"
case slices.ContainsFunc(out.Record, func(r assign.RecordNeed) bool { return r.Node == node }):
why = "its move waits for the operator's yes on its record ask"
case putOn(m, node):
why = "its put waits on its own steps and the operator's answer"
case m.FP == "":
why = "the trunk has no build of it"
case ruled:
why = h.Reason()
default:
why = "the node does not run it"
}
into[node+"."+m.Name] = why
}
}
// putOn says whether an operator's put on node owns m's want there: one still on its steps (ADR 0546), so the rules write none
// meanwhile. An ended put (done, refused, released) stays in the bucket and owns nothing.
func putOn(m assign.Module, node string) bool {
p, put := m.Puts[node]
_, owns := assign.PutOwns(p)
return put && owns
}
// startClears adds, for the first pass after a start, every node of m that is not on neither list after the pass: a neither
// condition a previous process raised for it is cleared. That covers a node with a want and one with none that is not on neither
// list, such as a held module (issue 1945: the hold is said on its own, so a neither raised for it before the restart would stay up
@@ -258,7 +389,9 @@ func startClears(into map[string]neitherOne, m assign.Module, out assign.Outcome
// once, the condition of each one said before that has a want again (issue 1945). A say that fails is made again at the next pass.
// stale is, on the first pass after a start, the modules whose condition a previous process may have raised and that have a want:
// each is cleared once.
func (p *passer) sayNeither(now, stale map[string]neitherOne) {
// left is, by node and module, why each assigned node not on neither list after this pass has no want ("" when it has one); a key
// it does not hold is no longer assigned.
func (p *passer) sayNeither(now, stale map[string]neitherOne, left map[string]string) {
if p.neither == nil {
p.neither = map[string]neitherOne{}
for key, one := range stale {
@@ -278,21 +411,29 @@ func (p *passer) sayNeither(now, stale map[string]neitherOne) {
continue
}
}
p.logf("%s is assigned to %s and on neither list: no want, and not in the node's own composition; rule %d holds its first want back: %s",
one.module, one.node, one.holding.Rule, one.holding.Words)
p.logf("%s is assigned to %s and on neither list: no want, and not in the node's own composition; what holds its first want back: %s",
one.module, one.node, one.holding.Reason())
p.neither[key] = one
}
for key, one := range p.neither {
if _, still := now[key]; still {
continue
}
why, assigned := left[key]
if !assigned {
why = "it is no longer assigned there"
}
if p.say != nil {
if err := p.say(assign.NeitherCondition(one.node, one.module, assign.Holding{}, false)); err != nil {
if err := p.say(assign.NeitherLeftCondition(one.node, one.module, why)); err != nil {
p.logf("the condition of %s on neither list on %s could not be cleared, and is cleared at the next pass: %v", one.module, one.node, err)
continue
}
}
p.logf("%s on %s has a want again", one.module, one.node)
if why == "" {
p.logf("%s on %s has a want again", one.module, one.node)
} else {
p.logf("%s on %s is no longer on neither list and has no want: %s", one.module, one.node, why)
}
delete(p.neither, key)
}
}
@@ -481,13 +622,19 @@ func (p *passer) staleAsks(ctx context.Context, trunk map[string]delivery.Module
}
for key := range p.w.read(delivery.AssignerAsksBucket) {
if p.staleAsk(key, trunk) {
// The condition is cleared first and the ask goes only once that was said: the ask is what brings this key back at the
// next pass, so a clear that failed once is made again, with no state of its own (issue 1906, note 29).
k, kerr := delivery.ParseAnyAskKey(key)
held := kerr == nil && (k.Kind == delivery.AskKindHold || k.Kind == delivery.AskKindRestore)
if held && !p.clearHeld(k.Module, k.FP) {
continue
}
p.logf("the ask %s is about a fingerprint or composition the trunk has left behind and is taken away", key)
p.remove(ctx, delivery.AssignerAsksBucket, key)
// The held record stays: it is the memory that this fingerprint failed (ADR 0408 §4), and a revert returns the trunk to
// exactly it. Only the condition, which would else ask for an answer that can no longer be given, is cleared, and it is
// raised again if the trunk returns to the fingerprint.
if k, err := delivery.ParseAnyAskKey(key); err == nil && (k.Kind == delivery.AskKindHold || k.Kind == delivery.AskKindRestore) {
p.clearHeld(k.Module, k.FP)
if held {
delete(p.said, assign.HeldCondition(k.Module, k.FP, "", "", false, false).Key+".raised")
}
}
@@ -566,16 +713,18 @@ func (p *passer) answeredWith(key, option string) bool {
}
// clearHeld clears, once, the condition of a hold the operator released.
func (p *passer) clearHeld(module, fp string) {
// It says whether nothing is left to say: the clear was said now or before, or there is no router to say it to.
func (p *passer) clearHeld(module, fp string) bool {
c := assign.HeldCondition(module, fp, "", "", false, false)
if p.say == nil || p.said[c.Key+".cleared"] {
return
return true
}
if err := p.say(c); err != nil {
p.logf("the condition of the released %s could not be cleared, and is cleared at the next pass: %v", module, err)
return
return false
}
p.said[c.Key+".cleared"] = true
return true
}
// recordAsk writes, once, the ask about a move of a module of policy record that waits for the operator's yes.
@@ -667,3 +816,83 @@ func (p *passer) backupAsk(ctx context.Context, key, node, module string, want d
}
p.said[c.Key] = true
}
// sayPlacement raises, once, the condition of every pair the controller's assignments and the catalogue's placement do not
// both name, clears it once they agree, and answers those pairs by node and module (ADR 0549 §5): the pass writes no want
// for them and keeps the one written before. A say that fails is made again at the next pass.
func (p *passer) sayPlacement(differ []assign.Pair) map[string]bool {
frozen := map[string]bool{}
now := map[string]bool{}
for _, d := range differ {
frozen[d.Node+"."+d.Module] = true
c := assign.PlacementCondition(d, true)
now[c.Key] = true
if p.said[c.Key+".raised"] {
continue
}
p.logf("%s on %s is named by %s alone: no want is written for it there until both sources name it (ADR 0549 §5)", d.Module, d.Node, d.Only)
if err := p.say(c); err != nil {
p.logf("the placement condition of %s on %s could not be said, and is said at the next pass: %v", d.Module, d.Node, err)
continue
}
p.said[c.Key+".raised"] = true
}
prefix := delivery.AssignerModule + ".placement."
for key := range p.said {
c, raised := strings.CutSuffix(key, ".raised")
if !raised || !strings.HasPrefix(c, prefix) || now[c] {
continue
}
node, module, _ := strings.Cut(strings.TrimPrefix(c, prefix), ".")
if err := p.say(assign.PlacementCondition(assign.Pair{Node: node, Module: module}, false)); err != nil {
p.logf("the placement condition of %s on %s could not be cleared, and is cleared at the next pass: %v", module, node, err)
continue
}
delete(p.said, key)
}
return frozen
}
// yeses writes a placed node's wants naming the build alone, and sees to the moves a node that composes its own judged a
// person must allow (ADR 0547 §3): for a have of phase
// waiting-yes it raises the record ask once, naming the composition waiting and the node's judgement, and on a yes writes
// the want again naming it; a yes once written is kept while the want stays at its build, so the node is not asked again.
func (p *passer) yeses(ctx context.Context, module string, m assign.Module, placed map[string]bool, before map[string]delivery.Want, out *assign.Outcome) {
for _, node := range m.Nodes {
if !placed[node] {
continue
}
was, had := before[node]
w, written := out.Wants[node]
if !written {
if !had || was.Removed {
continue
}
w = was
}
if w.Removed {
continue
}
yes := ""
if had && was.FP == w.FP {
yes = was.Yes
}
key, err := delivery.HaveKey(node, module)
if err != nil {
continue
}
if h, err := delivery.DecodeHave(p.w.value(delivery.HaveBucket, key)); err == nil && h.Phase == delivery.PhaseWaitingYes && h.Waits != yes {
if m.Allowed != nil && m.Allowed(node, h.Waits) {
yes = h.Waits
} else {
p.recordAsk(ctx, module, w.FP, assign.RecordNeed{Node: node, Digest: h.Waits, Judgement: h.Judgement, Words: h.Detail})
}
}
// A want carried over from before the node composed its own may still name the controller's composition; a placed
// node's want names the build alone (ADR 0547 §2).
if yes != w.Yes || w.Composed != "" || !w.Self {
w.Yes, w.Composed, w.Self = yes, "", true
out.Wants[node] = w
}
}
}
@@ -342,6 +342,9 @@ func putOfTheBus(t *testing.T) {
// again after a say that failed, and clears it once when the module has a want again; a module on neither list is also written to
// the journal, so it is loud with no operator channel.
func TestAModuleOnNeitherListIsRaisedOnceAndClearedOnce(t *testing.T) {
// A pair only one source names is a condition of the same kind (issue 1126): folded here to hold the count.
aPairOneSourceNamesAloneGetsNoWantAndOneCondition(t)
t.Run("gathers", testThePassGathersTheModulesOnNeitherList)
p := newPasser(buckets{}, time.Date(2026, 10, 25, 15, 0, 0, 0, time.UTC))
var said []asks.Condition
failing := true
@@ -357,30 +360,66 @@ func TestAModuleOnNeitherListIsRaisedOnceAndClearedOnce(t *testing.T) {
p.logf = func(f string, a ...any) { logged = append(logged, fmt.Sprintf(f, a...)) }
one := map[string]neitherOne{"g14.screen-lock": {node: "g14", module: "screen-lock",
holding: assign.Holding{Rule: delivery.RuleProviders, Words: "no provider of x11-display runs healthy on g14 or at its canary"}}}
p.sayNeither(one, nil) // the say fails
p.sayNeither(one, nil)
p.sayNeither(one, nil)
p.sayNeither(one, nil, nil) // the say fails
p.sayNeither(one, nil, nil)
p.sayNeither(one, nil, nil)
if len(said) != 1 || said[0].State != asks.StateRaised || said[0].Subject.Machine != "g14" || said[0].Subject.ID != "screen-lock" {
t.Fatalf("said = %+v, want one raised condition of screen-lock on g14", said)
}
if !strings.Contains(strings.Join(logged, "\n"), "screen-lock") {
t.Errorf("the journal does not name the module on neither list: %q", logged)
}
p.sayNeither(map[string]neitherOne{}, nil)
p.sayNeither(map[string]neitherOne{}, nil)
wants := map[string]string{"g14.screen-lock": ""}
p.sayNeither(map[string]neitherOne{}, nil, wants)
p.sayNeither(map[string]neitherOne{}, nil, wants)
if len(said) != 2 || said[1].State != asks.StateCleared || said[1].Key != said[0].Key {
t.Errorf("said = %+v, want the condition cleared once under its key", said)
}
if last := logged[len(logged)-1]; last != "screen-lock on g14 has a want again" {
t.Errorf("the clear said %q", last)
}
// 1945 (the bus bar): a module that leaves neither list with no want is cleared in words saying so, never "has a want again".
bus := map[string]neitherOne{"novox.nats": {node: "novox", module: "nats", holding: assign.Holding{Bus: true, Words: "the bus moves only by the operator's put"}}}
p.sayNeither(bus, nil, nil)
if c := said[len(said)-1]; c.State != asks.StateRaised || !strings.HasPrefix(c.Explanation, asks.NeedsYou) || !strings.Contains(c.Explanation, "the bus moves only by the operator's put") {
t.Errorf("the bus's neither condition = %+v", c)
}
p.sayNeither(map[string]neitherOne{}, nil, map[string]string{"novox.nats": "it is held, and its held condition says so"})
if c := said[len(said)-1]; c.State != asks.StateCleared || strings.Contains(c.Resolved, "has a want") {
t.Errorf("a module that left neither list with no want was cleared as %+v", c)
}
if last := logged[len(logged)-1]; last != "nats on novox is no longer on neither list and has no want: it is held, and its held condition says so" {
t.Errorf("the clear without a want said %q", last)
}
}
// 1945/AC21: the pass gathers the modules on neither list of every module's outcome, keyed by node and module.
func TestThePassGathersTheModulesOnNeitherList(t *testing.T) {
func testThePassGathersTheModulesOnNeitherList(t *testing.T) {
got := map[string]neitherOne{}
gatherNeither(got, "screen-lock", assign.Outcome{Neither: map[string]assign.Holding{"g14": {Rule: delivery.RuleGrant, Words: "w"}}})
gatherNeither(got, "rofi", assign.Outcome{})
if len(got) != 1 || got["g14.screen-lock"].node != "g14" || got["g14.screen-lock"].module != "screen-lock" || got["g14.screen-lock"].holding.Rule != delivery.RuleGrant {
t.Errorf("gathered = %+v", got)
}
// 1945 (the bus bar): a node off neither list is cleared in words that say why it has no want; an operator's put running
// through its steps owns the want, so it is named, never "the node does not run it".
for _, c := range []struct {
name string
puts map[string]delivery.Put
before map[string]delivery.Want
why string
}{
{"a live want", nil, map[string]delivery.Want{"novox": {FP: fpTest}}, ""},
{"a put on its steps", map[string]delivery.Put{"novox": {FP: fpTest, Step: delivery.PutPerson}}, nil, "its put waits on its own steps and the operator's answer"},
{"nothing", nil, nil, "the node does not run it"},
{"an ended put", map[string]delivery.Put{"novox": {FP: fpTest, Step: delivery.PutRefused}}, nil, "the node does not run it"},
} {
left := map[string]string{}
leftNeither(left, assign.Module{Name: "nats", FP: fpTest, Nodes: []string{"novox"}, Puts: c.puts}, c.before, assign.Outcome{})
if why, there := left["novox.nats"]; !there || why != c.why {
t.Errorf("%s: left %q (%v), want %q", c.name, why, there, c.why)
}
}
// 1945 (a first want held back silently): the journal says why a node gets no want when the reason appears or changes, and
// when it has a want again: one line per change, not per pass.
p := newPasser(buckets{}, time.Date(2026, 10, 25, 21, 30, 0, 0, time.UTC))
@@ -399,6 +438,7 @@ func TestThePassGathersTheModulesOnNeitherList(t *testing.T) {
{nil, assign.Outcome{Rules: map[string]assign.Holding{"shanks": {Rule: delivery.RuleGrant, Words: "no credential"}}}, "rofi on shanks gets no want: rule 6: no credential"},
{nil, wanted, "rofi on shanks has a want"},
{map[string]delivery.Want{"shanks": wanted.Wants["shanks"]}, assign.Outcome{}, ""},
{nil, assign.Outcome{Rules: map[string]assign.Holding{"shanks": {Bus: true, Words: "the bus moves only by the operator's put"}}}, "rofi on shanks gets no want: the bus moves only by the operator's put"},
} {
logged = nil
p.sayNoWant("rofi", []string{"shanks"}, c.before, c.out)
@@ -406,12 +446,71 @@ func TestThePassGathersTheModulesOnNeitherList(t *testing.T) {
t.Errorf("pass %d said %q, want %q", i, got, c.line)
}
}
// 1496, ADR 0557 §1: a failure that is the node's own (its fingerprint healthy elsewhere, not held) raises the node's condition
// once, naming the node, the module, F and the have's words, and clears it once when it ends.
var conds []asks.Condition
p.say = func(c asks.Condition) error { conds = append(conds, c); return nil }
failing := map[string]assign.NodeFailure{"shanks": {Others: []string{"ace"}, Words: "health failed, phase running: w"}}
for i, c := range []struct {
failed map[string]assign.NodeFailure
state string // the condition said in this pass, "" for none
}{
{failing, asks.StateRaised},
{failing, ""},
{nil, asks.StateCleared},
{nil, ""},
} {
conds = nil
p.sayNodeFailed("rofi", fpTest, []string{"shanks"}, assign.Nodes{}, assign.Outcome{NodeFailed: c.failed}, false)
switch {
case c.state == "" && len(conds) != 0, c.state != "" && (len(conds) != 1 || conds[0].State != c.state):
t.Errorf("node failed, pass %d said %+v, want %q", i, conds, c.state)
case c.state == asks.StateRaised && (conds[0].Key != "mesh-delivery-assigner.failed-alone.shanks.rofi" ||
!strings.Contains(conds[0].Explanation, "failed on shanks (health failed, phase running: w), while it runs healthy on ace")):
t.Errorf("node failed, raised %+v", conds[0])
}
}
// D2: a node that leaves the module's nodes while it failed alone is cleared once; the first pass after a start clears every
// node the presence bucket knows, for every module, failed have or not (a node that recovered unwatched leaves none).
conds = nil
p.sayNodeFailed("rofi", fpTest, []string{"shanks"}, assign.Nodes{}, assign.Outcome{NodeFailed: failing}, false)
p.sayNodeFailed("rofi", fpTest, nil, assign.Nodes{}, assign.Outcome{}, false)
p.sayNodeFailed("rofi", fpTest, nil, assign.Nodes{}, assign.Outcome{}, false)
if len(conds) != 2 || conds[1].State != asks.StateCleared || conds[1].Key != "mesh-delivery-assigner.failed-alone.shanks.rofi" {
t.Errorf("a node leaving the module said %+v, want one raise and one clear", conds)
}
known := assign.Nodes{Online: map[string]bool{"ace": true, "g14": false},
Haves: map[string]map[string]delivery.Have{"g14": {"rofi": {FP: fpTest, Health: delivery.HealthFailed}}}}
for _, c := range []struct {
name string
known assign.Nodes
keys []string
}{
{"a failed have", known, []string{"mesh-delivery-assigner.failed-alone.ace.rofi", "mesh-delivery-assigner.failed-alone.g14.rofi"}},
{"no failed have", assign.Nodes{Online: known.Online}, []string{"mesh-delivery-assigner.failed-alone.ace.rofi", "mesh-delivery-assigner.failed-alone.g14.rofi"}},
} {
conds = nil
fresh := newPasser(buckets{}, time.Date(2026, 10, 26, 22, 0, 0, 0, time.UTC))
fresh.logf = func(string, ...any) {}
fresh.say = func(c asks.Condition) error { conds = append(conds, c); return nil }
fresh.sayNodeFailed("rofi", fpTest, nil, c.known, assign.Outcome{}, true)
var keys []string
for _, cond := range conds {
if cond.State == asks.StateCleared {
keys = append(keys, cond.Key)
}
}
if !slices.Equal(keys, c.keys) || len(conds) != len(c.keys) {
t.Errorf("first pass, %s: said %+v, want clears of %q", c.name, conds, c.keys)
}
}
}
// A neither or held condition a previous process raised is cleared by the first pass after a start: a fresh pass remembers
// nothing it said, so it clears, once, the neither condition of every assigned module not on neither list and the held condition of
// every module that is not held. Clearing a condition that is not open is quiet at the router.
func TestTheFirstPassAfterAStartClearsWhatAPreviousProcessRaised(t *testing.T) {
t.Run("a held module", testAHeldModuleRaisesNoNeitherAndClearsAStandingOne)
wantsKV := &memKV{values: map[string][]byte{}}
p := newPasser(buckets{delivery.AssignerWantsBucket: wantsKV, delivery.AssignerAsksBucket: &memKV{values: map[string][]byte{}}}, time.Now())
var said []asks.Condition
@@ -449,7 +548,7 @@ func TestTheFirstPassAfterAStartClearsWhatAPreviousProcessRaised(t *testing.T) {
// 1945/AC24: given a module held at F under rule 2 on a node that runs it with no want, when the assigner passes, then it raises no
// neither condition for it, a neither condition this process raised for it before is cleared, and so is one a previous process
// raised (the first pass after a start), each once.
func TestAHeldModuleRaisesNoNeitherAndClearsAStandingOne(t *testing.T) {
func testAHeldModuleRaisesNoNeitherAndClearsAStandingOne(t *testing.T) {
setup := func() (*passer, *[]asks.Condition) {
p := newPasser(buckets{delivery.AssignerWantsBucket: &memKV{values: map[string][]byte{}}, delivery.AssignerAsksBucket: &memKV{values: map[string][]byte{}}}, time.Now())
said := &[]asks.Condition{}
@@ -606,6 +705,36 @@ func TestAsksOfAFingerprintTheTrunkLeftBehindAreTakenAway(t *testing.T) {
})
}
t.Run("a clear that failed to be said is said at the next pass, and the ask goes after it", func(t *testing.T) {
asksKV, wantsKV := &memKV{values: map[string][]byte{}}, &memKV{values: map[string][]byte{}}
p := newPasser(buckets{delivery.AssignerAsksBucket: asksKV, delivery.AssignerWantsBucket: wantsKV}, time.Now())
var said []asks.Condition
fail := true
p.say = func(c asks.Condition) error {
if fail {
fail = false
return errors.New("router unreachable")
}
said = append(said, c)
return nil
}
a := assign.HoldAsk(holdOld, "web", fpOld, "novox", "w", 1, time.Now())
v, err := delivery.EncodeAssignerAsk(holdOld, a)
if err != nil {
t.Fatal(err)
}
asksKV.values[holdOld] = v
p.w.put(delivery.AssignerAsksBucket, holdOld, v)
p.staleAsks(context.Background(), trunkOf(fpNew))
if _, there := asksKV.values[holdOld]; !there || len(said) != 0 {
t.Fatalf("after a failed clear: ask left = %v, conditions said = %d; want the ask kept and nothing said", there, len(said))
}
p.staleAsks(context.Background(), trunkOf(fpNew))
if _, there := asksKV.values[holdOld]; there || len(said) != 1 || said[0].State != asks.StateCleared {
t.Errorf("at the next pass: ask left = %v, conditions said = %v; want the ask gone and the clear said", there, said)
}
})
t.Run("a lift takes the restore asks of its fingerprint on every node", func(t *testing.T) {
asksKV, wantsKV := &memKV{values: map[string][]byte{}}, &memKV{values: map[string][]byte{}}
p := newPasser(buckets{delivery.AssignerAsksBucket: asksKV, delivery.AssignerWantsBucket: wantsKV}, time.Now())
@@ -635,6 +764,8 @@ func mustKey(key string, err error) string {
// (ADR 0545). That record ask is not about the trunk's fingerprint: it survives every pass, so the operator is asked, and on a
// yes the want names the revised digest.
func TestARecordAskOfAHeldBackNodeSurvivesPassesAndItsYesMovesTheWant(t *testing.T) {
// A placed node's yes comes from its have (issue 1126): folded here to hold the count.
aPlacedNodesWantsNameTheBuildAloneAndItsYesComesFromItsHave(t)
fpOld := delivery.SourcePrefix2 + strings.Repeat("a", 64)
revised, was := "sha256:"+strings.Repeat("e", 64), "sha256:"+strings.Repeat("d", 64)
asksKV, wantsKV := &memKV{values: map[string][]byte{}}, &memKV{values: map[string][]byte{}}
@@ -683,3 +814,133 @@ func TestARecordAskOfAHeldBackNodeSurvivesPassesAndItsYesMovesTheWant(t *testing
t.Error("the answered record ask was taken away")
}
}
// 1126 (ADR 0549 §5): until the third milestone a pair the controller's assignments and the catalogue's placement do not
// both name gets no want and one condition, raised once and cleared once both name it; before any placement was read the
// assignments are the one source and nothing is raised.
func aPairOneSourceNamesAloneGetsNoWantAndOneCondition(t *testing.T) {
p := newPasser(buckets{}, time.Date(2026, 10, 26, 18, 0, 0, 0, time.UTC))
var said []asks.Condition
p.say = func(c asks.Condition) error { said = append(said, c); return nil }
assigned := func(node, module string) {
raw, _ := json.Marshal(map[string]string{"node": node, "module": module})
p.w.put(provider.AssignmentsBucket, node+"."+module, raw)
}
place := func(node string, modules ...string) {
pl := delivery.Placement{Node: node}
for _, m := range modules {
pl.Modules = append(pl.Modules, delivery.PlacedModule{Module: m})
}
raw, err := delivery.EncodePlacement(pl)
if err != nil {
t.Fatal(err)
}
key, _ := delivery.PlacementKey(node)
p.w.put(delivery.PlacementBucket, key, raw)
}
assigned("g14", "a")
assigned("g14", "b")
if got, _ := p.w.assigned(); !slices.Equal(got["g14"], []string{"a", "b"}) || len(p.sayPlacement(p.w.placementDiffers())) != 0 || len(said) != 0 {
t.Fatalf("before any placement: assigned %v, said %v", got, said)
}
none := func(context.Context, string) ([]byte, error) { return nil, errNotThere }
if m := p.w.mesh(context.Background(), none); m.Input != nil || m.InputUnread != "" {
t.Errorf("before any placement the check composes over %+v (%q)", m.Input, m.InputUnread)
}
place("g14", "a", "c")
got, _ := p.w.assigned()
frozen := p.sayPlacement(p.w.placementDiffers())
p.sayPlacement(p.w.placementDiffers())
if !slices.Equal(got["g14"], []string{"a"}) || !frozen["g14.b"] || !frozen["g14.c"] || frozen["g14.a"] || len(said) != 2 ||
said[0].State != asks.StateRaised || !strings.Contains(said[0].Explanation, "the controller's assignments names b on g14") ||
!strings.Contains(said[1].Explanation, "the catalogue's placement names c on g14") {
t.Fatalf("one source alone: assigned %v, frozen %v, said %+v", got, frozen, said)
}
place("g14", "a", "b", "c")
assigned("g14", "c")
if frozen := p.sayPlacement(p.w.placementDiffers()); len(frozen) != 0 || len(said) != 4 || said[2].State != asks.StateCleared || said[3].State != asks.StateCleared {
t.Errorf("both agree again: frozen %v, said %+v", frozen, said)
}
// 1126 (ADR 0547 §2): placed, the check composes over the input the nodes compose from (frombus), once each placed node
// wrote its record; until then it says why and composes the assignments alone.
if m := p.w.mesh(context.Background(), none); m.Input != nil || !strings.Contains(m.InputUnread, "g14") {
t.Errorf("a placed node with no record: input %+v, unread %q", m.Input, m.InputUnread)
}
rec, _ := delivery.EncodeHaveNode(delivery.HaveNode{Record: &delivery.NodeRecord{Node: "g14", TunnelKey: "k"}})
p.w.put(delivery.HaveBucket, "g14.node", rec)
m := p.w.mesh(context.Background(), none)
if m.Input == nil || len(m.Input.Nodes) != 1 || !slices.Equal(m.Input.Nodes[0].Assigned, []string{"a", "b", "c"}) || m.Input.Nodes[0].Place.Key != "k" {
t.Errorf("the check's input from the bus: %+v (%q)", m.Input, m.InputUnread)
}
}
// 1126 (ADR 0547 §2 to §4): a node the catalogue places composes its own: its wants name the build alone and its own want
// names no composition; a have waiting for a yes raises the record ask once, and the yes is written into the want and
// kept there while the want stays at its build.
func aPlacedNodesWantsNameTheBuildAloneAndItsYesComesFromItsHave(t *testing.T) {
fpOld := delivery.SourcePrefix2 + strings.Repeat("a", 64)
waiting, was := "sha256:"+strings.Repeat("e", 64), "sha256:"+strings.Repeat("d", 64)
asksKV, wantsKV := &memKV{values: map[string][]byte{}}, &memKV{values: map[string][]byte{}}
p := newPasser(buckets{delivery.AssignerAsksBucket: asksKV, delivery.AssignerWantsBucket: wantsKV}, time.Now())
p.logf = t.Logf
put := func(bucket, key string, v []byte) { p.w.put(bucket, key, v) }
mod, _ := delivery.EncodeModule(delivery.Module{FP2: fpTest, Repository: "novox/web", Commit: strings.Repeat("1", 40), Path: delivery.RootPath})
put(delivery.ModulesBucket, "web", mod) // the trunk's build is not done: rule 1 holds g14 at fpOld
put(provider.AssignmentsBucket, "g14.web", []byte(`{"node":"g14","module":"web","build":"b-1"}`))
pl, _ := delivery.EncodePlacement(delivery.Placement{Node: "g14", Modules: []delivery.PlacedModule{{Module: "web"}}})
put(delivery.PlacementBucket, "placement.g14", pl)
on, _ := delivery.EncodePresence(delivery.Presence{Online: true})
put(delivery.PresenceBucket, "g14", on)
have, _ := delivery.EncodeHave(delivery.Have{FP: fpOld, Composed: was, Phase: delivery.PhaseRunning, Health: delivery.HealthOK})
put(delivery.HaveBucket, "g14.web", have)
wkey, _ := delivery.WantKey("g14", "web")
want, _ := delivery.EncodeWant(delivery.Want{FP: fpOld, Composed: was})
wantsKV.values[wkey] = want
put(delivery.AssignerWantsBucket, wkey, want)
sync := func() {
for k, v := range wantsKV.values {
put(delivery.AssignerWantsBucket, k, v)
}
}
p.wants(context.Background())
p.nodeWants(context.Background())
sync()
got, err := delivery.DecodeWant(wantsKV.values[wkey])
if err != nil || got.FP != fpOld || got.Composed != "" || !got.Self {
t.Fatalf("the placed node's want is %+v (%v), want it at %s, the node composing it", got, err, fpOld)
}
nkey, _ := delivery.WantNodeKey("g14")
if nw, err := delivery.DecodeNodeWant(wantsKV.values[nkey]); err != nil || nw.Composed != "" || !nw.Self {
t.Fatalf("the placed node's own want is %+v (%v), want one naming no composition", nw, err)
}
waits, _ := delivery.EncodeHave(delivery.Have{FP: fpOld, Composed: was, Phase: delivery.PhaseWaitingYes, Health: delivery.HealthOK,
Judgement: delivery.JudgedRecreates, Waits: waiting, Detail: "it recreates web.server"})
put(delivery.HaveBucket, "g14.web", waits)
akey, _ := delivery.RecordAskKey("g14", "web", waiting)
p.wants(context.Background())
if _, there := asksKV.values[akey]; !there {
t.Fatal("a have waiting for a yes raised no record ask")
}
a, err := delivery.DecodeAssignerAsk(akey, asksKV.values[akey])
if err != nil {
t.Fatal(err)
}
put(delivery.AssignerAsksBucket, akey, asksKV.values[akey])
yes, _ := delivery.EncodeAnswer(delivery.Answer{Option: assign.OptionYes, Warrant: a.Ask.ID + "@2026-10-26T19:00:00Z", Words: "the operator"})
put(delivery.AnswersBucket, akey, yes)
p.wants(context.Background())
sync()
if got, err = delivery.DecodeWant(wantsKV.values[wkey]); err != nil || got.Yes != waiting || got.FP != fpOld {
t.Fatalf("after the yes the want is %+v (%v), want it to name the yes %s", got, err, waiting)
}
// The node applied it: its have runs the composition; the yes stays while the want stays at its build.
ran, _ := delivery.EncodeHave(delivery.Have{FP: fpOld, Composed: waiting, Phase: delivery.PhaseRunning, Health: delivery.HealthOK})
put(delivery.HaveBucket, "g14.web", ran)
p.wants(context.Background())
if got, _ = delivery.DecodeWant(wantsKV.values[wkey]); got.Yes != waiting {
t.Errorf("a pass after the move dropped the yes: %+v", got)
}
}
@@ -14,6 +14,8 @@ import (
"github.com/nats-io/nats.go/jetstream"
"git.novox.be/novox/mesh-sdk/go/catalogue"
"git.novox.be/novox/mesh-sdk/go/compose"
"git.novox.be/novox/mesh-sdk/go/compose/frombus"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/provider"
@@ -162,8 +164,56 @@ func (w *world) manifest(ctx context.Context, get getter, digest string) parsed
return p
}
// assigned is each node's modules with an open assignment, read from the controller's assignments, in name order.
// assigned is each node's modules both the controller's assignments and the indexer's placement name, in name order
// (ADR 0549 §5; the assignments alone before any placement was read), with every open assignment by node and module.
func (w *world) assigned() (map[string][]string, map[string]assignment) {
byAssignment, all := w.assignedByController()
placed, seeded := w.placed()
both, _ := assign.Placed(byAssignment, placed, seeded)
return both, all
}
// placementDiffers is every pair the controller's assignments and the indexer's placement do not both name.
func (w *world) placementDiffers() []assign.Pair {
byAssignment, _ := w.assignedByController()
placed, seeded := w.placed()
_, differ := assign.Placed(byAssignment, placed, seeded)
return differ
}
// placed is each node's modules in the indexer's placement (ADR 0549 §2), and whether any placement key was read: a key
// that does not read places nothing on its node and is said by the indexer's own read tool.
func (w *world) placed() (map[string][]string, bool) {
out := map[string][]string{}
values := w.read(delivery.PlacementBucket)
for key, value := range values {
node, err := delivery.ParsePlacementKey(key)
if err != nil {
continue
}
p, err := delivery.DecodePlacement(value)
if err != nil || p.Node != node {
continue
}
out[node] = p.Assigned()
}
return out, len(values) > 0
}
// placedNodes is every node the indexer's placement names (ADR 0549 §2): those compose their own modules (ADR 0547 §2),
// so their wants name the build alone and the seventh rule does not hold them.
func (w *world) placedNodes() map[string]bool {
placed, _ := w.placed()
out := make(map[string]bool, len(placed))
for node := range placed {
out[node] = true
}
return out
}
// assignedByController is each node's modules with an open assignment, read from the controller's assignments, in
// name order.
func (w *world) assignedByController() (map[string][]string, map[string]assignment) {
out, all := map[string][]string{}, map[string]assignment{}
for _, value := range w.read(provider.AssignmentsBucket) {
var a assignment
@@ -210,9 +260,86 @@ func (w *world) mesh(ctx context.Context, manifests getter) assign.Mesh {
}
m.Manifests[name] = p.manifest
}
// Once the catalogue places the nodes, the check composes over the input they compose from (ADR 0547 §2, ADR 0549
// §3): one composer's input for the check and the nodes. An input that cannot be read is said, and the nodes and their
// assignments alone are composed, as before any placement.
if _, seeded := w.placed(); seeded {
in, err := frombus.Input(ctx, worldReader{w: w, manifests: manifests}, frombus.Mesh{}, frombus.Scope{Every: true})
if err != nil {
m.InputUnread = err.Error()
} else {
m.Input = &in
}
}
return m
}
// worldReader is what the assigner watched as frombus reads the bus: the indexer's placement and modules, each node's
// record in its own have key, the manifests by digest; the check reads no config value, own secret or credential (the
// composition it asks resolves, and stands in for what it does not carry).
type worldReader struct {
w *world
manifests getter
}
func (r worldReader) Placements(context.Context) (map[string]delivery.Placement, error) {
out := map[string]delivery.Placement{}
for key, value := range r.w.read(delivery.PlacementBucket) {
node, err := delivery.ParsePlacementKey(key)
if err != nil {
continue
}
if p, err := delivery.DecodePlacement(value); err == nil && p.Node == node {
out[node] = p
}
}
return out, nil
}
func (r worldReader) Records(context.Context) (map[string]delivery.NodeRecord, error) {
out := map[string]delivery.NodeRecord{}
for key, value := range r.w.read(delivery.HaveBucket) {
node, _, isNode, err := delivery.ParseHaveKey(key)
if err != nil || !isNode {
continue
}
if h, err := delivery.DecodeHaveNode(value); err == nil && h.Record != nil {
out[node] = *h.Record
}
}
return out, nil
}
func (r worldReader) Modules(context.Context) (map[string]delivery.Module, error) {
return r.w.trunk(), nil
}
func (r worldReader) Manifest(ctx context.Context, digest string) ([]byte, error) {
key, err := delivery.ManifestsKey(digest)
if err != nil {
return nil, err
}
raw, err := r.manifests(ctx, key)
if err != nil {
return nil, err
}
m, err := delivery.DecodeManifest(raw)
if err != nil {
return nil, err
}
return []byte(m.JSON), nil
}
func (worldReader) Values(context.Context, string, string) (map[string]any, error) { return nil, nil }
func (worldReader) OwnSecrets(context.Context, string) ([]compose.OwnSecretRow, error) {
return nil, nil
}
func (worldReader) Credentials(context.Context, string, []compose.CredentialKey, []compose.CredentialKey) ([]compose.Credential, []compose.Issued, error) {
return nil, nil, nil
}
// change is a candidate's change for the composition check, or why the composition waits for it.
func (w *world) change(ctx context.Context, manifests getter, c delivery.Candidate, trunk map[string]delivery.Module) (assign.Change, string) {
ch := assign.Change{Group: c.Group, Heads: c.HeadsHash, Manifests: map[string]catalogue.Manifest{}, Unreadable: map[string]string{},
+1 -1
View File
@@ -3,7 +3,7 @@ module mesh-delivery-assigner
go 1.26.0
require (
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117
github.com/nats-io/nats-server/v2 v2.11.17
github.com/nats-io/nats.go v1.54.0
)
+2 -10
View File
@@ -1,13 +1,5 @@
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261024210118-748c564cdf7c h1:pY2YpHSSsfDPBrOEiWTNSIPn0zdnAayRmtmmyQ3hCHY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261024210118-748c564cdf7c/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025195051-20a6dc4920f8 h1:jA7V7hlfJAigVUvJDpKwfgd9ylQ9bqppg3rK9XgLWoA=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025195051-20a6dc4920f8/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025200810-d4ae4b6790b8 h1:vhYk/iv43u2WGSTrFyQGDvGBxLaRITgMAiiI9DjuA10=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025200810-d4ae4b6790b8/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025203528-db26f7e8b6f5 h1:gE80bR0tPME7MTMBLcdP62v+F/9soVXmdcBYNMOye44=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025203528-db26f7e8b6f5/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585 h1:7QoFH5qjXRpx5M4uPdYsVJ5qKl2082gq8n0lFxhoNIg=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117 h1:88CNZOoQgu7YPYKW16qkWT8Gw9Ucg07aTrIImA2nYUk=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op/go.mod h1:FQyySiasQQM8735Ddel3MRojmy4dA1IqCeyJ5jmPMbI=
github.com/google/go-tpm v0.9.8 h1:slArAR9Ft+1ybZu0lBwpSmpwhRXaa85hWtMinMyRAWo=
@@ -81,6 +81,29 @@ func HeldCondition(module, fp, node, words string, raised, needsYou bool) asks.C
return c
}
// NodeFailedCondition is the condition the assigner raises for a module whose newest fingerprint fp failed its health check on
// node while other nodes run it healthy (ADR 0557 §1, issue 1496): the failure is the node's, so nothing is held or set back and
// the node keeps its want. It names the node, the module, fp and the have's words, keyed by node and module, and is cleared
// (raised false) once the node no longer fails alone.
func NodeFailedCondition(node, module, fp string, f NodeFailure, raised bool) asks.Condition {
key := delivery.AssignerModule + ".failed-alone." + node + "." + module
subject := asks.Scope{Scope: "module", ID: module, Machine: node}
if !raised {
return asks.Condition{Key: key, Subject: subject, State: asks.StateCleared,
Resolved: fmt.Sprintf("%s on %s no longer fails its health check alone.", module, node)}
}
headline := fmt.Sprintf("%s fails on %s alone", module, node)
if n := len([]rune(headline)); n > asks.HeadlineLength {
headline = string([]rune(headline)[:asks.HeadlineLength])
}
return asks.Condition{Key: key, Severity: asks.SeverityWarning, Subject: subject, State: asks.StateRaised,
Summary: headline, Headline: headline,
Explanation: fmt.Sprintf("%s The health check of %s at %s failed on %s (%s), while it runs healthy on %s. The failure is %s's, "+
"so the assigner holds nothing and sets no node back, and %s keeps its want. This clears once %s runs it healthy or "+
"another version is the newest.", asks.NothingToDo, module, short(fp), node, f.Words, strings.Join(f.Others, ", "), node,
node, node)}
}
// NeitherCondition is the condition of a module assigned to node that holds no want for it (issue 1945): the node's own composition
// gives up every module with a fingerprint of the new form, so nothing applies it there and its files are gone. It is raised (urgent)
// naming the module, the node and the rule holding its first want back with the rule's words, under the assigner's own name keyed by
@@ -96,6 +119,14 @@ func NeitherCondition(node, module string, h Holding, raised bool) asks.Conditio
if n := len([]rune(headline)); n > asks.HeadlineLength {
headline = string([]rune(headline)[:asks.HeadlineLength])
}
if h.Bus {
needs := fmt.Sprintf("put %s on %s with assigner_put", module, node)
return asks.Condition{Key: key, Severity: asks.SeverityUrgent, Subject: subject, State: asks.StateRaised,
Summary: headline, Headline: headline, Needs: needs,
Explanation: fmt.Sprintf("%s%s. %s is assigned to %s, but the node holds no want for it and the node's own composition does not "+
"hold it either, so nothing applies it there. The assigner never moves the bus: %s. This clears once the node has a want for it.",
asks.NeedsYou, needs, module, node, h.Words)}
}
return asks.Condition{Key: key, Severity: asks.SeverityUrgent, Subject: subject, State: asks.StateRaised,
Summary: headline, Headline: headline,
Explanation: fmt.Sprintf("%s %s is assigned to %s, but the node holds no want for it and the node's own composition does not "+
@@ -103,6 +134,17 @@ func NeitherCondition(node, module string, h Holding, raised bool) asks.Conditio
"rule holds, and clears this.", asks.NothingToDo, module, node, h.Rule, h.Words)}
}
// NeitherLeftCondition clears the neither condition of module on node when it is no longer on neither list: why is "" when the
// node has a live want for it again, else why it has none (it is held, its move waits for an ask, the node no longer runs it), so
// the clear never says a want the node does not hold (issue 1945: "nats on novox has a want again" while it had none).
func NeitherLeftCondition(node, module, why string) asks.Condition {
c := NeitherCondition(node, module, Holding{}, false)
if why != "" {
c.Resolved = fmt.Sprintf("%s on %s is no longer on neither list and has no want: %s.", module, node, why)
}
return c
}
// LiftedCondition clears the condition of a held fingerprint the assigner lifted by itself, saying why (issue 1906).
func LiftedCondition(module, fp string) asks.Condition {
c := HeldCondition(module, fp, "", "", false, false)
@@ -35,6 +35,13 @@ type Mesh struct {
// Digests is every trunk module's manifest digest in the indexer's manifests bucket, readable or not: what a
// composer at another mesh-sdk reads (issue 1622).
Digests map[string]string
// Input is the rest of the mesh as the nodes compose from it, filled by mesh-sdk's frombus from the catalogue's
// placement and the nodes' records (novox/hq ADR 0547 §2, ADR 0549 §3, issue 1126), so the check and the nodes share
// one composer's input; nil before any placement is read, when the nodes and their assignments alone are composed.
Input *compose.Input
// InputUnread says why that input could not be read from the bus, when it could not: the check then composes the
// nodes and their assignments alone, and says so in its warnings.
InputUnread string
}
// Change is a group's change as the indexer's candidate holds it: the manifest of every module it moves or adds, those
@@ -42,8 +42,15 @@ func HeadJob(m Mesh, c Change) (delivery.ComposeJob, error) {
for _, name := range c.Removed {
delete(change, name)
}
req := composer.Request{V: composer.Version, Input: plainInput(m.Assigned), Depth: composer.DepthResolve,
HoldingsFromClaims: true, Machines: Touched(m, c),
in, fromClaims := plainInput(m.Assigned), true
if m.Input != nil {
// The input the nodes compose from, without its catalogue: the cases carry the manifests (issue 1126).
in = *m.Input
in.Catalogue = nil
fromClaims = len(in.Holdings) == 0
}
req := composer.Request{V: composer.Version, Input: in, Depth: composer.DepthResolve,
HoldingsFromClaims: fromClaims, Machines: Touched(m, c),
Cases: []composer.Case{{Name: CaseTrunk, Manifests: trunk}, {Name: CaseChange, Manifests: change}}}
return job(SDKRepository, c.SDKHead, SDKDir, req, "the mesh-sdk head of "+c.Group)
}
@@ -195,6 +202,10 @@ func FromComposer(m Mesh, c Change, head Ran, readers []Reader, took time.Durati
readerFails, readerWarnings := judgeReaders(c, change, readers)
fails = append(fails, readerFails...)
v.Warnings = append(v.Warnings, readerWarnings...)
if m.InputUnread != "" {
v.Warnings = append(v.Warnings, "composed over the nodes and their assignments alone: the input the nodes compose from "+
"cannot be read from the bus ("+m.InputUnread+")")
}
return finish(v, m, c, machines, fails)
}
@@ -8,6 +8,7 @@ import (
"time"
"git.novox.be/novox/mesh-sdk/go/catalogue"
"git.novox.be/novox/mesh-sdk/go/compose"
"git.novox.be/novox/mesh-sdk/go/composer"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
@@ -100,6 +101,25 @@ func TestAGroupWithAnSDKMemberAsksTheComposerAtItsHead(t *testing.T) {
if !slices.Equal(req.Machines, []string{"g14"}) || req.Depth != composer.DepthResolve || !req.HoldingsFromClaims {
t.Errorf("request = machines %v, depth %q, holdings from claims %v", req.Machines, req.Depth, req.HoldingsFromClaims)
}
// 1126 (ADR 0547 §2, ADR 0549 §3): once the catalogue places the nodes, the check composes over the input the nodes
// compose from (frombus), its catalogue left to the cases, and the seats held as placement says.
m := digestedMesh()
m.Input = &compose.Input{Catalogue: map[string]catalogue.Manifest{"web": {Module: "web"}},
Nodes: []compose.Node{{Name: "g14", Assigned: []string{"web"}, Place: compose.Place{Endpoint: "g14.example:51820"}}},
Holdings: []catalogue.Held{{Claim: "web-seat", Scope: "mesh", Node: "g14", Module: "web"}}}
placedJob, err := HeadJob(m, c)
if err != nil {
t.Fatal(err)
}
placed, err := composer.Decode(placedJob.Request)
if err != nil {
t.Fatal(err)
}
if placed.Input.Catalogue != nil || len(placed.Input.Nodes) != 1 || placed.Input.Nodes[0].Place.Endpoint != "g14.example:51820" ||
len(placed.Input.Holdings) != 1 || placed.HoldingsFromClaims {
t.Errorf("the placed check's request = %+v, holdings from claims %v", placed.Input, placed.HoldingsFromClaims)
}
}
// 1622/AC2, AC3: a catalogue member using a field only the SDK head reads passes with the head's composer, and the
@@ -0,0 +1,99 @@
package assign
import (
"fmt"
"sort"
"git.novox.be/novox/mesh-sdk/go/asks"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// Until ADR 0547's third milestone two sources name which module runs on which node: the controller's assignments and
// the indexer's placement, read from the catalogue (ADR 0549 §5). The assigner writes wants only for a pair both name:
// for a pair one source names alone it writes no want and touches none it wrote before, so nothing moves and nothing is
// removed on one source's word, and it raises a condition naming the pair until both agree. Before the first placement
// file reaches the indexer (no placement key at all) the controller's assignments are the one source, as before.
// Pair is a module on a node that one source names and the other does not; Only says which: OnlyAssignments or
// OnlyPlacement.
type Pair struct {
Node, Module, Only string
}
// The source that names a pair alone.
const (
OnlyAssignments = "the controller's assignments"
OnlyPlacement = "the catalogue's placement"
)
// Placed is, by node, the modules both sources name, in name order, and every pair one names alone, by node then
// module. With no placement read yet (seeded false) it is the assignments, and no pair differs.
func Placed(assigned, placed map[string][]string, seeded bool) (map[string][]string, []Pair) {
if !seeded {
return assigned, nil
}
set := func(m map[string][]string) map[string]bool {
out := map[string]bool{}
for node, modules := range m {
for _, module := range modules {
out[node+"\x00"+module] = true
}
}
return out
}
inAssigned, inPlaced := set(assigned), set(placed)
both := map[string][]string{}
var differ []Pair
for node, modules := range assigned {
for _, module := range modules {
if inPlaced[node+"\x00"+module] {
both[node] = append(both[node], module)
} else {
differ = append(differ, Pair{Node: node, Module: module, Only: OnlyAssignments})
}
}
}
for node, modules := range placed {
for _, module := range modules {
if !inAssigned[node+"\x00"+module] {
differ = append(differ, Pair{Node: node, Module: module, Only: OnlyPlacement})
}
}
}
for _, modules := range both {
sort.Strings(modules)
}
sort.Slice(differ, func(i, j int) bool {
if differ[i].Node != differ[j].Node {
return differ[i].Node < differ[j].Node
}
return differ[i].Module < differ[j].Module
})
return both, differ
}
// PlacementCondition is the condition of a pair the two sources do not both name: raised while they differ, cleared once
// both name it or neither does.
func PlacementCondition(p Pair, raised bool) asks.Condition {
key := delivery.AssignerModule + ".placement." + p.Node + "." + p.Module
subject := asks.Scope{Scope: "module", ID: p.Module, Machine: p.Node}
if !raised {
return asks.Condition{Key: key, Subject: subject, State: asks.StateCleared,
Resolved: fmt.Sprintf("the controller's assignments and the catalogue's placement agree on %s on %s.", p.Module, p.Node)}
}
headline := fmt.Sprintf("%s on %s is placed by one source only", p.Module, p.Node)
if n := len([]rune(headline)); n > asks.HeadlineLength {
headline = string([]rune(headline)[:asks.HeadlineLength])
}
var needs string
if p.Only == OnlyAssignments {
needs = fmt.Sprintf("add %s to the catalogue's placement file of %s, or unassign it at the controller", p.Module, p.Node)
} else {
needs = fmt.Sprintf("assign %s to %s at the controller, as the merged placement file says", p.Module, p.Node)
}
return asks.Condition{Key: key, Severity: asks.SeverityUrgent, Subject: subject, State: asks.StateRaised,
Summary: headline, Headline: headline, Needs: needs,
Explanation: fmt.Sprintf("%s%s. %s names %s on %s and the other source does not. Until ADR 0547's third milestone both must "+
"name a module on a node (ADR 0549 §5): the assigner writes no want for it there and leaves the one it wrote before, so "+
"nothing moves or is removed. This clears once both name it or neither does.", asks.NeedsYou, needs, p.Only, p.Module, p.Node)}
}
@@ -2,6 +2,7 @@ package assign
import (
"fmt"
"maps"
"slices"
"sort"
"strings"
@@ -169,6 +170,9 @@ type Outcome struct {
// Lifted is the hold of m.FP lifted by itself: every node that runs the module runs m.FP healthy (issue 1906). The held
// record and its ask are taken away, and the wants move by the rules again in this pass.
Lifted bool
// NodeFailed is, by node, a failed health check of m.FP on a node while another node runs m.FP healthy: the failure is
// that node's, so m.FP is not held and nothing is set back (ADR 0557 §1, issue 1496). The pass raises its condition once.
NodeFailed map[string]NodeFailure
// Record is the moves a person must allow before the want names them: each is asked once (ADR 0539 §14).
Record []RecordNeed
// Restore is the nodes of a module holding data whose held fingerprint failed its health check there and that are not set
@@ -192,10 +196,27 @@ type RecordNeed struct {
Words string
}
// Holding is a rule holding a node back, and why.
// Holding is a rule holding a node back, and why. Bus marks the bar on the bus instead of one of the seven rules (Rule is then
// 0): the bus moves only by the operator's put (ADR 0539 §15, ADR 0546 §8), never by the assigner (issue 1945: nats on novox,
// running an old-form build, got no want and the pass said nothing).
type Holding struct {
Rule int
Words string
Bus bool
}
// Reason is the holding in words: the rule and its words, or for the bus bar its words alone.
func (h Holding) Reason() string {
if h.Bus {
return h.Words
}
return fmt.Sprintf("rule %d: %s", h.Rule, h.Words)
}
// BusBarWords say why the assigner writes no want moving the bus on node: it runs prev, the trunk builds fp, and only the
// operator's put moves it (ADR 0539 §15, ADR 0546 §8). The words reach the operator, so they name no record.
func BusBarWords(node, prev, fp string) string {
return fmt.Sprintf("only your put moves the bus: %s runs %s, the trunk builds %s; put it with assigner_put", node, short(prev), short(fp))
}
// Wants is the pass of one module m: for each of its nodes, the want by the six rules, starting from the want written
@@ -211,11 +232,25 @@ func Wants(m Module, n Nodes, wants map[string]delivery.Want, lastHealthy map[st
}
if !held && !out.Lifted && !m.Released && m.FP != "" {
for _, node := range slices.Sorted(slices.Values(m.Nodes)) {
if h, ok := n.Haves[node][m.Name]; ok && h.FP == m.FP && failed(h) {
out.Held = &delivery.Held{Node: node, Words: HeldWords(h, m.Now)}
held = true
break
h, ok := n.Haves[node][m.Name]
if !ok || h.FP != m.FP || !failed(h) {
continue
}
if others := healthyElsewhere(m, n, node); len(others) > 0 && !failedElsewhere(m, n, node) {
// F runs healthy on another node, so its failure here is this node's, not F's: F is not held, no node is set back,
// and the failing node keeps its want (ADR 0557 §1; issue 1496: on ace a newly assigned mesh-build-node failed its
// first check and the hold moved shanks, g14 and novox, healthy on F for an hour, back). The pass raises the node's
// condition. With no online healthy witness, or with F failed on a second node, the hold and the set-back stand (ADR 0557
// §2, ADR 0429 §6).
if out.NodeFailed == nil {
out.NodeFailed = map[string]NodeFailure{}
}
out.NodeFailed[node] = NodeFailure{Others: others, Words: HeldWords(h, time.Time{})} // said once: no clock in it
continue
}
out.Held = &delivery.Held{Node: node, Words: HeldWords(h, m.Now)}
held = true
break
}
}
mm := m
@@ -281,6 +316,13 @@ func Wants(m Module, n Nodes, wants map[string]delivery.Want, lastHealthy map[st
ok, target = true, have.FP
}
}
if ok && target == m.FP && len(out.NodeFailed) > 0 && !(had && !before.Removed && before.FP == m.FP) &&
n.Haves[node][m.Name].FP != m.FP {
// While a node fails F alone, F reaches no further node: a node that neither wants nor runs F gets no new want
// naming it, and wants already written stay (ADR 0557 §2). A node that runs F with no live want gets its want at F, or
// it would be on neither list and its node-engine would give the module up (issue 1945).
ok, rule, words = false, delivery.RuleNotHeld, FailedAloneWords(m.Name, m.FP, slices.Sorted(maps.Keys(out.NodeFailed)))
}
switch {
case ok:
d, _ := m.composition(node, target)
@@ -303,6 +345,14 @@ func Wants(m Module, n Nodes, wants map[string]delivery.Want, lastHealthy map[st
if need, wait := at.mustAsk(node, asked, hadLive, ran, running.FP, running.Composed == "", d); wait {
if need != nil {
out.Record = append(out.Record, *need)
} else if prev, barred := at.busBar(asked, hadLive, running.FP); barred {
// The bus is barred: no want moves it but the operator's put. Said, never silent (issue 1945): its holding is named,
// and a node that runs it with no live want has it on neither list, so its condition stays raised.
h := Holding{Bus: true, Words: BusBarWords(node, prev, at.FP)}
out.Rules[node] = h
if !hadLive && running.Phase != delivery.PhaseRemoved && !held {
out.Neither[node] = h
}
}
continue
}
@@ -394,6 +444,43 @@ func failed(h delivery.Have) bool {
return h.Health == delivery.HealthFailed && (h.Phase == delivery.PhaseRunning || h.Phase == delivery.PhaseFailed)
}
// healthyElsewhere is every online node of m but node that runs m.FP healthy, by name.
func healthyElsewhere(m Module, n Nodes, node string) []string {
var out []string
for _, other := range slices.Sorted(slices.Values(m.Nodes)) {
// A witness is online: a node that sleeps keeps the have it last wrote, and a stale have is no evidence that F runs
// healthy now.
if other != node && n.Online[other] && healthy(n.Haves[other][m.Name], m.FP) {
out = append(out, other)
}
}
return out
}
// failedElsewhere says whether another node of m has a have naming m.FP failed: then the failure is not one node's alone, and
// the hold and the set-back stand at the second failure (ADR 0557 §2).
func failedElsewhere(m Module, n Nodes, node string) bool {
for _, other := range m.Nodes {
if h, ok := n.Haves[other][m.Name]; ok && other != node && h.FP == m.FP && failed(h) {
return true
}
}
return false
}
// FailedAloneWords say why a node gets no new want naming fp: fp fails its health check on the nodes named, alone.
func FailedAloneWords(module, fp string, failing []string) string {
return fmt.Sprintf("%s at %s fails its health check on %s, so no further node moves to it until that is over", module, short(fp),
strings.Join(failing, ", "))
}
// NodeFailure is a failed health check of a module's newest fingerprint on one node while it runs healthy on Others: the have's
// health, phase and words (HeldWords without a time), and the healthy witnesses by name.
type NodeFailure struct {
Others []string
Words string
}
// runs says whether a have is of a module the node runs: running, whatever its health.
func runs(h delivery.Have) bool {
return h.Phase == delivery.PhaseRunning
@@ -466,12 +553,7 @@ func (m Module) mustAsk(node string, before delivery.Want, had, ran bool, runnin
if !had && ran && running == m.FP && haveNamesNoComposition && d.Judgement == delivery.JudgedUnchanged {
return nil, false
}
// The fingerprint the bus ran before: its want's, or with no readable want the one its have names (ADR 0539 §15 bars a new one).
prev := before.FP
if !had {
prev = running
}
if catalogue.ProvidesBus(m.Manifest) && prev != m.FP {
if _, barred := m.busBar(before, had, running); barred {
return nil, true // the bus is never moved by the assigner; the operator writes its want (§15)
}
if !m.recorded(d) || d.Judgement == delivery.JudgedNeither || d.Judgement == delivery.JudgedUnchanged {
@@ -483,6 +565,16 @@ func (m Module) mustAsk(node string, before delivery.Want, had, ran bool, runnin
return &RecordNeed{Node: node, Digest: d.Digest, Judgement: d.Judgement, Words: d.Words}, true
}
// busBar says whether m is the bus and the move would take it off the fingerprint it ran before, prev: its want's, or with no
// readable want the one its have names (ADR 0539 §15 bars a new one).
func (m Module) busBar(before delivery.Want, had bool, running string) (prev string, barred bool) {
prev = before.FP
if !had {
prev = running
}
return prev, catalogue.ProvidesBus(m.Manifest) && prev != m.FP
}
// recorded says whether the module is of policy record: as the controller holds it beside the composition's digest, the
// operator's own choice included (ADR 0539 §14); on a composition made by hand, which carries none, as the manifest derives it.
func (m Module) recorded(d delivery.ComposedDigest) bool {
@@ -150,6 +150,72 @@ func TestAFailedHealthHoldsAndSetsBack(t *testing.T) {
if w, ok := out.Wants["g14"]; ok && w.FP == fNew {
t.Errorf("g14 moved to a held fingerprint: %+v", w)
}
// 1496, ADR 0557 §1, §2: the same failure while F runs healthy on another node is that node's: F is not held, nothing is set
// back, the failing node keeps its want, and the outcome names it (on ace a newly assigned module's first check held a version healthy for an hour on three nodes).
n.Haves["g14"]["mail"] = delivery.Have{FP: fNew, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
before := map[string]delivery.Want{"ace": {FP: fNew, Composed: dig(fNew)}, "g14": {FP: fNew, Composed: dig(fNew)}}
out = Wants(m, n, before, map[string]string{"ace": fOld, "g14": fOld})
if out.Held != nil || len(out.Wants) != 0 {
t.Fatalf("held %+v, wants %+v: want no hold and no want changed", out.Held, out.Wants)
}
if f := out.NodeFailed["ace"]; !slices.Equal(f.Others, []string{"g14"}) || f.Words != "health failed, phase running: connection refused" ||
len(out.NodeFailed) != 1 {
t.Errorf("node failed = %+v", out.NodeFailed)
}
// ADR 0557 §2 (version 9): while one node fails F alone, novox, whose want names fOld, gets no new want naming F; the wants
// at F stay. The canary ace runs F healthy here and g14 fails it, so the canary rule (3) would let novox move: only §2 holds it.
three := m
three.Nodes = []string{"ace", "g14", "novox"}
failedHave := n.Haves["ace"]["mail"]
n.Haves["ace"]["mail"], n.Haves["g14"]["mail"] = n.Haves["g14"]["mail"], failedHave
n.Online["novox"] = true
n.Haves["novox"] = map[string]delivery.Have{"mail": {FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}}
before["novox"] = delivery.Want{FP: fOld, Composed: dig(fOld)}
out = Wants(three, n, before, map[string]string{"ace": fOld, "g14": fOld, "novox": fOld})
if _, alone := out.NodeFailed["g14"]; !alone {
t.Errorf("g14 does not fail alone: %+v", out.NodeFailed)
}
if w, written := out.Wants["novox"]; out.Held != nil || len(out.NodeFailed) != 1 || (written && w.FP == fNew) || out.Rules["novox"].Rule != delivery.RuleNotHeld {
t.Errorf("held %+v, node failed %+v, novox's want %+v (%+v): want novox kept off F", out.Held, out.NodeFailed, out.Wants["novox"], out.Rules["novox"])
}
if _, written := out.Wants["g14"]; written {
t.Errorf("g14's want at F was rewritten: %+v", out.Wants["g14"])
}
// A node that runs F healthy with no want, while g14 fails it alone, gets its want at F and is on no neither list (issue 1945).
n.Haves["novox"]["mail"] = delivery.Have{FP: fNew, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
delete(before, "novox")
out = Wants(three, n, before, map[string]string{"ace": fOld, "g14": fOld, "novox": fOld})
if w := out.Wants["novox"]; w.FP != fNew || len(out.Neither) != 0 {
t.Errorf("novox running F with no want: want %+v, neither %+v; want its want at F and no neither", w, out.Neither)
}
n.Haves["ace"]["mail"], n.Haves["g14"]["mail"] = failedHave, n.Haves["ace"]["mail"]
// ADR 0557 §2: a witness that is offline keeps the have it last wrote and witnesses nothing (D1); F failing on a second node is
// not one node's failure (D3). Either way F is held and ace is set back.
for _, c := range []struct {
name string
online map[string]bool
novox delivery.Health
}{
{"the witness sleeps", map[string]bool{"ace": true, "g14": false, "novox": true}, delivery.HealthOK},
{"F fails on a second node", map[string]bool{"ace": true, "g14": true, "novox": true}, delivery.HealthFailed},
} {
n.Online = c.online
n.Haves["novox"] = map[string]delivery.Have{"mail": {FP: fNew, Phase: delivery.PhaseRunning, Health: c.novox}}
if c.novox == delivery.HealthOK {
n.Haves["novox"]["mail"] = delivery.Have{FP: fOld, Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
}
before["novox"] = delivery.Want{FP: fNew, Composed: dig(fNew)}
out = Wants(three, n, before, map[string]string{"ace": fOld, "g14": fOld, "novox": fOld})
if out.Held == nil || out.Held.Node != "ace" || len(out.NodeFailed) != 0 {
t.Errorf("%s: held %+v, node failed %+v: want F held at ace", c.name, out.Held, out.NodeFailed)
}
if w := out.Wants["ace"]; w.FP != fOld {
t.Errorf("%s: ace's want = %+v, want set back", c.name, w)
}
}
}
// 1297/B4: a module holding data is not set back; its want stays and says why.
@@ -427,6 +493,23 @@ func TestTheBusIsNotMovedButItsUserListRevisionIsWrittenAtOnce(t *testing.T) {
t.Errorf("a held bus did not take its user-list revision as a put: %+v", w)
}
m.Puts = nil
// 1945 (the bus bar made loud): a node that runs the bus at an old-form build and holds no want for it gets none, since only
// the operator's put moves the bus, and the pass says so: a holding naming the bar, and the node on neither list so its
// condition stays raised. A node with a live want keeps it and is not on neither list.
m.Composed = composedEverywhere
n.Haves["ace"]["mail"] = delivery.Have{FP: "b-1700000000-oldform", Phase: delivery.PhaseRunning, Health: delivery.HealthOK}
out = Wants(m, n, map[string]delivery.Want{"g14": {FP: fOld, Composed: dig(fOld)}}, nil)
if w, wrote := out.Wants["ace"]; wrote {
t.Errorf("the bus bar wrote ace's want %+v", w)
}
h, on := out.Neither["ace"]
if !on || !h.Bus || out.Rules["ace"] != h || !strings.Contains(h.Words, "only your put moves the bus") || !strings.Contains(h.Words, "ace runs b-1700000000-oldform") ||
!strings.Contains(h.Words, "assigner_put") {
t.Errorf("the barred bus on ace: neither %+v (%v), rules %+v", h, on, out.Rules["ace"])
}
if _, on := out.Neither["g14"]; on || out.Wants["g14"].FP == fNew {
t.Errorf("g14 has a live want: neither %+v, its want %+v", out.Neither, out.Wants["g14"])
}
stepsOfAPut(t)
}
@@ -34,7 +34,7 @@ func TestManifest(t *testing.T) {
slices.Sort(reads)
want := []string{"mesh-build-go.steps", "mesh-controller.assignments", "mesh-controller.builds", "mesh-controller.composed", "mesh-controller.grants",
"mesh-delivery-builder.builds", "mesh-delivery-indexer.candidates", "mesh-delivery-indexer.manifests",
"mesh-delivery-indexer.modules", "mesh-delivery-teller.answers", "mesh-delivery-watchman.presence", "node-engine.have"}
"mesh-delivery-indexer.modules", "mesh-delivery-indexer.placement", "mesh-delivery-teller.answers", "mesh-delivery-watchman.presence", "node-engine.have"}
if !reflect.DeepEqual(reads, want) {
t.Errorf("reads = %v, want exactly %v", reads, want)
}
@@ -12,6 +12,7 @@
"mesh-delivery-indexer.candidates",
"mesh-delivery-indexer.manifests",
"mesh-delivery-indexer.modules",
"mesh-delivery-indexer.placement",
"mesh-delivery-teller.answers",
"mesh-delivery-watchman.presence",
"node-engine.have"
@@ -15,6 +15,7 @@ import (
stdio "git.novox.be/novox/mesh-sdk/go"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
"mesh-delivery-builder/internal/build"
)
@@ -31,7 +32,8 @@ func Tools(r build.StuckReader, askAgain func(key string, value []byte) error, q
Description: "Every build key that waits (building or queued), oldest first: its module, fingerprint, step words and since when, and each ask it " +
"waits on by language with its id, when it was put, its asking (1 for the first; the builder asks an ask again when it can no longer be " +
"answered, at most 3 times) and what its build module's steps hold of it (none, or its state on its node since its start). The " +
"seat workers' delivery counts are not readable by the builder. With module, that module's keys only. Read-only.",
"seat workers' delivery counts are not readable by the builder. dead_letters says whether the builder hears the asks the build seats give up " +
"on (issue 734), and why not: heard, a given-up ask is asked again at once; deaf, at its check bound. With module, that module's keys only. Read-only.",
Input: map[string]any{"type": "object", "properties": map[string]any{
"module": map[string]any{"type": "string", "description": "only this module's keys"}}},
Run: func(args map[string]any) (any, error) { return r.Stuck(str(args, "module")) }},
@@ -166,7 +168,8 @@ func main() {
steps[bm] = read(stdio.State(bm + ".steps"))
}
reasks := stdio.State(build.ReAsksState)
r := build.StuckReader{Logf: func(f string, a ...any) { fmt.Fprintf(os.Stderr, "[mesh-delivery-builder-tools] "+f+"\n", a...) }, Builds: read(stdio.State("builds")), ReAsks: read(reasks), Steps: steps}
r := build.StuckReader{Logf: func(f string, a ...any) { fmt.Fprintf(os.Stderr, "[mesh-delivery-builder-tools] "+f+"\n", a...) }, Builds: read(stdio.State("builds")), ReAsks: read(reasks), Steps: steps,
Hearing: read(stdio.State(buildrun.HearingState))}
put := func(key string, value []byte) error {
_, err := reasks.Put(key, json.RawMessage(value))
return err
@@ -149,6 +149,18 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
// Each build module's part of a build is an ask on its build seat's stream, pulled by a build node with a free slot
// (novox/hq ADR 0525): the builder chooses no node, so it reads neither presence nor have and holds no capacity.
asks := buildrun.JetStreamAsks{JS: js}
// The asks the build seats gave up on (issue 734): heard from DEAD_LETTERS, each ends the wait for its step at once,
// and the lane asks its part again (Builder.waitOnce) instead of waiting for the check bound.
// Heard in the background, and again at a reconnect or an ask while it is not; whether it is heard is kept in the
// state hearing, which builder_stuck says (dead_letters).
dead := &buildrun.DeadLetters{}
// The state may not be on the bus yet when a delivery started the builder (the grants part creates it, issue 1790): it
// is bound when the bus says it was created, and the latest hearing is written then.
hearingKV := process.AwaitBucket(ctx, js, conn, delivery.BuilderModule+"_"+buildrun.HearingState, "builder_stuck cannot say whether DEAD_LETTERS is heard", logger.Printf)
hearing := buildrun.NewDeadLetterHearing(ctx, buildrun.JetStreamDeadLetters{JS: js}, delivery.AskLanguages, dead, logger.Printf,
buildrun.SayHearing(ctx, hearingKV, logger.Printf))
defer hearing.Stop()
conn.SetReconnectHandler(func(*nats.Conn) { hearing.Again() })
// The builder asks the controller, as its own account, whether the artifact store still holds a done key's blobs
// (the seat verb held, novox/hq issue 1937) before it leaves the key done.
account, err := process.ReadAccount(path)
@@ -162,7 +174,7 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
defer calls.Close()
b := &build.Builder{
Claimer: build.Claimer{Store: kvStore{builds}, By: cred.Node + "." + cred.Module},
Asks: buildrun.Asker{Publisher: asks, Waiter: asks, Bound: bound},
Asks: buildrun.Asker{Publisher: asks, Waiter: buildrun.LostAskWaiter{Inner: asks, Dead: dead, Hearing: hearing}, Bound: bound},
Held: heldOnTheBus{js},
// An ask that can no longer be answered is asked again (issue 1945), read at the step bound: a context's bound.
Gone: heldOnTheBus{js},
@@ -280,24 +292,25 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
}
// A person asks a key's build again with builder_ask_again, which writes the reasks state (issue 1945): watched, as
// every other input. The builder asks again by itself without it.
if kv := process.OptionalBucket(ctx, js, build.ReAsksBucket, "builder_ask_again is not heard until the builder starts again; an ask that can no longer be answered is still asked again by itself", logger.Printf); kv != nil {
// A state not on the bus yet at this start is watched once it appears (issue 1790).
process.AwaitBucket(ctx, js, conn, build.ReAsksBucket, "builder_ask_again is not heard; an ask that can no longer be answered is still asked again by itself", logger.Printf).Ready(func(kv jetstream.KeyValue) {
eachReAsk := func(e jetstream.KeyValueEntry) { reAsked(ctx, b, kv, e, logger.Printf) }
if rw, err := settle(ctx, kv, eachReAsk); err != nil {
process.NotWatched(logger.Printf, build.ReAsksBucket, err, "builder_ask_again is not heard until the builder starts again")
} else {
go func() { errs <- follow(ctx, rw, eachReAsk) }()
}
}
})
// A person reads a build seat's stream with builder_seat_queue, which writes the queues state (issue 1945 note 62): the
// builder alone may read the stream, so it answers there.
if kv := process.OptionalBucket(ctx, js, build.QueuesBucket, "builder_seat_queue is not answered until the builder starts again", logger.Printf); kv != nil {
process.AwaitBucket(ctx, js, conn, build.QueuesBucket, "builder_seat_queue is not answered", logger.Printf).Ready(func(kv jetstream.KeyValue) {
eachRead := func(e jetstream.KeyValueEntry) { queueRead(ctx, heldOnTheBus{js}, kv, e, logger.Printf) }
if qw, err := settle(ctx, kv, eachRead); err != nil {
process.NotWatched(logger.Printf, build.QueuesBucket, err, "builder_seat_queue is not answered until the builder starts again")
} else {
go func() { errs <- follow(ctx, qw, eachRead) }()
}
}
})
logger.Printf("started on %s", cred.Node)
go func() { errs <- follow(ctx, w, each) }()
select {
+1 -1
View File
@@ -3,7 +3,7 @@ module mesh-delivery-builder
go 1.26.0
require (
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026142434-65ee43f07583
github.com/nats-io/nats-server/v2 v2.11.17
github.com/nats-io/nats.go v1.54.0
)
+2 -4
View File
@@ -1,7 +1,5 @@
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261024020722-34f210e5eba1 h1:fzOkrz9CSp3mEjeV7NfZx8yoivI1sD1AOtIPtyyF8RI=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261024020722-34f210e5eba1/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585 h1:7QoFH5qjXRpx5M4uPdYsVJ5qKl2082gq8n0lFxhoNIg=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026142434-65ee43f07583 h1:ikD/ggf9/5k23sxJOX1PFcGzb4M3+QeC66AT0MyQ9Zs=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026142434-65ee43f07583/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op/go.mod h1:FQyySiasQQM8735Ddel3MRojmy4dA1IqCeyJ5jmPMbI=
github.com/google/go-tpm v0.9.8 h1:slArAR9Ft+1ybZu0lBwpSmpwhRXaa85hWtMinMyRAWo=
@@ -12,6 +12,7 @@ import (
"git.novox.be/novox/mesh-sdk/go/buildstep"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
)
// **An ask that can no longer be answered is asked again** (novox/hq issue 1945). A build seat's stream gives an ask up
@@ -20,7 +21,8 @@ import (
// restart take over the same id and wait again. The builder notices it at the check bound (CheckAfter, the step bound),
// a context's bound and no timer of its own: it asks Gone, which reads the bus, and asks such an ask again under a new
// id, written into the key, said in its log and as the event build.asked-again, at most Askings times; then the key is
// failed with why. A person asks the same by hand with the tool builder_ask_again (AskAgain).
// failed with why. An ask DEAD_LETTERS says the seat gave up (buildrun.LostAskWaiter, issue 734) is asked again at once.
// A person asks the same by hand with the tool builder_ask_again (AskAgain).
// DefaultAskings is how many times a build module's part of a build is asked, the first included, when Builder.Askings is 0.
const DefaultAskings = 3
@@ -315,6 +317,11 @@ func (b *Builder) waitOnce(ctx context.Context, ask delivery.StepAsk, again <-ch
defer stop()
}
step, err := b.Asks.Wait(wait, ask)
if err == nil && buildrun.IsLost(step) {
// The build seat gave the ask up (issue 734): it can no longer be answered, so it is asked again now, with the
// words of why, not at the check bound.
return step, step.Reason, errAgain
}
if err == nil || ctx.Err() != nil {
return step, "", err
}
@@ -690,12 +690,14 @@ func TestABuildAskWaitingInItsStreamIsNotBounded(t *testing.T) {
step string
byHand bool
always bool // Gone says every ask cannot be answered, not only the first
dead bool // DEAD_LETTERS keeps the first ask: the seat gave it up (issue 734)
puts int
done bool
asking int // the asking the event says, 0 when none is said
}{
{name: "given up by its stream: asked again under a new id and done from it", puts: 2, done: true, asking: 2},
{name: "asked again by hand: the same, with no check bound", byHand: true, puts: 2, done: true, asking: 2},
{name: "given up into DEAD_LETTERS: its wait ends failed with the words and it is asked again at once", dead: true, puts: 2, done: true, asking: 2},
{name: "never answerable: failed after its askings", askings: 2, always: true, puts: 2},
{name: "taken over at its last asking: failed, not asked again", askings: 3, always: true, step: "asked mesh-build-go again (asking 3): x", puts: 0},
} {
@@ -714,7 +716,11 @@ func TestABuildAskWaitingInItsStreamIsNotBounded(t *testing.T) {
r.st.vals[r.key] = data
}
gone := &goneAsks{}
if !c.byHand {
dead := &buildrun.DeadLetters{}
if c.dead {
r.b.Asks = buildrun.Asker{Publisher: r.asks, Waiter: buildrun.LostAskWaiter{Inner: r.asks, Dead: dead}, Bound: 2 * time.Second}
}
if !c.byHand && !c.dead {
gone.why = func(a delivery.StepAsk) string {
if c.always || a.ID == first || first == "" {
return "it waits first in line while a node took a later ask"
@@ -724,6 +730,10 @@ func TestABuildAskWaitingInItsStreamIsNotBounded(t *testing.T) {
r.b.Gone, r.b.CheckAfter = gone, 5*time.Millisecond
}
r.b.Module(context.Background(), "m", moduleValue(t, "modules/m"))
if c.dead {
dead.Heard(buildrun.GivenUp{AskID: waitForAsks(t, r, 1)[0].ID, Stream: "SEAT_NODE_BUILD_GO", Deliveries: 5,
At: time.Date(2026, 10, 26, 12, 26, 40, 0, time.UTC)})
}
if c.byHand {
waitForAsks(t, r, 1)
for deadline := time.Now().Add(2 * time.Second); len(r.b.AskAgain(r.key, "stuck since 19:06")) == 0; time.Sleep(time.Millisecond) {
@@ -757,13 +767,16 @@ func TestABuildAskWaitingInItsStreamIsNotBounded(t *testing.T) {
mu.Lock()
defer mu.Unlock()
// Note 62: the old message of an ask asked again because its stream gave it up is dropped from the stream.
if gone.mu.Lock(); c.done && !c.byHand && (len(gone.dropped) != 1 || gone.dropped[0] != first) {
if gone.mu.Lock(); c.done && !c.byHand && !c.dead && (len(gone.dropped) != 1 || gone.dropped[0] != first) {
t.Errorf("dropped %v from the stream, want the old ask %s", gone.dropped, first)
}
gone.mu.Unlock()
if c.asking > 0 && (len(said) != 1 || said[0].Asking != c.asking || said[0].New == said[0].Old || said[0].Key != r.key || said[0].Why == "") {
t.Errorf("said %+v, want once at asking %d", said, c.asking)
}
if c.dead && (len(said) != 1 || !strings.Contains(said[0].Why, buildrun.LostWords+" after 5 deliveries at 2026-10-26T12:26:40Z")) {
t.Errorf("said %+v, want the ask asked again saying the seat gave it up", said)
}
})
}
t.Run("what the bus says of an ask", func(t *testing.T) { // 1945
@@ -836,12 +849,17 @@ func TestABuildAskWaitingInItsStreamIsNotBounded(t *testing.T) {
running, _ := json.Marshal(delivery.Step{V: delivery.Version, ID: "mvo53jcm70b5f8", Tool: "delivery ask", Node: "g14", State: delivery.StepRunning, Started: at})
_, reask, _ := NewReAsk("m", fp1, "stuck", "shanks.lead", at)
r := StuckReader{Builds: state(map[string]string{key: string(building), done: string(finished)}), ReAsks: state(map[string]string{key: string(reask)}),
Steps: map[string]StateRead{buildrun.GoBuildModule: state(map[string]string{}), ImageBuildModule: state(map[string]string{"g14.mvo53jcm70b5f8": string(running)})},
Now: func() time.Time { return at.Add(time.Hour) }}
Steps: map[string]StateRead{buildrun.GoBuildModule: state(map[string]string{}), ImageBuildModule: state(map[string]string{"g14.mvo53jcm70b5f8": string(running)})},
Hearing: state(map[string]string{buildrun.HearingKey: `{"heard":false,"streams":["SEAT_NODE_BUILD_GO"],"why":"DEAD_LETTERS is not heard: SEAT_NODE_BUILD_GO: permissions violation","at":"2026-10-26T13:00:00Z"}`}),
Now: func() time.Time { return at.Add(time.Hour) }}
out, err := r.Stuck("")
if err != nil {
t.Fatal(err)
}
// Issue 734, review round 1: a builder that does not hear DEAD_LETTERS says so, with why.
if dl := out["dead_letters"].(map[string]any); dl["heard"] != false || !strings.Contains(dl["why"].(string), "permissions violation") {
t.Errorf("dead_letters = %v, want not heard, saying why", dl)
}
stuck := out["stuck"].([]Stuck)
if len(stuck) != 1 || stuck[0].Key != key || stuck[0].Age != "1h0m0s" || stuck[0].NoneStep != 1 || stuck[0].ReAsk == nil || stuck[0].ReAsk.Why != "stuck" {
t.Fatalf("stuck = %+v, want the building key alone, an hour old, one ask with no step, with its ask again", stuck)
@@ -106,7 +106,9 @@ type StuckReader struct {
ReAsks StateRead
// Steps are the build modules' steps states, by build module.
Steps map[string]StateRead
Now func() time.Time
// Hearing is the builder's state hearing: whether it hears DEAD_LETTERS (issue 734); without it none is said.
Hearing StateRead
Now func() time.Time
// Logf, when set, gets the one line a call logs: the values it read and the time it took.
Logf func(format string, args ...any)
}
@@ -149,6 +151,20 @@ func parallel(n int, read func(i int) error) error {
}
// Stuck is every key building or queued with its asks, oldest first; with module, that module's only.
// deadLetters says whether the builder hears the asks the build seats give up on (issue 734), and why not: deaf, a lane
// whose ask is given up waits for its check bound (buildrun.HearingHealth over the kept hearing).
func (r StuckReader) deadLetters() map[string]any {
if r.Hearing.Get == nil {
return map[string]any{"heard": false, "why": "the builder's state hearing is not readable here"}
}
raw, _, err := r.Hearing.Get(buildrun.HearingKey)
if err != nil {
return map[string]any{"heard": false, "why": "the builder's state hearing cannot be read: " + err.Error()}
}
a := buildrun.HearingHealth(raw)
return map[string]any{"heard": a.Healthy, "why": a.Why}
}
func (r StuckReader) Stuck(module string) (map[string]any, error) {
started := time.Now()
reads := 0
@@ -268,6 +284,7 @@ func (r StuckReader) Stuck(module string) (map[string]any, error) {
})
answer := map[string]any{"stuck": out, "count": len(out),
"delivery_counts": "not readable by the builder: the build seats' workers' information is granted to the delivery view alone (its backlog); an ask given up on is told by the builder at its check bound and asked again"}
answer["dead_letters"] = r.deadLetters()
if cut {
answer["cut"], answer["note"] = true, fmt.Sprintf("read %d of %d build keys in key order and stopped; give a module to read its keys alone", MaxReads, total)
}
@@ -87,7 +87,7 @@ func TestManifestDeclarations(t *testing.T) { // 1607/AC1
if m.Module != "mesh-delivery-builder" {
t.Errorf("module = %q", m.Module)
}
if want := []state{{Name: "builds", History: 5}, {Name: "instances", History: 1}, {Name: "queues", History: 1}, {Name: "reasks", History: 1}}; !reflect.DeepEqual(m.State, want) { // delivery.BuilderBuildsHistory
if want := []state{{Name: "builds", History: 5}, {Name: "hearing", History: 1}, {Name: "instances", History: 1}, {Name: "queues", History: 1}, {Name: "reasks", History: 1}}; !reflect.DeepEqual(m.State, want) { // delivery.BuilderBuildsHistory
t.Errorf("state = %+v, want %+v", m.State, want)
}
// 1887/A4c: the builder chooses no node, so it reads neither presence nor have, invokes no build module's tool and
@@ -20,6 +20,7 @@
],
"state": [
{ "name": "builds", "history": 5 },
{ "name": "hearing", "history": 1 },
{ "name": "instances", "history": 1 },
{ "name": "queues", "history": 1 },
{ "name": "reasks", "history": 1 }
@@ -12,6 +12,7 @@ import (
stdio "git.novox.be/novox/mesh-sdk/go"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
"mesh-delivery-checker/internal/check"
)
@@ -53,10 +54,36 @@ func read(s stdio.KeptState) check.StateRead {
}}
}
// runsOf says which commit the mesh runs of a module from the indexer's modules state, as the checker's own record of it does
// (the commit the module's record was computed at); "" when the record is not readable.
type runsOf struct{ modules check.StateRead }
func (r runsOf) CommitOf(module string) string {
if r.modules.Get == nil {
return ""
}
raw, err := r.modules.Get(module)
if err != nil || len(raw) == 0 {
return ""
}
m, err := delivery.DecodeModule(raw)
if err != nil {
return ""
}
return m.Commit
}
// stateStore is a state the tools write: the tries.
type stateStore struct {
check.StateRead
create func(key string, value []byte) error
entry func(key string) (json.RawMessage, uint64, error)
update func(key string, value []byte, revision uint64) error
}
func (s stateStore) Entry(key string) (json.RawMessage, uint64, error) { return s.entry(key) }
func (s stateStore) Update(key string, value []byte, revision uint64) error {
return s.update(key, value, revision)
}
func (s stateStore) Keys() ([]string, error) { return s.StateRead.Keys() }
@@ -66,19 +93,21 @@ func (s stateStore) Create(key string, value []byte) error { return s.create(k
// Trying is what the try's tools work on: the tries, and the gate's trunk heads.
type Trying struct {
Tries check.TryStore
Ender check.TryEnder
Trunks func() (map[string]string, error)
Now func() time.Time
}
// TryTools are the try's two tools (ADR 0501 §5): ask one, read one. mesh-controller's try calls them.
// TryTools are the try's tools (ADR 0501 §5): ask one, read one, and end one by hand (issue 734). mesh-controller's try calls them.
func TryTools(tr Trying) []stdio.Tool {
str := func(args map[string]any, k string) string { s, _ := args[k].(string); return s }
return []stdio.Tool{
{Name: check.TryTool,
Description: "Try a pushed head before its pull request is opened: the checker judges it as it judges a pull request's head, merged onto its trunk's head, " +
"its unit and integration levels through the build modules, and owes the forge nothing. Answers the try's id at once; " + check.TryReadTool + " reads " +
"where it stands. A try of the same head that waits or runs is the answer to a second ask; at most 4 wait or run at once, mesh-wide. A repository " +
"the checker checks no level of (one whose trunk the gate does not name) is refused.",
"where it stands. A try of the same head that waits or runs is the answer to a second ask; a new try is refused while 4 tries run on a build node or 8 wait for one, mesh-wide; tries let in while waiting may later all run at once, and each node's build capacity bounds what runs on it. A repository " +
"the checker checks no level of (one whose trunk the gate does not name) is refused. A try whose ask no build node took within the step " +
"bound ends as error; its ask stays in the build seat's stream and may still run later, judged by nobody (issue 734).",
Input: map[string]any{"type": "object", "required": []string{"repository", "head"}, "properties": map[string]any{
"repository": map[string]any{"type": "string", "description": "<owner>/<name>"},
"head": map[string]any{"type": "string", "description": "the full commit to try, 40 or 64 hex digits"},
@@ -106,7 +135,8 @@ func TryTools(tr Trying) []stdio.Tool {
}},
{Name: check.TryReadTool,
Description: "Where a try stands, as stored: queued, running, or its end (pass, fail, or error when it could not run), with what was said of it, " +
"the head and the trunk commit it was merged onto, the time it took and what failed first. Read-only.",
"the head and the trunk commit it was merged onto, the time it took and what failed first; while it runs, each ask it put as soon as it is put, " +
"with the node that took it, its state and the phase its step says (fetch and merge, a level's digest, lint, tests) and since when. Read-only.",
Input: map[string]any{"type": "object", "required": []string{"id"}, "properties": map[string]any{
"id": map[string]any{"type": "string", "description": "the try's id, try-<digits>"}}},
Run: func(args map[string]any) (any, error) {
@@ -115,6 +145,28 @@ func TryTools(tr Trying) []stdio.Tool {
}
return check.ReadTry(tr.Tries, str(args, "id"))
}},
{Name: check.TryEndTool,
Description: "End a try that waits or runs, as error, by hand: say why (kept in what it says). Its place among the 4 is free at once, " +
"the checker stops waiting for its asks, and the next " + check.TryTool + " of its head asks anew with a new id. For a try whose ask " +
"was lost (no step runs for it in checker_status); the checker ends one by itself when the build seat gives its ask up, or no node " +
"takes it within the step bound. A try that ended already is refused.",
Input: map[string]any{"type": "object", "required": []string{"id", "why"}, "properties": map[string]any{
"id": map[string]any{"type": "string", "description": "the try's id, try-<digits>"},
"why": map[string]any{"type": "string", "description": "why it is ended, in words"}}},
Run: func(args map[string]any) (any, error) {
if tr.Ender == nil {
return nil, fmt.Errorf("the tries state cannot be written here")
}
now := time.Now
if tr.Now != nil {
now = tr.Now
}
t, err := check.EndTry(tr.Ender, str(args, "id"), str(args, "why"), now())
if err != nil {
return nil, err
}
return map[string]any{"id": t.ID, "state": t.State, "said": t.Said}, nil
}},
}
}
@@ -174,7 +226,10 @@ func Tools(r check.Reader) []stdio.Tool {
"and the candidates whose check is owed by kind (waiting_by_kind: new, nodes, and trunk, a re-judge pulled after the other two). A key a " +
"build module left running (started past the step bound, or its build module not running on its node) is listed apart under " +
"left_running with why, and not counted. Nothing holds the checker and it has no capacity of its own: a check is an ask on the " +
"build seats' streams, whose backlog the delivery view shows (novox/hq ADR 0525). Read-only.",
"build seats' streams, whose backlog the delivery view shows (novox/hq ADR 0525). held_by_turn names the candidates whose check an approved " +
"merge's turn holds off (novox/hq ADR 0548 §3), with the repository and the merge that waits; the gate's turns tool reads every turn and its " +
"hand-now tool steers one. dead_letters says whether the checker hears the asks the build seats give up on (issue 734), and why not: " +
"deaf, a try or check whose ask is given up waits until its bound. Read-only.",
Input: map[string]any{"type": "object", "properties": map[string]any{}},
Run: func(map[string]any) (any, error) { return r.Status() }},
}
@@ -193,7 +248,9 @@ func main() {
r := check.Reader{VerdictState: read(stdio.State("verdicts")), KeptState: read(stdio.State("kept")), CandidateState: read(stdio.State("mesh-delivery-indexer.candidates")),
PreviewState: read(stdio.State("previews")),
StepStates: map[string]check.StateRead{"mesh-build-go": read(stdio.State("mesh-build-go.steps")), "mesh-build-node": read(stdio.State("mesh-build-node.steps"))},
HaveState: read(stdio.State("node-engine.have")), StepBound: stepBound()}
HaveState: read(stdio.State("node-engine.have")), StepBound: stepBound(),
TurnState: read(stdio.State("mesh-delivery-gate.turns")), HearingState: read(stdio.State(buildrun.HearingState))}
r.Runs = runsOf{read(stdio.State("mesh-delivery-indexer.modules"))} // the node-engine at the commit the mesh runs, as the checker reads it
tries := stdio.State(check.TriesState)
source := stdio.State("mesh-delivery-gate.source")
trunks := func() (map[string]string, error) {
@@ -205,13 +262,21 @@ func main() {
return s.Heads, err
}
r.Trunks = trunks
trying := Trying{
Tries: stateStore{StateRead: read(tries), create: func(key string, value []byte) error {
_, err := tries.Create(key, json.RawMessage(value))
return err
}},
Trunks: trunks,
}
store := stateStore{StateRead: read(tries), create: func(key string, value []byte) error {
_, err := tries.Create(key, json.RawMessage(value))
return err
}, entry: func(key string) (json.RawMessage, uint64, error) {
reads.Add(1)
e, err := tries.Get(key)
if err != nil || e == nil {
return nil, 0, err
}
return e.Value, e.Revision, nil
}, update: func(key string, value []byte, revision uint64) error {
_, err := tries.Update(key, json.RawMessage(value), revision)
return err
}}
trying := Trying{Tries: store, Ender: store, Trunks: trunks}
if err := stdio.Serve("", logged(append(Tools(r), TryTools(trying)...))); err != nil {
fmt.Fprintf(os.Stderr, "[mesh-delivery-checker-tools] %v\n", err)
os.Exit(1)
@@ -53,13 +53,19 @@ func (m *memStore) Create(key string, value []byte) error {
return nil
}
func (m *memStore) Entry(key string) (json.RawMessage, uint64, error) { return m.vals[key], 1, nil }
func (m *memStore) Update(key string, value []byte, revision uint64) error {
m.vals[key] = value
return nil
}
func TestTheTryToolsAskAndReadATry(t *testing.T) { // ADR 0501 §5, the agreed interface with mesh-controller's try
store := &memStore{vals: map[string]json.RawMessage{}}
tools := TryTools(Trying{Tries: store, Now: func() time.Time { return time.Unix(1, 5) },
tools := TryTools(Trying{Tries: store, Ender: store, Now: func() time.Time { return time.Unix(1, 5) },
Trunks: func() (map[string]string, error) {
return map[string]string{"novox/mesh-delivery": strings.Repeat("a", 40)}, nil
}})
if len(tools) != 2 || tools[0].Name != "checker_try" || tools[1].Name != "checker_try_read" {
if len(tools) != 3 || tools[0].Name != "checker_try" || tools[1].Name != "checker_try_read" || tools[2].Name != "checker_try_end" {
t.Fatalf("tools = %+v", tools)
}
out, err := tools[0].Run(map[string]any{"repository": "novox/mesh-delivery", "head": strings.Repeat("1", 40), "trunk": "main"})
@@ -74,6 +80,14 @@ func TestTheTryToolsAskAndReadATry(t *testing.T) { // ADR 0501 §5, the agreed i
if err != nil || !strings.Contains(string(raw.(json.RawMessage)), `"state":"queued"`) {
t.Errorf("read = %s, %v", raw, err)
}
// Issue 734: a try whose ask was lost is ended by hand with why, and its head is asked anew.
ended, err := tools[2].Run(map[string]any{"id": id, "why": "its ask was lost"})
if err != nil || ended.(map[string]any)["state"] != "error" {
t.Errorf("end = %v, %v", ended, err)
}
if _, err := tools[2].Run(map[string]any{"id": id, "why": "again"}); err == nil {
t.Error("an ended try was ended again")
}
if _, err := tools[0].Run(map[string]any{"repository": "novox/hq", "head": strings.Repeat("1", 40)}); err == nil ||
err.Error() != "no level of novox/hq is checked by the checker" {
t.Errorf("hq: %v", err)
@@ -24,6 +24,7 @@ import (
"git.novox.be/novox/mesh-sdk/go/announce"
"git.novox.be/novox/mesh-sdk/go/buildstep"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
"mesh-delivery-checker/internal/check"
)
@@ -79,6 +80,8 @@ type fakeBuildModules struct {
js jetstream.JetStream
ctx context.Context
conflict atomic.Bool
// hold is a commit whose fetch never answers: its ask stays taken, as one lost in a bus reload (issue 734).
hold atomic.Value
mu sync.Mutex
calls map[string]int // by `<module>.<tool>`
testTool *testing.T
@@ -98,6 +101,9 @@ func (f *fakeBuildModules) result(t *testing.T, tool string, args map[string]any
var v any
switch {
case strings.HasSuffix(tool, "_fetch"):
if h, _ := f.hold.Load().(string); h != "" && args["commit"] == h {
<-f.ctx.Done()
}
name := "go-fetch-clean"
if f.conflict.Load() {
name = "go-fetch-conflict"
@@ -298,6 +304,10 @@ func startBed(t *testing.T) *bed {
}
b.tools[module] = kv
}
// What the controller keeps of the messages a consumer gave up on (issue 734).
if _, err := js.CreateStream(ctx, jetstream.StreamConfig{Name: buildrun.DeadLettersStream, Subjects: []string{"mesh.events.dead.>"}}); err != nil {
t.Fatal(err)
}
// The checker's stream of events, which the indexer and the forge bridge hear.
if _, err := js.CreateStream(ctx, jetstream.StreamConfig{Name: "CHECKS", Subjects: []string{"mesh.stream.mesh-delivery-checker.event.>"}}); err != nil {
t.Fatal(err)
@@ -305,7 +315,8 @@ func startBed(t *testing.T) *bed {
for name, history := range map[string]int{delivery.CheckerVerdictsBucket: delivery.CheckerVerdictsHistory, delivery.CheckerKeptBucket: delivery.CheckerKeptHistory,
delivery.CheckerOwedBucket: delivery.CheckerOwedHistory, delivery.PreviewsBucket: delivery.PreviewsHistory, delivery.CandidatesBucket: delivery.CandidatesHistory,
delivery.ModulesBucket: delivery.ModulesHistory, delivery.ManifestsBucket: delivery.ManifestsHistory, delivery.SourceBucket: delivery.SourceHistory,
delivery.PresenceBucket: delivery.PresenceHistory, delivery.HaveBucket: delivery.HaveHistory, check.TriesBucket: 1} {
delivery.PresenceBucket: delivery.PresenceHistory, delivery.HaveBucket: delivery.HaveHistory, check.TriesBucket: 1,
delivery.TurnsBucket: delivery.TurnsHistory} {
kv, err := js.CreateKeyValue(ctx, jetstream.KeyValueConfig{Bucket: name, History: uint8(history)})
if err != nil {
t.Fatal(err)
@@ -416,7 +427,8 @@ func (b *bed) start(t *testing.T, env map[string]string) {
if err := os.WriteFile(broker, cred, 0o600); err != nil {
t.Fatal(err)
}
e := map[string]string{"MESH_BROKER_FILE": broker}
// The settle window of a group's heads (check.DefaultSettle) is cut short: each bed's candidate is written whole.
e := map[string]string{"MESH_BROKER_FILE": broker, "MESH_CHECKER_SETTLE": "10ms"}
for k, v := range env {
e[k] = v
}
@@ -507,9 +519,12 @@ func TestCheckerJudgesACandidateAndOwesItsStatus(t *testing.T) { // N6/AC5, 1300
}
}
// TestACheckOfEqualInputsRunsNothing: a second candidate (another head, the same digests) is checked with every package
// taken from the first's kept verdicts: no test step runs and no second kept key appears.
func TestACheckOfEqualInputsRunsNothing(t *testing.T) { // 1287/AC5, 1288/AC6, 1929/AC2
// TestACheckOrACarryOfEqualInputsRunsNothing: a second candidate (another head, the same digests) is checked with every
// package taken from the first's kept verdicts: no test step runs and no second kept key appears. Then, through the real bus
// (novox/hq ADR 0548 §3, §5): a third candidate is not claimed while the gate's merge turn on its repository stands, and is
// claimed when the turn's key is deleted; and a move of the trunk carries the first verdict on the recheck stream by its
// unchanged digests, running no test, the verdict naming the trunk it was carried from.
func TestACheckOrACarryOfEqualInputsRunsNothing(t *testing.T) { // 1287/AC5, 1288/AC6, 1929/AC2, ADR 0548 §3, §5
b := startBed(t)
first := b.candidate(t, "feat/x", pull(repo, 7, head1))
b.start(t, nil)
@@ -526,6 +541,42 @@ func TestACheckOfEqualInputsRunsNothing(t *testing.T) { // 1287/AC5, 1288/AC6, 1
if got := keptCount(t, b); got != kept {
t.Errorf("%d kept keys, was %d", got, kept)
}
// §3: the gate's turn on the repository holds a new check off; its delete claims it.
at := time.Now().UTC()
turn, err := delivery.EncodeTurn(delivery.Turn{Repository: repo, Waiting: []delivery.TurnWait{{Group: "fix/other", Heads: "0123456789ab", Approved: at, Since: at}}})
if err != nil {
t.Fatal(err)
}
tk, _ := delivery.TurnsKey(repo)
if _, err := b.buckets[delivery.TurnsBucket].Put(b.ctx, tk, turn); err != nil {
t.Fatal(err)
}
time.Sleep(300 * time.Millisecond) // the checker hears the turn
third := b.candidate(t, "feat/z", pull(repo, 9, "3333333333333333333333333333333333333333"))
time.Sleep(time.Second)
if _, err := b.buckets[delivery.CheckerVerdictsBucket].Get(b.ctx, third); !errors.Is(err, jetstream.ErrKeyNotFound) {
t.Fatalf("verdicts.%s exists (%v) while a merge turn holds its repository", third, err)
}
if err := b.buckets[delivery.TurnsBucket].Delete(b.ctx, tk); err != nil {
t.Fatal(err)
}
b.verdict(t, third, ended)
// §5: the trunk moves; the first verdict is carried by its digests, nothing runs.
tests = b.fake.count("mesh-build-go.build_go_test")
const moved = "abababababababababababababababababababab"
b.setTrunk(t, map[string]string{repo: moved})
c := b.verdict(t, first, func(v delivery.Verdict) bool { return ended(v) && v.Trunk[repo] == moved })
if c.State != delivery.VerdictPass || c.CarriedFrom[repo] != trunk || c.Levels[delivery.LevelUnit].Run != 0 || c.Levels[delivery.LevelIntegration].Run != 0 {
t.Fatalf("the carried verdict = %+v, want a pass carried from %s with no test run", c, trunk[:12])
}
if got := b.fake.count("mesh-build-go.build_go_test"); got != tests {
t.Errorf("build_go_test called %d times on the carry, was %d: a carry of unchanged digests runs nothing", got, tests)
}
if o := b.owed(t, 7, head1, "success"); !strings.HasPrefix(o.Text, "pass 111111111111 on abababababab | carried from aaaaaaaaaaaa") {
t.Errorf("the carried status says %q", o.Text)
}
}
func keptCount(t *testing.T, b *bed) int {
@@ -676,6 +727,49 @@ func TestATryIsJudgedAndItsEndWrittenWithNoStatusOwed(t *testing.T) {
t.Errorf("a try wrote %s: %v", name, keys)
}
}
// Issue 734: a try whose ask the build seat gives up on (the controller keeps its copy in DEAD_LETTERS with the ask's
// message id) ends as error saying so, and its head asked again is a new try.
b.fake.hold.Store(head2)
lost, err := check.AskTry(store, map[string]string{repo: trunk}, check.TryAsk{Repository: repo, Head: head2}, time.Now())
if err != nil {
t.Fatal(err)
}
steps, err := b.js.KeyValue(b.ctx, delivery.StepsBucket("mesh-build-go"))
if err != nil {
t.Fatal(err)
}
askID := ""
for deadline := time.Now().Add(20 * time.Second); askID == "" && time.Now().Before(deadline); time.Sleep(50 * time.Millisecond) {
keys, _ := steps.Keys(b.ctx)
for _, k := range keys {
if e, err := steps.Get(b.ctx, k); err == nil && strings.Contains(string(e.Value()), `"running"`) {
askID = strings.TrimPrefix(k, "n1.")
}
}
}
if askID == "" {
t.Fatal("the held try's ask was never taken")
}
dead := &nats.Msg{Subject: "mesh.events.dead.SEAT_NODE_BUILD_GO.SEAT_NODE_BUILD_GO_CHECK_worker", Header: nats.Header{}}
dead.Header.Set(buildrun.DeadHeaderMsgID, askID)
dead.Header.Set(buildrun.DeadHeaderDeliveries, "5")
dead.Header.Set(buildrun.DeadHeaderStream, "SEAT_NODE_BUILD_GO")
dead.Header.Set(buildrun.DeadHeaderGaveUp, time.Now().UTC().Format(time.RFC3339Nano))
if _, err := b.js.PublishMsg(b.ctx, dead); err != nil {
t.Fatal(err)
}
for deadline := time.Now().Add(20 * time.Second); time.Now().Before(deadline); time.Sleep(50 * time.Millisecond) {
raw, _ := store.Get(lost)
if last, err = check.DecodeTry(raw); err == nil && last.Over() {
break
}
}
if last.State != check.TryError || !strings.Contains(last.Said, buildrun.LostWords+" after 5 deliveries") {
t.Fatalf("the try whose ask was given up = %+v, want error saying so", last)
}
if again, err := check.AskTry(store, map[string]string{repo: trunk}, check.TryAsk{Repository: repo, Head: head2}, time.Now()); err != nil || again == lost {
t.Errorf("its head asked again answered %q, %v, want a new try", again, err)
}
}
// TestTheCheckerSaysOwedReadableOnTheBus: started, the checker says on its own states subject what it keeps, owed among
@@ -702,7 +796,7 @@ func TestTheCheckerSaysOwedReadableOnTheBus(t *testing.T) { // R2-1
break
}
if said.Module != "mesh-delivery-checker" || said.Node != "n1" || !said.Keeps("owed") || !said.Readable("owed") ||
!said.Readable("previews") || !said.Readable("verdicts") || !said.Readable("kept") || said.Readable("tries") || len(said.States) != 5 {
t.Fatalf("the checker said %+v, want its five states with kept, owed, previews and verdicts readable (1929/AC4; 1887 A5b)", said)
!said.Readable("previews") || !said.Readable("verdicts") || !said.Readable("kept") || !said.Readable("tries") || said.Readable("hearing") || len(said.States) != 6 {
t.Fatalf("the checker said %+v, want its six states with kept, owed, previews, tries and verdicts readable, tries by the delivery view (1929/AC4; 1887 A5b; build_usage)", said)
}
}
@@ -37,8 +37,8 @@ func TestTheCheckerAnnouncesEveryStateItKeepsAndOwedReadable(t *testing.T) { //
if !equal(manifest, announced) {
t.Errorf("announced %v, the manifest keeps %v", announced, manifest)
}
// kept is read by the Go build module for a check ask (issue 1887 design limit 1).
if want := []string{"kept", "owed", "previews", "verdicts"}; !equal(readable, want) {
// kept is read by the Go build module for a check ask (issue 1887 design limit 1); tries by the delivery view (build_usage).
if want := []string{"kept", "owed", "previews", "tries", "verdicts"}; !equal(readable, want) {
t.Errorf("readable %v, want %v", readable, want)
}
}
@@ -0,0 +1,50 @@
package main
import (
"sync"
"git.novox.be/novox/mesh-sdk/go/delivery"
"mesh-delivery-checker/internal/check"
)
// lazyTurns is the gate's merge turns as the checker reads them (novox/hq ADR 0548 §3), opened when the gate's bucket is
// there: before each claim it is opened again until it is, so a checker that started before the gate made the bucket holds
// checks from the gate's first turn on, with no timer and no restart. Until then it holds nothing, and says so once, loudly,
// and once again when it opens.
type lazyTurns struct {
turns *check.Turns
open func() error
logf func(format string, args ...any)
mu sync.Mutex
opened bool
said bool
}
// ensure opens the bucket once it is there; a failure is said once.
func (l *lazyTurns) ensure() {
l.mu.Lock()
defer l.mu.Unlock()
if l.opened || l.open == nil {
return
}
if err := l.open(); err != nil {
if !l.said {
l.said = true
l.logf("NO MERGE TURN HOLDS A CHECK: the gate's bucket %s cannot be read (%v); every check is claimed as if no approved merge waited, "+
"and the bucket is opened again before each claim until it is there", delivery.TurnsBucket, err)
}
return
}
l.opened = true
if l.said {
l.logf("the gate's merge turns are read from %s from now on: an approved merge's turn holds checks off its repositories again", delivery.TurnsBucket)
}
}
// Holding opens the bucket when it is not yet, then answers what the turns hold.
func (l *lazyTurns) Holding(repositories []string, heads string) (string, delivery.TurnWait, bool) {
l.ensure()
return l.turns.Holding(repositories, heads)
}
@@ -112,14 +112,17 @@ func durationEnv(getenv func(string) string, name string, def time.Duration) (ti
// run is the program: it returns when ctx ends or the bus cannot be used.
// keptStates is every state the checker keeps and which others may read (ADR 0495 §2-§3): owed by the forge module,
// which writes mesh/check from it (ADR 0521), and by the forge bridge; previews by the indexer; verdicts by the gate.
// which writes mesh/check from it (ADR 0521), and by the forge bridge; previews by the indexer; verdicts by the gate; tries by the delivery view.
var keptStates = []announce.State{
// The kept verdicts are read by the Go build module, which takes them for a check ask's unchanged packages (issue 1887's
// design limit 1); the checker alone writes them (ADR 0444 clause 5).
// Whether DEAD_LETTERS is heard (issue 734): read by the checker's own status tool alone.
{Name: "hearing"},
{Name: "kept", Readable: true},
{Name: "owed", Readable: true},
{Name: "previews", Readable: true},
{Name: "tries"},
// The tries are read by the delivery view, which tells a try's asks from a candidate's by the asks each names (build_usage).
{Name: "tries", Readable: true},
{Name: "verdicts", Readable: true},
}
@@ -186,6 +189,10 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
if err != nil {
return err
}
settleWindow, err := durationEnv(getenv, "MESH_CHECKER_SETTLE", check.DefaultSettle)
if err != nil {
return err
}
link := strings.TrimSpace(getenv("MESH_CHECKER_LINK"))
if link == "" {
link = defaultLink
@@ -193,16 +200,32 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
tools := &buildrun.ToolchainSet{}
presence := &buildrun.PresenceSet{}
trunks := &check.Trunks{}
turns := &check.Turns{}
lazy := &lazyTurns{turns: turns, logf: logger.Printf}
// A check is one ask per pull request and build module on the build seats' check stream, pulled by a build node with a
// free slot (novox/hq ADR 0525 §1, §3): the checker chooses no node and starts no step.
asks := buildrun.JetStreamAsks{JS: js}
// The asks the build seats gave up on (issue 734): heard from DEAD_LETTERS, each ends the step its asker waits for.
// Heard in the background, and again at a reconnect or an ask while it is not; whether it is heard is kept in the
// state hearing, which checker_status says (dead_letters).
dead := &buildrun.DeadLetters{}
// The state may not be on the bus yet when a delivery started the checker (the grants part creates it, issue 1790): it
// is bound when the bus says it was created, and the latest hearing is written then.
hearingKV := process.AwaitBucket(ctx, js, conn, check.HearingBucket, "checker_status cannot say whether DEAD_LETTERS is heard", logger.Printf)
hearing := buildrun.NewDeadLetterHearing(ctx, buildrun.JetStreamDeadLetters{JS: js}, []string{delivery.AskGo, delivery.AskNode}, dead,
logger.Printf, buildrun.SayHearing(ctx, hearingKV, logger.Printf))
defer hearing.Stop()
conn.SetReconnectHandler(func(*nats.Conn) { hearing.Again() })
waiter := buildrun.LostAskWaiter{Inner: asks, Dead: dead, Hearing: hearing}
ch := &check.Checker{
Verdicts: kvStore{buckets[delivery.CheckerVerdictsBucket]},
Owed: kvStore{buckets[delivery.CheckerOwedBucket]},
Trunks: trunks,
Turns: lazy,
Bound: checkBound,
Link: link,
Logf: logger.Printf,
Settle: settleWindow,
}
// The builder's builds say how the candidates' fingerprints stand (issue 1684); where its bucket is not there yet the line
// says nothing of the builds, until the checker starts again.
@@ -228,7 +251,7 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
return cw, nil
}}
// The tries (ADR 0501 §5) are judged by the same judge, held by the same deliveries.
tryer := &check.Tryer{Tries: kvStore{buckets[check.TriesBucket]}, Bound: checkBound, Logf: logger.Printf}
tryer := &check.Tryer{Tries: kvStore{buckets[check.TriesBucket]}, Bound: checkBound, NotTaken: stepBound, Logf: logger.Printf}
announce := func(ctx context.Context, event, key string) {
body, _ := json.Marshal(map[string]string{"key": key})
if _, err := js.Publish(ctx, eventSubject(event), body); err != nil {
@@ -236,8 +259,9 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
}
}
repos := &repoModules{manifests: kvStore{buckets[delivery.ManifestsBucket]}, logf: logger.Printf}
ch.Runs = repos // the node-engine a check reads beside at its running commit (ADR 0548 §6)
ch.Judge = newJudge(judgeDeps{
asks: buildrun.Asker{Publisher: asks, Waiter: asks, Bound: stepBound}, tools: tools, trunks: trunks,
asks: buildrun.Asker{Publisher: asks, Waiter: waiter, Bound: stepBound}, tools: tools, trunks: trunks,
kept: kvStore{buckets[delivery.CheckerKeptBucket]}, previews: kvStore{buckets[delivery.PreviewsBucket]},
repos: repos, builds: builds, forge: forgeReads{call: caller.Call}, logf: logger.Printf,
announce: announce, compositions: compositions,
@@ -313,13 +337,42 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
}()
}
}
// The gate's merge turns (novox/hq ADR 0548 §3): no check is claimed of a repository an approved merge waits on, and the
// delete of its turn claims it. Read before the candidates, so no check starts past a turn written before this start.
// Where the gate's bucket is not there yet it is read as empty, and opened again before each claim until it is there
// (lazyTurns), so a gate delivered after the checker started holds checks from its first turn on.
lazy.open = func() error {
kv, err := js.KeyValue(ctx, delivery.TurnsBucket)
if err != nil {
return err
}
tw, err := settle(ctx, kv, func(e jetstream.KeyValueEntry) { turns.Put(e.Key(), e.Value(), !opOf(e)) })
if err != nil {
return err
}
go func() {
errs <- follow(ctx, tw, func(e jetstream.KeyValueEntry) {
if turns.Put(e.Key(), e.Value(), !opOf(e)) {
ch.Reconsider(ctx) // a turn went or changed: a check it held is claimed now
}
})
}()
return nil
}
lazy.ensure()
// The modules say which repository holds a TypeScript module, so the Node build module judges it too.
mw, err := settle(ctx, buckets[delivery.ModulesBucket], func(e jetstream.KeyValueEntry) { repos.Put(ctx, e.Key(), e.Value(), !opOf(e)) })
if err != nil {
return err
}
go func() {
errs <- follow(ctx, mw, func(e jetstream.KeyValueEntry) { repos.Put(ctx, e.Key(), e.Value(), !opOf(e)) })
errs <- follow(ctx, mw, func(e jetstream.KeyValueEntry) {
before := repos.CommitOf(check.NodeEngineModule)
repos.Put(ctx, e.Key(), e.Value(), !opOf(e))
if after := repos.CommitOf(check.NodeEngineModule); after != before {
ch.TrunkMoved(ctx) // the node-engine the mesh runs moved: a verdict that read it beside is carried (ADR 0548 §6)
}
})
}()
each := func(e jetstream.KeyValueEntry) {
if opOf(e) {
+1 -1
View File
@@ -3,7 +3,7 @@ module mesh-delivery-checker
go 1.26.0
require (
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026233910-a2759ec92ec9
github.com/nats-io/nats-server/v2 v2.11.17
github.com/nats-io/nats.go v1.54.0
)
+2 -4
View File
@@ -1,7 +1,5 @@
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261023213200-403a5d72ab93 h1:FAiFY2KBKEoZSH49OZTt7/zRBIQxyl5g/KhYrpggbN4=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261023213200-403a5d72ab93/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585 h1:7QoFH5qjXRpx5M4uPdYsVJ5qKl2082gq8n0lFxhoNIg=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026233910-a2759ec92ec9 h1:MS7i4NVNQHMAMqRRFMdzbto31BHZY+BuEZY1ql3p4Rs=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026233910-a2759ec92ec9/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op/go.mod h1:FQyySiasQQM8735Ddel3MRojmy4dA1IqCeyJ5jmPMbI=
github.com/google/go-tpm v0.9.8 h1:slArAR9Ft+1ybZu0lBwpSmpwhRXaa85hWtMinMyRAWo=
@@ -0,0 +1,355 @@
package check
import (
"context"
"strings"
"testing"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// One check per delivery (novox/hq ADR 0548): the checker's part. A merge turn holds new checks off its repository (§3);
// a trunk or sibling move never ends a running check and carries its verdict once it ended (§1, §5, §6); only a head git
// refused gets a whole check after a move (§7); the line of a carried verdict says so (§8).
// turnOf is a turn's value on a repository for a delivery of the heads hash.
func turnOf(t testing.TB, repo, group, heads string) []byte {
t.Helper()
at := time.Date(2026, 10, 26, 9, 0, 0, 0, time.UTC)
data, err := delivery.EncodeTurn(delivery.Turn{Repository: repo, Waiting: []delivery.TurnWait{{Group: group, Heads: heads, Approved: at, Since: at}}})
if err != nil {
t.Fatal(err)
}
return data
}
func turnKey(t testing.TB, repo string) string {
t.Helper()
k, err := delivery.TurnsKey(repo)
if err != nil {
t.Fatal(err)
}
return k
}
// waitFor polls a condition the checker's goroutines make true, failing after two seconds.
func waitFor(t testing.TB, what string, ok func() bool) {
t.Helper()
for deadline := time.Now().Add(2 * time.Second); !ok(); time.Sleep(time.Millisecond) {
if time.Now().After(deadline) {
t.Fatalf("%s did not happen", what)
}
}
}
// pullWithFP is a pull request touching the module m at the fingerprint `src2:` and 64 times the hex digit d.
func pullWithFP(repo string, number int, head string, d string) delivery.CandidatePull {
p := pullOf(repo, number, head)
p.Modules = []string{"m"}
p.Fps = map[string]delivery.CandidateModule{"m": {FP2: "src2:" + strings.Repeat(d, 64)}}
return p
}
func TestOneCheckPerDelivery(t *testing.T) {
ctx := context.Background()
for name, run := range map[string]func(t *testing.T){
// §3: no claim while a turn names one of its repositories; a claim on the turn's delete event.
"a turn on its repository holds a new check off until the turn goes": func(t *testing.T) {
j := passJudge()
r := newRig(t, j)
turns := &Turns{}
r.ch.Turns = turns
turns.Put(turnKey(t, repoA), turnOf(t, repoA, "fix/other", "0123456789ab"), false)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
if j.count() != 0 || len(r.verdicts.vals) != 0 {
t.Fatalf("judged %d times with a turn on %s, verdicts %v", j.count(), repoA, r.verdicts.vals)
}
if held := r.ch.Held()[keyOf(t, "feat/x", cand)]; !strings.Contains(held, repoA) || !strings.Contains(held, "fix/other") {
t.Errorf("the hold is said as %q, want the turn's repository and group", held)
}
if !turns.Put(turnKey(t, repoA), nil, true) {
t.Fatal("the turn's delete changed nothing")
}
r.ch.Reconsider(ctx) // what the watch of turns does on the delete
r.ch.Wait()
if j.count() != 1 || r.verdict(t, keyOf(t, "feat/x", cand)).State != delivery.VerdictPass {
t.Errorf("judged %d times after the turn went, want once", j.count())
}
},
"a turn on another repository, or of its own heads, holds nothing": func(t *testing.T) {
j := passJudge()
r := newRig(t, j)
turns := &Turns{}
r.ch.Turns = turns
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
c, _ := delivery.DecodeCandidate(cand)
turns.Put(turnKey(t, repoSDK), turnOf(t, repoSDK, "fix/other", "0123456789ab"), false)
turns.Put(turnKey(t, repoA), turnOf(t, repoA, "feat/x", GateHeads(c)), false)
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
if j.count() != 1 {
t.Errorf("judged %d times, want once: the turns hold another repository and its own merge", j.count())
}
},
// §3, the race: a check claimed before the gate's turn is judged on the commit it read, never restarted.
"a check claimed before a turn was written runs to its verdict": func(t *testing.T) {
release, started := make(chan struct{}), make(chan struct{})
j := &judgeFunc{f: func(context.Context, *Case, int) (Outcome, error) {
close(started)
<-release
return Outcome{Verdict: passing()}, nil
}}
r := newRig(t, j)
turns := &Turns{}
r.ch.Turns = turns
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
<-started
turns.Put(turnKey(t, repoA), turnOf(t, repoA, "fix/other", "0123456789ab"), false)
r.ch.Reconsider(ctx)
close(release)
r.ch.Wait()
if j.count() != 1 || r.verdict(t, keyOf(t, "feat/x", cand)).State != delivery.VerdictPass {
t.Errorf("judged %d times; the running check is not ended by the turn", j.count())
}
},
// §1, §5: a trunk move during a running check calls the judge once, writes its verdict, keeps the status pending (a pass
// on an old trunk is never shown), then asks one carry, on the recheck stream, and no check.
"a trunk move during a check ends nothing and asks one carry after it": func(t *testing.T) {
var r *rig
var kinds []string
var from []map[string]string
judged := make(chan struct{})
j := &judgeFunc{f: func(ctx context.Context, c *Case, call int) (Outcome, error) {
kinds, from = append(kinds, c.Kind), append(from, c.CarriedFrom)
if call == 1 {
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS}) // the trunk moves while it runs
r.ch.TrunkMoved(ctx) // as the watch of the source says it
close(judged)
} else if s := owedState(t, r, repoA, 7, head1); s != "pending" {
t.Errorf("the status during the carry is %q, want pending: the pass on the old trunk is never shown", s)
}
return Outcome{Verdict: passing()}, nil
}}
r = newRig(t, j)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
<-judged
r.ch.Wait()
v := r.verdict(t, keyOf(t, "feat/x", cand))
if j.count() != 2 || kinds[0] != KindNew || kinds[1] != KindTrunk || from[0] != nil || from[1][repoA] != trunkA {
t.Fatalf("judged %d times as %v carried from %v; want a whole check, then one carry from %s", j.count(), kinds, from, trunkA[:12])
}
if v.State != delivery.VerdictPass || v.Trunk[repoA] != trunkA2 || v.CarriedFrom[repoA] != trunkA {
t.Errorf("the verdict is %s on %v carried from %v; want a pass on the new trunk carried from the old", v.State, v.Trunk, v.CarriedFrom)
}
if s := owedState(t, r, repoA, 7, head1); s != "success" {
t.Errorf("the status after the carry is %q, want success", s)
}
if r.verdicts.creates[keyOf(t, "feat/x", cand)] != 1 {
t.Error("the verdict key was made again: a carry rewrites it")
}
},
// §5: the composition is kept when no fingerprint of a module it composed changed, and runs again otherwise.
"a carry keeps a passed composition while the fingerprints stay the same": func(t *testing.T) {
var kept []bool
j := &judgeFunc{f: func(_ context.Context, c *Case, _ int) (Outcome, error) {
kept = append(kept, c.Composition != nil)
v := passing()
v.Composition = &delivery.Composition{State: delivery.CompositionPass, SDK: "v0.1.14-x", FPSet: c.FPSet}
return Outcome{Verdict: v}, nil
}}
r := newRig(t, j)
cand := candidateOf(t, "feat/x", pullWithFP(repoA, 7, head1, "a"))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS})
r.ch.TrunkMoved(ctx) // the indexer's fingerprints of the candidate stay the same
r.ch.Wait()
r.ch.Candidate(ctx, "feat%2Fx", candidateOf(t, "feat/x", pullWithFP(repoA, 7, head1, "b"))) // a fingerprint changed with the move
r.setTrunks(t, map[string]string{repoA: trunkA, repoSDK: trunkS})
r.ch.TrunkMoved(ctx)
r.ch.Wait()
if len(kept) != 3 || kept[0] || !kept[1] || kept[2] {
t.Errorf("the composition was kept in the checks %v; want composed, kept on the carry, composed again once a fingerprint changed", kept)
}
},
// §6: a sibling's move puts the verdict out of date, its status pending saying it is carried, and the carry's verdict
// names the sibling's old commit; the status follows the carry's verdict. Which packages run is the digest's (only
// those that name MESH_CHECK_BESIDE change with a sibling), the build module's part.
"a sibling's move is carried and the status follows the carry": func(t *testing.T) {
var r *rig
var from map[string]string
j := &judgeFunc{f: func(_ context.Context, c *Case, call int) (Outcome, error) {
if call == 2 {
from = c.CarriedFrom
if s := owedState(t, r, repoA, 7, head1); s != "pending" {
t.Errorf("the status during the sibling's carry is %q, want pending", s)
}
}
return Outcome{Verdict: passing(), Beside: map[string]string{catalogueRepository: r.trunks.Heads()[catalogueRepository]}}, nil
}}
r = newRig(t, j)
r.setTrunks(t, map[string]string{repoA: trunkA, repoSDK: trunkS, catalogueRepository: trunkS})
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
if b := r.verdict(t, keyOf(t, "feat/x", cand)).Beside; b[catalogueRepository] != trunkS {
t.Fatalf("the verdict names the siblings %v, want the catalogue at %s", b, trunkS[:12])
}
r.setTrunks(t, map[string]string{repoA: trunkA, repoSDK: trunkS, catalogueRepository: trunkS2})
r.ch.TrunkMoved(ctx)
r.ch.Wait()
v := r.verdict(t, keyOf(t, "feat/x", cand))
if j.count() != 2 || from[catalogueRepository] != trunkS || v.CarriedFrom[catalogueRepository] != trunkS || v.Beside[catalogueRepository] != trunkS2 {
t.Errorf("judged %d times, carried from %v, verdict beside %v carried from %v; want one carry over the catalogue's move", j.count(), from, v.Beside, v.CarriedFrom)
}
if s := owedState(t, r, repoA, 7, head1); s != "success" {
t.Errorf("the status after the carry is %q, want success", s)
}
},
// §7: a carry whose merge conflicts fails naming the files and asks no check; a later move onto which the head merges
// cleanly asks a whole check, not a carry.
"a carry git refuses fails naming the files, and the next move asks a whole check": func(t *testing.T) {
var kinds []string
var from []map[string]string
j := &judgeFunc{f: func(_ context.Context, c *Case, call int) (Outcome, error) {
kinds, from = append(kinds, c.Kind), append(from, c.CarriedFrom)
if call == 2 {
return Outcome{Verdict: delivery.Verdict{State: delivery.VerdictFail, Failed: delivery.FailedFirst{Machine: "ace",
Words: repoA + "#7 " + ConflictWords + "go.mod"}}}, nil
}
return Outcome{Verdict: passing()}, nil
}}
r := newRig(t, j)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS})
r.ch.TrunkMoved(ctx)
r.ch.Wait()
r.ch.Reconsider(ctx) // anything else that looks again asks nothing of a conflict
r.ch.Wait()
if v := r.verdict(t, keyOf(t, "feat/x", cand)); j.count() != 2 || !IsConflictFailure(v) || !strings.Contains(v.Failed.Words, "go.mod") {
t.Fatalf("judged %d times, verdict %+v; want the carry failed naming go.mod and nothing asked after it", j.count(), v.Failed)
}
r.setTrunks(t, map[string]string{repoA: trunkA, repoSDK: trunkS})
r.ch.TrunkMoved(ctx)
r.ch.Wait()
if j.count() != 3 || kinds[1] != KindTrunk || from[1] == nil || kinds[2] != KindNew || from[2] != nil {
t.Errorf("checks %v carried from %v; want a whole check, a carry, then a whole check after the conflict", kinds, from)
}
},
// §8, §6: the line of a carried verdict names the head on the new trunk commit first, as the gate and the merge tool
// read a line, then where it was carried from and what the carry ran and kept; while it is carried it is pending and says so.
"the line of a carry says where it was carried from": func(t *testing.T) {
cs, p := lineCase(t, pullOf(repoA, 7, head1))
v := passVerdict()
v.Trunk = map[string]string{repoA: trunkA}
v.Beside = map[string]string{catalogueRepository: trunkS2}
v.CarriedFrom = map[string]string{repoA: trunkA2, catalogueRepository: trunkS}
line := Line(v, cs, p, nil, nil, nil)
if !strings.HasPrefix(line, "pass 111111111111 on aaaaaaaaaaaa | carried from abababababab, mesh-catalog 555555555555 | unit 12 run") {
t.Errorf("the carried line = %q", line)
}
if head, trunk := checkWordsOf(line); head != head1[:12] || trunk != trunkA[:12] {
t.Errorf("the line names %s on %s first, want the head on the new trunk", head, trunk)
}
cs.CarriedFrom = map[string]string{repoA: trunkA2}
pending := Line(delivery.Verdict{State: delivery.VerdictRunning}, cs, p, nil, nil, nil)
if !strings.HasPrefix(pending, "being carried to 111111111111 on aaaaaaaaaaaa | from abababababab") {
t.Errorf("the pending line of a carry = %q", pending)
}
},
// §3 (review round 1): two approved merges in one turn whose trunk moved each get their carry; neither holds the other's.
"two merges waiting in one turn each get their carry": func(t *testing.T) {
var r *rig
j := &judgeFunc{f: func(context.Context, *Case, int) (Outcome, error) { return Outcome{Verdict: passing()}, nil }}
r = newRig(t, j)
candA, candC := candidateOf(t, "feat/a", pullOf(repoA, 7, head1)), candidateOf(t, "feat/c", pullOf(repoA, 8, head2))
r.ch.Candidate(ctx, "feat%2Fa", candA)
r.ch.Candidate(ctx, "feat%2Fc", candC)
r.ch.Wait()
a, _ := delivery.DecodeCandidate(candA)
c, _ := delivery.DecodeCandidate(candC)
at := time.Date(2026, 10, 26, 9, 0, 0, 0, time.UTC)
turn, err := delivery.EncodeTurn(delivery.Turn{Repository: repoA, Waiting: []delivery.TurnWait{
{Group: "feat/a", Heads: GateHeads(a), Approved: at, Since: at}, {Group: "feat/c", Heads: GateHeads(c), Approved: at.Add(time.Minute), Since: at}}})
if err != nil {
t.Fatal(err)
}
turns := &Turns{}
turns.Put(turnKey(t, repoA), turn, false)
r.ch.Turns = turns
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS}) // a third merge moves the trunk under both
r.ch.TrunkMoved(ctx)
r.ch.Wait()
for _, cand := range [][]byte{candA, candC} {
group := map[bool]string{true: "feat/a", false: "feat/c"}[string(cand) == string(candA)]
if v := r.verdict(t, keyOf(t, group, cand)); v.Trunk[repoA] != trunkA2 || v.CarriedFrom[repoA] != trunkA {
t.Errorf("%s: verdict on %v carried from %v; want its carry claimed and done", group, v.Trunk, v.CarriedFrom)
}
}
},
// §6 (review round 1): the node-engine read at the commit the mesh runs is named, and a new running commit is carried.
"a new running node-engine is carried": func(t *testing.T) {
runs := runsAt{nodeEngineModule: trunkS}
j := &judgeFunc{f: func(_ context.Context, c *Case, _ int) (Outcome, error) {
return Outcome{Verdict: passing(), Beside: map[string]string{nodeEngineRepo: runs[nodeEngineModule]}}, nil
}}
r := newRig(t, j)
r.ch.Runs = runs
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
r.ch.TrunkMoved(ctx) // nothing moved: nothing is carried
r.ch.Wait()
runs[nodeEngineModule] = trunkS2
r.ch.TrunkMoved(ctx) // what the indexer's watch does when mesh-host's running commit changes
r.ch.Wait()
v := r.verdict(t, keyOf(t, "feat/x", cand))
if j.count() != 2 || v.CarriedFrom[nodeEngineRepo] != trunkS || v.Beside[nodeEngineRepo] != trunkS2 {
t.Errorf("judged %d times, verdict beside %v carried from %v; want one carry over the node-engine's new running commit", j.count(), v.Beside, v.CarriedFrom)
}
},
// §5: a carry that fails names the move it failed on.
"a carry that fails names the move": func(t *testing.T) {
j := &judgeFunc{f: func(_ context.Context, c *Case, call int) (Outcome, error) {
if call == 1 {
return Outcome{Verdict: passing()}, nil
}
return Outcome{Verdict: delivery.Verdict{State: delivery.VerdictFail, Failed: delivery.FailedFirst{Test: "unit: m/a.TestX", Words: "want 4"}}}, nil
}}
r := newRig(t, j)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS})
r.ch.TrunkMoved(ctx)
r.ch.Wait()
v := r.verdict(t, keyOf(t, "feat/x", cand))
if v.State != delivery.VerdictFail || v.CarriedFrom != nil || !strings.HasPrefix(v.Failed.Words, "want 4") ||
!strings.Contains(v.Failed.Words, repoA+" "+trunkA[:12]+" to "+trunkA2[:12]) {
t.Errorf("the failed carry is %s carried from %v with words %q; want the test's words, then the move", v.State, v.CarriedFrom, v.Failed.Words)
}
},
} {
t.Run(name, run)
}
}
// checkWordsOf is the head and trunk a line names first, as the gate reads it (mesh-delivery-gate onestatus.go checkLine).
func checkWordsOf(line string) (string, string) {
for _, part := range strings.Split(line, " | ") {
f := strings.Fields(part)
for i := 0; i+2 < len(f); i++ {
if f[i+1] == "on" && len(f[i]) == 12 && len(f[i+2]) == 12 {
return f[i], f[i+2]
}
}
}
return "", ""
}
@@ -127,6 +127,9 @@ func (ch *Checker) Composed(ctx context.Context, comp delivery.CompositionVerdic
return
}
withComposition(&v, comp)
if v.Composition != nil && v.Composition.State != delivery.CompositionNotRun {
v.Composition.FPSet = cs.FPSet
}
data, err := delivery.EncodeVerdict(v)
if err != nil {
ch.logf("%s: the verdict with its composition cannot be written: %v", cs.Key, err)
@@ -12,8 +12,18 @@ type BesideRef struct {
Repository string `json:"repository"`
Commit string `json:"commit"`
Ref string `json:"ref,omitempty"`
// atTrunk says the sibling was read at its trunk's head, so a move of that trunk puts the verdict out of date and is
// carried (novox/hq ADR 0548 §6). Never sent: the node-engine at the commit the mesh runs, mesh-sdk at a pin and a
// group's member at its head do not move with a trunk.
atTrunk bool
// runs says the sibling was read at the commit the mesh runs (the node-engine, the indexer's record of mesh-host): a new
// running commit puts the verdict out of date and is carried (ADR 0548 §6). Never sent.
runs bool
}
// NodeEngineModule is the module whose running commit the node-engine sibling is read at.
const NodeEngineModule = nodeEngineModule
const (
catalogueRepository = "novox/mesh-catalog"
nodeEngineRepo = "novox/mesh-host"
@@ -51,18 +61,21 @@ func (j *Examiner) besideFor(ctx context.Context, c *Case, p Pull) ([]BesideRef,
switch {
case repo == catalogueRepository:
if head := heads[repo]; head != "" {
out = append(out, BesideRef{Repository: repo, Commit: head})
out = append(out, BesideRef{Repository: repo, Commit: head, atTrunk: c.member(repo) == nil})
}
case repo == nodeEngineRepo:
commit := ""
if j.Runs != nil {
commit = j.Runs.CommitOf(nodeEngineModule)
}
atTrunk := false
if commit == "" {
commit = heads[repo]
commit, atTrunk = heads[repo], true
}
if commit != "" {
out = append(out, BesideRef{Repository: repo, Commit: commit})
// Named in the verdict either way (ADR 0548 §6, issue 568's own example): at the commit the mesh runs, a new
// running commit puts the verdict out of date; at the trunk, a trunk move does (Checker.besideNow).
out = append(out, BesideRef{Repository: repo, Commit: commit, atTrunk: atTrunk, runs: !atTrunk})
}
case repo == SDKRepository && p.Repository == controllerRepo && j.Forge != nil:
text, err := j.Forge.File(ctx, p.Repository, "go.mod", p.Head)
@@ -82,7 +95,7 @@ func (j *Examiner) besideFor(ctx context.Context, c *Case, p Pull) ([]BesideRef,
if member := c.member(repo); member != nil {
out = append(out, BesideRef{Repository: repo, Commit: member.Head, Ref: refOf(member.Number)})
} else if head := heads[repo]; head != "" {
out = append(out, BesideRef{Repository: repo, Commit: head})
out = append(out, BesideRef{Repository: repo, Commit: head, atTrunk: true})
} else if repo == controllerRepo && p.Repository == nodeEngineRepo {
// The node-engine's step-nine test judges the controller's own manifest and fails without it (issue 1295): a
// checker that cannot place it says so, where placing none left that test to skip.
@@ -108,3 +121,13 @@ func refOf(number int) string {
}
return "refs/pull/" + strconv.Itoa(number) + "/head"
}
// besideAtTrunk are the siblings of refs read at their trunk's head that the case does not merge, by repository: what a
// verdict names as Beside (ADR 0548 §6).
func besideAtTrunk(c *Case, refs []BesideRef, into map[string]string) {
for _, r := range refs {
if _, merged := c.Trunk[r.Repository]; (r.atTrunk || r.runs) && !merged {
into[r.Repository] = r.Commit
}
}
}
@@ -0,0 +1,42 @@
package check
import (
"fmt"
"strings"
)
// A change that adds or modifies a compiled program fails (issue 1577): it is build output, `go build` in a module's folder
// writes one beside go.mod, and `git add -A` takes it. The build module's fetch step, which holds the merged clone, reads the
// first bytes of each changed file and reports the programs (mesh-sdk buildstep.MergeInto); the checker judges the answer.
// NoProgramsReported is the warning of a build module whose fetch answered no programs field: an older module, which cannot
// tell, so the check does not refuse a program it did not look for.
const NoProgramsReported = "the build module did not report programs: a compiled program in the change is not refused"
// checkBinariesOf fails a pull request whose change adds or modifies a compiled program, naming each file. It judges the
// first fetch answer of the pull request, whichever build module gave it; a build module that reports no programs leaves a
// warning on the check and passes it.
func (st *state) checkBinariesOf(t *target) verdictPtr {
key := fmt.Sprintf("%s#%d", t.pull.Repository, t.pull.Number)
if t.fetched.Merge == "" {
return nil // no merge was made: nothing was changed to judge
}
if t.fetched.Programs == nil {
st.c.note(fmt.Sprintf("%s: %s (%s)", key, NoProgramsReported, t.module))
st.c.warn(key, fmt.Sprintf("a compiled program is not refused: %s reported no programs", t.module))
st.j.logf("%s: %s (%s)", key, NoProgramsReported, t.module)
return nil
}
var found []string
for _, p := range *t.fetched.Programs {
if p.Commit != "" {
found = append(found, fmt.Sprintf("%s: a compiled program added in %s and removed later: it stays in history; squash the branch (issue 1577)", p.Path, short12(p.Commit)))
continue
}
found = append(found, p.Path+": a compiled program is build output: remove it and add it to .gitignore (issue 1577)")
}
if len(found) > 0 {
return failed(t.node, fmt.Sprintf("%s: %s", key, strings.Join(found, "; ")))
}
return nil
}
@@ -0,0 +1,63 @@
package check
import (
"slices"
"strings"
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
func TestAChangeWhoseFetchReportsProgramsFailsNamingEachFile(t *testing.T) { // 1577
prog := func(p ...string) *[]ProgramFile {
out := []ProgramFile{}
for _, f := range p {
path, commit, _ := strings.Cut(f, "@") // `path@commit`: added by a commit of the branch and removed later
out = append(out, ProgramFile{Path: path, Kind: "ELF", Commit: commit})
}
return &out
}
for _, c := range []struct {
name string
programs *[]ProgramFile
fail []string // paths named; none passes
warned bool // the build module reported no programs field
}{
{"none found", prog(), nil, false},
{"one program", prog("modules/m/m"), []string{"modules/m/m"}, false},
{"two programs", prog("a/x", "b/y"), []string{"a/x", "b/y"}, false},
{"added and removed in the branch", prog("gone/prog@0123456789abcdef0123"), []string{"gone/prog: a compiled program added in 0123456789ab and removed later: it stays in history; squash the branch (issue 1577)"}, false},
{"an older build module reports nothing", nil, nil, true},
} {
b := newBench(t)
b.fetch.Merge, b.fetch.Programs = "mmmm", c.programs
cs := caseOf(t, pullOf(repoA, 7, head1))
v := b.judge(t, cs).Verdict
if got := slices.ContainsFunc(cs.Notes(), func(n string) bool { return strings.Contains(n, NoProgramsReported) }); got != c.warned {
t.Errorf("%s: warned = %v, want %v (%v)", c.name, got, c.warned, cs.Notes())
}
line := Line(v, cs, cs.Pulls[0], nil, nil, nil)
if got := strings.Contains(line, "a compiled program is not refused: mesh-build-go reported no programs"); got != c.warned {
t.Errorf("%s: the status line holds the warning = %v, want %v: %s", c.name, got, c.warned, line)
}
if len(c.fail) == 0 {
if v.State != delivery.VerdictPass {
t.Errorf("%s: verdict = %+v", c.name, v)
}
continue
}
if v.State != delivery.VerdictFail {
t.Errorf("%s: verdict = %+v, want a failure", c.name, v)
continue
}
for _, p := range c.fail {
want := p
if !strings.Contains(p, " added in ") {
want = p + ": a compiled program is build output: remove it and add it to .gitignore (issue 1577)"
}
if !strings.Contains(v.Failed.Words, want) {
t.Errorf("%s: words %q do not name %s", c.name, v.Failed.Words, p)
}
}
}
}
@@ -43,12 +43,43 @@ type Case struct {
Try *TryOptions
notesMu sync.Mutex
notes []string
// warnings are, by pull request, what part of its check did not run (Case.warn).
warnings map[string]string
notes []string
// asks are the ids of the asks the check put on the build seats' streams, in order: a try names them (Try.Asks).
asks []string
// Watch, when set, hears each ask the check puts, as it is put (step nil), and each value its step key takes while
// the check waits for it: the node that took it, the phase it says, its end. A try writes them into its record, so a
// running try names its asks and where each stands (issue 1890). Called from the check's goroutines; it must not block
// for long.
Watch func(askID string, step *delivery.Step)
// Progress says what the check is doing, so a check that ran past its bound names the step it was stuck at.
Progress *Progress
// Kind is what asked the check (KindNew, KindNodes, KindTrunk): a re-judge after a trunk move is asked on the build
// seats' recheck stream, every other check on check (ADR 0536 §4).
// Kind is what asked the check (KindNew, KindNodes, KindTrunk): a carry after a trunk or sibling move is asked on the
// build seats' recheck stream, every other check on check (ADR 0536 §4, ADR 0548 §5).
Kind string
// CarriedFrom are, for the carry of a verdict that passed, the trunk and sibling commits it was judged on (ADR 0548 §5);
// nil for a whole check.
CarriedFrom map[string]string
// FPSet is the candidate's fingerprint set (Candidate.FPSet): a composition names the set it judged.
FPSet string
// Composition is, for a carry, the composition that passed on the same fingerprint set: it is kept and not asked again,
// since no fingerprint of a module it composed changed (ADR 0548 §5). Nil composes.
Composition *delivery.Composition
}
// Asked records the id of an ask the check put.
func (c *Case) Asked(id string) {
c.notesMu.Lock()
c.asks = append(c.asks, id)
c.notesMu.Unlock()
}
// Asks are the ids of the asks the check put, in order.
func (c *Case) Asks() []string {
c.notesMu.Lock()
defer c.notesMu.Unlock()
return append([]string(nil), c.asks...)
}
// Progress is the step a check is at, written by the check and read when its bound passes.
@@ -80,7 +111,7 @@ func CaseOf(c delivery.Candidate, trunk map[string]string) *Case {
if err != nil {
return nil
}
cs := &Case{Key: key, Group: c.Group, HeadsHash: c.HeadsHash, Heads: GateHeads(c), Trunk: map[string]string{}, Progress: &Progress{}}
cs := &Case{Key: key, Group: c.Group, HeadsHash: c.HeadsHash, Heads: GateHeads(c), Trunk: map[string]string{}, Progress: &Progress{}, FPSet: c.FPSet}
for _, p := range c.Pulls {
switch head, known := trunk[p.Repository]; {
case !known || head == "":
@@ -162,6 +193,23 @@ func (c *Case) note(s string) {
}
}
// warn says of a pull request (pullKey) that a part of its check did not run, in words for its status line (issue 1577).
func (c *Case) warn(key, words string) {
c.notesMu.Lock()
defer c.notesMu.Unlock()
if c.warnings == nil {
c.warnings = map[string]string{}
}
c.warnings[key] = words
}
// Warning is what warn said of a pull request, or "".
func (c *Case) Warning(key string) string {
c.notesMu.Lock()
defer c.notesMu.Unlock()
return c.warnings[key]
}
// Notes are the words noted on the case, in the order they were noted.
func (c *Case) Notes() []string {
c.notesMu.Lock()
@@ -4,6 +4,7 @@ import (
"context"
"errors"
"fmt"
"sort"
"strings"
"sync"
"time"
@@ -42,6 +43,9 @@ type Outcome struct {
// Judged says, by pull request (pullKey), which program judged a change to a build module (ADR 0535 §5); a pull request
// that changes none is not named.
Judged map[string]string
// Beside are the siblings the check read at their trunk's head, by repository (ADR 0548 §6): the verdict names them, and
// a move of one is carried.
Beside map[string]string
}
// BuildNote is the state of a candidate's module build, by module: built, failed or waiting (ADR 0484 clause 7).
@@ -65,6 +69,14 @@ type Checker struct {
Owed KV
Judge Judge
Trunks interface{ Heads() map[string]string }
// Runs says which commit the mesh runs of a module (the indexer's modules): the node-engine a verdict read beside at its
// running commit is out of date when that commit moves (ADR 0548 §6); nil reads every sibling at its trunk.
Runs interface{ CommitOf(module string) string }
// Turns are the gate's merge turns (ADR 0548 §3): no check is claimed of a repository a turn names, but the check of the
// waiting delivery's own heads; nil holds none.
Turns interface {
Holding(repositories []string, heads string) (string, delivery.TurnWait, bool)
}
// Bound is how long one check may run; none when 0 (DefaultBound).
Bound time.Duration
// Link is the base of the address a status points at; the verdict key follows.
@@ -73,6 +85,10 @@ type Checker struct {
Announce Announcer
Now func() time.Time
Logf func(format string, args ...any)
// Settle is how long a group's heads must stay the same after the candidate last changed them before its check starts
// (issue 1890): the members of a group are announced one after another, and a check of each partial set is thrown
// away at the next. None when 0; DefaultSettle in the module.
Settle time.Duration
once sync.Once
wg sync.WaitGroup
@@ -83,15 +99,78 @@ type Checker struct {
tried map[string]bool // by verdict key: looked at since this process started
epoch uint64 // changes of the nodes seen since this process started
nodesFail map[string]uint64 // by verdict key: the epoch a failure of the nodes was written at
held map[string]string // by verdict key: the turn that holds its check off, as last said
live map[string]liveCheck // by verdict key: the check running in this process, ended when its heads go
settling map[string]*window // by candidates key: the window its heads settle in
}
// liveCheck is a check running in this process: the candidates key of its group and the end of its run.
type liveCheck struct {
group string
end context.CancelCauseFunc
}
// window is a group's settle window: aborted with errUnsettled when its heads change again or the group goes, and with
// errSettled to end it at once.
type window struct{ abort context.CancelCauseFunc }
var (
errUnsettled = errors.New("the group's heads changed again")
errSettled = errors.New("the group's heads settled")
)
// settleNow ends the settle window of every group at once, as if Settle had passed.
func (ch *Checker) settleNow() {
ch.mu.Lock()
defer ch.mu.Unlock()
for _, w := range ch.settling {
w.abort(errSettled)
}
}
// DefaultSettle is the module's Settle (issue 1890). On 2026-10-26 the three members of a group were announced within 2.8
// seconds of each other (20:31:52.3, :53.7, :55.1 UTC), a session opening or pushing them one API call after another;
// fifteen seconds holds five times that, and costs a check of 10 to 18 minutes under 2.5 % of its time. A member that comes
// later still is no waste beyond its asks already taken: the check of the heads it replaced is ended at once.
const DefaultSettle = 15 * time.Second
// EndedWords open the words of a verdict the checker ended without judging it, its heads no candidate's any more (issue
// 1890): never a failure of the change, and judged whole if a candidate holds those heads again.
const EndedWords = "ended: "
// IsEnded says whether a verdict was ended without being judged (EndedWords).
func IsEnded(v delivery.Verdict) bool {
return v.State == delivery.VerdictFail && v.Failed.Machine == "checker" && strings.HasPrefix(v.Failed.Words, EndedWords)
}
// superseded is the cause a running check is ended with: the heads that replaced its own, or none when its group went.
type superseded struct{ by string }
func (s superseded) Error() string {
if s.by == "" {
return EndedWords + "no candidate holds these heads any more: the group was closed or merged"
}
return EndedWords + "superseded by " + s.by
}
func (ch *Checker) init() {
ch.once.Do(func() {
ch.known, ch.inflight, ch.again, ch.tried = map[string]delivery.Candidate{}, map[string]bool{}, map[string]bool{}, map[string]bool{}
ch.nodesFail = map[string]uint64{}
ch.nodesFail, ch.held = map[string]uint64{}, map[string]string{}
ch.live, ch.settling = map[string]liveCheck{}, map[string]*window{}
})
}
// endOthers ends every check of the group running in this process whose verdict key is not keep: no candidate holds its
// heads any more (issue 1890). Called with ch.mu held.
func (ch *Checker) endOthers(group, keep, by string) {
for k, l := range ch.live {
if l.group == group && k != keep {
l.end(superseded{by: by})
}
}
}
func (ch *Checker) logf(format string, args ...any) {
if ch.Logf != nil {
ch.Logf(format, args...)
@@ -124,17 +203,62 @@ func (ch *Checker) Candidate(ctx context.Context, key string, value []byte) {
ch.logf("candidates.%s holds what is not a candidate, left: %v", key, err)
return
}
vkey, _ := delivery.CheckerVerdictKey(c.Group, c.HeadsHash)
ch.mu.Lock()
prev, had := ch.known[key]
ch.known[key] = c
// A check of heads this candidate no longer holds is ended at once (issue 1890): a member's new head, a member joining
// or leaving.
ch.endOthers(key, vkey, GateHeads(c))
moved := !had || prev.HeadsHash != c.HeadsHash
if ch.Settle > 0 && (moved || ch.settling[key] != nil) {
// The group's heads changed: its check waits until they have been the same for Settle, so a group whose members
// are announced one after another is checked once, whole. A later change starts the window again. The wait is a
// context's bound, no timer (ADR 0437 §8).
if w := ch.settling[key]; w != nil {
w.abort(errUnsettled)
}
base, abort := context.WithCancelCause(ctx)
w := &window{abort: abort}
ch.settling[key] = w
ch.wg.Add(1)
go func() {
defer ch.wg.Done()
wait, done := context.WithTimeout(base, ch.Settle)
defer done()
<-wait.Done()
settled := errors.Is(wait.Err(), context.DeadlineExceeded) || errors.Is(context.Cause(wait), errSettled)
ch.mu.Lock()
if !settled || ch.settling[key] != w {
ch.mu.Unlock()
return // changed again, forgotten, or the checker stops
}
delete(ch.settling, key)
now, ok := ch.known[key]
ch.mu.Unlock()
abort(nil)
if ok {
ch.consider(ctx, key, now)
}
}()
ch.mu.Unlock()
return
}
ch.mu.Unlock()
ch.consider(ctx, key, c)
}
// Forget takes a candidate out (its key was deleted).
// Forget takes a candidate out (its key was deleted): a check of its heads still running is ended, as no candidate holds
// them any more.
func (ch *Checker) Forget(key string) {
ch.init()
ch.mu.Lock()
delete(ch.known, key)
if w := ch.settling[key]; w != nil {
w.abort(errUnsettled)
delete(ch.settling, key)
}
ch.endOthers(key, "", "")
ch.mu.Unlock()
}
@@ -177,32 +301,49 @@ func (ch *Checker) consider(ctx context.Context, key string, c delivery.Candidat
}
cs.Queued = ch.now()
ch.mu.Lock()
if ch.settling[key] != nil {
ch.mu.Unlock()
return // its heads are settling: the window's end considers it (Settle)
}
if ch.inflight[cs.Key] {
ch.again[cs.Key] = true // looked at once more when the check ends, so a change of the trunk is never lost
ch.mu.Unlock()
return
}
ch.inflight[cs.Key] = true
run, end := context.WithCancelCause(ctx)
ch.live[cs.Key] = liveCheck{group: key, end: end}
ch.mu.Unlock()
ch.wg.Add(1)
go func() {
defer ch.wg.Done()
vkey := cs.Key
defer func() {
ch.mu.Lock()
delete(ch.inflight, vkey)
delete(ch.again, vkey)
delete(ch.live, vkey)
ch.mu.Unlock()
end(nil)
}()
for {
// Every check is asked at once: the build seats' streams hold the backlog, a new head's check and a check after a
// failure of the nodes on check, a re-judge after a trunk move on recheck, pulled after them (ADR 0525 §4, §9;
// ADR 0536 §4). The checker holds no place and no capacity of its own.
cs.Kind = ch.kindOf(ctx, cs)
ch.try(ctx, key, cs)
ch.prepare(ctx, cs)
ch.try(ctx, run, key, cs)
ch.mu.Lock()
more := ch.again[cs.Key]
delete(ch.again, cs.Key)
if !more || ctx.Err() != nil {
delete(ch.inflight, cs.Key)
ch.mu.Unlock()
more := ch.again[vkey]
delete(ch.again, vkey)
ch.mu.Unlock()
if !more || run.Err() != nil {
return
}
ch.mu.Unlock()
// Looked at again for the same heads only: a candidate that moved to other heads has its own check (issue 1890).
if next := CaseOf(ch.candidate(key, c), ch.heads()); next != nil {
if next.Key != vkey {
return
}
cs = next
}
cs.Queued = ch.now()
@@ -211,29 +352,132 @@ func (ch *Checker) consider(ctx context.Context, key string, c delivery.Candidat
}
// The kinds of a check (ADR 0536 §4): a head never judged and a head asked again after a failure of the nodes are asked on
// the build seats' check stream, a head asked again only because its trunk moved on the recheck stream, pulled after.
// the build seats' check stream; a carry, a verdict moved to a newer trunk or sibling commit by its package digests (ADR 0548
// §5), on the recheck stream, pulled after.
const (
KindNew = "new"
KindNodes = "nodes"
KindTrunk = "trunk"
)
// kindOf is the kind of a case's check, read from its verdict key: none (or one left running) is a head never judged; a
// failure of the nodes is asked again after the nodes changed; a verdict on other trunk heads is a re-judge after a trunk
// move. A verdict that stands is ranked with the re-judges: its claim finds nothing to do.
func (ch *Checker) kindOf(ctx context.Context, cs *Case) string {
// prepare reads the case's verdict key and says what its check is (ADR 0548 §1, §5, §7): none (or one left running) is a
// head never judged, a whole check; a failure of the nodes is asked again after the nodes changed; a verdict judged on trunk
// or sibling commits that moved since is carried, its packages run only where the move changed their digest, and a pass's
// commits are named as the carry's CarriedFrom. A verdict that stands is ranked with the carries: its claim finds nothing
// to do.
func (ch *Checker) prepare(ctx context.Context, cs *Case) {
cs.Kind, cs.CarriedFrom, cs.Composition = KindNew, nil, nil
value, _, found, err := ch.Verdicts.Get(ctx, cs.Key)
if err != nil || !found {
return KindNew
return
}
v, err := delivery.DecodeVerdict(value)
switch {
case err != nil || v.State == delivery.VerdictRunning:
return KindNew
return
case IsNodesFailure(v):
return KindNodes
cs.Kind = KindNodes
return
case IsConflictFailure(v):
return // a head git refused: its next check is whole (ADR 0548 §7)
case IsEnded(v):
return // ended unjudged when its heads went (issue 1890): held again, they are checked whole
}
return KindTrunk
cs.Kind = KindTrunk
if v.State == delivery.VerdictPass && ch.stale(v, cs) {
cs.CarriedFrom = judgedOn(v)
if c := v.Composition; c != nil && c.State == delivery.CompositionPass && c.FPSet != "" && c.FPSet == cs.FPSet {
kept := *c
cs.Composition = &kept
}
}
}
// stale says whether a verdict was judged on commits that are no longer the case's: a trunk it merged moved, or a sibling it
// read at its trunk's head (ADR 0548 §6).
func (ch *Checker) stale(v delivery.Verdict, cs *Case) bool {
return !sameTrunk(v.Trunk, cs.Trunk) || besideMoved(v.Beside, ch.besideNow())
}
// besideNow are the commits a sibling is read at now, by repository: its trunk head, and for the node-engine the commit the
// mesh runs when the indexer names one, as besideFor places them.
func (ch *Checker) besideNow() map[string]string {
return besideAt(ch.heads(), ch.Runs)
}
// besideAt places the commits a sibling is read at: the trunk heads, with the node-engine at the commit the mesh runs when
// runs names one (ADR 0548 §6). The checker's decision and the status read the same commits through it.
func besideAt(now map[string]string, runs interface{ CommitOf(module string) string }) map[string]string {
if runs != nil {
if c := runs.CommitOf(nodeEngineModule); c != "" {
out := make(map[string]string, len(now)+1)
for r, h := range now {
out[r] = h
}
out[nodeEngineRepo] = c
return out
}
}
return now
}
// besideMoved says whether a sibling read at these commits moved: the commit it is read at now is known and another.
func besideMoved(beside, now map[string]string) bool {
for r, c := range beside {
if h := now[r]; h != "" && h != c {
return true
}
}
return false
}
// judgedOn are the trunk and sibling commits a verdict was judged on, by repository.
func judgedOn(v delivery.Verdict) map[string]string {
out := make(map[string]string, len(v.Trunk)+len(v.Beside))
for r, c := range v.Beside {
out[r] = c
}
for r, c := range v.Trunk {
out[r] = c
}
return out
}
// carried is what a carry's verdict names as carried from (ADR 0548 §5): the commits it was judged on before, of the
// repositories the new verdict still judges or reads; nil when none of them moved.
func carried(from map[string]string, v delivery.Verdict) map[string]string {
out, moved := map[string]string{}, false
for r, c := range from {
now, ok := v.Trunk[r]
if !ok {
now, ok = v.Beside[r]
}
if ok && now != "" && c != "" {
out[r] = c
moved = moved || now != c
}
}
if !moved {
return nil
}
return out
}
// moveWords say a carry's moves, for the words of a failure: "carried from <old12> to <new12> of <repository>", each moved
// repository, sorted.
func moveWords(from map[string]string, v delivery.Verdict) string {
var moves []string
for r, c := range from {
now, ok := v.Trunk[r]
if !ok {
now = v.Beside[r]
}
if now != "" && now != c {
moves = append(moves, fmt.Sprintf("%s %s to %s", r, short12(c), short12(now)))
}
}
sort.Strings(moves)
return strings.Join(moves, ", ")
}
func (ch *Checker) candidate(key string, fallback delivery.Candidate) delivery.Candidate {
@@ -265,6 +509,9 @@ func (ch *Checker) claim(ctx context.Context, cs *Case, first bool, epoch uint64
return nil, fmt.Errorf("%s: %w", cs.Key, err)
}
if !found {
if ch.heldByTurn(cs) {
return nil, nil
}
rev, err := ch.Verdicts.Create(ctx, cs.Key, data)
if errors.Is(err, ErrConflict) {
return nil, nil // another checker made it first
@@ -285,15 +532,15 @@ func (ch *Checker) claim(ctx context.Context, cs *Case, first bool, epoch uint64
return &claimed{rev}, nil
}
return nil, nil
case !sameTrunk(cur.Trunk, cs.Trunk):
retake = true
case IsEnded(cur), ch.stale(cur, cs):
retake = true // heads held again after their check was ended unjudged (issue 1890), or a trunk that moved
case cur.State == delivery.VerdictFail && IsNodesFailure(cur):
ch.mu.Lock()
at, ok := ch.nodesFail[cs.Key]
ch.mu.Unlock()
retake = !ok || epoch > at
}
if !retake {
if !retake || ch.heldByTurn(cs) {
return nil, nil
}
rev, err = ch.Verdicts.Update(ctx, cs.Key, data, rev)
@@ -306,6 +553,59 @@ func (ch *Checker) claim(ctx context.Context, cs *Case, first bool, epoch uint64
return &claimed{rev}, nil
}
// heldByTurn says whether a merge turn holds the case's check off (ADR 0548 §3): an approved merge waits on one of the
// repositories it merges heads into, and it is not the merge of the case's own heads. The check is claimed when the turn's
// key goes (Reconsider on the turns watch). Said once per turn in the log. A check already claimed is never held: one
// claimed in the moment before the gate wrote its turn is judged on the commit it read and carried once after.
func (ch *Checker) heldByTurn(cs *Case) bool {
if ch.Turns == nil {
return false
}
repos := make([]string, 0, len(cs.Trunk))
for r := range cs.Trunk {
repos = append(repos, r)
}
repo, w, held := ch.Turns.Holding(repos, cs.Heads)
why := ""
if held {
why = repo + " " + w.Group + " " + w.Heads
}
ch.mu.Lock()
said := ch.held[cs.Key]
if why == "" {
delete(ch.held, cs.Key)
} else {
ch.held[cs.Key] = why
}
ch.mu.Unlock()
if held && said != why {
ch.logf("%s: not claimed: the approved merge of %s (heads %s) holds a turn on %s, so its check waits for that merge (ADR 0548 §3)",
cs.Key, w.Group, w.Heads, repo)
}
return held
}
// Held is every check a merge turn holds off now, by verdict key, with the turn's repository, group and heads.
func (ch *Checker) Held() map[string]string {
ch.init()
ch.mu.Lock()
defer ch.mu.Unlock()
out := make(map[string]string, len(ch.held))
for k, v := range ch.held {
out[k] = v
}
return out
}
// ConflictWords are what the words of a head that git would not merge onto its trunk say before the files in conflict.
const ConflictWords = "does not merge into the trunk: files in conflict: "
// IsConflictFailure says whether a verdict failed because its head did not merge onto its trunk (ADR 0548 §7): it has no
// pass to carry, so its next check, at a push or at a later trunk move, is a whole check of its head.
func IsConflictFailure(v delivery.Verdict) bool {
return v.State == delivery.VerdictFail && strings.Contains(v.Failed.Words, ConflictWords)
}
// IsNodesFailure says whether a verdict failed because a part of the mesh the check reads was not heard (UnheardWords):
// it is judged again when the nodes change. No check fails for want of a build node any more: its asks wait in the build
// seats' streams (ADR 0525 §6, issue 1912).
@@ -313,9 +613,13 @@ func IsNodesFailure(v delivery.Verdict) bool {
return v.State == delivery.VerdictFail && strings.Contains(v.Failed.Words, UnheardWords)
}
// try claims the check, judges it within its bound and writes the verdict and the owed statuses. A check that ends
// because ctx did is left running: the checker takes it again when it starts.
func (ch *Checker) try(ctx context.Context, key string, cs *Case) {
// try claims the check, judges it within its bound under run and writes the verdict and the owed statuses. A check that ends
// because ctx did is left running: the checker takes it again when it starts. A check whose run was ended because no
// candidate holds its heads any more (issue 1890) is ended: its verdict says so, and no status is written.
func (ch *Checker) try(ctx, run context.Context, key string, cs *Case) {
if run.Err() != nil {
return // ended before it was claimed: nothing to write
}
ch.mu.Lock()
first := !ch.tried[cs.Key]
ch.tried[cs.Key] = true
@@ -331,12 +635,47 @@ func (ch *Checker) try(ctx context.Context, key string, cs *Case) {
}
began := ch.now()
ch.writeOwed(ctx, cs, delivery.Verdict{State: delivery.VerdictRunning}, Outcome{})
out := ch.judge(ctx, cs)
out := ch.judge(run, cs)
// An ask the build seat gave up on judged nothing (issue 734): the check is asked again under the same claim, its status
// pending, and after LostAskTries it fails as not heard, judged again when the nodes change.
for try := 1; IsLostAsk(out.Verdict) && run.Err() == nil; try++ {
if try >= LostAskTries {
out.Verdict.Failed.Words = strings.TrimSpace(out.Verdict.Failed.Words + "\nasked " + fmt.Sprint(LostAskTries) +
" times and given up each time: " + UnheardWords)
break
}
ch.logf("%s: %s; asked again (%d of %d)", cs.Key, out.Verdict.Failed.Words, try+1, LostAskTries)
out = ch.judge(run, cs)
}
if ctx.Err() != nil {
return
}
var why superseded
if errors.As(context.Cause(run), &why) {
ch.end(ctx, cs, cl, why)
return
}
v := out.Verdict
v.Group, v.Heads, v.Trunk = cs.Group, cs.Heads, cs.Trunk
v.Group, v.Heads, v.Trunk, v.Beside = cs.Group, cs.Heads, cs.Trunk, nil
for r, c := range out.Beside {
if _, merged := cs.Trunk[r]; !merged && c != "" {
if v.Beside == nil {
v.Beside = map[string]string{}
}
v.Beside[r] = c
}
}
if cs.CarriedFrom != nil {
// A carry (ADR 0548 §5): a pass names the commits it was carried from; a failure names the move it failed on.
switch v.State {
case delivery.VerdictPass:
v.CarriedFrom = carried(cs.CarriedFrom, v)
case delivery.VerdictFail:
if moves := moveWords(cs.CarriedFrom, v); moves != "" {
v.Failed.Words = strings.TrimSpace(v.Failed.Words + "\nfailed on the carry over the move of " + moves + ": the move changed what these tests read")
}
}
}
if v.Took.Queue == 0 && !cs.Queued.IsZero() && began.After(cs.Queued) {
v.Took.Queue = began.Sub(cs.Queued).Milliseconds()
}
@@ -358,17 +697,37 @@ func (ch *Checker) try(ctx context.Context, key string, cs *Case) {
ch.nodesFail[cs.Key] = epoch
ch.mu.Unlock()
}
if ch.movedSince(cs) {
if ch.movedSince(cs) || besideMoved(v.Beside, ch.besideNow()) {
// Judged on a trunk that moved while it ran: the verdict is kept as judged, and the status stays pending until the
// check of the new trunk, which Reconsider starts, ends.
ch.writeOwed(ctx, cs, delivery.Verdict{State: delivery.VerdictRunning}, Outcome{})
ch.logf("%s: %s on a trunk that moved while it ran; the status stays pending", cs.Key, v.State)
ch.owePending(ctx, cs, v)
ch.logf("%s: %s on a trunk that moved while it ran; the status stays pending until it is carried", cs.Key, v.State)
return
}
ch.writeOwed(ctx, cs, v, out)
ch.logf("%s: %s", cs.Key, v.State)
}
// end writes the verdict of a check whose heads no candidate holds any more (issue 1890), by compare-and-set on its claim,
// and says it once in the log with the asks it had put: they stay in the build seats' streams, since an ask once put cannot
// be taken back (buildrun.Asker has no withdraw), and are no longer waited for. No status is written: the heads that
// replaced these have their own check, whose pending status stands.
func (ch *Checker) end(ctx context.Context, cs *Case, cl *claimed, why superseded) {
v := delivery.Verdict{Group: cs.Group, Heads: cs.Heads, Trunk: cs.Trunk, State: delivery.VerdictFail,
Failed: delivery.FailedFirst{Machine: "checker", Words: why.Error()}}
data, err := delivery.EncodeVerdict(v)
if err == nil {
_, err = ch.Verdicts.Update(ctx, cs.Key, data, cl.rev)
}
if err != nil {
ch.logf("%s: %s, but the verdict could not say so: %v", cs.Key, why.Error(), err)
return
}
stage, _ := cs.Progress.Now()
asks := cs.Asks()
ch.logf("%s: %s; it was at %q with %d asks put %v, which are no longer waited for", cs.Key, why.Error(), stage, len(asks), asks)
}
// judge runs the case within the bound. A judge that fails to judge, and a check that ran past its bound, are verdicts of
// failure naming the machine and the step.
func (ch *Checker) judge(ctx context.Context, cs *Case) Outcome {
@@ -467,7 +826,7 @@ func (ch *Checker) Sweep(ctx context.Context) error {
continue
}
cur.State = delivery.VerdictFail
cur.Failed = delivery.FailedFirst{Machine: "checker", Words: "closed at start: left running, and no candidate holds these heads any more, so nothing will check them"}
cur.Failed = delivery.FailedFirst{Machine: "checker", Words: EndedWords + "closed at start: left running, and no candidate holds these heads any more, so nothing will check them"}
data, err := delivery.EncodeVerdict(cur)
if err != nil {
return err
@@ -502,14 +861,37 @@ func (ch *Checker) TrunkMoved(ctx context.Context) {
if err != nil || !found {
continue // never judged: no status to put out of date
}
if v, err := delivery.DecodeVerdict(value); err == nil && sameTrunk(v.Trunk, cs.Trunk) {
v, err := delivery.DecodeVerdict(value)
if err == nil && !ch.stale(v, cs) {
continue
}
if err == nil && v.State == delivery.VerdictPass {
cs.CarriedFrom = judgedOn(v) // the status says it is being carried to the new commit (ADR 0548 §6)
}
ch.writeOwed(ctx, cs, delivery.Verdict{State: delivery.VerdictRunning}, Outcome{})
}
ch.Reconsider(ctx)
}
// owePending owes mesh/check pending on the heads of a verdict judged on commits that moved while it ran: a pass says it is
// being carried to the trunk's head now (ADR 0548 §6), anything else that it is checked again.
func (ch *Checker) owePending(ctx context.Context, judged *Case, v delivery.Verdict) {
next := judged
ckey, err := delivery.CandidatesKey(v.Group)
ch.mu.Lock()
c, ok := ch.known[ckey]
ch.mu.Unlock()
if err == nil && ok {
if now := CaseOf(c, ch.heads()); now != nil && now.Key == judged.Key {
next = now
}
}
if v.State == delivery.VerdictPass {
next.CarriedFrom = judgedOn(v)
}
ch.writeOwed(ctx, next, delivery.Verdict{State: delivery.VerdictRunning}, Outcome{})
}
// movedSince says whether a trunk the case was judged on is no longer the trunk's head.
func (ch *Checker) movedSince(cs *Case) bool {
now := ch.heads()
@@ -3,6 +3,7 @@ package check
import (
"context"
"encoding/json"
"fmt"
"strings"
"sync"
"testing"
@@ -333,7 +334,10 @@ func TestTheStatusIsPendingWhileTheCheckRunsAndEndsSuccessOrFailure(t *testing.T
}
}
func TestSweepClosesARunningKeyNoCandidateHoldsAndLeavesOneItDoes(t *testing.T) {
// No check of heads no candidate holds is left running (issue 1890): at start the sweep ends a key left running; while the
// checker runs, a check whose group moved to other heads is ended at once, unjudged, and writes no status; heads held again
// are checked whole; and a group's check starts only once its heads settled, so a partial set of members is never checked.
func TestHeadsNoCandidateHoldsAreNeverLeftRunning(t *testing.T) {
ctx, j := context.Background(), passJudge()
r := newRig(t, j)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
@@ -351,12 +355,91 @@ func TestSweepClosesARunningKeyNoCandidateHoldsAndLeavesOneItDoes(t *testing.T)
if err := r.ch.Sweep(ctx); err != nil {
t.Fatal(err)
}
if v := r.verdict(t, dead); v.State != delivery.VerdictFail || !strings.Contains(v.Failed.Words, "closed at start") {
if v := r.verdict(t, dead); !IsEnded(v) || !strings.Contains(v.Failed.Words, "closed at start") {
t.Errorf("the dead key = %+v", v)
}
if v := r.verdict(t, live); v.State != delivery.VerdictRunning {
t.Errorf("the live key = %+v, want it left for the check to take over", v)
}
// A member joins while the check of the lone member's heads runs: that check is ended, its judge's run cancelled.
started, ended := make(chan struct{}), make(chan error, 1)
j = &judgeFunc{f: func(run context.Context, _ *Case, call int) (Outcome, error) {
if call == 1 {
close(started)
<-run.Done()
ended <- context.Cause(run)
return Outcome{}, run.Err()
}
return Outcome{Verdict: passing()}, nil
}}
r = newRig(t, j)
var logged []string
var logMu sync.Mutex
r.ch.Logf = func(f string, a ...any) { logMu.Lock(); logged = append(logged, fmt.Sprintf(f, a...)); logMu.Unlock() }
alone, both := cand, candidateOf(t, "feat/x", pullOf(repoA, 7, head1), pullOf(repoSDK, 3, head2))
r.ch.Candidate(ctx, "feat%2Fx", alone)
<-started
r.ch.Candidate(ctx, "feat%2Fx", both)
var bothC delivery.Candidate
_ = json.Unmarshal(both, &bothC)
select {
case err := <-ended:
if err == nil || !strings.Contains(err.Error(), "superseded by "+GateHeads(bothC)) {
t.Errorf("the old check's run ended with %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("the check of the heads the candidate left still runs")
}
r.ch.Wait()
if v := r.verdict(t, live); !IsEnded(v) || v.Failed.Words != EndedWords+"superseded by "+GateHeads(bothC) {
t.Errorf("the superseded verdict = %+v", v.Failed)
}
if v := r.verdict(t, keyOf(t, "feat/x", both)); v.State != delivery.VerdictPass {
t.Errorf("the group's verdict = %s", v.State)
}
okey, _ := delivery.CheckerOwedKey(repoA, 7, head1)
if raw, _, _, _ := r.owed.Get(ctx, okey); !strings.Contains(string(raw), `"success"`) {
t.Errorf("the head's status = %s, want the group's success, never the ended check's", raw)
}
logMu.Lock()
said := 0
for _, l := range logged {
if strings.Contains(l, "ended: superseded") {
said++
}
}
logMu.Unlock()
if said != 1 {
t.Errorf("the end was said %d times: %v", said, logged)
}
// The member leaves again: the ended heads are held again and checked whole.
r.ch.Candidate(ctx, "feat%2Fx", alone)
r.ch.Wait()
if v := r.verdict(t, live); v.State != delivery.VerdictPass || j.count() != 3 {
t.Errorf("the heads held again = %s after %d judgings, want a pass of a third", v.State, j.count())
}
// Members announced one after another: only the settled, whole set is checked.
j = passJudge()
r = newRig(t, j)
r.ch.Settle = time.Hour
sdk := candidateOf(t, "feat/y", pullOf(repoSDK, 4, head2))
whole := candidateOf(t, "feat/y", pullOf(repoSDK, 4, head2), pullOf(repoA, 8, head1))
r.ch.Candidate(ctx, "feat%2Fy", sdk)
r.ch.Candidate(ctx, "feat%2Fy", whole)
r.ch.Reconsider(ctx) // a trunk move or a turn while the heads settle starts nothing
if j.count() != 0 {
t.Fatalf("judged %d times before the heads settled", j.count())
}
r.ch.settleNow()
r.ch.Wait()
if _, _, found, _ := r.verdicts.Get(ctx, keyOf(t, "feat/y", sdk)); found || j.count() != 1 {
t.Errorf("the partial set has a verdict (%v) or the judge ran %d times, want the whole set alone", found, j.count())
}
if v := r.verdict(t, keyOf(t, "feat/y", whole)); v.State != delivery.VerdictPass {
t.Errorf("the whole set = %s", v.State)
}
}
func TestTheForgeBridgeIsToldOnceForEachChangeOfAStatus(t *testing.T) { // 1683 (the checker's half of AC5)
@@ -28,6 +28,7 @@ func TestAPassingCheckTakesTheAssignersComposition(t *testing.T) {
f := &fakeCompositions{v: delivery.CompositionVerdict{Group: "feat/x", Heads: "0123456789ab", State: delivery.CompositionPass, SDK: "v0.1.14-x", Machines: []string{"g14"}, Took: 40}}
b.ex.Compositions = f
cs := caseOf(t, pullOf(repoA, 7, head1))
cs.FPSet = "fps-of-the-candidate"
v := b.judge(t, cs).Verdict
if v.State != delivery.VerdictPass || v.Composition == nil || v.Composition.State != delivery.CompositionPass || v.Composition.SDK != "v0.1.14-x" || v.Took.Composition != 40 {
t.Fatalf("verdict = %+v, composition %+v", v, v.Composition)
@@ -38,6 +39,15 @@ func TestAPassingCheckTakesTheAssignersComposition(t *testing.T) {
if !compositionMentions(v, "passed") {
t.Errorf("the line says %q", compositionWords(v.Composition, v.Took.Composition))
}
if v.Composition.FPSet == "" || v.Composition.FPSet != cs.FPSet {
t.Errorf("the composition names the fingerprint set %q, want the candidate's %q (ADR 0548 §5)", v.Composition.FPSet, cs.FPSet)
}
// ADR 0548 §5: a carry whose fingerprints did not change keeps the composition it passed and asks the assigner nothing.
carry := caseOf(t, pullOf(repoA, 7, head1))
carry.CarriedFrom, carry.Composition = map[string]string{repoA: trunkA2}, v.Composition
if kept := b.judge(t, carry).Verdict; kept.Composition == nil || kept.Composition.SDK != "v0.1.14-x" || len(f.asked) != 1 {
t.Errorf("the carry's composition is %+v and the assigner was asked %d times; want the kept one and no second ask", kept.Composition, len(f.asked))
}
}
func compositionMentions(v delivery.Verdict, word string) bool {
@@ -0,0 +1,90 @@
package check
import (
"encoding/json"
"fmt"
"strings"
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// failingResult is a level's test result in which the named tests failed, each in its own package.
func failingResult(t *testing.T, tests int, failed ...Failure) json.RawMessage {
t.Helper()
raw, err := json.Marshal(Tested{Level: delivery.LevelIntegration, Tests: tests, Elapsed: 2, Failed: failed, Passed: len(failed) == 0})
if err != nil {
t.Fatal(err)
}
return raw
}
func judgeLevel(t *testing.T, tests ...json.RawMessage) (*delivery.Verdict, *state) {
t.Helper()
st := &state{j: &Examiner{}, c: &Case{}, levels: map[string]*acc{delivery.LevelUnit: {}, delivery.LevelIntegration: {}}, nodes: map[string]bool{}}
v, _ := st.level(t.Context(), &target{pull: Pull{Repository: "novox/x"}}, delivery.LevelAnswer{Level: delivery.LevelIntegration, Tests: tests})
return v, st
}
// Issue 1921: two failing tests in two packages are two entries of the verdict's Failures and two lines of a try's
// failed list, each with its own first line (its cause, issue 1901), whether the build module answered them in one test
// result (one go test run over both packages) or in two (the changed packages first, then the rest); every result is
// counted, the ones after a failure too; and Failed keeps the first one's words whole.
func TestTwoFailingTestsInTwoPackagesAreTwoEntries(t *testing.T) {
a := Failure{Package: "x/a", Test: "TestA", Cause: "a_test.go:9: want 1, got 2", Output: " a_test.go:3: logged\n a_test.go:9: want 1, got 2\n--- FAIL: TestA (0.00s)"}
b := Failure{Package: "x/b", Test: "TestB", Cause: "b_test.go:4: boom", Output: " b_test.go:4: boom\n--- FAIL: TestB (0.00s)"}
for name, tests := range map[string][]json.RawMessage{
"one result": {failingResult(t, 10, a, b)},
"two results": {failingResult(t, 4, a), failingResult(t, 3), failingResult(t, 3, b)},
} {
v, st := judgeLevel(t, tests...)
if v == nil {
t.Fatalf("%s: the level passed", name)
}
want := []string{"integration: x/a.TestA: a_test.go:9: want 1, got 2", "integration: x/b.TestB: b_test.go:4: boom"}
if got := failedLines(*v); strings.Join(got, "\n") != strings.Join(want, "\n") {
t.Errorf("%s: failed lines\n%s\nwant\n%s", name, strings.Join(got, "\n"), strings.Join(want, "\n"))
}
if len(v.Failures) != 2 || v.Failed.Test != "integration: x/a.TestA" || !strings.Contains(v.Failed.Words, "a_test.go:3: logged") {
t.Errorf("%s: verdict %+v", name, v)
}
if run := st.levels[delivery.LevelIntegration].run; run != 10 {
t.Errorf("%s: %d tests counted, want all 10", name, run)
}
if _, err := delivery.EncodeVerdict(delivery.Verdict{Group: "g", Heads: "0123456789ab", State: v.State, Failed: v.Failed, Failures: v.Failures}); err != nil {
t.Errorf("%s: the verdict is not written: %v", name, err)
}
}
}
// Issue 1921: past delivery.MaxFailures the list names the first fifty and how many more, the status line says how many
// failed, and each entry's line is cut to MaxFailureLine.
func TestFailuresAreCappedWithHowManyMore(t *testing.T) {
var fs []Failure
for i := range 60 {
fs = append(fs, Failure{Package: fmt.Sprintf("x/p%d", i), Test: "TestT", Cause: strings.Repeat("y", 500)})
}
v, _ := judgeLevel(t, failingResult(t, 60, fs...))
if len(v.Failures) != delivery.MaxFailures+1 || v.Failures[delivery.MaxFailures].Test != "and 10 more" {
t.Fatalf("%d entries, last %+v", len(v.Failures), v.Failures[len(v.Failures)-1])
}
if len(v.Failures[0].Words) > MaxFailureLine {
t.Errorf("an entry's line is %d bytes", len(v.Failures[0].Words))
}
if n := failedCount(*v); n != 60 {
t.Errorf("failedCount = %d, want 60", n)
}
cs := &Case{Pulls: []Pull{{Repository: "novox/x", Head: "0123456789abcdef", Trunk: "fedcba9876543210"}}}
if line := Line(*v, cs, cs.Pulls[0], nil, nil, nil); !strings.HasPrefix(line, "failed (60 tests): integration: x/p0.TestT") {
t.Errorf("line %q", line)
}
}
// A verdict that names no list (a conflict, a refused step, a verdict written before issue 1921) is one failed line, as
// before.
func TestAVerdictWithoutFailuresIsOneLine(t *testing.T) {
v := delivery.Verdict{State: delivery.VerdictFail, Failed: delivery.FailedFirst{Machine: "g14", Words: "files in conflict: go.mod\nmore"}}
if got := failedLines(v); len(got) != 1 || got[0] != "g14: files in conflict: go.mod" {
t.Fatalf("%q", got)
}
}
@@ -116,6 +116,9 @@ type bench struct {
fetch Fetched
digest map[string]Digested // by level
test func(c stepCall) (any, error)
// declines is how many asks naming needs a node lacking them declines before shanks, which has every need, takes
// one (issue 890); an ask naming none runs on shanks, which then has none.
declines int
}
// refsOf reads the siblings of a fetch's arguments, as they arrive in an ask (JSON) or as the checker gave them.
@@ -151,8 +154,17 @@ func (b *bench) answer(a delivery.StepAsk) delivery.Step {
if module == buildrun.GoBuildModule {
steps.Digest = tool("digest")
}
out, err := buildstep.RunAsk(context.Background(), steps, a)
began := time.Date(2026, 10, 21, 19, 0, 0, 0, time.UTC)
if len(a.Needs) > 0 {
if b.declines > 0 {
b.declines--
result, _ := json.Marshal(delivery.StepDeclined{Declined: true, Needs: a.Needs, Lacks: a.Needs})
return delivery.Step{ID: a.ID, Tool: a.Kind + " ask", Node: "ace", State: delivery.StepRefused, Started: began, Ended: began,
Reason: "declined", Result: result}
}
steps.Has = func(string) bool { return true }
}
out, err := buildstep.RunAsk(context.Background(), steps, a)
step := delivery.Step{ID: a.ID, Tool: a.Kind + " ask", Node: node, State: delivery.StepPassed, Started: began, Ended: began.Add(2 * time.Second)}
if err != nil {
step.State, step.Reason = delivery.StepFailed, err.Error()
@@ -276,6 +288,49 @@ func TestOnlyThePackageWithoutAKeptVerdictRuns(t *testing.T) { // 1288/AC1
if u := out.Verdict.Levels[delivery.LevelUnit]; u.Run != 2 || u.Kept != 40 {
t.Errorf("unit = %+v, want 2 run and 40 kept", u)
}
// Issue 890/AC5: a package needing node-build-node is left unmet where the node lacks it; the check keeps the passes
// and asks the integration level again for a node with it, puts it again when a node declines it, and that node tests
// the package alone, its kept passes not counted twice.
// fixture/outside has no kept verdict after its pass (it reads outside itself: its digest moves every call), so a
// placed ask that ran more than its unmet packages would run and count it again.
b = newBench(t)
b.declines = 1
outside := 0
b.steps.handlers["build_go_digest"] = func(c stepCall) (any, error) {
d := echoBeside(b, b.digest[c.args["level"].(string)])
if d.Level == "integration" {
outside++
d.Packages = map[string]string{"fixture/slow": digB, "fixture/node": digC, "fixture/outside": fmt.Sprintf("%064x", outside)}
}
return d, nil
}
b.digest[delivery.LevelIntegration] = Digested{Level: "integration",
Needs: []delivery.PackageNeeds{{Package: "fixture/node", Needs: []string{delivery.NeedNodeBuildNode}}}}
out = b.judge(t, caseOf(t, pullOf(repoA, 7, head1)))
asks := b.asks.Asks()
if out.Verdict.State != delivery.VerdictPass || len(asks) != 3 || len(asks[0].Needs) != 0 {
t.Fatalf("verdict %+v, asks %+v: want a pass of three asks, the first naming no need", out.Verdict, asks)
}
for _, a := range asks[1:] {
if strings.Join(a.Needs, ",") != delivery.NeedNodeBuildNode || strings.Join(a.Check.Levels, ",") != delivery.LevelIntegration || strings.Join(a.Check.Packages, ",") != "fixture/node" {
t.Errorf("a placed ask %+v, want the integration level needing node-build-node", a)
}
}
tests = b.steps.tool("test")
if last := tests[len(tests)-1]; len(tests) != 3 || strings.Join(packagesOf(last), ",") != "fixture/node" || last.args["level"] != "integration" {
t.Errorf("test calls = %+v, want the unit level, the integration level of fixture/slow and fixture/outside, then fixture/node alone", tests)
}
slow := captured[Tested](t, "go-test-slow")
if i := out.Verdict.Levels[delivery.LevelIntegration]; i.Run != 2*slow.Tests || i.Kept != 0 {
t.Errorf("integration = %+v, want %d run and nothing counted kept", i, 2*slow.Tests)
}
// And where no node has the need, the check fails once the declines outlast the bound, naming the need.
b.kept, b.declines = newMemKV(), 1000
b.ex.Kept, b.ex.DeclineBound = Kept{Store: b.kept}, 200*time.Millisecond
if v := b.judge(t, caseOf(t, pullOf(repoA, 7, head1))).Verdict; v.State != delivery.VerdictFail || !strings.Contains(fmt.Sprint(v), "needs a node with node-build-node") {
t.Errorf("with no node having the need: %+v", v)
}
}
func TestAReAskRunsOnlyWhatFailedAndWhatChanged(t *testing.T) { // 1288/AC2, AC3
@@ -341,7 +396,9 @@ func TestANodePackageIsNeverKeptAndRunsEveryCheck(t *testing.T) { // 1288/AC4, 1
}
}
func TestTheChangesOwnFailingTestRunsFirstAndNothingElseRuns(t *testing.T) { // 1289/AC1, AC2
// 1289/AC1, AC2 as issue 1921 changed it: the change's own failing test runs first, the level's other packages still run
// after it (so one check names every failure), and no later level runs.
func TestTheChangesOwnFailingTestRunsFirstThenTheRestAndNoLaterLevel(t *testing.T) {
b := newBench(t)
b.fetch.Changed = []string{"modules/m/internal/b/b_test.go"}
b.digest[delivery.LevelUnit] = Digested{Level: "unit", Packages: map[string]string{"m/internal/a": digA, "m/internal/b": digB}}
@@ -357,8 +414,8 @@ func TestTheChangesOwnFailingTestRunsFirstAndNothingElseRuns(t *testing.T) { //
t.Fatalf("verdict = %+v", out.Verdict)
}
tests := b.steps.tool("test")
if len(tests) != 1 || strings.Join(packagesOf(tests[0]), ",") != "m/internal/b" {
t.Fatalf("test calls = %+v, want only m/internal/b, once", tests)
if len(tests) != 2 || strings.Join(packagesOf(tests[0]), ",") != "m/internal/b" || strings.Join(packagesOf(tests[1]), ",") != "m/internal/a" {
t.Fatalf("test calls = %+v, want m/internal/b, then m/internal/a, each once", tests)
}
if n := len(b.steps.tool("digest")); n != 1 {
t.Errorf("%d digest calls, want the unit level's alone (the check stopped)", n)
@@ -618,9 +675,8 @@ func TestNothingIsKeptForAWorkspaceGivenSiblingsWhenTheDigestDoesNotSayItCoversT
// Beside): every check now places the change's own repository at its trunk (issue 1887 A12).
func echoBeside(b *bench, d Digested) Digested {
var lines []string
for _, f := range b.steps.tool("fetch") {
refs, _ := refsOf(f.args["beside"]), true
for _, r := range refs {
if fetches := b.steps.tool("fetch"); len(fetches) > 0 { // the workspace digested is the last fetch's (issue 890: a placed ask fetches again)
for _, r := range refsOf(fetches[len(fetches)-1].args["beside"]) {
lines = append(lines, r.Repository+" "+r.Commit)
}
}
@@ -42,7 +42,7 @@ func TestTheStatusLineNamesTheFailureNotTheFirstLoggedLine(t *testing.T) { // 19
f := fatalAfterLog()
f.Cause = ""
v = levelFailure(&target{}, "unit", Tested{Failed: []Failure{f}})
if !strings.HasPrefix(v.Failed.Words, "agreement_catalogue_test.go:130") {
if !strings.HasPrefix(afterHid(v.Failed.Words), "agreement_catalogue_test.go:130") {
t.Errorf("words = %q", v.Failed.Words)
}
}
@@ -56,18 +56,45 @@ func TestWordsLongerThanTheBoundKeepTheirLastLines(t *testing.T) { // 1901/AC1
_ = delivery.VerdictFail
}
// An older build module sends no "cause": its answer is still accepted and the verdict's words are its output, as before.
func TestAnOlderBuildModulesFailureIsAcceptedWithItsWholeOutput(t *testing.T) { // 1901/AC3
// An older build module sends no "cause": its answer is still accepted and the verdict's words are its output, as before
// (1901/AC3). A failure that hides its cause (none, or a passing line) is said loudly first in the words and in the status
// line, naming the build module and the version it says it is, and the level still fails; a failure that keeps the
// contract is worded as before (1901/AC8).
func TestAnOlderBuildModulesFailureIsAcceptedWithItsWholeOutput(t *testing.T) { // 1901/AC3, AC8
old := `{"package":"p","test":"TestX","output":"x_test.go:1: logged\nx_test.go:2: the old kept lines"}`
var f Failure
if err := json.Unmarshal([]byte(old), &f); err != nil || f.Cause != "" {
t.Fatalf("1901/AC3: err = %v, cause = %q", err, f.Cause)
}
v := levelFailure(&target{}, "unit", Tested{Failed: []Failure{f}})
if v.Failed.Words != "x_test.go:1: logged\nx_test.go:2: the old kept lines" {
if afterHid(v.Failed.Words) != "x_test.go:1: logged\nx_test.go:2: the old kept lines" {
t.Errorf("1901/AC3: words = %q", v.Failed.Words)
}
if got := firstFailed(*v); !strings.Contains(got, "logged") || strings.Contains(got, "\n") {
t.Errorf("1901/AC3: status = %q", got)
if got := firstFailed(*v); !strings.Contains(got, "the build module hid the failure") || strings.Contains(got, "\n") {
t.Errorf("1901/AC3, AC8: status = %q", got)
}
by := "mesh-build-node with mesh-sdk v0.1.14-0.20261026184535-e40dd0f56a78"
for _, c := range []struct {
name, module, by, cause, output, loud string
}{
{"no cause", "mesh-build-node", by, "", "test/b.test.ts: something", "the build module hid the failure (mesh-build-node at mesh-sdk v0.1.14-0.20261026184535-e40dd0f56a78): it sent no cause"},
{"a passing line as its cause", "mesh-build-node", by, "\u2714 the same merge has the same id (8.006913ms)", "", "the build module hid the failure (mesh-build-node at mesh-sdk v0.1.14-0.20261026184535-e40dd0f56a78): its cause is a passing line"},
{"from before the contract", "mesh-build-node", "", "", "\u2714 the same merge has the same id (8.006913ms)", "the build module hid the failure (mesh-build-node at a version before the failure contract): it sent no cause"},
{"a real cause", "mesh-build-node", by, "one plus one is not three", "not ok 3 - fails", ""},
} {
t.Run(c.name, func(t *testing.T) {
v := levelFailure(&target{module: c.module}, "unit", Tested{By: c.by, Failed: []Failure{{Package: "modules/gitea", Test: "fails", Cause: c.cause, Output: c.output}}})
if v.State != delivery.VerdictFail {
t.Errorf("1901/AC8: the verdict is %q, want it failed", v.State)
}
line := firstFailed(*v)
switch {
case c.loud == "" && strings.Contains(v.Failed.Words, "hid the failure"):
t.Errorf("1901/AC8: a failure that keeps the contract is said hidden: %q", v.Failed.Words)
case c.loud != "" && (!strings.HasPrefix(v.Failed.Words, c.loud) || !strings.Contains(line, c.loud) || !strings.Contains(v.Failures[0].Words, "hid the failure")):
t.Errorf("1901/AC8: words %q, status %q, want both to start with %q", v.Failed.Words, line, c.loud)
}
})
}
}
@@ -4,7 +4,6 @@ import (
"context"
"strings"
"testing"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
@@ -88,29 +87,3 @@ func TestATrunkMoveMarksEveryOpenPullRequestIntoItPendingAtOnce(t *testing.T) {
t.Errorf("mesh-sdk#13, alone into a trunk that did not move, is %q, want its success to stand", s)
}
}
func TestACheckThatEndsAfterItsTrunkMovedOwesPendingNotItsVerdict(t *testing.T) { // gate (a): a pass on an old trunk is never shown
ctx := context.Background()
var r *rig
release, judged := make(chan struct{}), make(chan struct{})
j := &judgeFunc{f: func(ctx context.Context, c *Case, call int) (Outcome, error) {
if call == 1 {
r.setTrunks(t, map[string]string{repoA: trunkA2, repoSDK: trunkS}) // the trunk moves while it runs
defer close(judged)
} else {
<-release // the check asked again for the new trunk waits in the stream meanwhile
}
return Outcome{Verdict: passing()}, nil
}}
r = newRig(t, j)
r.ch.Candidate(ctx, "feat%2Fx", candidateOf(t, "feat/x", pullOf(repoA, 7, head1)))
<-judged
for deadline := time.Now().Add(2 * time.Second); owedState(t, r, repoA, 7, head1) == "" && time.Now().Before(deadline); time.Sleep(time.Millisecond) {
}
time.Sleep(20 * time.Millisecond)
if s := owedState(t, r, repoA, 7, head1); s != "pending" {
t.Errorf("the status is %q, want pending: the pass was judged on a trunk that moved", s)
}
close(release)
r.ch.Wait()
}
@@ -80,6 +80,9 @@ var askLanguage = map[string]string{buildrun.GoBuildModule: delivery.AskGo, buil
// stopping at the first failure; it judges the step's answer as it judged the tools' answers.
type Examiner struct {
Asks Asks
// DeclineBound is how long a placed ask may go on being declined, from its first decline, before the check fails
// saying no node with its need took it (issue 890, ADR 0552); DefaultDeclineBound when zero.
DeclineBound time.Duration
// Toolchains are the nodes' toolchains: a kept verdict names the toolchain the nodes agree on.
Toolchains ToolchainEntries
Kept Kept
@@ -203,12 +206,15 @@ type state struct {
nodes map[string]bool
clone time.Duration
slow []string
// beside are the siblings read at their trunk's head, by repository (ADR 0548 §6).
beside map[string]string
}
// Judge runs the case (it is the Judge of a Checker). A failure of the change, of a merge or of a step is a verdict; an error is only a case that cannot
// be judged at all.
func (j *Examiner) Judge(ctx context.Context, c *Case) (Outcome, error) {
st := &state{j: j, c: c, levels: map[string]*acc{delivery.LevelUnit: {}, delivery.LevelIntegration: {}}, nodes: map[string]bool{}}
st := &state{j: j, c: c, levels: map[string]*acc{delivery.LevelUnit: {}, delivery.LevelIntegration: {}}, nodes: map[string]bool{},
beside: map[string]string{}}
if v := st.judgeAll(ctx); v != nil {
return st.finish(ctx, *v), nil
}
@@ -230,6 +236,13 @@ func (st *state) finish(ctx context.Context, v delivery.Verdict) Outcome {
// fails the check, naming the machine it failed at. A try is not composed (it has no candidate); a composition that
// cannot be had leaves it not run, said; with no assigner running it is not judged, said at once.
func (st *state) composed(ctx context.Context, v *delivery.Verdict) {
if st.c.Composition != nil && st.c.Try == nil {
// A carry whose candidate's fingerprints did not change keeps the composition it passed (ADR 0548 §5).
kept := *st.c.Composition
v.Composition = &kept
st.c.note("the composition is kept: no fingerprint of a module it composed changed")
return
}
if st.j.Compositions == nil || st.c.Try != nil {
return
}
@@ -243,6 +256,9 @@ func (st *state) composed(ctx context.Context, v *delivery.Verdict) {
return
}
withComposition(v, comp)
if v.Composition != nil && v.Composition.State != delivery.CompositionNotRun {
v.Composition.FPSet = st.c.FPSet // a carry keeps it while the fingerprints stay the same (ADR 0548 §5)
}
}
// outcome fills the verdict's levels, times, machines and composition from what ran.
@@ -282,6 +298,9 @@ func (st *state) outcome(v delivery.Verdict) Outcome {
}
}
out := Outcome{Verdict: v, Slowest: slow}
if len(st.beside) > 0 {
out.Beside = st.beside
}
for _, t := range st.targets {
if t.judged != "" {
if out.Judged == nil {
@@ -343,6 +362,7 @@ func (st *state) judgeAll(ctx context.Context) *delivery.Verdict {
if len(beside) > 0 {
args["beside"] = beside
t.beside = besideLines(beside)
besideAtTrunk(st.c, beside, st.beside)
}
if p.Repository == catalogueRepository {
t.runner = RunnerWorkspace // used by the build module when the merge changes it (ADR 0535 §1)
@@ -364,11 +384,25 @@ func (st *state) judgeAll(ctx context.Context) *delivery.Verdict {
return nil
}
// DefaultDeclineBound is an Examiner's DeclineBound when it names none: longer than a check runs, so a node with the
// need that is busy is waited for, and short enough that a mesh where no node has it says so.
const DefaultDeclineBound = 30 * time.Minute
// ask puts the target's check ask, waits for the step of the node that pulled it and judges its answer: the fetch, the
// pins of its go.mod files, then each level's digest, kept verdicts and tests.
// pins of its go.mod files, then each level's digest, kept verdicts and tests. When the answer leaves packages unmet,
// their tests needing what that node lacks (issue 890), the passes are kept and the levels from the first such one are
// asked again of a node that has it (placed).
func (st *state) ask(ctx context.Context, t *target, merged *[]Fetched) *delivery.Verdict {
return st.askNeeding(ctx, t, merged, st.levelsOf(), nil, nil)
}
// askNeeding is ask with the levels to run, what the node must have and the packages it tests: none for the first ask
// of a target; for the placed one the unmet packages' needs and those packages alone, so no package the first ask judged
// is run or counted again (ADR 0552 §4).
func (st *state) askNeeding(ctx context.Context, t *target, merged *[]Fetched, levels, needs, packages []string) *delivery.Verdict {
p := t.pull
job := &delivery.CheckJob{Fetch: t.fetchArgs, Levels: st.levelsOf(), Kept: true}
placed := len(needs) > 0
job := &delivery.CheckJob{Fetch: t.fetchArgs, Levels: levels, Kept: true, Packages: packages}
if t.runner != "" {
job.Runner, job.RunnerWhen = t.runner, buildGoDir
}
@@ -380,17 +414,53 @@ func (st *state) ask(ctx context.Context, t *target, merged *[]Fetched) *deliver
kind = delivery.AskRecheck // a re-judge after a trunk move goes after every new head's check (ADR 0536 §4)
}
a := delivery.StepAsk{ID: buildstep.NewID(), Kind: kind, Language: askLanguage[t.module], Asker: delivery.CheckerModule,
At: st.j.now().UTC(), Check: job}
st.c.Progress.Set(fmt.Sprintf("the %s ask %s of %s#%d", kind, a.ID, p.Repository, p.Number), "")
if err := st.j.Asks.Put(ctx, a); err != nil {
return fail(t.module, fmt.Sprintf("the check of %s#%d could not be asked: %v", p.Repository, p.Number, err))
}
step, err := st.j.Asks.Wait(ctx, a)
if err != nil {
if ctx.Err() != nil {
return fail(t.module, ctx.Err().Error())
At: st.j.now().UTC(), Check: job, Needs: needs}
var step delivery.Step
var firstDecline time.Time
for {
what := fmt.Sprintf("the %s ask %s of %s#%d", kind, a.ID, p.Repository, p.Number)
if placed {
what += " for a node with " + strings.Join(needs, ", ")
}
return fail(t.module, fmt.Sprintf("the check of %s#%d: %v", p.Repository, p.Number, err))
st.c.Progress.Set(what, "")
st.c.Asked(a.ID)
if err := st.j.Asks.Put(ctx, a); err != nil {
return fail(t.module, fmt.Sprintf("the check of %s#%d could not be asked: %v", p.Repository, p.Number, err))
}
waiting := ctx
if watch := st.c.Watch; watch != nil {
id := a.ID
watch(id, nil) // put: a try names it at once, not only when it ends (issue 1890)
waiting = buildrun.WithStepSeen(ctx, func(s delivery.Step) { watch(id, &s) })
}
var err error
step, err = st.j.Asks.Wait(waiting, a)
if err != nil {
if ctx.Err() != nil {
return fail(t.module, ctx.Err().Error())
}
return fail(t.module, fmt.Sprintf("the check of %s#%d: %v", p.Repository, p.Number, err))
}
d, declined := delivery.DeclinedOf(step)
if !declined {
break
}
// A node that lacks the need dropped the ask (issue 890): it is put again, for a node that has it to take; the
// node that declined looks at the stream again only after its next pull's wait.
now := st.j.now()
if firstDecline.IsZero() {
firstDecline = now
}
bound := st.j.DeclineBound
if bound <= 0 {
bound = DefaultDeclineBound
}
if now.Sub(firstDecline) > bound {
return fail(step.Node, fmt.Sprintf("the check of %s#%d needs a node with %s, and none took it in %s: every node that pulled it lacks it (last %s)",
p.Repository, p.Number, strings.Join(needs, ", "), bound, step.Node))
}
st.j.logf("%s#%d: the ask %s was declined on %s, which lacks %s; asked again", p.Repository, p.Number, a.ID, step.Node, strings.Join(d.Lacks, ", "))
a.ID, a.At = buildstep.NewID(), now.UTC()
}
// The fetch and the digests are what of the ask's time its levels' running did not take (Took.Clone).
took := time.Duration(0)
@@ -414,18 +484,31 @@ func (st *state) ask(ctx context.Context, t *target, merged *[]Fetched) *deliver
if st.c.Try == nil {
st.j.writePreview(ctx, p, append(*merged, t.fetched))
}
return fail(step.Node, fmt.Sprintf("%s#%d does not merge into the trunk: files in conflict: %s", p.Repository, p.Number, strings.Join(t.fetched.Conflicts, ", ")))
return fail(step.Node, fmt.Sprintf("%s#%d %s%s", p.Repository, p.Number, ConflictWords, strings.Join(t.fetched.Conflicts, ", ")))
}
t.changed = t.fetched.Changed
if !changesBuildGo(p.Repository, t.changed) {
t.runner = "" // the build module used no runner: the merge does not change it
}
*merged = append(*merged, t.fetched)
if v := st.checkPinsOf(ctx, t); v != nil {
return v
if !placed { // the placed ask fetched the same merge: its preview and pins were read from the first
*merged = append(*merged, t.fetched)
if v := st.checkPinsOf(ctx, t); v != nil {
return v
}
if v := st.checkBinariesOf(t); v != nil {
return v
}
}
ran := 0.0
for _, la := range answer.Levels {
if placed {
if len(la.Unmet) > 0 {
st.clone += max(0, took-time.Duration(ran*float64(time.Second)))
return fail(step.Node, fmt.Sprintf("the %s level of %s#%d left %s untested on %s, which was asked for a node with %s",
la.Level, p.Repository, p.Number, unmetNames(la.Unmet), step.Node, strings.Join(needs, ", ")))
}
la.Kept = nil // it names its packages and takes no other's kept verdict; this is only a guard against an older module
}
v, elapsed := st.level(ctx, t, la)
ran += elapsed
if v != nil {
@@ -434,12 +517,52 @@ func (st *state) ask(ctx context.Context, t *target, merged *[]Fetched) *deliver
}
}
st.clone += max(0, took-time.Duration(ran*float64(time.Second)))
if unmet := delivery.UnmetNeeds(answer.Levels); len(unmet) > 0 && !placed {
// Every level passed, its passes kept, and some packages were left to a node with what their tests need: the levels
// from the first that left one are asked of such a node, which takes the kept passes and tests those (issue 890).
var again, packages []string
seen := map[string]bool{}
for _, la := range answer.Levels {
if len(again) > 0 || len(la.Unmet) > 0 {
again = append(again, la.Level)
}
for _, u := range la.Unmet {
if !seen[u.Package] {
seen[u.Package] = true
packages = append(packages, u.Package)
}
}
}
st.j.logf("%s#%d: %s on %s left untested; asked of a node with %s", p.Repository, p.Number, unmetOf(answer.Levels), step.Node, strings.Join(unmet, ", "))
return st.askNeeding(ctx, t, merged, again, unmet, packages)
}
return nil
}
// unmetNames names the unmet packages and their needs.
func unmetNames(unmet []delivery.PackageNeeds) string {
var out []string
for _, u := range unmet {
out = append(out, u.Package+" (needs "+strings.Join(u.Needs, ", ")+")")
}
return strings.Join(out, ", ")
}
// unmetOf names the unmet packages of every level.
func unmetOf(levels []delivery.LevelAnswer) string {
var out []string
for _, la := range levels {
if len(la.Unmet) > 0 {
out = append(out, "at "+la.Level+" "+unmetNames(la.Unmet))
}
}
return strings.Join(out, "; ")
}
// level judges one level of a check ask's answer as the tools' answers were judged: the digest keys the kept verdicts
// of the packages that passed, the kept ones taken are counted, and the first test result that did not pass is the
// verdict. It answers the level's running time (its tests' and setup's).
// of the packages that passed, the kept ones taken are counted, and every test result is read and counted, also after
// one that did not pass, so the verdict names every failed test of the level (issue 1921). It answers the level's
// running time (its tests' and setup's).
func (st *state) level(ctx context.Context, t *target, la delivery.LevelAnswer) (*delivery.Verdict, float64) {
level := la.Level
a := st.levels[level]
@@ -461,6 +584,7 @@ func (st *state) level(ctx context.Context, t *target, la delivery.LevelAnswer)
a.kept += k.Tests
}
elapsed := 0.0
var failedResults []Tested
for _, raw := range la.Tests {
var res Tested
if err := json.Unmarshal(raw, &res); err != nil {
@@ -489,9 +613,12 @@ func (st *state) level(ctx context.Context, t *target, la delivery.LevelAnswer)
}
}
if !res.Passed {
return levelFailure(t, level, res), elapsed
failedResults = append(failedResults, res)
}
}
if len(failedResults) > 0 {
return levelFailure(t, level, failedResults...), elapsed
}
return nil, elapsed
}
@@ -536,6 +663,12 @@ const MaxWords = 4096
// no cause told apart (an older build module) the cause is the output's first "panic:", "fatal error:" or "signal:"
// line, read before the output is cut, or the words say the tests passed and the program failed after them.
func failureWords(f Failure) string {
return failureWordsAfter(f, "")
}
// failureWordsAfter is failureWords with the line loud put first, within the same MaxWords: what the checker itself says of
// the failure (that the build module hid it), which the status line then names.
func failureWordsAfter(f Failure, loud string) string {
out, passed := dropSummary(strings.TrimSpace(f.Output))
cause := strings.TrimSpace(f.Cause)
if cause == "" {
@@ -551,11 +684,14 @@ func failureWords(f Failure) string {
if len(cause) > MaxCause {
cause = strings.ToValidUTF8(cause[:MaxCause], "") + "..."
}
// The words keep MaxWords exactly: the cause and its newline, the marker and the kept lines (review round 1).
// The words keep MaxWords exactly: the loud line, the cause and their newlines, the marker and the kept lines.
budget := MaxWords
if cause != "" {
budget -= len(cause) + 1
}
if loud != "" {
budget -= len(loud) + 1
}
if len(out) > budget {
const marker = "... "
tail := out[len(out)-(budget-len(marker)):]
@@ -568,9 +704,31 @@ func failureWords(f Failure) string {
if cause != "" && !strings.HasPrefix(firstLine(out), cause) {
out = strings.TrimRight(cause+"\n"+out, "\n")
}
if loud != "" {
out = strings.TrimRight(loud+"\n"+out, "\n")
}
return out
}
// hidWords is the checker's loud line for a failure the build module sent hiding its cause (issue 1901, its thorough
// fix): with no cause, or a passing line as its cause, under the failure contract (buildstep.Hidden). It names the build
// module and the version it says it is (Tested.By, "<module> with mesh-sdk <version>"); one from before the contract
// says nothing of itself. "" when the failure keeps the contract. The check fails either way: only its words change.
func hidWords(module, by string, f Failure) string {
why := buildstep.Hidden(f)
if why == "" {
return ""
}
version := "a version before the failure contract"
if module == "" {
module = "an unnamed build module"
}
if m, v, ok := strings.Cut(by, " with "); ok && m != "" {
module, version = m, v
}
return fmt.Sprintf("the build module hid the failure (%s at %s): %s", module, version, why)
}
// MaxCause is the most a cause in a failure's words holds: one long line (the 1901 review: a cause of a kilobyte could
// push the words past MaxWords).
const MaxCause = 1024
@@ -633,23 +791,60 @@ func firstLine(s string) string {
return ""
}
// failure turns a level that did not pass into a verdict naming what failed first.
func levelFailure(t *target, level string, res Tested) *delivery.Verdict {
// MaxFailureLine is the most bytes of a failure's line in a verdict's Failures: the first line of its words, which is
// its cause where the build module told one apart (issue 1901). Failed keeps the first failure's words whole.
const MaxFailureLine = 300
// levelFailure turns a level that did not pass into a verdict: Failed names what failed first with its words whole,
// and Failures every failed test of the level's results with its first line, at most delivery.MaxFailures and an entry
// saying how many more (issue 1921).
func levelFailure(t *target, level string, results ...Tested) *delivery.Verdict {
v := delivery.Verdict{State: delivery.VerdictFail}
switch {
case len(res.Failed) > 0:
f := res.Failed[0]
name := failedName(f)
if t.module == buildrun.NodeBuildModule && f.Package == "" {
name = res.Module
var all []delivery.FailedFirst
for _, res := range results {
if len(res.Failed) == 0 {
why := res.Why
if why == "" {
why = "the " + level + " level did not pass"
}
all = append(all, delivery.FailedFirst{Test: level + ": " + t.pull.Repository, Words: why})
continue
}
v.Failed = delivery.FailedFirst{Test: level + ": " + name, Words: failureWords(f)}
default:
why := res.Why
if why == "" {
why = "the " + level + " level did not pass"
for _, f := range res.Failed {
name := failedName(f)
if t.module == buildrun.NodeBuildModule && f.Package == "" {
name = res.Module
}
all = append(all, delivery.FailedFirst{Test: level + ": " + name, Words: failureWordsAfter(f, hidWords(t.module, res.By, f))})
}
v.Failed = delivery.FailedFirst{Test: level + ": " + t.pull.Repository, Words: why}
}
if len(all) == 0 { // no result: a level said failed with nothing to name
all = append(all, delivery.FailedFirst{Test: level + ": " + t.pull.Repository, Words: "the " + level + " level did not pass"})
}
v.Failed = all[0]
v.Failures = failureLines(all)
return &v
}
// failureLines is every failure with the first line of its words, cut to MaxFailureLine, at most delivery.MaxFailures
// and then one entry saying how many more failed.
func failureLines(all []delivery.FailedFirst) []delivery.FailedFirst {
n := min(len(all), delivery.MaxFailures)
out := make([]delivery.FailedFirst, 0, n+1)
for _, f := range all[:n] {
out = append(out, delivery.FailedFirst{Test: f.Test, Machine: f.Machine, Words: clipLine(firstLine(f.Words), MaxFailureLine)})
}
if more := len(all) - n; more > 0 {
out = append(out, delivery.FailedFirst{Test: fmt.Sprintf("and %d more", more)})
}
return out
}
// clipLine cuts a line to at most n bytes on a whole character, marking the cut.
func clipLine(s string, n int) string {
if len(s) <= n {
return s
}
s = strings.ToValidUTF8(s[:n-3], "")
return s + "..."
}
@@ -56,12 +56,22 @@ func (k Kept) Keep(ctx context.Context, level, digest string, v delivery.Kept) (
// KeepPasses keeps, for the packages of a test result that passed with tests, a verdict under the digest the build module
// named for the package at the level. A package that failed, has no tests or has no digest is never kept (ADR 0444 clause
// 1): it runs every check. It answers the packages it kept now.
// 1): it runs every check. A package with skipped tests is not kept either: a pass that skipped (a test needing a
// toolchain the node lacks) would be reused for its digest and never run anywhere (issue 890). A build module that
// counts skips only per result, not per package, keeps nothing of a result in which a test skipped. It answers the
// packages it kept now.
func (k Kept) KeepPasses(ctx context.Context, repository, level, toolchain string, res Tested, digests map[string]string, at time.Time) ([]string, error) {
var kept []string
perPackage := false
for _, p := range res.Packages {
perPackage = perPackage || p.Skipped > 0
}
if res.Skipped > 0 && !perPackage {
return kept, nil // an older build module: the skip cannot be placed in a package
}
for _, p := range res.Packages {
digest, ok := digests[p.Package]
if !ok || !p.Passed || p.NoTests || p.Tests == 0 {
if !ok || !p.Passed || p.NoTests || p.Tests == 0 || p.Skipped > 0 {
continue
}
if failedIn(res, p.Package) {
@@ -74,6 +74,32 @@ func TestOnlyPassingPackagesWithADigestAreKept(t *testing.T) { // 1287/AC3, AC4
}
}
func TestAResultWithSkippedTestsKeepsNothing(t *testing.T) { // issue 890: a pass that skipped proved nothing
ctx, st := context.Background(), newMemKV()
k := Kept{Store: st}
res := Tested{Level: "integration", Passed: true, Skipped: 2, Packages: []GoPackage{{Package: "m/a", Passed: true, Tests: 4}}}
kept, err := k.KeepPasses(ctx, "novox/mesh-catalog", "integration", "go1.26.0", res, map[string]string{"m/a": digA}, at0)
if err != nil || len(kept) != 0 {
t.Fatalf("kept %v, %v, want nothing", kept, err)
}
if keys, _ := st.Keys(ctx); len(keys) != 0 {
t.Errorf("the bucket holds %v, want none", keys)
}
}
func TestOnlyTheSkippingPackageOfAResultIsWithheld(t *testing.T) { // issue 890
ctx, st := context.Background(), newMemKV()
k := Kept{Store: st}
res := Tested{Level: "integration", Passed: true, Skipped: 2, Packages: []GoPackage{
{Package: "m/a", Passed: true, Tests: 4, Skipped: 2},
{Package: "m/b", Passed: true, Tests: 3},
}}
kept, err := k.KeepPasses(ctx, "novox/mesh-catalog", "integration", "go1.26.0", res, map[string]string{"m/a": digA, "m/b": digB}, at0)
if err != nil || strings.Join(kept, ",") != "m/b" {
t.Fatalf("kept %v, %v, want only m/b", kept, err)
}
}
func TestAKeptValueOfAnotherDigestIsNotRead(t *testing.T) { // 1287/AC7
ctx, st := context.Background(), newMemKV()
k := Kept{Store: st}
@@ -49,6 +49,34 @@ func firstFailed(v delivery.Verdict) string {
return firstWordsLine(f.Words)
}
// failedLines is everything a verdict names as failed, one line each: every failed test with its first line (issue
// 1921), or what failed first when the verdict names no list (a conflict, a refused step, an older verdict).
func failedLines(v delivery.Verdict) []string {
if len(v.Failures) == 0 {
if f := firstFailed(v); f != "" {
return []string{f}
}
return nil
}
lines := make([]string, 0, len(v.Failures))
for _, f := range v.Failures {
lines = append(lines, firstFailed(delivery.Verdict{Failed: f}))
}
return lines
}
// failedCount is how many tests a verdict names as failed, the ones after delivery.MaxFailures counted too.
func failedCount(v delivery.Verdict) int {
n := len(v.Failures)
if n > delivery.MaxFailures {
var more int
if _, err := fmt.Sscanf(v.Failures[n-1].Test, "and %d more", &more); err == nil {
return delivery.MaxFailures + more
}
}
return n
}
// firstWordsLine is the first line of a failure's words: the words are several lines (issue 1901) and the status line one.
func firstWordsLine(w string) string { return firstLine(w) }
@@ -57,18 +85,22 @@ type pieces struct {
head, commits, counts, shortCounts, times, shortTimes, over, builds, composition, shortComposition, group string
// judged names the program that judged a change to a build module (ADR 0535 §5); never dropped.
judged string
// carried says, of a carried verdict, the commits it was carried from (ADR 0548 §8); never dropped.
carried string
// warning says that a part of the check did not run, for instance that an older build module could not refuse a compiled program; never dropped.
warning string
}
// pullKey names a pull request of a case: `<owner/repo>#<number>`.
func pullKey(p Pull) string { return fmt.Sprintf("%s#%d", p.Repository, p.Number) }
func (p pieces) join(drop int) string {
parts := []string{p.head, p.commits, p.counts, p.judged}
parts := []string{p.head, p.commits, p.carried, p.warning, p.counts, p.judged}
switch {
case drop < 3 && p.times != "":
parts = append(parts, p.times)
case drop == 3 && p.shortTimes != "":
parts = []string{p.head, p.commits, p.shortCounts, p.judged, p.shortTimes}
parts = []string{p.head, p.commits, p.carried, p.warning, p.shortCounts, p.judged, p.shortTimes}
}
if p.over != "" {
parts = append(parts, p.over)
@@ -143,18 +175,67 @@ func buildWords(builds []BuildNote) string {
return "builds: " + strings.Join(parts, ", ")
}
// carriedWords say the commits a verdict was carried from, for the line of pull request p (ADR 0548 §8): its own trunk's
// commit as twelve characters, a sibling's with its name before it; only those that moved, sorted. Empty when none did.
func carriedWords(from map[string]string, v delivery.Verdict, cs *Case, p Pull) string {
var own string
var others []string
for r, old := range from {
now, ok := v.Trunk[r]
if cs != nil && !ok {
now, ok = cs.Trunk[r]
}
if !ok {
now = v.Beside[r]
}
if now == "" || now == old {
continue
}
if strings.EqualFold(r, p.Repository) {
own = short12(old)
continue
}
name := r
if i := strings.LastIndex(r, "/"); i >= 0 {
name = r[i+1:]
}
others = append(others, name+" "+short12(old))
}
slices.Sort(others)
if own != "" {
others = append([]string{own}, others...)
}
return strings.Join(others, ", ")
}
// parts makes the pieces of the line of a pull request of a case.
func parts(v delivery.Verdict, cs *Case, p Pull, builds []BuildNote, slowest []string, judged map[string]string) pieces {
var pc pieces
pc.judged = judged[pullKey(p)]
if cs != nil {
pc.warning = cs.Warning(pullKey(p))
}
commits := short12(p.Head) + " on " + short12(p.Trunk)
switch v.State {
case delivery.VerdictRunning:
pc.head = "checking " + commits
if cs != nil && cs.CarriedFrom != nil {
// Out of date and being carried to the new commit (ADR 0548 §6): pending, and saying so.
pc.head = "being carried to " + commits
if w := carriedWords(cs.CarriedFrom, v, cs, p); w != "" {
pc.carried = "from " + w
}
}
case delivery.VerdictPass:
pc.head = "pass " + commits
if w := carriedWords(v.CarriedFrom, v, cs, p); w != "" {
pc.carried = "carried from " + w // the counts that follow say what the carry ran and kept
}
default:
pc.head = "failed: " + firstFailed(v)
if n := failedCount(v); n > 1 {
pc.head = fmt.Sprintf("failed (%d tests): %s", n, firstFailed(v))
}
pc.commits = commits
}
var counts []string
@@ -0,0 +1,34 @@
package check
import (
"strings"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
)
// **An ask the build seat gave up on ends its step** (novox/hq issue 734): mesh-sdk buildrun.LostAskWaiter hands the
// waiting asker a failed step on the node buildrun.LostNode when DEAD_LETTERS keeps the ask's copy (buildrun.HearDeadLetters,
// granted by buscompose DeadLetterReads), and, under buildrun.NotTakenBound, when no build node took the ask within the
// bound. What follows is the checker's: a try ends as error (Tryer.run), with the not-taken bound at the step bound
// (Tryer.NotTaken); a check is asked again under its claim, at most LostAskTries times, then fails as not heard and is
// judged again when the nodes change (Checker.try). A check's ask has no not-taken bound: it waits in the stream as long as
// the backlog is (ADR 0525 §9), and a check ended for a backlog would owe its pull request a failure.
// LostAskTries is how often a check asks again whose ask was given up, before it fails as not heard (UnheardWords) and is
// judged again when the nodes change.
const LostAskTries = 3
// IsLostAsk says whether a verdict failed because the build seat gave its ask up (buildrun.LostWords).
func IsLostAsk(v delivery.Verdict) bool {
return v.State == delivery.VerdictFail && strings.Contains(v.Failed.Words, buildrun.LostWords)
}
// IsNotTaken says whether a verdict failed because no build node took its ask within the bound (buildrun.NotTakenWords).
func IsNotTaken(v delivery.Verdict) bool {
return v.State == delivery.VerdictFail && strings.Contains(v.Failed.Words, buildrun.NotTakenWords)
}
// HearingBucket is the checker's state hearing: whether it hears DEAD_LETTERS (buildrun.Hearing under
// buildrun.HearingKey), which checker_status says (dead_letters).
const HearingBucket = delivery.CheckerModule + "_" + buildrun.HearingState
@@ -0,0 +1,170 @@
package check
import (
"context"
"encoding/json"
"strings"
"sync"
"testing"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
)
// losingAsks is the build seat over a bus that gives up the first `lose` asks put (DEAD_LETTERS keeps them, as on
// 2026-10-26 during a bus reload: try-1793017340003993788) and answers the rest; with never set it neither answers nor
// gives up (an ask no node takes).
type losingAsks struct {
*buildrun.FakeAsks
answer func(delivery.StepAsk) delivery.Step
dead *buildrun.DeadLetters
never bool
mu sync.Mutex
lose int
}
func (l *losingAsks) PublishAsk(ctx context.Context, subject, id string, data []byte) error {
if err := l.FakeAsks.PublishAsk(ctx, subject, id, data); err != nil {
return err
}
l.mu.Lock()
lost := l.lose > 0
if lost {
l.lose--
}
l.mu.Unlock()
switch {
case l.never:
case lost:
h := map[string][]string{buildrun.DeadHeaderMsgID: {id}, buildrun.DeadHeaderDeliveries: {"5"}, buildrun.DeadHeaderStream: {"SEAT_NODE_BUILD_GO"},
buildrun.DeadHeaderGaveUp: {"2026-10-26T12:26:40Z"}}
g, ok := buildrun.GivenUpOf(h)
if !ok {
return nil
}
go l.dead.Heard(g)
default:
a, err := delivery.DecodeStepAsk(data)
if err != nil {
return err
}
l.Write(l.answer(a))
}
return nil
}
func losing(b *bench, lose int) *losingAsks {
l := &losingAsks{FakeAsks: &buildrun.FakeAsks{}, answer: b.answer, dead: &buildrun.DeadLetters{}, lose: lose}
b.ex.Asks = buildrun.Asker{Publisher: l, Waiter: buildrun.LostAskWaiter{Inner: l.FakeAsks, Dead: l.dead}, Bound: 5 * time.Second}
return l
}
// endable is the tries state as the end tool sees it.
type endable struct{ kv *memKV }
func (e endable) Entry(key string) (json.RawMessage, uint64, error) {
v, rev, _, err := e.kv.Get(context.Background(), key)
return v, rev, err
}
func (e endable) Update(key string, value []byte, revision uint64) error {
_, err := e.kv.Update(context.Background(), key, value, revision)
return err
}
// Issue 734: a try whose ask the build seat gave up on ends as error saying so, and the next ask of its head is a new try;
// one no node takes ends at the step bound; one ended by hand frees its place and stops being waited for.
func TestAnAskGivenUpOrNeverTakenEndsItsTryAsErrorAndItsHeadIsAskedAnew(t *testing.T) {
ctx := context.Background()
for _, row := range []struct {
name string
lose int
never bool
want string
}{
{"given up by the build seat", 1, false, buildrun.LostWords + " after 5 deliveries at 2026-10-26T12:26:40Z"},
{"taken by no build node within the bound", 0, true, buildrun.NotTakenWords + " within 200ms"},
} {
t.Run(row.name, func(t *testing.T) {
kv := newMemKV()
at := time.Date(2026, 10, 26, 12, 22, 20, 0, time.UTC)
id, err := AskTry(memTries{kv}, tryTrunks, TryAsk{Repository: repoA, Head: head1}, at)
if err != nil {
t.Fatal(err)
}
b := newBench(t)
l := losing(b, row.lose)
l.never = row.never
tr := &Tryer{Tries: kv, Judge: b.ex, NotTaken: 200 * time.Millisecond}
raw, _, _, _ := kv.Get(ctx, id)
tr.Put(ctx, id, raw)
tr.Wait()
got := readTry(t, kv, id)
if got.State != TryError || len(got.Failed) != 1 || !strings.Contains(got.Failed[0], row.want) || !strings.Contains(got.Said, row.want) {
t.Fatalf("try = %+v, want error saying %q", got, row.want)
}
again, err := AskTry(memTries{kv}, tryTrunks, TryAsk{Repository: repoA, Head: head1}, at.Add(time.Minute))
if err != nil || again == id {
t.Errorf("the head asked again answered %q, %v, want a new try", again, err)
}
})
}
t.Run("ended by hand", func(t *testing.T) {
kv := newMemKV()
id, _ := AskTry(memTries{kv}, tryTrunks, TryAsk{Repository: repoA, Head: head1}, time.Now())
b := newBench(t)
losing(b, 0).never = true
tr := &Tryer{Tries: kv, Judge: b.ex} // no bound: only the hand ends it
raw, _, _, _ := kv.Get(ctx, id)
tr.Put(ctx, id, raw)
deadline := time.Now().Add(2 * time.Second)
for readTry(t, kv, id).State != TryRunning && time.Now().Before(deadline) {
time.Sleep(5 * time.Millisecond)
}
if _, err := EndTry(endable{kv}, id, "", time.Now()); err == nil {
t.Error("an end without a why was taken")
}
ended, err := EndTry(endable{kv}, id, "its ask was lost in the bus reload", time.Now())
if err != nil || ended.State != TryError || !strings.HasPrefix(ended.Said, "ended by hand ("+TryEndTool+"): its ask was lost") {
t.Fatalf("ended = %+v, %v", ended, err)
}
raw, _, _, _ = kv.Get(ctx, id)
tr.Put(ctx, id, raw) // the watch of the tries hands the end over
tr.Wait() // returns: the try's judging stopped
if got := readTry(t, kv, id); got.State != TryError || got.Said != ended.Said {
t.Errorf("try = %+v, want the hand's end kept", got)
}
if _, err := EndTry(endable{kv}, id, "again", time.Now()); err == nil || !strings.Contains(err.Error(), "ended already") {
t.Errorf("a second end answered %v", err)
}
})
}
// Issue 734: a pull request's check whose ask the build seat gave up on is asked again under its claim, its status
// pending, and judged; one given up every time fails as not heard and is judged again when the nodes change.
func TestACheckWhoseAskWasGivenUpIsAskedAgain(t *testing.T) {
ctx := context.Background()
for _, row := range []struct {
lose int
state string
asks int
}{
{1, delivery.VerdictPass, 2},
{LostAskTries, delivery.VerdictFail, LostAskTries},
} {
b := newBench(t)
l := losing(b, row.lose)
r := newRig(t, b.ex)
cand := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
r.ch.Candidate(ctx, "feat%2Fx", cand)
r.ch.Wait()
v := r.verdict(t, keyOf(t, "feat/x", cand))
if v.State != row.state || len(l.Asks()) != row.asks {
t.Errorf("lose %d: verdict %s (%s) after %d asks, want %s after %d", row.lose, v.State, v.Failed.Words, len(l.Asks()), row.state, row.asks)
}
if row.state == delivery.VerdictFail && (!IsLostAsk(v) || !IsNodesFailure(v)) {
t.Errorf("a check given up every time = %q, want lost and judged again when the nodes change", v.Failed.Words)
}
}
}
@@ -38,11 +38,11 @@ func TestAPackageFailureIsNeverWordedPass(t *testing.T) { // 1901/AC5
t.Errorf("1901/AC5: the words keep %q: %q", l, v.Failed.Words)
}
}
if !strings.HasPrefix(v.Failed.Words, c.want) {
t.Errorf("1901/AC5: words start %q, want %q", firstLine(v.Failed.Words), c.want)
if !strings.HasPrefix(afterHid(v.Failed.Words), c.want) {
t.Errorf("1901/AC5: words after the loud line start %q, want %q", firstLine(afterHid(v.Failed.Words)), c.want)
}
if got := firstFailed(*v); !strings.Contains(got, c.want) || strings.HasSuffix(got, ": PASS") {
t.Errorf("1901/AC5: status = %q", got)
if got := firstFailed(*v); !strings.Contains(got, "the build module hid the failure") || strings.HasSuffix(got, ": PASS") {
t.Errorf("1901/AC5, AC8: status = %q", got)
}
})
}
@@ -63,7 +63,7 @@ func TestALongCauseIsCutToALinesBound(t *testing.T) { // 1901/AC5
func TestAPackageStoppedWithOnlyItsFailLineIsStillWorded(t *testing.T) { // 1901/AC5
for _, out := range []string{"FAIL\texample.com/timeout\t2.017s", "FAIL\texample.com/exiting\t0.006s", "FAIL\texample.com/killed\t0.006s"} {
v := levelFailure(&target{}, "integration", Tested{Failed: []Failure{{Package: "example.com/x", Output: out}}})
if !strings.HasPrefix(v.Failed.Words, "the package failed with no test to name and no panic, signal or build error in its output") {
if !strings.HasPrefix(afterHid(v.Failed.Words), "the package failed with no test to name and no panic, signal or build error in its output") {
t.Errorf("1901/AC5: %q is worded %q", out, v.Failed.Words)
}
}
@@ -74,7 +74,7 @@ func TestAPackageStoppedWithOnlyItsFailLineIsStillWorded(t *testing.T) { // 1901
func TestOnlyGosExactFormsAreSummaryOrCause(t *testing.T) { // 1901/AC5
out := " x_test.go:3: ok so far\nok then: the seat is free\n panic: child retried and recovered\npanic: the real one\nFAIL\tp\t1.0s"
v := levelFailure(&target{}, "unit", Tested{Failed: []Failure{{Package: "p", Output: out}}})
if !strings.HasPrefix(v.Failed.Words, "panic: the real one") || !strings.Contains(v.Failed.Words, "ok then: the seat is free") {
if !strings.HasPrefix(afterHid(v.Failed.Words), "panic: the real one") || !strings.Contains(v.Failed.Words, "ok then: the seat is free") {
t.Errorf("1901/AC5: words = %q", v.Failed.Words)
}
}
@@ -90,3 +90,13 @@ func TestTheWordsKeepTheirBoundExactly(t *testing.T) { // 1901/AC5
}
}
}
// afterHid is a verdict's words after the checker's loud line, which a failure with no cause from an older build module
// now starts with (1901/AC8).
func afterHid(w string) string {
if strings.HasPrefix(w, "the build module hid the failure") {
_, rest, _ := strings.Cut(w, "\n")
return rest
}
return w
}
@@ -240,7 +240,12 @@ func TestTheSiblingsAreWhereTheTestsStateThem(t *testing.T) { // review D2, ADR
catalogue + " go.mod " + head2: "module git.novox.be/novox/mesh-catalog\n"}, onTrunk: map[string]bool{"cb54756c3865": true}}
b.ex.Forge = forge
// A group of the controller and the catalogue: the controller's tests read the catalogue's trunk, not the group's head.
b.judge(t, caseOf(t, pullOf(controller, 7, head1), pullOf(catalogue, 12, head2)))
out := b.judge(t, caseOf(t, pullOf(controller, 7, head1), pullOf(catalogue, 12, head2)))
// ADR 0548 §6, issue 568's own example: the verdict names the node-engine at the commit the mesh runs, and mesh-sdk at the
// trunk the catalogue's tests read it at; the catalogue, a member here, is not named.
if out.Beside[host] != hostRuns || out.Beside[catalogue] != "" || out.Beside[SDKRepository] != "9999999999999999999999999999999999999999" {
t.Errorf("the siblings named for the verdict = %v, want the node-engine at %s and mesh-sdk at its trunk", out.Beside, hostRuns[:12])
}
got := besideOf(t, b, controller)
if c := got[catalogue]; c.Commit != trunkS || c.Ref != "" {
t.Errorf("the catalogue beside the controller = %+v, want its trunk %s", c, trunkS)
@@ -324,7 +329,11 @@ func TestTheChangesOwnRepositoryIsBesideItAtItsTrunk(t *testing.T) {
func TestACatalogueChangeHasTheCatalogueBesideItAtItsTrunk(t *testing.T) {
b := newBench(t)
b.ex.Trunks = trunksOf(t, map[string]string{catalogueRepository: trunkA, SDKRepository: trunkS, controllerRepo: trunkA2})
b.judge(t, caseOf(t, pullOf(catalogueRepository, 12, head1)))
out := b.judge(t, caseOf(t, pullOf(catalogueRepository, 12, head1)))
// ADR 0548 §6: the outcome names the siblings read at their trunk's head, never the repository it merges.
if out.Beside[SDKRepository] != trunkS || out.Beside[controllerRepo] != trunkA2 || out.Beside[catalogueRepository] != "" {
t.Errorf("the siblings named for the verdict = %v, want mesh-sdk and mesh-controller at their trunks and not the catalogue", out.Beside)
}
got := besideOf(t, b, catalogueRepository)
if c := got[catalogueRepository]; c.Commit != trunkA || c.Ref != "" {
t.Fatalf("the catalogue beside its own change = %+v, want its trunk %s", c, trunkA)
@@ -75,3 +75,40 @@ func TestCheckerStatusNamesTheWaitingChecksByKind(t *testing.T) { // 1897/AC4
t.Errorf("waiting = %v, want %v", st.Waiting, want)
}
}
// 568 note 21: a verdict that read the node-engine at its running commit is not waiting for a trunk move while that commit
// is still the one the mesh runs, though the node-engine's trunk has moved on (ADR 0548 §6).
func TestCheckerStatusReadsTheNodeEngineAtItsRunningCommit(t *testing.T) { // 568/21
cand := candidateOf(t, group(1), pullOf(repoA, 1, head1))
ckey, _ := delivery.CandidatesKey(group(1))
vkey := keyOf(t, group(1), cand)
dc, _ := delivery.DecodeCandidate(cand)
judged, err := delivery.EncodeVerdict(delivery.Verdict{Group: group(1), Heads: dc.HeadsHash, Trunk: map[string]string{repoA: trunkA}, Beside: map[string]string{nodeEngineRepo: "run111"}, State: delivery.VerdictPass, Levels: passing().Levels})
if err != nil {
t.Fatal(err)
}
heads := func() (map[string]string, error) {
return map[string]string{repoA: trunkA, nodeEngineRepo: "trunk222"}, nil
}
for _, row := range []struct {
name string
runs runsAt
want int
}{
{"the running commit the verdict read", runsAt{nodeEngineModule: "run111"}, 0},
{"the running commit moved", runsAt{nodeEngineModule: "run999"}, 1},
{"no running commit known: the trunk's", nil, 1},
} {
rd := Reader{CandidateState: fakeState{ckey: cand}.read(), VerdictState: fakeState{vkey: judged}.read(), Trunks: heads}
if row.runs != nil {
rd.Runs = row.runs
}
st, err := rd.Status()
if err != nil {
t.Fatal(err)
}
if got := len(st.Waiting[KindTrunk]); got != row.want {
t.Errorf("%s: waiting for a trunk move = %v, want %d", row.name, st.Waiting[KindTrunk], row.want)
}
}
}
@@ -8,6 +8,7 @@ import (
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
)
// The read tools (novox/hq ADR 0484 clauses 8 and 13, issue 1686): what the checker stored, as stored, with the line of the
@@ -39,7 +40,14 @@ type Reader struct {
StepBound time.Duration // how long the checker waits for one step; DefaultStepBound when 0
// Trunks are the trunk heads of every repository (the gate's source); nil names no re-judge after a trunk move.
Trunks func() (map[string]string, error)
Now func() time.Time
// TurnState is the gate's merge turns (ADR 0548 §3); without it no check is named as held by a turn.
TurnState StateRead
// Runs says which commit the mesh runs of a module (the indexer's modules), as the checker's decision reads it: the
// node-engine a verdict names is compared at its running commit, not at the trunk's (ADR 0548 §6); nil knows none.
Runs interface{ CommitOf(module string) string }
// HearingState is the checker's state hearing: whether it hears DEAD_LETTERS (issue 734).
HearingState StateRead
Now func() time.Time
}
// VerdictRead is a verdict as stored with the full report of the status of each pull request it judged (Report), when the
@@ -312,7 +320,68 @@ type Status struct {
// Waiting are the candidates whose check waits, by kind (ADR 0536 §4): new (no verdict for its heads) and nodes (failed
// for the nodes, asked again when they change) start before trunk (judged on trunk heads that moved since).
Waiting map[string][]string `json:"waiting_by_kind"`
Notes []string `json:"notes,omitempty"`
// HeldByTurn are the candidates whose check an approved merge's turn holds off (ADR 0548 §3), by group: the repository and
// the group of the merge that waits. Their check is claimed when that turn goes.
HeldByTurn map[string]string `json:"held_by_turn"`
// DeadLetters says whether the checker hears the asks the build seats give up on (issue 734): deaf, a try or check
// whose ask is given up waits until its bound.
DeadLetters DeadLettersHeard `json:"dead_letters"`
Notes []string `json:"notes,omitempty"`
}
// DeadLettersHeard is the checker's hearing of DEAD_LETTERS as the status says it: healthy or not, and why.
type DeadLettersHeard struct {
Heard bool `json:"heard"`
Why string `json:"why"`
}
// deadLetters reads the checker's kept hearing (buildrun.HearingHealth).
func (r Reader) deadLetters() DeadLettersHeard {
if r.HearingState.Get == nil {
return DeadLettersHeard{Why: "the checker's state hearing is not readable here"}
}
raw, err := r.HearingState.Get(buildrun.HearingKey)
if err != nil {
return DeadLettersHeard{Why: "the checker's state hearing cannot be read: " + err.Error()}
}
a := buildrun.HearingHealth(raw)
return DeadLettersHeard{Heard: a.Healthy, Why: a.Why}
}
// turns reads the gate's merge turns, by repository; nil when they cannot be read, said in a note.
func (r Reader) turns(st *Status) *Turns {
if r.TurnState.Keys == nil || r.TurnState.Get == nil {
st.Notes = append(st.Notes, "the gate's merge turns are not readable here, so no check is named as held by one")
return nil
}
keys, err := r.TurnState.Keys()
if err != nil {
st.Notes = append(st.Notes, "the gate's merge turns cannot be listed, so no check is named as held by one: "+err.Error())
return nil
}
t := &Turns{}
for _, k := range keys {
if raw, err := r.TurnState.Get(k); err == nil && len(raw) > 0 {
t.Put(k, raw, false)
}
}
return t
}
// heldBy says the turn that holds a candidate's check off, or "".
func heldBy(t *Turns, c delivery.Candidate, heads map[string]string) string {
cs := CaseOf(c, heads)
if t == nil || cs == nil {
return ""
}
repos := make([]string, 0, len(cs.Trunk))
for r := range cs.Trunk {
repos = append(repos, r)
}
if repo, w, ok := t.Holding(repos, cs.Heads); ok {
return repo + ": the approved merge of " + w.Group + " waits on it"
}
return ""
}
// Status reads the build modules' steps and the candidates against their verdicts.
@@ -383,6 +452,8 @@ func (r Reader) Status() (Status, error) {
}
st.Unjudged = []string{}
st.Waiting = map[string][]string{KindNew: {}, KindNodes: {}, KindTrunk: {}}
st.HeldByTurn = map[string]string{}
turns := r.turns(&st)
var heads map[string]string
if r.Trunks == nil {
st.Notes = append(st.Notes, "the trunk heads are not readable here, so no check is named as waiting for a trunk move")
@@ -435,6 +506,9 @@ func (r Reader) Status() (Status, error) {
if len(raw) == 0 {
st.Unjudged = append(st.Unjudged, c.Group)
st.Waiting[KindNew] = append(st.Waiting[KindNew], c.Group)
if why := heldBy(turns, c, heads); why != "" {
st.HeldByTurn[c.Group] = why
}
continue
}
v, err := delivery.DecodeVerdict(raw)
@@ -443,12 +517,16 @@ func (r Reader) Status() (Status, error) {
case IsNodesFailure(v):
st.Waiting[KindNodes] = append(st.Waiting[KindNodes], c.Group)
case heads != nil:
if cs := CaseOf(c, heads); cs != nil && !sameTrunk(v.Trunk, cs.Trunk) {
if cs := CaseOf(c, heads); cs != nil && (!sameTrunk(v.Trunk, cs.Trunk) || besideMoved(v.Beside, besideAt(heads, r.Runs))) {
st.Waiting[KindTrunk] = append(st.Waiting[KindTrunk], c.Group)
if why := heldBy(turns, c, heads); why != "" {
st.HeldByTurn[c.Group] = why
}
}
}
}
}
st.DeadLetters = r.deadLetters()
st.Notes = append(st.Notes, "nothing holds the checker and it holds no place: running_steps are the build modules' keys that can still be running (started within the step bound, their build module running on the node); a candidate without a verdict has its check ask in a build seat's stream, waiting for a node with a free slot, whose backlog the delivery view shows (ADR 0525 §4, §9)")
return st, nil
}
@@ -144,7 +144,10 @@ func TestTheStatusToolReadsRunningSteps(t *testing.T) {
CandidateState: fakeState{ckey: cand}.read(),
VerdictState: fakeState{}.read(),
HaveState: fakeState{"shanks.mesh-build-go": haveOf(t, delivery.PhaseRunning)}.read(),
Now: func() time.Time { return time.Date(2026, 10, 22, 15, 10, 0, 0, time.UTC) },
HearingState: fakeState{"dead-letters": []byte(`{"heard":false,"streams":["SEAT_NODE_BUILD_GO"],` +
`"why":"DEAD_LETTERS is not heard, so an ask a build seat gives up on is waited for until its bound: SEAT_NODE_BUILD_GO: permissions violation",` +
`"at":"2026-10-26T13:00:00Z"}`)}.read(),
Now: func() time.Time { return time.Date(2026, 10, 22, 15, 10, 0, 0, time.UTC) },
}
st, err := rd.Status()
if err != nil {
@@ -156,9 +159,13 @@ func TestTheStatusToolReadsRunningSteps(t *testing.T) {
if len(st.Unjudged) != 1 || st.Unjudged[0] != "feat/x" {
t.Errorf("unjudged = %v", st.Unjudged)
}
// Issue 734, review round 1: a checker that does not hear DEAD_LETTERS says so in its status, with why.
if st.DeadLetters.Heard || !strings.Contains(st.DeadLetters.Why, "permissions violation") {
t.Errorf("dead letters = %+v, want not heard, saying why", st.DeadLetters)
}
// What cannot be read is said, not guessed.
st, err = Reader{}.Status()
if err != nil || len(st.Notes) == 0 {
if err != nil || len(st.Notes) == 0 || st.DeadLetters.Heard {
t.Errorf("empty status = %+v, %v", st, err)
}
}
@@ -234,6 +241,18 @@ func TestTheStatusCountsAnHQOnlyGroupAsUnjudged(t *testing.T) {
if len(st.Unjudged) != 1 || st.Unjudged[0] != "pb-x" {
t.Errorf("unjudged = %v, want pb-x", st.Unjudged)
}
// ADR 0548 §3: a candidate whose check an approved merge's turn holds off is named with the turn.
held := candidateOf(t, "feat/x", pullOf(repoA, 7, head1))
hkey, _ := delivery.CandidatesKey("feat/x")
st, err = Reader{CandidateState: fakeState{hkey: held}.read(), VerdictState: fakeState{}.read(),
Trunks: func() (map[string]string, error) { return map[string]string{repoA: trunkA}, nil },
TurnState: fakeState{turnKey(t, repoA): turnOf(t, repoA, "fix/other", "0123456789ab")}.read()}.Status()
if err != nil {
t.Fatal(err)
}
if why := st.HeldByTurn["feat/x"]; !strings.Contains(why, repoA) || !strings.Contains(why, "fix/other") {
t.Errorf("held_by_turn = %v, want feat/x held by the merge of fix/other on %s", st.HeldByTurn, repoA)
}
}
// slowBucket wraps a state with a Get that takes delay and counts the values read, the lists and the most Gets in flight.
@@ -3,6 +3,9 @@ package check
import (
"encoding/json"
"strings"
"git.novox.be/novox/mesh-sdk/go/buildstep"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// The shapes of what the build modules' steps answer, as their code writes them (mesh-catalog modules/mesh-build-go
@@ -42,6 +45,17 @@ type Fetched struct {
Conflicts []string `json:"conflicts,omitempty"`
Changed []string `json:"changed,omitempty"`
Trees map[string]MergeTree `json:"trees,omitempty"`
// Programs are the changed files whose first bytes are an executable binary, read by the build module in its workspace
// (issue 1577); nil for a build module that does not report it, a pointer to an empty list for one that found none.
Programs *[]ProgramFile `json:"programs,omitempty"`
}
// ProgramFile is a changed file that is a compiled program, and its format (ELF, Mach-O or PE).
type ProgramFile struct {
Path string `json:"path"`
Kind string `json:"kind"`
// Commit is set for a program a commit of the branch added and a later one removed: it stays in history once merged.
Commit string `json:"commit,omitempty"`
}
// GoPackage is one package of a Go test result.
@@ -52,16 +66,13 @@ type GoPackage struct {
Elapsed float64 `json:"elapsed"`
Tests int `json:"tests"`
NoTests bool `json:"no_tests,omitempty"`
// Skipped is how many of the package's tests skipped; an older build module names none.
Skipped int `json:"skipped,omitempty"`
}
// Failure is a test that failed, or a package that failed with no test to name.
type Failure struct {
Package string `json:"package"`
Test string `json:"test,omitempty"`
Output string `json:"output,omitempty"`
// Cause is the first line of what the test reported as its error, when the build module told it apart (issue 1901).
Cause string `json:"cause,omitempty"`
}
// Failure is a test that failed, or a package that failed with no test to name: the build modules' failure contract
// (buildstep.Failure, issue 1901), its cause never empty and never a passing line from a build module that keeps it.
type Failure = buildstep.Failure
// TestTime is one test and the seconds it took.
type TestTime struct {
@@ -94,6 +105,9 @@ type Tested struct {
// JudgedBy is the program that ran the step (ADR 0535 §5): "installed", or "candidate <merge commit>" under runner
// workspace; absent from a build module that does not take the runner.
JudgedBy string `json:"judged_by,omitempty"`
// By is the build module that answered and the SDK it was built with (buildstep.By); absent from a build module from
// before the failure contract.
By string `json:"by,omitempty"`
}
// Digested is the answer of the Go build module's digest step: one digest per package that has tests at the level, by import
@@ -107,4 +121,6 @@ type Digested struct {
WithoutTests []string `json:"without_tests,omitempty"`
// JudgedBy is the program that ran the digest, as Tested.JudgedBy says.
JudgedBy string `json:"judged_by,omitempty"`
// Needs are the packages whose tests need something of the node at this level (issue 890).
Needs []delivery.PackageNeeds `json:"needs,omitempty"`
}
@@ -10,7 +10,7 @@ TestFetchWithABaseMergesAndAnswersThePreview and TestFetchWithAConflictEndsPasse
into a throwaway test that wrote them out, at mesh-catalog feat/r4c5-checker d5804927 (the git commands are the test's fake,
so the shas are its constants; the shape and the field names are the module's own marshalling). node-fetch-clean.json is the
FetchResult json.Marshal of mesh-build-node's fetch (fetch.go, preview.go) in its test TestFetchWithABaseMergesAndAnswersThePreview
(cmd/mesh-build-node/preview_test.go) at the same commit. Captured on 2026-10-21 for issue 1679.
(cmd/mesh-build-node/preview_test.go) at the same commit. Captured on 2026-10-21 for issue 1679. Both clean fetches carry `"programs":[]`, which a merging fetch of the build modules answers since issue 1577 (mesh-sdk buildstep.MergeInto's Programs), added by hand to the captured bytes: a fixture without it is an older build module's, and its line warns that no compiled program is refused.
forge-get-file-go-mod.json is the answer of the forge's gitea_get_file (mesh-catalog modules/gitea, tools/index.ts) for modules/mesh-delivery-builder/go.mod of
novox/mesh-delivery at main, and forge-compare-trunk.json the answer of gitea_compare for novox/mesh-sdk base main, head cb54756c3865 (a commit on the
@@ -1 +1 @@
{"workspace":"mvii1iwu5414e4","commit":"0123456789abcdef0123456789abcdef01234567","modules":["."],"base":"abcdef0123456789abcdef0123456789abcdef01","merge":"fedcba9876543210fedcba9876543210fedcba98","changed":["a.go","modules/m/main.go"],"trees":{"modules/m":{"tree":"1111111111111111111111111111111111111111","manifest":"ca3d163bab055381827226140568f3bef7eaac187cebd76878e0b63e9e442356","locks":{"go.sum":"c5fc83c01e92404452b986527d239140ccf9a48b88e0c268fbf38c2e1429e9c9"}}}}
{"workspace":"mvii1iwu5414e4","commit":"0123456789abcdef0123456789abcdef01234567","modules":["."],"base":"abcdef0123456789abcdef0123456789abcdef01","merge":"fedcba9876543210fedcba9876543210fedcba98","changed":["a.go","modules/m/main.go"],"programs":[],"trees":{"modules/m":{"tree":"1111111111111111111111111111111111111111","manifest":"ca3d163bab055381827226140568f3bef7eaac187cebd76878e0b63e9e442356","locks":{"go.sum":"c5fc83c01e92404452b986527d239140ccf9a48b88e0c268fbf38c2e1429e9c9"}}}}
@@ -1 +1 @@
{"workspace":"mvii1sxt979f67","commit":"0123456789abcdef0123456789abcdef01234567","modules":["modules/m"],"installed":[],"refused":[],"base":"abcdef0123456789abcdef0123456789abcdef01","merge":"0123456789abcdef0123456789abcdef01234567","changed":["modules/m/package.json"],"trees":{"modules/m":{"tree":"0123456789abcdef0123456789abcdef01234567","manifest":"0581b699d9d05517e69b21f5cedbeb2eee8aba2b205280b561be731594d7feef"}}}
{"workspace":"mvii1sxt979f67","commit":"0123456789abcdef0123456789abcdef01234567","modules":["modules/m"],"installed":[],"refused":[],"base":"abcdef0123456789abcdef0123456789abcdef01","merge":"0123456789abcdef0123456789abcdef01234567","changed":["modules/m/package.json"],"programs":[],"trees":{"modules/m":{"tree":"0123456789abcdef0123456789abcdef01234567","manifest":"0581b699d9d05517e69b21f5cedbeb2eee8aba2b205280b561be731594d7feef"}}}
@@ -6,6 +6,7 @@ import (
"errors"
"fmt"
"regexp"
"slices"
"sort"
"strconv"
"strings"
@@ -13,6 +14,7 @@ import (
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/delivery/buildrun"
)
// A try is a pushed head judged by the checker as a pull request's head is, before its pull request is opened (ADR 0501
@@ -24,11 +26,17 @@ import (
const (
TryTool = "checker_try"
TryReadTool = "checker_try_read"
TryEndTool = "checker_try_end"
// TriesState is the checker's state of tries: the bucket mesh-delivery-checker_tries, keys try-<unix nanoseconds>.
TriesState = "tries"
TriesBucket = delivery.CheckerModule + "_" + TriesState
// MaxTries is how many tries may wait or run at once, mesh-wide: each holds a build node for minutes.
MaxTries = 4
// MaxTries and MaxTriesWaiting are checked when a try is asked, mesh-wide: a new try is refused while MaxTries tries run
// on a build node or MaxTriesWaiting wait for one (queued, or their asks not yet taken by a node). It is no bound on
// what runs later: tries let in while waiting may later all run at once, and each node's build capacity bounds what
// runs on it. A try waiting behind the seat's other asks holds no node, so it no longer blocks the next try (issue
// 1890: at 19:53 on 2026-10-26 four tries sat queued behind 44 checks for up to 47 minutes and no fifth could be asked).
MaxTries = 4
MaxTriesWaiting = 8
// TryScan is how many of the newest tries a new ask reads to find the ones that wait or run (issue 1827).
TryScan = 64
// TryKeep is how long an ended try's key is kept before the checker removes it.
@@ -69,13 +77,110 @@ type Try struct {
// Said is the try's report in words: what failed first, the tests run, kept and skipped per level, the time of each part.
Said string `json:"said,omitempty"`
// Took is the time from its start to its end, as Go writes a duration.
Took string `json:"took,omitempty"`
Took string `json:"took,omitempty"`
// Failed is every test that failed, each with its first line, at most delivery.MaxFailures and then how many more
// (issue 1921); or what else failed first (a conflict, a refused step).
Failed []string `json:"failed,omitempty"`
Levels map[string]delivery.LevelResult `json:"levels,omitempty"`
Machines []string `json:"machines,omitempty"`
Asked time.Time `json:"asked"`
Started time.Time `json:"started,omitzero"`
Ended time.Time `json:"ended,omitzero"`
// Asks are the ids of the asks the try put on the build seats' streams, each ending on the step key <node>.<id> of its
// build module: the delivery view counts those steps as a try's, not a candidate's check (build_usage).
Asks []string `json:"asks,omitempty"`
// Steps say where each ask stands while the try runs and after: asked until a node takes it, then the node, its
// state, the phase its step says and when (issue 1890: a 31-minute try could not be told queued from running).
Steps []TryStep `json:"steps,omitempty"`
Asked time.Time `json:"asked"`
Started time.Time `json:"started,omitzero"`
Ended time.Time `json:"ended,omitzero"`
}
// TryAsked is a TryStep's state before a build node took its ask: it waits in the seat's stream.
const TryAsked = "asked"
// TryStep is one ask of a try as its step key last said it.
type TryStep struct {
Ask string `json:"ask"`
Node string `json:"node,omitempty"`
State string `json:"state"`
Phase string `json:"phase,omitempty"`
Started time.Time `json:"started,omitzero"`
PhaseAt time.Time `json:"phase-at,omitzero"`
Ended time.Time `json:"ended,omitzero"`
}
// tryWriter writes a running try's record again as its asks are put and their steps move: one writer per try, holding
// the revision it wrote last, so the try's end is written at that revision. A write refused (the try ended by hand)
// stops the writes; the end's own write then says it.
type tryWriter struct {
mu sync.Mutex
tr *Tryer
ctx context.Context
key string
t Try
rev uint64
stopped bool
}
// heard takes an ask put (step nil) or a value of its step key in, and writes the record when it changed.
func (w *tryWriter) heard(askID string, step *delivery.Step) {
w.mu.Lock()
defer w.mu.Unlock()
if w.stopped {
return
}
changed := false
if !slices.Contains(w.t.Asks, askID) {
w.t.Asks, changed = append(w.t.Asks, askID), true
}
i := slices.IndexFunc(w.t.Steps, func(s TryStep) bool { return s.Ask == askID })
if i < 0 {
w.t.Steps, changed = append(w.t.Steps, TryStep{Ask: askID, State: TryAsked}), true
i = len(w.t.Steps) - 1
}
if step != nil {
now := TryStep{Ask: askID, Node: step.Node, State: step.State, Phase: step.Phase, Started: step.Started,
PhaseAt: step.PhaseAt, Ended: step.Ended}
changed = changed || now != w.t.Steps[i]
w.t.Steps[i] = now
}
if changed {
w.write()
}
}
// write writes the record at the revision written last; a refusal stops the writes.
func (w *tryWriter) write() {
data, err := encodeTry(w.t)
if err != nil {
return
}
rev, err := w.tr.Tries.Update(w.ctx, w.key, data, w.rev)
if err != nil {
w.stopped = true
w.tr.logf("%s: its asks' progress is no longer written: %v", w.key, err)
return
}
w.rev = rev
}
// end stops the writes and answers the record and its revision, for the try's end.
func (w *tryWriter) end() (Try, uint64) {
w.mu.Lock()
defer w.mu.Unlock()
w.stopped = true
return w.t, w.rev
}
// HoldsANode says whether the try holds a build node now: it runs, and a step of its asks runs on a node, or it names
// no step (a checker that writes its asks only at the end, or one about to put its first).
func (t Try) HoldsANode() bool {
if t.State != TryRunning {
return false
}
if len(t.Steps) == 0 {
return true
}
return slices.ContainsFunc(t.Steps, func(s TryStep) bool { return s.State == delivery.StepRunning })
}
// Over says whether the try came to its end.
@@ -117,8 +222,8 @@ func NotChecked(repository string) error {
}
// AskTry writes a new try, queued, and answers its id: a try of the same head that waits or runs is the answer to a second
// ask, and at most MaxTries wait or run at once. trunks are the gate's trunk heads of every repository (source.main): a
// repository with none is no repository the checker checks. A trunk asked by name is its trunk head now.
// ask, and a new try is refused while MaxTries tries run on a build node or MaxTriesWaiting wait for one. trunks are the
// gate's trunk heads of every repository (source.main): a repository with none is no repository the checker checks. A trunk asked by name is its trunk head now.
func AskTry(store TryStore, trunks map[string]string, a TryAsk, now time.Time) (string, error) {
repo := strings.ToLower(strings.TrimSuffix(strings.TrimSpace(a.Repository), ".git"))
head := strings.ToLower(strings.TrimSpace(a.Head))
@@ -181,7 +286,7 @@ func AskTry(store TryStore, trunks map[string]string, a TryAsk, now time.Time) (
}); err != nil {
return "", err
}
var open []string
var holding, waiting []string
for i, cur := range tries {
if !live[i] {
continue
@@ -189,10 +294,18 @@ func AskTry(store TryStore, trunks map[string]string, a TryAsk, now time.Time) (
if cur.Repository == t.Repository && cur.Head == t.Head && cur.Trunk == t.Trunk && cur.Tests == t.Tests && cur.Level == t.Level {
return cur.ID, nil
}
open = append(open, fmt.Sprintf("%s (%s@%.12s, %s)", cur.ID, cur.Repository, cur.Head, cur.State))
said := fmt.Sprintf("%s (%s@%.12s, %s)", cur.ID, cur.Repository, cur.Head, cur.State)
if cur.HoldsANode() {
holding = append(holding, said)
} else {
waiting = append(waiting, said)
}
}
if len(open) >= MaxTries {
return "", fmt.Errorf("%d tries wait or run, the most at once: %s; ask again when one ends", len(open), strings.Join(open, ", "))
// Checked at this ask only: a try queued behind the seat's checks holds no node; tries let in while waiting may later
// all run at once, and each node's build capacity bounds what runs on it.
if len(holding) >= MaxTries || len(waiting) >= MaxTriesWaiting {
return "", fmt.Errorf("a new try is refused while %d tries run on a build node or %d wait for one; now %d run and %d wait: running %s; waiting %s; ask again when one ends",
MaxTries, MaxTriesWaiting, len(holding), len(waiting), orNone(holding), orNone(waiting))
}
t.ID = "try-" + strconv.FormatInt(now.UnixNano(), 10)
data, err := encodeTry(t)
@@ -220,6 +333,52 @@ func ReadTry(store TryStore, id string) (json.RawMessage, error) {
return raw, nil
}
// TryEnder is the tries state as the end tool sees it: one key with its revision, and a write at that revision.
type TryEnder interface {
Entry(key string) (json.RawMessage, uint64, error)
Update(key string, value []byte, revision uint64) error
}
// EndTry ends a try that waits or runs as error, saying why and that it was ended by hand (issue 734): its place is
// free at once, and the next checker_try of its head asks anew. A try that ended already is refused, naming its end.
func EndTry(store TryEnder, id, why string, now time.Time) (Try, error) {
if !tryKeyOK.MatchString(id) {
return Try{}, fmt.Errorf("%q is not a try's id, try-<digits>", id)
}
why = strings.TrimSpace(why)
if why == "" {
return Try{}, errors.New("say why the try is ended: the why is kept in what it says")
}
raw, rev, err := store.Entry(id)
if err != nil {
return Try{}, err
}
if len(raw) == 0 {
return Try{}, fmt.Errorf("no try %s: it was never asked, or it ended more than %s ago and was removed", id, TryKeep)
}
t, err := DecodeTry(raw)
if err != nil {
return Try{}, err
}
if t.Over() {
return Try{}, fmt.Errorf("%s ended already, %s at %s", id, t.State, t.Ended.Format(time.RFC3339))
}
t.State, t.Ended = TryError, now.UTC()
if !t.Started.IsZero() {
t.Took = t.Ended.Sub(t.Started).Round(time.Second).String()
}
t.Said = strings.TrimSpace("ended by hand (" + TryEndTool + "): " + why + " | " + t.Said)
t.Said = strings.TrimSuffix(t.Said, " |")
data, err := encodeTry(t)
if err != nil {
return Try{}, err
}
if err := store.Update(id, data, rev); err != nil {
return Try{}, fmt.Errorf("%s changed while it was ended, read it and end it again: %w", id, err)
}
return t, nil
}
// TryOptions narrow what a try runs; a pull request's check has none.
type TryOptions struct {
Level string
@@ -232,12 +391,14 @@ type Tryer struct {
Tries KV
Judge Judge
Bound time.Duration
Now func() time.Time
Logf func(format string, args ...any)
// NotTaken ends a try whose ask no build node took within it as error (issue 734); none when 0.
NotTaken time.Duration
Now func() time.Time
Logf func(format string, args ...any)
mu sync.Mutex
wg sync.WaitGroup
inflight map[string]bool
inflight map[string]context.CancelFunc
seen map[string]bool // looked at since this process started: a try left running is taken over the first time
}
@@ -265,25 +426,31 @@ func (tr *Tryer) Put(ctx context.Context, key string, value []byte) {
tr.logf("tries.%s holds what is not a try, left: %v", key, err)
return
}
if t.Over() {
return
}
tr.mu.Lock()
if tr.inflight == nil {
tr.inflight, tr.seen = map[string]bool{}, map[string]bool{}
tr.inflight, tr.seen = map[string]context.CancelFunc{}, map[string]bool{}
}
if tr.inflight[key] || (t.State == TryRunning && tr.seen[key]) {
if t.Over() {
// Ended while it ran (checker_try_end): its judging stops, and its asks are no longer waited for.
if stop := tr.inflight[key]; stop != nil {
stop()
}
tr.mu.Unlock()
return
}
if tr.inflight[key] != nil || (t.State == TryRunning && tr.seen[key]) {
tr.mu.Unlock()
return
}
run, stop := context.WithCancel(ctx)
tr.seen[key] = true
tr.inflight[key] = true
tr.inflight[key] = stop
tr.mu.Unlock()
tr.wg.Add(1)
go func() {
defer tr.wg.Done()
defer func() { tr.mu.Lock(); delete(tr.inflight, key); tr.mu.Unlock() }()
tr.run(ctx, key)
defer func() { tr.mu.Lock(); delete(tr.inflight, key); tr.mu.Unlock(); stop() }()
tr.run(run, key)
}()
}
@@ -334,34 +501,42 @@ func (tr *Tryer) run(ctx context.Context, key string) {
}
return
}
cs := &Case{Key: key, Group: key,
progress := &tryWriter{tr: tr, ctx: ctx, key: key, t: t, rev: rev}
cs := &Case{Key: key, Group: key, Watch: progress.heard,
Pulls: []Pull{{Repository: t.Repository, Head: t.Head, Trunk: t.Trunk}},
Trunk: map[string]string{t.Repository: t.Trunk},
Queued: t.Asked,
Progress: &Progress{},
Try: &TryOptions{Level: t.Level, Run: t.Tests}}
out, judged := judgeWithin(ctx, tr.Judge, cs, tr.Bound)
if ctx.Err() != nil {
return // left running: the checker takes it over when it starts
judging := ctx
if tr.NotTaken > 0 {
judging = buildrun.NotTakenBound(ctx, tr.NotTaken)
}
out, judged := judgeWithin(judging, tr.Judge, cs, tr.Bound)
if ctx.Err() != nil {
return // left running: the checker takes it over when it starts; or ended by hand, which wrote its end
}
t, rev = progress.end()
v := out.Verdict
t.Ended = tr.now().UTC()
t.Took = t.Ended.Sub(t.Started).Round(time.Second).String()
t.Levels, t.Machines = v.Levels, v.Machines
t.Levels, t.Machines, t.Asks = v.Levels, v.Machines, cs.Asks()
t.Said = Report(v, cs, cs.Pulls[0], out.Builds, out.Slowest, out.Judged)
for _, n := range cs.Notes() {
t.Said += " | " + n
}
switch {
case !judged || IsNodesFailure(v):
case !judged || IsNodesFailure(v) || IsLostAsk(v) || IsNotTaken(v):
// Not the change's: its ask was given up or never taken (issue 734), or a part of the mesh was not heard.
t.State = TryError
if IsLostAsk(v) || IsNotTaken(v) {
t.Failed = []string{v.Failed.Words}
}
case v.State == delivery.VerdictPass:
t.State = TryPass
default:
t.State = TryFail
if f := firstFailed(v); f != "" {
t.Failed = []string{f}
}
t.Failed = failedLines(v)
}
if t.Tests != "" {
t.Said += " | only the tests " + t.Tests + " (Go); nothing kept"
@@ -403,3 +578,11 @@ func judgeWithin(ctx context.Context, j Judge, cs *Case, bound time.Duration) (O
}
return out, true
}
// orNone joins the tries said, or says none.
func orNone(said []string) string {
if len(said) == 0 {
return "none"
}
return strings.Join(said, ", ")
}
@@ -3,6 +3,7 @@ package check
import (
"context"
"encoding/json"
"fmt"
"strings"
"testing"
"time"
@@ -23,6 +24,19 @@ func (m memTries) Create(key string, value []byte) error {
return err
}
// writtenTries is a memKV that keeps every running try it was given to write, in order.
type writtenTries struct {
*memKV
running []Try
}
func (w *writtenTries) Update(ctx context.Context, key string, value []byte, rev uint64) (uint64, error) {
if t, err := DecodeTry(value); err == nil && !t.Over() {
w.running = append(w.running, t)
}
return w.memKV.Update(ctx, key, value, rev)
}
var tryTrunks = map[string]string{repoA: trunkA, repoSDK: trunkS, hqRepo: trunkA}
func readTry(t testing.TB, kv *memKV, id string) Try {
@@ -54,14 +68,38 @@ func TestATryIsAskedQueuedOncePerHeadAndAtMostFourAtOnce(t *testing.T) { // ADR
if err != nil || again != id {
t.Errorf("a second ask of the same head answered %q, %v, want %q", again, err, id)
}
heads := []string{head2, trunkA2, trunkS2}
for i, h := range heads {
if _, err := AskTry(store, tryTrunks, TryAsk{Repository: repoA, Head: h, Trunk: trunkA}, at.Add(time.Duration(i+2)*time.Second)); err != nil {
t.Fatalf("try %d: %v", i+2, err)
}
// Issue 1890: tries on a build node (a step of theirs running, or none named yet) and tries waiting for one (queued, or
// their asks not taken) are bounded apart, MaxTries and MaxTriesWaiting; the refusal names both counts.
onNode := func(i int) Try {
return Try{ID: fmt.Sprintf("try-%d", i), Repository: repoA, Head: fmt.Sprintf("%040x", i), Trunk: trunkA, State: TryRunning,
Steps: []TryStep{{Ask: "a", State: delivery.StepRunning}}}
}
if _, err := AskTry(store, tryTrunks, TryAsk{Repository: repoSDK, Head: head1}, at.Add(time.Minute)); err == nil || !strings.Contains(err.Error(), "4 tries wait or run") {
t.Errorf("a fifth try answered %v, want refused naming the four", err)
waitingFor := func(i int) Try {
return Try{ID: fmt.Sprintf("try-%d", i), Repository: repoA, Head: fmt.Sprintf("%040x", i), Trunk: trunkA, State: TryRunning,
Steps: []TryStep{{Ask: "a", State: TryAsked}}}
}
for _, c := range []struct {
what string
holding, wait int
refused string
}{
{"four queued behind the checks, none on a node: a fifth is taken", 0, 4, ""},
{"four on a node", MaxTries, 0, "now 4 run and 0 wait"},
{"three on a node, eight waiting", 3, MaxTriesWaiting, "a new try is refused while 4 tries run on a build node or 8 wait for one; now 3 run and 8 wait"},
} {
kv := newMemKV()
for i := range c.holding + c.wait {
tr := waitingFor(i + 1)
if i < c.holding {
tr = onNode(i + 1)
}
data, _ := encodeTry(tr)
_, _ = kv.Create(context.Background(), tr.ID, data)
}
_, err := AskTry(memTries{kv}, tryTrunks, TryAsk{Repository: repoSDK, Head: head1}, at.Add(time.Minute))
if (c.refused == "") != (err == nil) || (err != nil && !strings.Contains(err.Error(), c.refused)) {
t.Errorf("%s: answered %v, want refused %q", c.what, err, c.refused)
}
}
}
@@ -89,7 +127,8 @@ func TestATryIsJudgedAsAPullRequestsHeadAndOwesTheForgeNothing(t *testing.T) { /
t.Fatal(err)
}
b := newBench(t)
tr := &Tryer{Tries: kv, Judge: b.ex}
written := &writtenTries{memKV: kv}
tr := &Tryer{Tries: written, Judge: b.ex}
raw, _, _, _ := kv.Get(ctx, id)
tr.Put(ctx, id, raw)
tr.Wait()
@@ -97,6 +136,21 @@ func TestATryIsJudgedAsAPullRequestsHeadAndOwesTheForgeNothing(t *testing.T) { /
if got.State != TryPass || got.Took == "" || got.Head != head1 || got.Trunk != trunkA || got.Said == "" || got.Levels[delivery.LevelUnit].Run != 2 {
t.Fatalf("try = %+v", got)
}
// Issue 1890: while it runs, the try names its ask as soon as it is put, and then where its step stands; the end
// keeps each ask's node and state.
if len(written.running) < 3 || len(written.running[1].Asks) != 1 || written.running[1].Steps[0].State != TryAsked ||
written.running[2].Steps[0].Node != "shanks" {
t.Errorf("the running try's writes = %+v, want its ask named while it ran, then its node", written.running)
}
if len(got.Steps) != 1 || got.Steps[0].Node != "shanks" || got.Steps[0].State != delivery.StepPassed {
t.Errorf("the ended try's steps = %+v", got.Steps)
}
// The try names the asks it put, so the delivery view counts their steps as a try's (build_usage).
if len(got.Asks) != 1 {
t.Errorf("asks = %v, want the one check ask the try put", got.Asks)
} else if _, err := delivery.AskStepKey("shanks", got.Asks[0]); err != nil {
t.Errorf("the try names %q, which is no ask id: %v", got.Asks[0], err)
}
fetch := b.steps.tool("fetch")
if len(fetch) != 1 || fetch[0].args["commit"] != head1 || fetch[0].args["base"] != trunkA || fetch[0].args["ref"] != "" {
t.Errorf("fetch = %+v, want the head merged onto the trunk, with no pull request's ref", fetch)
@@ -0,0 +1,78 @@
package check
import (
"sort"
"strings"
"sync"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// Turns holds the gate's merge turns (novox/hq ADR 0548 §3), by repository, as a watch of the gate's turns bucket puts them
// in: the checker claims no check of a repository a turn names, except the check of a waiting delivery's own heads, so an
// approved merge goes before a check not yet started. A bucket not there is read as empty.
type Turns struct {
mu sync.Mutex
byRepo map[string]delivery.Turn
}
// Put takes a value of the turns bucket in, or its delete, and answers whether what the checker holds changed. A value that
// is not a turn is dropped as a delete would be: a turn the checker cannot read holds nothing.
func (t *Turns) Put(key string, value []byte, deleted bool) bool {
repository, err := delivery.ParseTurnsKey(key)
if err != nil {
return false
}
t.mu.Lock()
defer t.mu.Unlock()
if t.byRepo == nil {
t.byRepo = map[string]delivery.Turn{}
}
cur, had := t.byRepo[repository]
if !deleted {
if turn, err := delivery.DecodeTurn(value); err == nil {
t.byRepo[repository] = turn
return !had || !delivery.SameTurn(cur, turn)
}
}
delete(t.byRepo, repository)
return had
}
// Holding names the first of the repositories (by name) a turn holds off the check of the heads, and the delivery that
// waits first in it; false when none does.
func (t *Turns) Holding(repositories []string, heads string) (string, delivery.TurnWait, bool) {
if t == nil {
return "", delivery.TurnWait{}, false
}
sorted := make([]string, 0, len(repositories))
for _, r := range repositories {
sorted = append(sorted, strings.ToLower(r))
}
sort.Strings(sorted)
t.mu.Lock()
defer t.mu.Unlock()
for _, r := range sorted {
turn, ok := t.byRepo[r]
if !ok || !delivery.TurnHolds(turn, heads) {
continue
}
for _, w := range turn.Waiting {
if w.Heads != heads {
return r, w, true
}
}
}
return "", delivery.TurnWait{}, false
}
// All is every turn held now, by repository (a copy).
func (t *Turns) All() map[string]delivery.Turn {
t.mu.Lock()
defer t.mu.Unlock()
out := make(map[string]delivery.Turn, len(t.byRepo))
for r, turn := range t.byRepo {
out[r] = turn
}
return out
}
@@ -93,11 +93,11 @@ func TestManifestDeclarations(t *testing.T) { // 1678/AC1
t.Errorf("module = %q", m.Module)
}
// delivery.CheckerKeptHistory, CheckerOwedHistory, PreviewsHistory, CheckerVerdictsHistory; the names follow the SDK's buckets.
if want := []state{{Name: "kept", History: 1}, {Name: "owed", History: 5}, {Name: "previews", History: 5}, {Name: "tries", History: 1}, {Name: "verdicts", History: 5}}; !reflect.DeepEqual(m.State, want) {
if want := []state{{Name: "hearing", History: 1}, {Name: "kept", History: 1}, {Name: "owed", History: 5}, {Name: "previews", History: 5}, {Name: "tries", History: 1}, {Name: "verdicts", History: 5}}; !reflect.DeepEqual(m.State, want) {
t.Errorf("state = %+v, want %+v", m.State, want)
}
if want := []string{"mesh-build-go.steps", "mesh-build-go.toolchains", "mesh-build-node.steps", "mesh-build-node.toolchains", "mesh-controller.assignments",
"mesh-delivery-assigner.compositions", "mesh-delivery-builder.builds", "mesh-delivery-gate.source", "mesh-delivery-indexer.candidates",
"mesh-delivery-assigner.compositions", "mesh-delivery-builder.builds", "mesh-delivery-gate.source", "mesh-delivery-gate.turns", "mesh-delivery-indexer.candidates",
"mesh-delivery-indexer.manifests", "mesh-delivery-indexer.modules", "mesh-delivery-watchman.presence", "node-engine.have"}; !reflect.DeepEqual(sorted(m.Reads), want) {
t.Errorf("reads = %v, want %v", m.Reads, want)
}
@@ -124,8 +124,8 @@ func TestManifestDeclarations(t *testing.T) { // 1678/AC1
t.Errorf("streams = %+v, want %+v", m.Streams, want)
}
// The read tools (issue 1686) are served by the tools bundle, not by the checker's process.
// The try's two tools (ADR 0501 §5) beside them: mesh-controller's try calls them.
if want := []string{"checker_kept", "checker_previews", "checker_pull", "checker_status", "checker_try", "checker_try_read", "checker_verdicts"}; !reflect.DeepEqual(sorted(m.Tools), want) {
// The try's tools (ADR 0501 §5; the end by hand, issue 734) beside them: mesh-controller's try calls them.
if want := []string{"checker_kept", "checker_previews", "checker_pull", "checker_status", "checker_try", "checker_try_end", "checker_try_read", "checker_verdicts"}; !reflect.DeepEqual(sorted(m.Tools), want) {
t.Errorf("tools = %v, want %v", m.Tools, want)
}
for _, k := range []string{"logs", "emits", "claims", "provides", "requires", "defines-seats", "seat", "seats", "bus", "config"} {
@@ -10,6 +10,7 @@
{ "name": "checks", "events": ["preview.**", "owed.**"] }
],
"state": [
{ "name": "hearing", "history": 1 },
{ "name": "kept", "history": 1 },
{ "name": "owed", "history": 5 },
{ "name": "previews", "history": 5 },
@@ -25,6 +26,7 @@
"mesh-delivery-assigner.compositions",
"mesh-delivery-builder.builds",
"mesh-delivery-gate.source",
"mesh-delivery-gate.turns",
"mesh-delivery-indexer.candidates",
"mesh-delivery-indexer.manifests",
"mesh-delivery-indexer.modules",
@@ -41,6 +43,7 @@
"checker_pull",
"checker_status",
"checker_try",
"checker_try_end",
"checker_try_read",
"checker_verdicts"
],
@@ -91,6 +91,28 @@ func treesVia(ctx context.Context, call caller) bridge.Trees {
}
}
// listedVia asks the forge for the files directly under one directory of a commit through call: gitea_commit_trees with
// listed (issue 1126), answer {listed: {directory: {name: blob id}}}.
func listedVia(ctx context.Context, call caller) bridge.Listed {
return func(owner, repo, sha, dir string) (map[string]string, error) {
raw, err := call(ctx, commitTreesTool, map[string]any{"owner": owner, "repo": repo, "sha": sha, "listed": []string{dir}})
if err != nil {
return nil, err
}
var a struct {
Listed map[string]map[string]string `json:"listed"`
}
if err := json.Unmarshal(raw, &a); err != nil {
return nil, fmt.Errorf("the answer of %s is not {listed}: %w", commitTreesTool, err)
}
if a.Listed == nil {
// A forge module from before listed answers none: placement cannot be read, never read as empty.
return nil, fmt.Errorf("the answer of %s lists no directory; the forge module does not list yet", commitTreesTool)
}
return a.Listed[dir], nil
}
}
// getFileTool is the forge's tool for one file at a ref (mesh-catalog modules/gitea tools/index.ts gitea_get_file):
// arguments owner, repo, path and ref, answer {file: {path, ref, sha, size, content}}, the content decoded to text
// (client.ts getFile). A path the forge does not have fails with `Gitea API <path>: 404 <body>` (client.ts request).
@@ -117,8 +117,8 @@ type publisher interface {
// handle turns one event into its message on FORGE and acknowledges it. An event it cannot read is terminated, never
// redelivered, and the log names its subject and why. A message it could not write stays unacknowledged: the server
// delivers it again (the server's ack wait is the only clock).
func handle(ctx context.Context, m delivered, pub publisher, trees bridge.Trees, files bridge.Files, logger *log.Logger) {
msg, err := bridge.Convert(m.Subject(), m.Data(), trees, files)
func handle(ctx context.Context, m delivered, pub publisher, trees bridge.Trees, files bridge.Files, list bridge.Listed, logger *log.Logger) {
msg, err := bridge.ConvertWith(m.Subject(), m.Data(), trees, files, list)
switch {
case errors.Is(err, bridge.ErrRetry):
logger.Printf("%v; heard again in %s", err, retryDelay)
@@ -237,7 +237,7 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
// Bounded: the turn is held, and no sync or other event goes on without it.
hctx, cancel := context.WithTimeout(ctx, syncDeadline)
defer cancel()
handle(hctx, m, rec, treesVia(hctx, call), filesVia(hctx, call), logger)
handle(hctx, m, rec, treesVia(hctx, call), filesVia(hctx, call), listedVia(hctx, call), logger)
})
},
jetstream.ConsumeErrHandler(func(_ jetstream.ConsumeContext, err error) { logger.Printf("hearing the bus: %v", err) }))
@@ -81,7 +81,7 @@ func TestReadableEventIsWrittenOnForgeAndAcked(t *testing.T) { // 1235/AC2 (unit
})
var logs bytes.Buffer
m, pub := &fakeMsg{subject: "mesh.mod.gitea.event.pull.updated", data: body(nil)}, &fakePub{}
handle(context.Background(), m, pub, nil, someFiles, log.New(&logs, "", 0))
handle(context.Background(), m, pub, nil, someFiles, nil, log.New(&logs, "", 0))
if len(pub.subjects) != 1 || pub.subjects[0] != "mesh.stream.mesh-delivery-forge-bridge.event.forge.novox.mesh-delivery.pull" {
t.Fatalf("published on %v", pub.subjects)
}
@@ -97,7 +97,7 @@ func TestReadableEventIsWrittenOnForgeAndAcked(t *testing.T) { // 1235/AC2 (unit
func TestUnreadableEventIsTerminatedAndLogged(t *testing.T) { // 1235/AC5
var logs bytes.Buffer
m, pub := &fakeMsg{subject: "mesh.mod.gitea.event.pull.updated", data: body(map[string]any{"head_sha": "short"})}, &fakePub{}
handle(context.Background(), m, pub, nil, someFiles, log.New(&logs, "", 0))
handle(context.Background(), m, pub, nil, someFiles, nil, log.New(&logs, "", 0))
if len(pub.subjects) != 0 {
t.Errorf("a message was written: %v", pub.subjects)
}
@@ -112,7 +112,7 @@ func TestUnreadableEventIsTerminatedAndLogged(t *testing.T) { // 1235/AC5
func TestUnwrittenMessageIsNotAcked(t *testing.T) { // 1235 (a failed write is redelivered by the server, not retried here)
var logs bytes.Buffer
m, pub := &fakeMsg{subject: "mesh.mod.gitea.event.pull.updated", data: body(nil)}, &fakePub{err: errors.New("no stream")}
handle(context.Background(), m, pub, nil, someFiles, log.New(&logs, "", 0))
handle(context.Background(), m, pub, nil, someFiles, nil, log.New(&logs, "", 0))
if m.acked != 0 || m.termd != 0 {
t.Errorf("acked %d terminated %d, want neither", m.acked, m.termd)
}
@@ -134,7 +134,7 @@ func TestTreesRefusedLeavesTheEventForTenSecondsAndLogs(t *testing.T) { // 1236/
call := func(context.Context, string, any) (json.RawMessage, error) {
return nil, errors.New("the tree listing of novox/mesh-delivery was cut")
}
handle(context.Background(), m, pub, treesVia(context.Background(), call), filesVia(context.Background(), call), log.New(&logs, "", 0))
handle(context.Background(), m, pub, treesVia(context.Background(), call), filesVia(context.Background(), call), nil, log.New(&logs, "", 0))
if len(pub.subjects) != 0 {
t.Errorf("a message was written: %v", pub.subjects)
}
@@ -163,7 +163,7 @@ func TestMergeCallsTheToolOnceAndWritesTheMergedMessage(t *testing.T) { // 1236/
}
return json.RawMessage(`{"root":"r0","trees":{"":"t0","modules/a":"t1"}}`), nil
}
handle(context.Background(), m, pub, treesVia(context.Background(), call), filesVia(context.Background(), call), log.New(&logs, "", 0))
handle(context.Background(), m, pub, treesVia(context.Background(), call), filesVia(context.Background(), call), nil, log.New(&logs, "", 0))
if len(keys) == 0 || keys[0] != "gitea.gitea_commit_trees" || slices.Index(keys[1:], "gitea.gitea_commit_trees") >= 0 {
t.Fatalf("calls %v, want the trees once and first", keys)
}
+1 -1
View File
@@ -3,7 +3,7 @@ module mesh-delivery-forge-bridge
go 1.26.0
require (
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117
github.com/nats-io/nats-server/v2 v2.11.17
github.com/nats-io/nats.go v1.54.0
)
+2 -4
View File
@@ -1,7 +1,5 @@
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261022190740-104533e8c0c9 h1:Ufmd2cYEXICQ817IGflccP9bexABE4WhIO49HPbaoSo=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261022190740-104533e8c0c9/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585 h1:7QoFH5qjXRpx5M4uPdYsVJ5qKl2082gq8n0lFxhoNIg=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117 h1:88CNZOoQgu7YPYKW16qkWT8Gw9Ucg07aTrIImA2nYUk=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026210326-979644420117/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op/go.mod h1:FQyySiasQQM8735Ddel3MRojmy4dA1IqCeyJ5jmPMbI=
github.com/google/go-tpm v0.9.8 h1:slArAR9Ft+1ybZu0lBwpSmpwhRXaa85hWtMinMyRAWo=
@@ -7,6 +7,7 @@ import (
"encoding/json"
"errors"
"fmt"
"maps"
"path"
"regexp"
"slices"
@@ -88,6 +89,17 @@ func unreadable(subject, why string, args ...any) error {
// the trunk asks trees once and becomes a merged message; a merge into any other branch is a closed pull message, as the
// pull request left its group. A forge that cannot say is ErrRetry; a merge with no module.json is ErrNoModules.
func Convert(subject string, data []byte, trees Trees, files Files) (Message, error) {
return ConvertWith(subject, data, trees, files, nil)
}
// Listed asks the forge for the files directly under one directory of a commit, by name to blob id (gitea_commit_trees
// with listed); a directory the commit lacks lists none.
type Listed func(owner, repo, sha, dir string) (map[string]string, error)
// ConvertWith is Convert; with list, a merge of the catalogue repository (delivery.CatalogueRepository) also carries every
// placement file of its merge commit (novox/hq ADR 0549 §2, issue 1126), read whole or not at all: a forge that cannot
// say is ErrRetry, as for the manifests, so the indexer never takes a part of placement for all of it.
func ConvertWith(subject string, data []byte, trees Trees, files Files, list Listed) (Message, error) {
var action string
switch subject {
case UpdatedSubject:
@@ -114,7 +126,7 @@ func Convert(subject string, data []byte, trees Trees, files Files) (Message, er
return Message{}, unreadable(subject, "no branch")
}
if subject == MergedSubject && p.Base == Trunk {
return merged(subject, p, trees, files)
return merged(subject, p, trees, files, list)
}
f := delivery.Forge{Kind: delivery.ForgePull, Repository: p.Owner + "/" + p.Repo, Branch: p.Head, Number: p.Number,
Head: p.HeadSHA, Action: action}
@@ -227,7 +239,7 @@ func inputsOf(owner, repo, ref string, dirs []string, files Files) (manifests ma
}
// merged is the merged message of a merge into the trunk: the trees of the merge commit, asked once.
func merged(subject string, p pull, trees Trees, files Files) (Message, error) {
func merged(subject string, p pull, trees Trees, files Files, list Listed) (Message, error) {
if !sha40.MatchString(p.MergeSHA) {
return Message{}, unreadable(subject, "the merge commit %q is not 40 hex digits", p.MergeSHA)
}
@@ -252,8 +264,15 @@ func merged(subject string, p pull, trees Trees, files Files) (Message, error) {
if err != nil {
return Message{}, unreadable(subject, "%v", err)
}
body, err := delivery.EncodeForge(delivery.Forge{Kind: delivery.ForgeMerged, Repository: repository, Branch: p.Head, Number: p.Number,
Head: p.HeadSHA, Merge: p.MergeSHA, Root: root, Trees: dirs, Manifests: manifests, Locks: locks})
f := delivery.Forge{Kind: delivery.ForgeMerged, Repository: repository, Branch: p.Head, Number: p.Number,
Head: p.HeadSHA, Merge: p.MergeSHA, Root: root, Trees: dirs, Manifests: manifests, Locks: locks}
if list != nil && delivery.SameRepository(repository, delivery.CatalogueRepository) {
if f.Placements, err = placementsOf(p.Owner, p.Repo, p.MergeSHA, list, files); err != nil {
return Message{}, fmt.Errorf("%w: %s: the placement files of %s at the merge commit %s: %v", ErrRetry, subject, repository, p.MergeSHA, err)
}
f.PlacementsRead = true
}
body, err := delivery.EncodeForge(f)
if err != nil {
return Message{}, unreadable(subject, "%v", err)
}
@@ -350,3 +369,30 @@ func OpenMessage(answer []byte) (Message, error) {
}
return Message{Event: delivery.OpenEvent(), Data: body}, nil
}
// placementsOf is every placement file of the catalogue at a commit, by node, its bytes exact: each `<node>.json` directly
// under delivery.PlacementDir. A file whose name is no node is left out and the rest read; the catalogue's check refuses
// such a file before it merges (ADR 0549 §1). Nil when there is none.
func placementsOf(owner, repo, sha string, list Listed, files Files) (map[string]string, error) {
names, err := list(owner, repo, sha, delivery.PlacementDir)
if err != nil {
return nil, err
}
var out map[string]string
for _, name := range slices.Sorted(maps.Keys(names)) {
file := delivery.PlacementDir + "/" + name
node, ok := delivery.PlacementFileNode(file)
if !ok {
continue
}
raw, err := files(owner, repo, file, sha)
if err != nil {
return nil, fmt.Errorf("%s: %w", file, err)
}
if out == nil {
out = map[string]string{}
}
out[node] = string(raw)
}
return out, nil
}
@@ -79,6 +79,47 @@ func TestMergedCarriesManifestsAndLocks(t *testing.T) { // 1448/AC1
if want := map[string]map[string]string{"modules/a": {"go.sum": hexSum(sum)}}; !reflect.DeepEqual(f.Locks, want) {
t.Errorf("locks = %v, want %v", f.Locks, want)
}
if f.Placements != nil || f.PlacementsRead {
t.Errorf("a merge of another repository than the catalogue carries placement: %v", f.Placements)
}
// Issue 1126 (ADR 0549 §2): a merge of the catalogue carries every placement file of its merge commit, by node; a file
// whose name is no node is left out; a listing or a file the forge cannot give is a retry, never a part.
catalogue := mergedBody(map[string]any{"repo": "mesh-catalog"})
forge.have["placement/ace.json"] = []byte(`{"v":1,"node":"ace","modules":[]}`)
forge.have["placement/g14.json"] = []byte(`{"v":1,"node":"g14","modules":[{"module":"zsh"}]}`)
listing := map[string]string{"ace.json": "b1", "g14.json": "b2", "README.md": "b3", "a.b.json": "b4"}
var listedAt []string
list := func(owner, repo, sha, dir string) (map[string]string, error) {
listedAt = append(listedAt, owner+"/"+repo+"@"+sha+":"+dir)
return listing, nil
}
msg, err = ConvertWith(mergedSubject, catalogue, treesOf(&calls, "r0", map[string]string{"modules/a": "t1"}, nil), forge.get, list)
if err != nil {
t.Fatal(err)
}
if f, err = delivery.DecodeForge(msg.Data); err != nil {
t.Fatal(err)
}
want := map[string]string{"ace": `{"v":1,"node":"ace","modules":[]}`, "g14": `{"v":1,"node":"g14","modules":[{"module":"zsh"}]}`}
if !f.PlacementsRead || !reflect.DeepEqual(f.Placements, want) || !reflect.DeepEqual(listedAt, []string{"novox/mesh-catalog@" + merge + ":placement"}) {
t.Errorf("placements = %v (read %v), listed at %v", f.Placements, f.PlacementsRead, listedAt)
}
listing = map[string]string{}
if msg, err = ConvertWith(mergedSubject, catalogue, treesOf(&calls, "r0", map[string]string{"modules/a": "t1"}, nil), forge.get, list); err != nil {
t.Fatal(err)
}
if f, _ = delivery.DecodeForge(msg.Data); !f.PlacementsRead || f.Placements != nil {
t.Errorf("a catalogue with no placement file says it read none: %v, %v", f.Placements, f.PlacementsRead)
}
listing = map[string]string{"gone.json": "b5"}
if _, err := ConvertWith(mergedSubject, catalogue, treesOf(&calls, "r0", map[string]string{"modules/a": "t1"}, nil), forge.get, list); !errors.Is(err, ErrRetry) || !strings.Contains(err.Error(), "placement/gone.json") {
t.Errorf("a placement file the forge cannot give: %v", err)
}
failing := func(_, _, _, _ string) (map[string]string, error) { return nil, errors.New("forge down") }
if _, err := ConvertWith(mergedSubject, catalogue, treesOf(&calls, "r0", map[string]string{"modules/a": "t1"}, nil), forge.get, failing); !errors.Is(err, ErrRetry) {
t.Errorf("a listing the forge cannot give: %v", err)
}
}
func TestPullWithKnownDirsCarriesTheSameAtTheHead(t *testing.T) { // 1448/AC2
@@ -0,0 +1,111 @@
// Command mesh-delivery-indexer-tools serves the indexer's doing tool (novox/hq issue 1945): the runtime launches it for a
// call. indexer_take_in writes one value of the indexer's takeins state and waits for the indexer's answer there. The
// indexer is the other program: it serves no request subject (ADR 0437 §4). The read tools of the indexer's states are
// mesh-delivery-view's (indexer_module, indexer_modules, indexer_candidate, indexer_candidates).
package main
import (
"context"
"encoding/json"
"fmt"
"os"
"strings"
"sync"
"time"
stdio "git.novox.be/novox/mesh-sdk/go"
"mesh-delivery-indexer/internal/index"
)
// asker writes a takeins value and waits for the indexer's answer to it.
type asker func(key string, value []byte) (index.TakeInAsk, error)
// Tools are the indexer's tools.
func Tools(ask asker, now func() time.Time) []stdio.Tool {
str := func(args map[string]any, k string) string { s, _ := args[k].(string); return strings.TrimSpace(s) }
return []stdio.Tool{
{Name: index.TakeInTool,
Description: "Take a registered module in now: the indexer reads its manifest from the repository and commit the controller's " +
"build records register it from (at the head the gate's source names, when it names one), whatever it tried before, and " +
"writes it to its modules state as a merge would. For a module registered from outside the indexed repositories whose " +
"repository the forge could not read (a 404 before the forge's account could read it). Answers stored, from which " +
"repository, commit and path, or why not (held already, no registration, the forge's error, no manifest there). " +
"The indexer reads such a module again by itself on the next source revision and merged message; this asks now. " +
"indexer_module (mesh-delivery-view) reads the result.",
Input: map[string]any{"type": "object", "required": []string{"module"}, "properties": map[string]any{
"module": map[string]any{"type": "string", "description": "the module's name, as the catalogue registers it"},
"why": map[string]any{"type": "string", "description": "why it is taken in now, in plain words"}}},
Run: func(args map[string]any) (any, error) {
key, value, err := index.NewTakeIn(str(args, "module"), str(args, "why"), stdio.CallerOf(args), now())
if err != nil {
return nil, err
}
a, err := ask(key, value)
if err != nil {
return nil, err
}
return a.Result, nil
}},
}
}
// askTakeIn writes the takeins value at key and waits, on a watch of the key, at most bound for the indexer's answer.
func askTakeIn(s stdio.KeptState, bound time.Duration) asker {
return func(key string, value []byte) (index.TakeInAsk, error) {
answered := make(chan index.TakeInAsk, 1)
var rev uint64
var mu sync.Mutex
take := func(revision uint64, v []byte) bool {
mu.Lock()
mine := rev
mu.Unlock()
if mine == 0 || revision <= mine {
return false
}
a, err := index.DecodeTakeIn(v)
if err != nil || a.Answered.IsZero() || a.Result == nil {
return false
}
select {
case answered <- a:
default:
}
return true
}
if err := s.Watch(key, func(c stdio.StateChange) error {
if c.Op == "put" {
take(c.Revision, c.Value)
}
return nil
}); err != nil {
return index.TakeInAsk{}, fmt.Errorf("watching %s.%s: %w", index.TakeInsState, key, err)
}
r, err := s.Put(key, json.RawMessage(value))
if err != nil {
return index.TakeInAsk{}, fmt.Errorf("the ask could not be written to %s: %w", index.TakeInsBucket, err)
}
mu.Lock()
rev = r
mu.Unlock()
// The watch may have heard the answer before the revision was known: read it once too.
if e, err := s.Get(key); err == nil && e != nil {
take(e.Revision, e.Value)
}
wait, cancel := context.WithTimeout(context.Background(), bound) // a context's bound, no timer (ADR 0437 §8)
defer cancel()
select {
case a := <-answered:
return a, nil
case <-wait.Done():
return index.TakeInAsk{}, fmt.Errorf("the indexer did not answer within %s: the take-in is still running (a forge read can take 30 s); its answer will be in %s.%s", bound, index.TakeInsBucket, key)
}
}
}
func main() {
if err := stdio.Serve("", Tools(askTakeIn(stdio.State(index.TakeInsState), 25*time.Second), time.Now)); err != nil {
fmt.Fprintf(os.Stderr, "[mesh-delivery-indexer-tools] %v\n", err)
os.Exit(1)
}
}
@@ -1,7 +1,8 @@
// Command mesh-delivery-indexer is delivery's indexer (novox/hq ADR 0437 §4, §5, issue 1238): it hears the forge
// bridge's messages on the stream FORGE and keeps the candidate groups (candidates.<group>): the pull requests that share
// a branch, their heads and the modules each touches. It reads the controller's builds to know which module a changed
// directory is. It serves no request subject and holds no timer: every reaction is a message or a watch (ADR 0437 §8).
// directory is. It serves no request subject and holds no timer: every reaction is a message or a watch (ADR 0437 §8). Its
// tool indexer_take_in (the program mesh-delivery-indexer-tools) asks through its state takeins, which it watches.
//
// Its build of 2026-10-24 00:37 named a program whose push never reached the artifact store, and a retry sends that
// same build again; this paragraph was added only to give the indexer a new build (issue 1933).
@@ -90,6 +91,58 @@ func pinned(want string) *tls.Config {
}
}
// awaitedStore is a state bound when it appears (process.AwaitBucket) as the indexer's store: until then every call is
// process.ErrNotYet.
type awaitedStore struct{ b *process.AwaitedBucket }
func (s awaitedStore) store() (kvStore, error) {
kv := s.b.KV()
if kv == nil {
return kvStore{}, process.ErrNotYet
}
return kvStore{kv}, nil
}
func (s awaitedStore) Get(ctx context.Context, key string) ([]byte, uint64, bool, error) {
st, err := s.store()
if err != nil {
return nil, 0, false, err
}
return st.Get(ctx, key)
}
func (s awaitedStore) Create(ctx context.Context, key string, value []byte) error {
st, err := s.store()
if err != nil {
return err
}
return st.Create(ctx, key, value)
}
func (s awaitedStore) Update(ctx context.Context, key string, value []byte, revision uint64) error {
st, err := s.store()
if err != nil {
return err
}
return st.Update(ctx, key, value, revision)
}
func (s awaitedStore) Delete(ctx context.Context, key string, revision uint64) error {
st, err := s.store()
if err != nil {
return err
}
return st.Delete(ctx, key, revision)
}
func (s awaitedStore) Keys(ctx context.Context) ([]string, error) {
st, err := s.store()
if err != nil {
return nil, err
}
return st.Keys(ctx)
}
// kvStore is the candidates bucket as the indexer's store: compare-and-set through Create, Update and Delete at a revision.
type kvStore struct{ kv jetstream.KeyValue }
@@ -139,6 +192,7 @@ type delivered interface {
Subject() string
Data() []byte
Ack() error
DoubleAck(ctx context.Context) error
Term() error
}
@@ -159,7 +213,12 @@ func handle(ctx context.Context, m delivered, ix *index.Indexer, logger *log.Log
case err != nil:
logger.Printf("%s: %s#%d left for redelivery: %v", m.Subject(), f.Repository, f.Number, err)
default:
_ = m.Ack()
// Waits for the server, so the consumer's info no longer counts the message as unacknowledged when the fill that
// follows asks whether the stream is applied (issue 1280); a failed ack is redelivered after the ack wait, which
// kicks the fill again.
if err := m.DoubleAck(ctx); err != nil {
logger.Printf("%s: %s#%d applied, its acknowledgement was not confirmed: %v", m.Subject(), f.Repository, f.Number, err)
}
}
}
@@ -252,6 +311,12 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
if err != nil {
return err
}
// Its taken state (ADR 0556 clause 1) is bound when it appears, as every state a delivery adds (issue 1790): until then
// a merge's taken key is said and not written, and the rest of the indexer runs.
taken := process.AwaitBucket(ctx, js, conn, delivery.TakenBucket, "no repository's taken key is written", logger.Printf)
// Its placement state (ADR 0549 §2) is bound when it appears, as every state a delivery adds (issue 1790): until then a
// catalogue merge's placement is said and not written, and the rest of the indexer runs.
placement := process.AwaitBucket(ctx, js, conn, delivery.PlacementBucket, "no catalogue merge's placement is written", logger.Printf)
// The controller's build records only name the directory of a module whose key names none (fillPaths): without them
// such a module waits for its path, every other is indexed (issue 1945).
builds := process.OptionalBucket(ctx, js, delivery.BuildsBucket, "a module whose key names no directory is not given one from the controller's build records until the indexer starts again", logger.Printf)
@@ -266,11 +331,12 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
}
}
errs := make(chan error, 1)
var buildsWatch jetstream.KeyWatcher
if builds != nil {
if w, err := settle(ctx, builds, each); err != nil {
process.NotWatched(logger.Printf, delivery.BuildsBucket, err, "a module whose key names no directory is not given one until the indexer starts again")
} else {
go func() { errs <- follow(ctx, w, each) }()
buildsWatch = w
}
}
@@ -284,8 +350,9 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
return err
}
defer calls.Close()
ix := &index.Indexer{Store: kvStore{candidates}, Modules: kvStore{modules}, Manifests: kvStore{manifests}, Builds: known, Logf: logger.Printf,
Named: namedVia(calls.Call), Trunk: trunkVia(calls.Call), Contains: containsVia(calls.Call)}
ix := &index.Indexer{Store: kvStore{candidates}, Modules: kvStore{modules}, Manifests: kvStore{manifests}, Taken: awaitedStore{taken}, Builds: known, Logf: logger.Printf,
Named: namedVia(calls.Call), Trunk: trunkVia(calls.Call), Contains: containsVia(calls.Call), Read: readVia(calls.Call),
Placement: awaitedStore{placement}, Say: func(event string, body any) error { return calls.Emit(event, body) }}
// The gate's source bucket (mesh-delivery-gate_source) comes with the gate (release 3e, novox/hq ADR 0450 §1): its key
// main is the mesh's source revision, which stamps the modules (issue 1344). Where it does not exist yet the indexer does
// not wait for it (that needs a timer, ADR 0437 §8): it says so, and is started again with the gate.
@@ -336,6 +403,40 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
if err := ix.RecomputeAtStart(ctx); err != nil {
logger.Printf("computing the new-form fingerprints at start: %v", err)
}
// A module registered in the catalogue from its own repository gets its manifest indexed when the controller's records
// name it and the bucket holds none (registered.go); the records that follow are heard below.
if err := ix.IndexRegisteredAll(ctx); err != nil {
logger.Printf("indexing the registered modules at start: %v", err)
}
if buildsWatch != nil {
go func() {
errs <- follow(ctx, buildsWatch, func(e jetstream.KeyValueEntry) {
each(e)
if !opOf(e) {
return
}
var r delivery.BuildRecord
if json.Unmarshal(e.Value(), &r) != nil || r.Module == "" {
return
}
if err := ix.IndexRegisteredModule(ctx, r.Module); err != nil {
logger.Printf("the registered module %s is not indexed: %v", r.Module, err)
}
})
}()
}
// A person takes a registered module in now with indexer_take_in, which writes the takeins state (issue 1945): watched,
// as every other input. A module the forge could not read is also read again by itself on the next source revision and
// merged message (index.RetryUnread).
// A state not on the bus yet at this start is watched once it appears (issue 1790).
process.AwaitBucket(ctx, js, conn, index.TakeInsBucket, "indexer_take_in is not answered; a module the forge could not read is still read again by itself", logger.Printf).Ready(func(kv jetstream.KeyValue) {
eachTakeIn := func(e jetstream.KeyValueEntry) { takenIn(ctx, ix, kv, e, logger.Printf) }
if tw, err := settle(ctx, kv, eachTakeIn); err != nil {
process.NotWatched(logger.Printf, index.TakeInsBucket, err, "indexer_take_in is not answered until the indexer starts again")
} else {
go func() { errs <- follow(ctx, tw, eachTakeIn) }()
}
})
// The modules the trunk no longer holds go from the bucket (issue 1900); a merge does the same from its own message.
if err := ix.Reconcile(ctx); err != nil {
logger.Printf("dropping the modules the trunk does not hold at start: %v", err)
@@ -373,7 +474,17 @@ func run(ctx context.Context, getenv func(string) string, logOut io.Writer) erro
// Essential: the forge's merges are what the indexer indexes.
return fmt.Errorf("the consumer %s_%s on %s cannot be read (%v): the indexer hears no merge without it, so it stops; where the bus refused it or did not answer, this module's consumes (made by the controller with its grant) have not reached the bus yet", cred.Node, cred.Module, forgeStream, err)
}
cc, err := cons.Consume(func(m jetstream.Msg) { handle(ctx, m, ix, logger) },
// A merge applied while the taken state was missing, and the gate's heads once the forge stream is applied whole, get
// their taken key (issue 1280): when the state appears (Ready runs once, now or then), and after each applied message.
ix.Drained = func(ctx context.Context) (bool, error) {
info, err := cons.Info(ctx)
if err != nil {
return false, err
}
return info.NumPending == 0 && info.NumAckPending == 0, nil
}
taken.Ready(func(jetstream.KeyValue) { ix.KickFill(ctx) })
cc, err := cons.Consume(func(m jetstream.Msg) { handle(ctx, m, ix, logger); ix.KickFill(ctx) },
jetstream.ConsumeErrHandler(func(_ jetstream.ConsumeContext, err error) { logger.Printf("hearing the bus: %v", err) }))
if err != nil {
return err
@@ -16,12 +16,18 @@ type fakeMsg struct {
subject string
data []byte
acked, termd int
doubled int // acked by DoubleAck, which waits for the server
}
func (f *fakeMsg) Subject() string { return f.subject }
func (f *fakeMsg) Data() []byte { return f.data }
func (f *fakeMsg) Ack() error { f.acked++; return nil }
func (f *fakeMsg) Term() error { f.termd++; return nil }
func (f *fakeMsg) DoubleAck(context.Context) error {
f.acked++
f.doubled++
return nil
}
func (f *fakeMsg) Term() error { f.termd++; return nil }
type memStore struct{ vals map[string][]byte }
@@ -53,7 +59,7 @@ func TestAMessageIsIndexedAndAcked(t *testing.T) { // 1238/AC2 (unit part)
m := &fakeMsg{subject: "mesh.stream.mesh-delivery-forge-bridge.event.forge.novox.mesh-delivery.pull", data: data}
handle(context.Background(), m, ix, log.New(&bytes.Buffer{}, "", 0))
c, err := delivery.DecodeCandidate(st.vals["feat/x"])
if err != nil || len(c.Pulls) != 1 || c.Pulls[0].Modules[0] != "a" || m.acked != 1 || m.termd != 0 {
if err != nil || len(c.Pulls) != 1 || c.Pulls[0].Modules[0] != "a" || m.acked != 1 || m.doubled != 1 || m.termd != 0 {
t.Errorf("candidate %+v err %v acked %d terminated %d", c, err, m.acked, m.termd)
}
}
@@ -2,11 +2,16 @@ package main
import (
"context"
"crypto/sha256"
"encoding/hex"
"encoding/json"
"fmt"
"path"
"strings"
"git.novox.be/novox/mesh-sdk/go/delivery"
"mesh-delivery-indexer/internal/index"
)
// commitTreesTool is the forge's tool for a commit's trees (mesh-catalog modules/gitea): with `paths`, by module directory
@@ -93,3 +98,75 @@ func containsVia(call toolCaller) func(ctx context.Context, repository, commit,
return *a.Contains, nil
}
}
// getFileTool is the forge's tool for one file at a ref (mesh-catalog modules/gitea gitea_get_file): arguments owner,
// repo, path and ref, answer {file: {content}}; a path it does not have fails with a 404 in its words.
const getFileTool = "gitea.gitea_get_file"
// lockFiles are read beside a manifest, as the forge bridge reads them (bridge.LockFiles).
var lockFiles = []string{"go.sum", "package-lock.json"}
// readVia reads a repository at a commit as the forge bridge does for a merged message: the trees by gitea_commit_trees,
// then the module.json and lock files of the one directory dir by gitea_get_file (a repository can hold hundreds). A manifest the
// forge does not have is an error; a directory with no module.json in the trees is answered with no manifest; a lock file it does not have is left out.
func readVia(call toolCaller) func(ctx context.Context, repository, sha, dir string) (index.RepoRead, error) {
return func(_ context.Context, repository, sha, dir string) (index.RepoRead, error) {
owner, repo, ok := strings.Cut(delivery.RepositoryName(repository), "/")
if !ok {
return index.RepoRead{}, fmt.Errorf("%q is not <owner>/<name>", repository)
}
raw, err := call(commitTreesTool, map[string]any{"owner": owner, "repo": repo, "sha": sha})
if err != nil {
return index.RepoRead{}, err
}
var a struct {
Root string `json:"root"`
Trees map[string]string `json:"trees"`
}
if err := json.Unmarshal(raw, &a); err != nil {
return index.RepoRead{}, fmt.Errorf("the answer of %s is not {root, trees}: %w", commitTreesTool, err)
}
out := index.RepoRead{Root: a.Root, Trees: a.Trees, Manifests: map[string]string{}, Locks: map[string]map[string]string{}}
file := func(dir, name string) (string, error) {
p := name
if dir != "" {
p = dir + "/" + name
}
raw, err := call(getFileTool, map[string]string{"owner": owner, "repo": repo, "path": p, "ref": sha})
if err != nil {
return "", err
}
var f struct {
File *struct {
Content *string `json:"content"`
} `json:"file"`
}
if err := json.Unmarshal(raw, &f); err != nil || f.File == nil || f.File.Content == nil {
return "", fmt.Errorf("the answer of %s is not {file: {content}}: %v", getFileTool, err)
}
return *f.File.Content, nil
}
if _, held := a.Trees[dir]; held {
m, err := file(dir, "module.json")
if err != nil {
return index.RepoRead{}, fmt.Errorf("%s: %w", path.Join(dir, "module.json"), err)
}
out.Manifests[dir] = m
for _, name := range lockFiles {
c, err := file(dir, name)
if err != nil {
if strings.Contains(err.Error(), ": 404 ") {
continue
}
return index.RepoRead{}, fmt.Errorf("%s: %w", path.Join(dir, name), err)
}
sum := sha256.Sum256([]byte(c))
if out.Locks[dir] == nil {
out.Locks[dir] = map[string]string{}
}
out.Locks[dir][name] = hex.EncodeToString(sum[:])
}
}
return out, nil
}
}
@@ -3,6 +3,7 @@ package main
import (
"context"
"encoding/json"
"errors"
"slices"
"strings"
"testing"
@@ -50,3 +51,33 @@ func TestTheTrunkIsTheDirectoriesTheForgeListsTrees(t *testing.T) { // 1900/AC3
t.Error("the trunk asked for paths")
}
}
func TestReadViaReadsManifestsAndLocks(t *testing.T) {
var files []string
call := func(key string, body any) (json.RawMessage, error) {
switch key {
case commitTreesTool:
return json.RawMessage(`{"root":"r0","trees":{"":"t1","modules/other":"t2"}}`), nil
case getFileTool:
files = append(files, body.(map[string]string)["path"])
switch body.(map[string]string)["path"] {
case "module.json":
return json.RawMessage(`{"file":{"content":"{\"module\":\"snake\"}"}}`), nil
case "go.sum":
return json.RawMessage(`{"file":{"content":"x"}}`), nil
}
return nil, errors.New("Gitea API /x: 404 not found")
}
return nil, errors.New("unexpected " + key)
}
got, err := readVia(call)(context.Background(), "http://novox.internal:20000/jschoubben/snake.git", "abc", "")
if err != nil {
t.Fatal(err)
}
if len(got.Manifests) != 1 || len(files) != 3 { // the one directory asked: its manifest and two lock files, never the others
t.Fatalf("manifests = %v, files asked = %v", got.Manifests, files)
}
if got.Root != "r0" || got.Manifests[""] != `{"module":"snake"}` || len(got.Locks[""]) != 1 || got.Locks[""]["go.sum"] == "" {
t.Fatalf("read = %+v", got)
}
}
@@ -0,0 +1,44 @@
package main
import (
"context"
"encoding/json"
"errors"
"time"
"github.com/nats-io/nats.go/jetstream"
"mesh-delivery-indexer/internal/index"
)
// takerIn is the part of the indexer a takeins value asks through.
type takerIn interface {
TakeIn(ctx context.Context, module string) index.TakenIn
}
// takenIn answers one value of the takeins state (the tool indexer_take_in, issue 1945): a value not yet answered takes the
// module in now and is written back, by compare-and-set at its revision, with what the indexer stored or why not. A value
// answered already (this indexer's own write) or deleted is left.
func takenIn(ctx context.Context, ix takerIn, kv jetstream.KeyValue, e jetstream.KeyValueEntry, logf func(string, ...any)) {
if e.Operation() != jetstream.KeyValuePut {
return
}
a, err := index.DecodeTakeIn(e.Value())
if err != nil {
logf("%s.%s holds what is not a take-in ask, left: %v", index.TakeInsState, e.Key(), err)
return
}
if !a.Answered.IsZero() {
return
}
r := ix.TakeIn(ctx, a.Module)
a.Result, a.Answered = &r, time.Now().UTC()
logf("%s: a take-in asked by %s (%s): %s", a.Module, a.By, a.Why, r.Answer)
data, err := json.Marshal(a)
if err != nil {
return
}
if _, err := kv.Update(ctx, e.Key(), data, e.Revision()); err != nil && !errors.Is(err, jetstream.ErrKeyExists) && !errors.Is(err, jetstream.ErrKeyRevisionMismatch) {
logf("%s.%s: the answer could not be written: %v", index.TakeInsState, e.Key(), err)
}
}
+1 -1
View File
@@ -3,7 +3,7 @@ module mesh-delivery-indexer
go 1.26.0
require (
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026211913-68d65e6bbf61
github.com/nats-io/nats-server/v2 v2.11.17
github.com/nats-io/nats.go v1.54.0
)
+2 -4
View File
@@ -1,7 +1,5 @@
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261023150321-a2c813a6b468 h1:UcMee+IV2ii5E4dRZdQUaSaeh9IedTFYumuQbBiB0k0=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261023150321-a2c813a6b468/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585 h1:7QoFH5qjXRpx5M4uPdYsVJ5qKl2082gq8n0lFxhoNIg=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261025211220-e0d1de047585/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026211913-68d65e6bbf61 h1:/c6ncICWjq55nBf8TlDM0ROe/PuH+qZIdnuVbtg8UuM=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261026211913-68d65e6bbf61/go.mod h1:ouERNpZizzwZ0HLEtAypsY9FR20FyUgUYRni6J7kVCY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op/go.mod h1:FQyySiasQQM8735Ddel3MRojmy4dA1IqCeyJ5jmPMbI=
github.com/google/go-tpm v0.9.8 h1:slArAR9Ft+1ybZu0lBwpSmpwhRXaa85hWtMinMyRAWo=
@@ -2,6 +2,8 @@ package index
import (
"encoding/json"
"maps"
"slices"
"strings"
"sync"
@@ -88,3 +90,102 @@ func (b *Builds) PathOf(module, repository string) (string, bool) {
}
return pathOf(strings.Trim(best.Path, "/")), true
}
// Registration is a module registered in the catalogue from its own repository, as the controller's newest outcome record
// of it names it: the repository, the directory of its manifest there ("" or "." for the root) and the commit it was
// built at.
type Registration struct {
Module, Repository, Path, Commit, Ref string
}
// registered is whether the record is an outcome that did not fail: what registers a module. Unlike IsTrunkBuild it
// asks for no recorded source inputs, which no build records since ADR 0530 clause 1, and for no particular branch: a
// repository's trunk is named main, master, or what its owner chose (the registered photos, de-spiegel and link2pay were
// built from nox-mesh, and every record of them names that ref).
func registered(r delivery.BuildRecord) bool {
return r.State == "" && r.Failed == ""
}
// RegistrationsOf are the places a module was registered from, by its outcome records: the newest record of each
// repository and directory (a module's name was once another repository's, as photos was a directory of the catalogue's).
// Records of a trunk's ref (main, master, trunk, none) come before those of another branch, then the newer first, so the
// first that holds the module's manifest is the one indexed.
func (b *Builds) RegistrationsOf(module string) []Registration {
type place struct{ repository, path string }
byPlace := map[place][]delivery.BuildRecord{}
b.mu.Lock()
for _, r := range b.records {
if r.Module == module && registered(r) {
k := place{delivery.RepositoryName(r.Repository), pathOf(strings.Trim(r.Path, "/"))}
byPlace[k] = append(byPlace[k], r)
}
}
b.mu.Unlock()
// Within one place the record is that of the ref a registration comes from, never a newer pull request's or try's: a
// trunk's ref (main, master, trunk, none), else the ref most of the place's records name (the registered photos, de-spiegel
// and link2pay were built from nox-mesh), the newer of two equal; then its newest record.
best := map[place]delivery.BuildRecord{}
for k, recs := range byPlace {
count, latest := map[string]int{}, map[string]delivery.BuildRecord{}
for _, r := range recs {
count[r.Ref]++
if cur, ok := latest[r.Ref]; !ok || newer(r, cur) {
latest[r.Ref] = r
}
}
var pick string
first := true
for ref, rec := range latest {
switch {
case first:
case trunkLike[ref] != trunkLike[pick]:
if !trunkLike[ref] {
continue
}
case count[ref] != count[pick]:
if count[ref] < count[pick] {
continue
}
case !newer(rec, latest[pick]):
continue
}
pick, first = ref, false
}
best[k] = latest[pick]
}
recs := slices.Collect(maps.Values(best))
slices.SortFunc(recs, func(x, y delivery.BuildRecord) int {
switch {
case trunkLike[x.Ref] != trunkLike[y.Ref]:
if trunkLike[x.Ref] {
return -1
}
return 1
case newer(x, y):
return -1
case newer(y, x):
return 1
}
return strings.Compare(x.Repository+x.Path, y.Repository+y.Path)
})
out := make([]Registration, 0, len(recs))
for _, r := range recs {
out = append(out, Registration{Module: r.Module, Repository: r.Repository, Path: r.Path, Commit: r.Commit, Ref: r.Ref})
}
return out
}
// Registered is the registrations of every module a record names, by module name.
func (b *Builds) Registered() []Registration {
b.mu.Lock()
names := map[string]bool{}
for _, r := range b.records {
names[r.Module] = true
}
b.mu.Unlock()
var out []Registration
for _, name := range slices.Sorted(maps.Keys(names)) {
out = append(out, b.RegistrationsOf(name)...)
}
return out
}
@@ -8,6 +8,8 @@ import (
"slices"
"strings"
"sync"
"sync/atomic"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
@@ -46,22 +48,44 @@ type Indexer struct {
Modules Store
// Manifests is the manifests bucket; nil writes no manifests (issue 1449).
Manifests Store
Builds *Builds
// Taken is the taken bucket: per repository the last merge taken in (taken.go, ADR 0556); nil writes none.
Taken Store
// Now is the clock of a taken value; nil is time.Now.
Now func() time.Time
Builds *Builds
// Named asks the forge for the git object of each named path of the module directories given, at a commit
// (gitea.gitea_commit_trees with paths; novox/hq ADR 0499 clauses 2 and 3): the indexer is the one reader of a
// manifest's paths (ADR 0495 clause 1, the lead's decision L8 of R4e). Nil asks nothing, and a module naming paths
// has none.
Named func(ctx context.Context, repository, sha string, paths map[string][]string) (map[string]map[string]string, error)
Trunk func(ctx context.Context, repository, sha string) ([]string, error)
// Read asks the forge for a repository's trees at a commit and the manifest and lock-file hashes of the one module directory dir (the
// root spelled ""), as the forge bridge reads them for a merged message (registered.go). Nil indexes no registered module.
Read func(ctx context.Context, repository, sha, dir string) (RepoRead, error)
// Contains says whether a commit is the second or an ancestor of it (gitea.gitea_contains).
Contains func(ctx context.Context, repository, commit, in string) (bool, error)
Logf func(format string, args ...any)
// Placement is the placement bucket (delivery.PlacementBucket, ADR 0549 §2); nil writes no placement.
Placement Store
// Say says one of the indexer's events (placement-changed); nil says none.
Say func(event string, body any) error
// Drained says whether the indexer has applied every message of the forge stream (the consumer has none pending and none
// unacknowledged); FillTaken writes the source's heads only then. Nil fills none from the source (taken.go).
Drained func(ctx context.Context) (bool, error)
mu sync.Mutex // guards source, toolchains and previews, which the watches and the forge messages all reach
source sourceNow
previews map[string]delivery.Preview // by previews key (previews.go)
toolchains map[string]map[string]delivery.Toolchain // language, then node (toolchains.go)
tried map[string]bool // repository, commit and directory a registered module's manifest was tried at (registered.go)
unread map[string]bool // registered modules whose last read the forge could not answer, read again by RetryUnread
retrying atomic.Bool // a RetryUnread runs in the background (kickRetry)
retryAgain atomic.Bool // an event asked for one more RetryUnread after the running one
retryWG sync.WaitGroup // the background RetryUnread, for tests to wait on
recompute sync.Mutex // one Recompute at a time
untaken map[string]string // repository to the latest merge applied while the taken state was missing (taken.go)
filling atomic.Bool // a FillTaken runs in the background (KickFill)
fillAgain atomic.Bool // a kick arrived while it ran
}
func (ix *Indexer) logf(format string, args ...any) {
@@ -85,6 +109,13 @@ func (ix *Indexer) Apply(ctx context.Context, f delivery.Forge) error {
if err := ix.applyMerged(ctx, f); err != nil {
return err
}
if err := ix.writeTaken(ctx, f); err != nil {
return err
}
ix.kickRetry(ctx) // a registered module the forge could not read is read again on a merge, in the background (issue 1945)
}
if err := ix.applyPlacement(ctx, f); err != nil {
return err
}
var digests map[string]string
var held map[string]delivery.Module
@@ -17,8 +17,9 @@ const (
// memStore is the candidates bucket's compare-and-set behaviour in memory: a revision per write, a conflict when the
// revision read is no longer the key's. Writes counts the writes that went through; beforeWrite lets a test change the
// key under the indexer.
// key under the indexer; unbound, when set, is every read's error (a state not yet on the bus).
type memStore struct {
unbound error
vals map[string][]byte
revs map[string]uint64
next uint64
@@ -29,6 +30,9 @@ type memStore struct {
func newMem() *memStore { return &memStore{vals: map[string][]byte{}, revs: map[string]uint64{}} }
func (m *memStore) Get(_ context.Context, key string) ([]byte, uint64, bool, error) {
if m.unbound != nil {
return nil, 0, false, m.unbound
}
v, ok := m.vals[key]
return v, m.revs[key], ok, nil
}
@@ -43,6 +43,8 @@ func withManifests(f delivery.Forge, manifests map[string]string, locks map[stri
}
func TestAMergedManifestIsKeptByItsDigestAndNamedByTheModule(t *testing.T) { // 1449/AC1
// A merge's placement files are written as its manifests are (issue 1126): folded here to hold the count.
placementIsWrittenAtAMergeOnlyWhereItChanged(t)
avahi := catalogueManifest(t, "avahi")
ix, mods, manifests := manifestsIndexer(t, rec(t, "build-1", "avahi", catalogueRepo, "modules/avahi", "main", "", "2026-10-19T09:00:00Z", aInputs))
f := withManifests(mergedMsg(catalogueRepo, map[string]string{"modules/avahi": "newtree"}), map[string]string{"modules/avahi": avahi}, nil)
@@ -25,6 +25,13 @@ import (
// A value the module already holds but for its commit is not written again: its commit stays the one its fingerprint first
// appeared at (writeOverlay), so a merge writes only the modules it changes.
func (ix *Indexer) applyMerged(ctx context.Context, f delivery.Forge) error {
return ix.applyManifests(ctx, f, true)
}
// applyManifests is applyMerged; drop says whether the modules of the repository the message holds no manifest for are
// deleted. A message that names only some of a repository's modules (a registered module read on its own, registered.go)
// drops none.
func (ix *Indexer) applyManifests(ctx context.Context, f delivery.Forge, drop bool) error {
digests, err := ix.keepManifests(ctx, f)
if err != nil {
return err
@@ -84,7 +91,7 @@ func (ix *Indexer) applyMerged(ctx context.Context, f delivery.Forge) error {
}
// A module of this repository the merge commit holds no module.json for is gone from the trunk, and so is its key
// (issue 1900). A message that carries no manifests (an older writer) or one that cannot be read whole says nothing.
if len(f.Manifests) > 0 && every {
if drop && len(f.Manifests) > 0 && every {
if err := ix.dropRemoved(ctx, f.Repository, held, func(key string, _ delivery.Module) bool { return named[key] }); err != nil {
return err
}
@@ -8,6 +8,7 @@ import (
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/process"
)
const merge = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
@@ -94,22 +95,92 @@ func TestAnUnchangedModuleWritesNothing(t *testing.T) { // 1239 (unchanged value
// writes is the values written by then; commit is each module's commit after.
writes int
commit map[string]string
// taken is the merge the repository's taken key names after then (issue 1280, ADR 0556 clause 1).
taken string
}{
{name: "the same merge again", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge},
{name: "the same merge again", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: merge,
then: func(ix *Indexer) error { return ix.Apply(context.Background(), mergedMsg(repo, three)) }},
{name: "a merge touching one of three modules", writes: 1, commit: map[string]string{"a": next, "b": merge, "c": merge},
{name: "a merge touching one of three modules", writes: 1, commit: map[string]string{"a": next, "b": merge, "c": merge}, taken: next,
then: func(ix *Indexer) error {
m := mergedMsg(repo, map[string]string{"modules/a": "treeA2", "modules/b": "treeB", "modules/c": "treeC"})
m.Merge = next
return ix.Apply(context.Background(), m)
}},
{name: "a later source revision of another repository's merge", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge},
{name: "a merge touching none of them is still taken in", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: next,
then: func(ix *Indexer) error {
m := mergedMsg(repo, three)
m.Merge = next
return ix.Apply(context.Background(), m)
}},
{name: "a merge while the taken state is not on the bus yet is still taken in", writes: 1, commit: map[string]string{"a": next, "b": merge, "c": merge}, taken: merge,
then: func(ix *Indexer) error { // issue 1790: the state is bound when it appears, so the indexer runs before it does
ix.Taken = &memStore{unbound: process.ErrNotYet}
m := mergedMsg(repo, map[string]string{"modules/a": "treeA2", "modules/b": "treeB", "modules/c": "treeC"})
m.Merge = next
return ix.Apply(context.Background(), m)
}},
{name: "a merge heard while the taken state was missing is filled when the state appears", writes: 2, commit: map[string]string{"a": next, "b": merge, "c": merge}, taken: next,
then: func(ix *Indexer) error { // issue 1280: the key was missing, the merge was taken in without it
on := ix.Taken.(*memStore)
key, _ := delivery.RepositoryKey(repo)
delete(on.vals, key)
ix.Taken = &memStore{unbound: process.ErrNotYet}
m := mergedMsg(repo, map[string]string{"modules/a": "treeA2", "modules/b": "treeB", "modules/c": "treeC"})
m.Merge = next
if err := ix.Apply(context.Background(), m); err != nil {
return err
}
if err := ix.ApplySource(context.Background(), 8, sourceOf(repo, next)); err != nil {
return err
}
ix.Taken = on
return ix.FillTaken(context.Background()) // no Drained: only the merge applied is written, not the source's word
}},
{name: "a source head whose merge was not applied is not a taken key", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: "",
then: func(ix *Indexer) error { // the forge stream still holds the merge: the consumer is not drained
delete(ix.Taken.(*memStore).vals, takenKey(repo))
ix.Drained = func(context.Context) (bool, error) { return false, nil }
if err := ix.ApplySource(context.Background(), 8, sourceOf(repo, next)); err != nil {
return err
}
return ix.FillTaken(context.Background())
}},
{name: "a source head of a drained forge stream fills a missing key", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: next,
then: func(ix *Indexer) error {
delete(ix.Taken.(*memStore).vals, takenKey(repo))
ix.Drained = func(context.Context) (bool, error) { return true, nil }
if err := ix.ApplySource(context.Background(), 8, sourceOf(repo, next)); err != nil {
return err
}
return ix.FillTaken(context.Background())
}},
{name: "filling from an older source head leaves a later key", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: next,
then: func(ix *Indexer) error {
ix.Contains = func(_ context.Context, _, commit, in string) (bool, error) { return commit == merge && in == next, nil }
m := mergedMsg(repo, three)
m.Merge = next
if err := ix.Apply(context.Background(), m); err != nil {
return err
}
ix.Drained = func(context.Context) (bool, error) { return true, nil }
return ix.FillTaken(context.Background()) // the source still names merge
}},
{name: "an earlier merge heard again after a later one leaves taken", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: merge,
then: func(ix *Indexer) error {
ix.Contains = func(_ context.Context, _, commit, in string) (bool, error) { return commit == next && in == merge, nil }
m := mergedMsg(repo, three)
m.Merge = next
return ix.Apply(context.Background(), m)
}},
{name: "a later source revision of another repository's merge", writes: 0, commit: map[string]string{"a": merge, "b": merge, "c": merge}, taken: merge,
then: func(ix *Indexer) error {
return ix.ApplySource(context.Background(), 9, delivery.Source{Group: "g", Heads: map[string]string{repo: merge, "novox/mesh-catalog": otherMerge}})
}},
} {
t.Run(tc.name, func(t *testing.T) {
ix, mods := modulesIndexer(t)
taken := newMem()
ix.Taken = taken
ctx := context.Background()
if err := ix.Apply(ctx, mergedMsg(repo, three)); err != nil {
t.Fatal(err)
@@ -129,6 +200,13 @@ func TestAnUnchangedModuleWritesNothing(t *testing.T) { // 1239 (unchanged value
t.Errorf("%s = %+v (%v), want commit %.8s", module, m, ok, want)
}
}
key, _ := delivery.RepositoryKey(repo)
if _, held := taken.vals[key]; !held && tc.taken == "" {
return
}
if got, err := delivery.DecodeTaken(taken.vals[key]); err != nil || got.Merge != tc.taken || got.Repository != repo {
t.Errorf("taken.%s = %+v (%v), want merge %.8s", key, got, err, tc.taken)
}
})
}
}
@@ -252,3 +330,8 @@ func TestTwoDirectoriesNamingOneModuleWriteOneAndSayTheOther(t *testing.T) { //
t.Errorf("module = %+v, want the held directory kept", m)
}
}
func takenKey(repo string) string {
key, _ := delivery.RepositoryKey(repo)
return key
}
@@ -0,0 +1,123 @@
package index
import (
"context"
"errors"
"fmt"
"maps"
"slices"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/process"
)
// Placement of modules on nodes (novox/hq ADR 0549 §2, issue 1126): at each merge of the catalogue the forge bridge
// carries every placement file of the merge commit, and the indexer writes each node's key `placement.<node>` in its
// state `placement` only when it differs (the revision alone is no change), deletes the key of a node whose file is
// gone, and says `placement-changed` for each node it wrote or deleted, naming the modules added, removed or changed.
// A merge that changes no placement writes nothing and says nothing.
//
// A file that does not read as a placement, or names another node than its file, is said in the log and its key is left
// as it was: the catalogue's own check refuses such a file before it merges (ADR 0549 §1), so it is a fault of that
// check, and a node keeps composing from the placement it had rather than from none.
// applyPlacement takes the placement files of a merged message in; a message that read none (another repository, an
// older bridge) changes nothing.
func (ix *Indexer) applyPlacement(ctx context.Context, f delivery.Forge) error {
if ix.Placement == nil || f.Kind != delivery.ForgeMerged || !f.PlacementsRead {
return nil
}
keys, err := ix.Placement.Keys(ctx)
if errors.Is(err, process.ErrNotYet) {
// The state is not on the bus yet: said, and the merge's candidates are still taken out (issue 1790's binding
// writes the next catalogue merge's placement once it appears).
ix.logf("%s at %s: its placement files are not written: %v", f.Repository, f.Merge, err)
return nil
}
if err != nil {
return fmt.Errorf("reading the placement keys: %w", err)
}
gone := map[string]bool{}
for _, key := range keys {
if node, err := delivery.ParsePlacementKey(key); err == nil {
gone[node] = true
}
}
var refused []string
for _, node := range slices.Sorted(maps.Keys(f.Placements)) {
delete(gone, node)
next, err := delivery.DecodePlacement([]byte(f.Placements[node]))
if err == nil && next.Node != node {
err = fmt.Errorf("the file of %s names the node %q", node, next.Node)
}
if err != nil {
refused = append(refused, node)
ix.logf("%s at %s: the placement file of %s is refused and placement.%s is left as it was (the catalogue's check should have refused it): %v",
f.Repository, f.Merge, node, node, err)
continue
}
next.Revision = f.Merge
if err := ix.writePlacement(ctx, node, &next); err != nil {
return err
}
}
for _, node := range slices.Sorted(maps.Keys(gone)) {
if err := ix.writePlacement(ctx, node, nil); err != nil {
return err
}
}
if len(refused) > 0 {
// Said, and the merge's candidates are still taken out: a refused file holds nobody's delivery.
ix.logf("%s at %s: the placement files of %v do not read; their keys are left as they were", f.Repository, f.Merge, refused)
}
return nil
}
// writePlacement writes one node's key, or deletes it for next nil, only when it differs, and says what changed.
func (ix *Indexer) writePlacement(ctx context.Context, node string, next *delivery.Placement) error {
key, err := delivery.PlacementKey(node)
if err != nil {
return err
}
var change *delivery.PlacementChange
err = write(ctx, ix.Placement, key, func(value []byte, found bool) ([]byte, bool, error) {
change = nil
var before delivery.Placement
if found {
cur, err := delivery.DecodePlacement(value)
if err != nil {
ix.logf("placement.%s holds what is not a placement, written again: %v", node, err)
} else {
before = cur
}
}
if next == nil {
if !found {
return nil, false, nil
}
c := delivery.ChangeOf(before, delivery.Placement{Node: node})
change = &c
return nil, true, nil
}
if found && delivery.SamePlacement(before, *next) {
return nil, false, nil
}
data, err := delivery.EncodePlacement(*next)
if err != nil {
return nil, false, err
}
c := delivery.ChangeOf(before, *next)
change = &c
return data, false, nil
})
if err != nil || change == nil {
return err
}
if ix.Say != nil {
if err := ix.Say(delivery.PlacementChangedEvent, *change); err != nil {
// The key is written: a reader that watches it hears the change; the event is said loudly as not said.
ix.logf("placement.%s is written, and %s could not be said: %v", node, delivery.PlacementChangedEvent, err)
}
}
return nil
}
@@ -0,0 +1,97 @@
package index
import (
"context"
"reflect"
"strings"
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// ADR 0549 §2 (issue 1126): a merge changing one node's file writes that key alone and says one event; a merge with no
// placement change writes nothing and says nothing; a node whose file is gone loses its key and says its modules
// removed; a file that does not read leaves its key; a message that read no placement changes none.
func placementIsWrittenAtAMergeOnlyWhereItChanged(t *testing.T) {
ctx := context.Background()
ix, _, logs := newIndexer(t)
place := newMem()
ix.Placement = place
var said []delivery.PlacementChange
ix.Say = func(event string, body any) error {
if event != delivery.PlacementChangedEvent {
t.Errorf("said %s", event)
}
said = append(said, body.(delivery.PlacementChange))
return nil
}
merged := func(merge string, files map[string]string) delivery.Forge {
return delivery.Forge{V: delivery.Version, Kind: delivery.ForgeMerged, Repository: delivery.CatalogueRepository, Branch: "main",
Number: 7, Head: "h", Merge: merge, Root: "r", Trees: map[string]string{"": "t"}, Placements: files, PlacementsRead: true}
}
ace := `{"v":1,"node":"ace","modules":[{"module":"plex"}]}`
g14 := `{"v":1,"node":"g14","modules":[{"module":"zsh"}]}`
written := func(node string) (delivery.Placement, uint64) {
t.Helper()
v, ok := place.vals["placement."+node]
if !ok {
return delivery.Placement{}, 0
}
p, err := delivery.DecodePlacement(v)
if err != nil {
t.Fatal(err)
}
return p, place.revs["placement."+node]
}
if err := ix.Apply(ctx, merged("m1", map[string]string{"ace": ace, "g14": g14})); err != nil {
t.Fatal(err)
}
if p, _ := written("ace"); p.Revision != "m1" || !reflect.DeepEqual(p.Assigned(), []string{"plex"}) || len(said) != 2 {
t.Fatalf("first merge: %+v, said %v", p, said)
}
_, aceRev := written("ace")
said = nil
if err := ix.Apply(ctx, merged("m2", map[string]string{"ace": ace, "g14": g14})); err != nil {
t.Fatal(err)
}
if p, rev := written("ace"); rev != aceRev || p.Revision != "m1" || len(said) != 0 {
t.Errorf("a merge with no placement change wrote %+v (revision %d, was %d) and said %v", p, rev, aceRev, said)
}
said = nil
if err := ix.Apply(ctx, merged("m3", map[string]string{"ace": ace, "g14": `{"v":1,"node":"g14","modules":[{"module":"zsh"},{"module":"xterm"}]}`})); err != nil {
t.Fatal(err)
}
if _, rev := written("ace"); rev != aceRev || len(said) != 1 || said[0].Node != "g14" || said[0].Revision != "m3" ||
!reflect.DeepEqual(said[0].Added, []string{"xterm"}) {
t.Errorf("one file changed: ace's revision %d (was %d), said %+v", rev, aceRev, said)
}
said = nil
if err := ix.Apply(ctx, merged("m4", map[string]string{"ace": ace, "g14": `{"v":1,"node":"ace","modules":[]}`})); err != nil {
t.Fatal(err)
}
if p, _ := written("g14"); len(p.Modules) != 2 || len(said) != 0 || !strings.Contains(logs.String(), "the placement file of %s is refused") {
t.Errorf("a file naming another node replaced the key: %+v, said %v, logs %s", p, said, logs)
}
said = nil
if err := ix.Apply(ctx, merged("m5", map[string]string{"ace": ace})); err != nil {
t.Fatal(err)
}
if _, rev := written("g14"); rev != 0 || len(said) != 1 || !reflect.DeepEqual(said[0].Removed, []string{"xterm", "zsh"}) {
t.Errorf("a file removed: g14's key at revision %d, said %+v", rev, said)
}
said = nil
unread := merged("m6", nil)
unread.PlacementsRead = false
if err := ix.Apply(ctx, unread); err != nil {
t.Fatal(err)
}
if p, _ := written("ace"); p.Revision != "m1" || len(said) != 0 {
t.Errorf("a message that read no placement changed it: %+v, %v", p, said)
}
}
@@ -164,6 +164,7 @@ func TestAPreviewHeardBeforeItsPullRequestGivesTheFingerprintWhenTheCandidateIsW
}
func TestAPreviewNamingNoManifestOfTheModulesDirectoryGivesAReason(t *testing.T) {
t.Run("the same preview again", testTheSamePreviewAppliedAgainWritesNothing)
ix, _, cands := previewBed(t, true)
if err := ix.Apply(context.Background(), avahiPull(t, h1)); err != nil {
t.Fatal(err)
@@ -176,7 +177,7 @@ func TestAPreviewNamingNoManifestOfTheModulesDirectoryGivesAReason(t *testing.T)
}
}
func TestTheSamePreviewAppliedAgainWritesNothing(t *testing.T) {
func testTheSamePreviewAppliedAgainWritesNothing(t *testing.T) {
ix, _, cands := previewBed(t, true)
if err := ix.Apply(context.Background(), avahiPull(t, h1)); err != nil {
t.Fatal(err)
@@ -0,0 +1,275 @@
package index
import (
"context"
"fmt"
"maps"
"slices"
"strings"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// A module registered in the catalogue from its own repository (the operator registers it with the repository and the
// path of its manifest) had no manifest here: the indexer learns a repository only from a merged message, and no merge
// followed the registration. The controller's outcome records name each registered module's repository, directory and
// commit (builds.go Registered), so the indexer reads that module's manifest at the repository's head (the gate's source names it) or, with none named, at the
// registered commit, and writes the same keys a merge writes. A later merge to the repository's trunk reaches the module through the forge bridge's merged
// message as for every other repository.
// RepoRead is a repository at a commit as the forge bridge reads it for a merged message: the root tree, the tree of each
// directory holding a module.json (the root spelled ""), their manifests and the hex sha256 of the lock files beside them.
type RepoRead struct {
Root string
Trees map[string]string
Manifests map[string]string
Locks map[string]map[string]string
}
// IndexRegistered writes the module of a registration as a merge of its repository would, when the modules bucket holds
// none of that name. Its commit is the head the gate's source names for the repository, else the registration's. The
// repository's other modules are neither written nor dropped: they are the merges'. A manifest that is not at the path or
// names another module is said and left. A forge that could not say (a timeout, or a 404 while the forge's account could
// not read the repository) is an error, and the module is kept as unread: it is read again at the next source revision
// and the next merged message (RetryUnread), not only when the indexer starts again (issue 1945).
func (ix *Indexer) IndexRegistered(ctx context.Context, r Registration) error {
_, err := ix.indexRegistered(ctx, r, false)
return err
}
// taken is what indexRegistered did with one registration: the module written, held already, or the reason it was not.
type taken struct {
stored, held bool
repository, commit, dir string
why string
}
func (ix *Indexer) indexRegistered(ctx context.Context, r Registration, force bool) (taken, error) {
if ix.Modules == nil || ix.Read == nil {
return taken{why: "the indexer reads no forge or writes no modules"}, nil
}
key, err := delivery.ModulesKey(r.Module)
if err != nil {
ix.logf("%s: the registered module has no usable key, skipped: %v", r.Module, err)
return taken{why: fmt.Sprintf("the module has no usable key: %v", err)}, nil
}
if _, _, found, err := ix.Modules.Get(ctx, key); err != nil {
return taken{}, fmt.Errorf("modules.%s: %w", key, err)
} else if found {
ix.markRead(r.Module)
return taken{held: true, why: "the modules state holds it already: a merge to its repository writes it from now on"}, nil
}
repository := delivery.RepositoryName(r.Repository)
commit := r.Commit
dir := strings.Trim(r.Path, "/")
if dir == delivery.RootPath {
dir = ""
}
// The head the gate's source names for the repository, when it names one, else the registration's commit. A module a
// merge removed is absent at the head and is said and left.
ix.mu.Lock()
for repo, head := range ix.source.heads {
if delivery.SameRepository(repo, r.Repository) && head != "" {
commit = head
}
}
ix.mu.Unlock()
t := taken{repository: repository, commit: commit, dir: dir}
if repository == "" || commit == "" {
ix.logf("%s: registered with no repository or commit, so its manifest is not read", r.Module)
t.why = "the controller's record names no repository or commit"
return t, nil
}
// One try for each repository, commit and directory: a repository with no manifest where it was registered is said once,
// not at every record of the controller's builds. A person's take-in (force) reads again.
tried := repository + "@" + commit + ":" + dir
ix.mu.Lock()
if ix.tried[tried] && !force {
ix.mu.Unlock()
t.why = "tried already at this commit and directory"
return t, nil
}
if ix.tried == nil {
ix.tried = map[string]bool{}
}
ix.tried[tried] = true
ix.mu.Unlock()
read, err := ix.Read(ctx, repository, commit, dir)
if err != nil {
ix.mu.Lock()
delete(ix.tried, tried) // a forge that could not say is asked again
if ix.unread == nil {
ix.unread = map[string]bool{}
}
ix.unread[r.Module] = true // and read again on the next source revision or merged message (RetryUnread)
ix.mu.Unlock()
return t, fmt.Errorf("%s: reading %s at %s: %w", r.Module, repository, commit, err)
}
ix.markRead(r.Module)
raw, ok := read.Manifests[dir]
if !ok || read.Trees[dir] == "" {
ix.logf("%s: %s at %s holds no module.json at %q, so the registered module is not indexed", r.Module, repository, commit, dir)
t.why = fmt.Sprintf("%s at %s holds no module.json at %q", repository, commit, dir)
return t, nil
}
if name := manifestModule(raw); name != r.Module {
ix.logf("%s: the manifest at %q of %s at %s names the module %q, so the registered module is not indexed", r.Module, dir, repository, commit, name)
t.why = fmt.Sprintf("the manifest at %q of %s at %s names the module %q", dir, repository, commit, name)
return t, nil
}
f := delivery.Forge{V: delivery.Version, Kind: delivery.ForgeMerged, Repository: repository, Merge: commit, Root: read.Root,
Trees: map[string]string{dir: read.Trees[dir]},
Manifests: map[string]string{dir: raw}}
if l := read.Locks[dir]; len(l) > 0 {
f.Locks = map[string]map[string]string{dir: l}
}
if err := ix.applyManifests(ctx, f, false); err != nil {
return t, err
}
if _, _, found, err := ix.Modules.Get(ctx, key); err != nil {
return t, fmt.Errorf("modules.%s: %w", key, err)
} else if !found {
t.why = "its manifest was read but the modules state holds no value of it (the indexer's log says why)"
return t, nil
}
t.stored = true
ix.logf("%s: the registered module is indexed from %s at %s", r.Module, repository, commit)
return t, nil
}
// markRead takes a module out of the unread ones: the forge answered for it.
func (ix *Indexer) markRead(module string) {
ix.mu.Lock()
delete(ix.unread, module)
ix.mu.Unlock()
}
// RetryUnread reads again every registered module whose last read the forge could not answer (a timeout, or a 404 while
// the forge's account could not read the repository; issue 1945). It is called on the events that already reach the
// indexer, a new source revision and a merged message, so a repository that became readable is taken in without a hand
// act and without a timer of its own (ADR 0437 §8). What still cannot be read is said and kept for the next.
func (ix *Indexer) RetryUnread(ctx context.Context) {
ix.mu.Lock()
names := slices.Sorted(maps.Keys(ix.unread))
ix.mu.Unlock()
for _, module := range names {
if err := ix.IndexRegisteredModule(ctx, module); err != nil {
ix.logf("the registered module %s is still not indexed: %v", module, err)
}
}
}
// kickRetry runs RetryUnread in one background goroutine, so the message or revision that asked it is acknowledged without
// waiting on the forge (each unread module costs up to two forge calls of 30 s each). A kick while one runs makes it run
// once more when it ends; never two at once.
func (ix *Indexer) kickRetry(ctx context.Context) {
ix.mu.Lock()
none := len(ix.unread) == 0
ix.mu.Unlock()
if none {
return
}
ix.retryAgain.Store(true)
if !ix.retrying.CompareAndSwap(false, true) {
return
}
ix.retryWG.Add(1)
go func() {
defer ix.retryWG.Done()
for ix.retryAgain.Swap(false) {
ix.RetryUnread(ctx)
}
ix.retrying.Store(false)
// A kick between the last Swap and the Store found it running and returned: run for it.
if ix.retryAgain.Load() {
ix.kickRetry(ctx)
}
}()
}
// TakenIn is the answer of a take-in (the tool indexer_take_in): whether the module is now in the modules state, from
// where, and otherwise why not.
type TakenIn struct {
Module string `json:"module"`
Stored bool `json:"stored"`
Held bool `json:"held,omitempty"`
Repository string `json:"repository,omitempty"`
Commit string `json:"commit,omitempty"`
Path string `json:"path,omitempty"`
Answer string `json:"answer"`
}
// TakeIn reads a registered module's manifest now, whatever was tried before, and writes it as a merge would (the tool
// indexer_take_in, issue 1945). A module the modules state holds is answered so and left; one with no registration in the
// controller's records, or whose every registration could not be read or holds no manifest of it, is answered with why.
func (ix *Indexer) TakeIn(ctx context.Context, module string) TakenIn {
out := TakenIn{Module: module}
if ix.Builds == nil {
out.Answer = "the indexer reads no controller's build records, so it knows no registration"
return out
}
regs := ix.Builds.RegistrationsOf(module)
if len(regs) == 0 {
if key, err := delivery.ModulesKey(module); err == nil && ix.Modules != nil {
if _, _, found, err := ix.Modules.Get(ctx, key); err == nil && found {
out.Held, out.Answer = true, "the modules state holds it already: a merge to its repository writes it from now on"
return out
}
}
out.Answer = "the controller's build records name no registration of " + module + ": a module from outside the indexed repositories is known only from them, and one inside them from a merge"
return out
}
var whys []string
for _, r := range regs {
t, err := ix.indexRegistered(ctx, r, true)
out.Repository, out.Commit, out.Path = t.repository, t.commit, pathOf(t.dir)
switch {
case err != nil:
whys = append(whys, err.Error())
continue
case t.held:
out.Held, out.Answer = true, t.why
out.Repository, out.Commit, out.Path = "", "", ""
return out
case t.stored:
out.Stored = true
out.Answer = fmt.Sprintf("written to the modules state from %s at %s", t.repository, t.commit)
return out
}
whys = append(whys, t.why)
}
out.Answer = "not taken in: " + strings.Join(whys, "; ")
return out
}
// IndexRegisteredModule indexes the registration the controller's records name for a module, when there is one.
func (ix *Indexer) IndexRegisteredModule(ctx context.Context, module string) error {
if ix.Builds == nil {
return nil
}
var first error
for _, r := range ix.Builds.RegistrationsOf(module) {
if err := ix.IndexRegistered(ctx, r); err != nil && first == nil {
first = err
}
}
return first
}
// IndexRegisteredAll indexes every registered module the modules bucket does not hold. A module that cannot be read is
// said and the others go on; the first error is answered.
func (ix *Indexer) IndexRegisteredAll(ctx context.Context) error {
if ix.Builds == nil {
return nil
}
var first error
for _, r := range ix.Builds.Registered() {
if err := ix.IndexRegistered(ctx, r); err != nil {
ix.logf("the registered module %s is not indexed: %v", r.Module, err)
if first == nil {
first = err
}
}
}
return first
}
@@ -0,0 +1,258 @@
package index
import (
"context"
"errors"
"strings"
"testing"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
func TestARegisteredModuleIsIndexedFromItsRepository(t *testing.T) { // the grants part's gaps: modules registered from their own repositories had no manifest here
const forge = "http://novox.internal:20000/jschoubben/snake.git"
snake := `{"module":"snake","version":"1"}`
for _, tc := range []struct {
name string
path string
read RepoRead
err error
ref string // the build record's ref
heads map[string]string // the gate source's heads
wantRead bool // the forge is read
held bool // modules.snake exists already
other bool // modules.other of the same repository exists
want bool // modules.snake is written
}{
{wantRead: true, name: "a module at the root of its repository", path: "", want: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a root spelled as a dot", path: ".", want: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a module in a directory", path: "modules/snake", want: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"modules/snake": "t1"}, Manifests: map[string]string{"modules/snake": snake}}},
{wantRead: true, name: "another module of the repository is neither written nor dropped", path: "", want: true, other: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{name: "a module the bucket holds is left", path: "", held: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "no manifest at the path", path: "modules/x",
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a manifest of another module", path: "",
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": `{"module":"other","version":"1"}`}}},
{wantRead: true, name: "a repository whose default branch is master", path: "", ref: "master", want: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a repository built from a branch of its own (nox-mesh)", path: "", ref: "nox-mesh", want: true,
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a repository the source names, the module unheld and present at the head", path: "", want: true, heads: map[string]string{"jschoubben/snake": h2},
read: RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}},
{wantRead: true, name: "a forge that cannot say", path: "", err: errors.New("down")},
} {
t.Run(tc.name, func(t *testing.T) {
ix, mods := modulesIndexer(t, rec(t, "build-1", "snake", forge, tc.path, tc.ref, "", "2026-10-19T09:00:00Z", ""))
if tc.held {
if err := mods.Create(context.Background(), "snake", mustModule(t, delivery.Module{Repository: "jschoubben/snake", Commit: h1, Path: ".", None2: "x"})); err != nil {
t.Fatal(err)
}
}
if tc.other {
if err := mods.Create(context.Background(), "other", mustModule(t, delivery.Module{Repository: "jschoubben/snake", Commit: h1, Path: "modules/other", None2: "x"})); err != nil {
t.Fatal(err)
}
}
ix.Manifests = newMem()
var asked struct {
repo, sha, dir string
n int
}
ix.Read = func(_ context.Context, repo, sha, dir string) (RepoRead, error) {
asked.repo, asked.sha, asked.dir = repo, sha, dir
asked.n++
return tc.read, tc.err
}
if tc.heads != nil {
ix.source = sourceNow{rev: 1, heads: tc.heads}
}
err := ix.IndexRegisteredAll(context.Background())
if (err != nil) != (tc.err != nil) {
t.Fatalf("error = %v", err)
}
wantCommit := "7a74cee83f88e6585a5511c3cc53bdf0f5303d20" // the record's
if tc.heads != nil {
wantCommit = h2 // the source's head
}
if tc.wantRead && asked.sha != wantCommit {
t.Errorf("the forge was asked at %s, want %s", asked.sha, wantCommit)
}
wantReads := 1
if !tc.wantRead {
wantReads = 0
}
if tc.wantRead && asked.dir != dirOf(tc.path) {
t.Errorf("the forge was asked for directory %q", asked.dir)
}
if asked.n != wantReads {
t.Errorf("forge reads = %d, want %d", asked.n, wantReads)
}
m, ok := moduleOf(t, mods, "snake")
if tc.want {
if !ok || m.Manifest == "" || m.Repository != "jschoubben/snake" || m.Commit != wantCommit || m.Path != pathOf(dirOf(tc.path)) {
t.Fatalf("modules.snake = %+v, found %v", m, ok)
}
if len(ix.Manifests.(*memStore).vals) != 1 {
t.Errorf("manifests written = %d, want 1", len(ix.Manifests.(*memStore).vals))
}
if asked.repo != "jschoubben/snake" {
t.Errorf("the forge was asked for %q", asked.repo)
}
} else if !tc.held && ok {
t.Fatalf("modules.snake written: %+v", m)
}
if _, ok := moduleOf(t, mods, "other"); ok != tc.other {
t.Errorf("modules.other held = %v, want %v", ok, tc.other)
}
})
}
}
func mustModule(t *testing.T, m delivery.Module) []byte {
t.Helper()
b, err := delivery.EncodeModule(withV(m))
if err != nil {
t.Fatal(err)
}
return b
}
func dirOf(p string) string {
if p == "." {
return ""
}
return p
}
func TestARegistrationOfAnotherRepositoryDoesNotHideTheRealOne(t *testing.T) { // photos: its newest trunk record was a directory of the catalogue's, then novox/photos at nox-mesh
const catalogue, own = "http://novox.internal:20000/novox/mesh-catalog.git", "http://novox.internal:20000/novox/Link2Pay.git"
ix, mods := modulesIndexer(t,
rec(t, "build-1", "link2pay", own, "", "nox-mesh", "", "2026-10-06T23:02:00Z", ""),
rec(t, "build-2", "link2pay", catalogue, "modules/link2pay", "main", "", "2026-10-04T16:02:00Z", ""),
rec(t, "build-3", "link2pay", catalogue, "modules/link2pay", "main", "", "2026-10-03T16:02:00Z", ""))
ix.Manifests = newMem()
var asked []string
ix.Read = func(_ context.Context, repo, _, dir string) (RepoRead, error) {
asked = append(asked, repo+":"+dir)
if repo == "novox/Link2Pay" {
return RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": `{"module":"link2pay","version":"1"}`}}, nil
}
return RepoRead{Root: "r0"}, nil
}
if err := ix.IndexRegisteredAll(context.Background()); err != nil {
t.Fatal(err)
}
m, ok := moduleOf(t, mods, "link2pay")
if !ok || m.Repository != "novox/Link2Pay" {
t.Fatalf("modules.link2pay = %+v, found %v", m, ok)
}
if len(asked) != 2 || asked[0] != "novox/mesh-catalog:modules/link2pay" || asked[1] != "novox/Link2Pay:." && asked[1] != "novox/Link2Pay:" {
t.Errorf("the forge was asked %v: want the catalogue's trunk record first, once, then the repository's own", asked)
}
}
func TestAForgeThatCannotSayLeavesTheModuleUnwrittenAndAsksAgain(t *testing.T) { // a timeout or a 404 at the head: no read at an older commit
// Issue 1945: car-hunter and snake answered 404 until the forge's account could read them, and were read again only when
// the indexer started again. Each row's trigger must now take the module in once the forge reads it.
const snake = `{"module":"snake","version":"1"}`
ctx := context.Background()
for _, tc := range []struct{ failure, trigger string }{
{"timeout", "the indexer starts again"},
{"Gitea API /repos/x/git/trees/y: 404 not found", "the indexer starts again"},
{"Gitea API /repos/x/git/trees/y: 404 not found", "a new source revision"},
{"Gitea API /repos/x/git/trees/y: 404 not found", "a merged message of another repository"},
{"Gitea API /repos/x/git/trees/y: 404 not found", "a person's take-in"},
{"timeout", "a person's take-in"},
} {
t.Run(tc.failure+", then "+tc.trigger, func(t *testing.T) {
ix, mods := modulesIndexer(t, rec(t, "build-1", "snake", "http://novox.internal:20000/jschoubben/snake.git", "", "master", "", "2026-10-19T09:00:00Z", ""))
ix.Manifests = newMem()
ix.source = sourceNow{rev: 1, heads: map[string]string{"jschoubben/snake": h2}}
var asked []string
readable := false
ix.Read = func(_ context.Context, _, sha, _ string) (RepoRead, error) {
asked = append(asked, sha)
if readable {
return RepoRead{Root: "r0", Trees: map[string]string{"": "t1"}, Manifests: map[string]string{"": snake}}, nil
}
return RepoRead{}, errors.New(tc.failure)
}
if err := ix.IndexRegisteredAll(ctx); err == nil {
t.Errorf("no error")
}
if _, ok := moduleOf(t, mods, "snake"); ok || len(asked) != 1 || asked[0] != h2 {
t.Errorf("written %v, asked %v", ok, asked)
}
if err := ix.IndexRegisteredAll(ctx); err == nil || len(asked) != 2 {
t.Errorf("not asked again (asked %v)", asked)
}
readable = true // the forge's account reads the repository now
var got TakenIn
switch tc.trigger {
case "the indexer starts again":
if err := ix.IndexRegisteredAll(ctx); err != nil {
t.Fatal(err)
}
case "a new source revision":
if err := ix.ApplySource(ctx, 2, delivery.Source{Heads: map[string]string{"jschoubben/snake": h2, "novox/other": h1}}); err != nil {
t.Fatal(err)
}
ix.retryWG.Wait() // read again in the background
case "a merged message of another repository":
if err := ix.Apply(ctx, mergedMsg("novox/other", map[string]string{"": "t9"})); err != nil {
t.Fatal(err)
}
ix.retryWG.Wait() // read again in the background
case "a person's take-in":
got = ix.TakeIn(ctx, "snake")
if !got.Stored || got.Repository != "jschoubben/snake" || got.Commit != h2 || got.Path != "." || got.Answer == "" {
t.Errorf("take-in answered %+v", got)
}
}
if m, ok := moduleOf(t, mods, "snake"); !ok || m.Manifest == "" || m.Commit != h2 {
t.Fatalf("after %s the readable module is not taken in: %+v, found %v (asked %v)", tc.trigger, m, ok, asked)
}
// Taken in, it is read no more: a further event asks the forge nothing.
n := len(asked)
if err := ix.ApplySource(ctx, 3, delivery.Source{Heads: map[string]string{"jschoubben/snake": h2}}); err != nil {
t.Fatal(err)
}
ix.retryWG.Wait()
if len(asked) != n {
t.Errorf("the forge was asked again after the module was taken in: %v", asked)
}
if again := ix.TakeIn(ctx, "snake"); again.Stored || !again.Held || len(asked) != n {
t.Errorf("a take-in of a held module answered %+v and asked %v", again, asked)
}
if none := ix.TakeIn(ctx, "nobody"); none.Stored || none.Held || !strings.Contains(none.Answer, "no registration") {
t.Errorf("a take-in of an unregistered module answered %+v", none)
}
})
}
}
func TestARecordOfAPullRequestDoesNotReplaceTheRegisteredRef(t *testing.T) { // a newer feat/x record of the same place
const repo = "http://novox.internal:20000/novox/photos.git"
b := NewBuilds()
for _, r := range [][2]any{
rec(t, "b1", "photos", repo, "", "nox-mesh", "", "2026-10-01T09:00:00Z", ""),
rec(t, "b2", "photos", repo, "", "nox-mesh", "", "2026-10-02T09:00:00Z", ""),
rec(t, "b3", "photos", repo, "", "feat/x", "", "2026-10-09T09:00:00Z", ""),
} {
b.Put(r[0].(string), r[1].([]byte))
}
got := b.RegistrationsOf("photos")
if len(got) != 1 || got[0].Ref != "nox-mesh" {
t.Fatalf("registrations = %+v, want the ref most records name, not the newer feat/x", got)
}
k, v := record(t, "b4", "photos", repo, "", "main", "", "2026-10-03T09:00:00Z", "")
b.Put(k, v) // a trunk's ref beats a branch of its own
if got := b.RegistrationsOf("photos"); len(got) != 1 || got[0].Ref != "main" {
t.Fatalf("registrations = %+v", got)
}
}
@@ -47,6 +47,10 @@ func (ix *Indexer) ApplySource(ctx context.Context, rev uint64, s delivery.Sourc
if ix.Modules == nil {
return ix.RecomputeCandidates(ctx)
}
// A registered module the forge could not read is read again at each new source revision (issue 1945), in the
// background: a forge that does not answer must not hold this revision. One taken in later is stamped when it is written
// (applyManifests reads srcOf).
ix.kickRetry(ctx)
held, err := ix.heldModules(ctx)
if err != nil {
return err
@@ -0,0 +1,64 @@
package index
import (
"encoding/json"
"errors"
"fmt"
"strings"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
)
// The indexer's doing tool (novox/hq issue 1945; the instruction "tools are how a module is spoken to"): indexer_take_in
// reads a registered module's manifest now and writes it to the modules state. The tools bundle serves it through the tool
// runner; the indexer's process serves no request subject (ADR 0437 §4), so the tool writes an ask into the indexer's
// state takeins, which the indexer watches and answers there, as the builder answers builder_ask_again. The read tools
// of the indexer's states stay in mesh-delivery-view.
const (
TakeInTool = "indexer_take_in"
// TakeInsState is the indexer's state of take-in asks: the bucket mesh-delivery-indexer_takeins, keyed as the modules state.
TakeInsState = "takeins"
TakeInsBucket = delivery.IndexerModule + "_" + TakeInsState
)
// TakeInAsk is a value of the takeins state: which module a person asked to be taken in, why, who and when, and, once the
// indexer read it, its answer.
type TakeInAsk struct {
V int `json:"v"`
Module string `json:"module"`
Why string `json:"why,omitempty"`
By string `json:"by,omitempty"`
At time.Time `json:"at"`
Result *TakenIn `json:"result,omitempty"`
Answered time.Time `json:"answered,omitzero"`
}
// DecodeTakeIn reads a takeins value.
func DecodeTakeIn(data []byte) (TakeInAsk, error) {
var a TakeInAsk
if err := json.Unmarshal(data, &a); err != nil {
return a, err
}
if a.V != 1 {
return a, fmt.Errorf("a takeins value of version %d, not 1", a.V)
}
if a.Module == "" {
return a, errors.New("a takeins value naming no module")
}
return a, nil
}
// NewTakeIn is the value indexer_take_in writes for a module: its key and the value.
func NewTakeIn(module, why, by string, at time.Time) (string, []byte, error) {
module = strings.TrimSpace(module)
if module == "" {
return "", nil, errors.New("name the module to take in (module)")
}
key, err := delivery.ModulesKey(module)
if err != nil {
return "", nil, err
}
data, err := json.Marshal(TakeInAsk{V: 1, Module: module, Why: strings.TrimSpace(why), By: by, At: at.UTC()})
return key, data, err
}
@@ -0,0 +1,152 @@
package index
import (
"context"
"errors"
"fmt"
"maps"
"slices"
"time"
"git.novox.be/novox/mesh-sdk/go/delivery"
"git.novox.be/novox/mesh-sdk/go/process"
)
// What the indexer took in (novox/hq ADR 0556 clause 1, issue 1280): after a merged message is applied whole, the key of
// its repository in the state `taken` names that merge, so a reader of the modules knows they say the repository at that
// merge or later. The indexer is the key's only writer. A merge heard again after a later one (a redelivery) does not put
// the key back: the key moves only to a merge that contains the one it names, as the forge says it; a forge that cannot
// say keeps the key as it is and says so.
// writeTaken writes the repository's taken key for the merge f carries, unless it names that merge or a later one.
func (ix *Indexer) writeTaken(ctx context.Context, f delivery.Forge) error {
if ix.Taken == nil || f.Merge == "" {
return nil
}
key, err := delivery.RepositoryKey(f.Repository)
if err != nil {
ix.logf("%s at %s: no taken key is written: %v", f.Repository, f.Merge, err)
return nil
}
err = write(ctx, ix.Taken, key, func(value []byte, found bool) ([]byte, bool, error) {
if found {
if cur, err := delivery.DecodeTaken(value); err == nil {
if cur.Merge == f.Merge {
return nil, false, nil
}
if ix.Contains != nil {
later, err := ix.Contains(ctx, f.Repository, f.Merge, cur.Merge)
if err != nil {
ix.logf("%s: whether %s is in %s cannot be read, so taken.%s stays at %s: %v", f.Repository, f.Merge, cur.Merge, key, cur.Merge, err)
return nil, false, nil
}
if later {
return nil, false, nil // a merge heard again after a later one
}
}
}
}
now := time.Now
if ix.Now != nil {
now = ix.Now
}
data, err := delivery.EncodeTaken(delivery.Taken{Repository: delivery.RepositoryName(f.Repository), Merge: f.Merge, At: now().UTC()})
return data, false, err
})
if errors.Is(err, process.ErrNotYet) {
// The state is not on the bus yet (a delivery that adds it, bound when it appears as issue 1790 says): said, and
// the merge's modules are still taken in; FillTaken writes its key once the state appears.
ix.logf("%s at %s: its taken key is not written yet, FillTaken writes it when the state appears: %v", f.Repository, f.Merge, err)
ix.mu.Lock()
if ix.untaken == nil {
ix.untaken = map[string]string{}
}
ix.untaken[f.Repository] = f.Merge // the latest heard; FillTaken keeps a later key over an earlier merge
ix.mu.Unlock()
return nil
}
return err
}
// FillTaken writes the taken keys that were left unwritten, never a merge the indexer has not applied whole (a key names
// a merge only after it was applied: ADR 0556 clause 1, and the view reads it as the upper bound of what is reached). Two
// sources, both through writeTaken as a live merge does (issue 1280):
// - the merges applied while the state was not on the bus (kept by writeTaken), whenever the state is there;
// - the gate's source heads, only when Drained says the indexer has applied every message of the forge stream that was
// there before the heads were read. Otherwise it says so, and the next kick of the fill (after a message is applied)
// tries again.
//
// It is idempotent: a key naming the merge or a later one is left. A state not on the bus yet writes nothing and says
// nothing (the fill runs again when it appears). The first error is returned after every repository was tried.
func (ix *Indexer) FillTaken(ctx context.Context) error {
if ix.Taken == nil {
return nil
}
if _, err := ix.Taken.Keys(ctx); errors.Is(err, process.ErrNotYet) {
return nil
}
ix.mu.Lock()
pending := maps.Clone(ix.untaken)
heads := maps.Clone(ix.source.heads)
ix.mu.Unlock()
var first error
note := func(repository string, err error) {
if err != nil && first == nil {
first = fmt.Errorf("%s: %w", repository, err)
}
}
for _, repository := range slices.Sorted(maps.Keys(pending)) {
err := ix.writeTaken(ctx, delivery.Forge{Repository: repository, Merge: pending[repository]})
if err == nil {
ix.mu.Lock()
if ix.untaken[repository] == pending[repository] {
delete(ix.untaken, repository)
}
ix.mu.Unlock()
}
note(repository, err)
}
if ix.Drained == nil {
return first
}
if len(heads) == 0 {
ix.logf("the source names no repository head, so no taken key is filled from it")
return first
}
done, err := ix.Drained(ctx)
if err != nil {
ix.logf("whether the forge stream is applied cannot be read, so no taken key is filled from the source: %v", err)
return first
}
if !done {
ix.logf("the forge stream is not applied whole yet, so no taken key is filled from the source; the next applied message tries again")
return first
}
for _, repository := range slices.Sorted(maps.Keys(heads)) {
note(repository, ix.writeTaken(ctx, delivery.Forge{Repository: repository, Merge: heads[repository]}))
}
return first
}
// KickFill runs FillTaken in the background after a message was applied, one at a time; a kick that arrives while one runs
// makes it run once more.
func (ix *Indexer) KickFill(ctx context.Context) {
if ix.filling.Swap(true) {
ix.fillAgain.Store(true)
return
}
go func() {
for {
ix.fillAgain.Store(false)
if err := ix.FillTaken(ctx); err != nil {
ix.logf("filling the taken keys: %v", err)
}
if !ix.fillAgain.Load() {
ix.filling.Store(false)
if !ix.fillAgain.Load() || ix.filling.Swap(true) {
return
}
}
}
}()
}
+29 -13
View File
@@ -24,18 +24,21 @@ type state struct {
}
type artifact struct {
Name string `json:"name"`
Kind string `json:"kind"`
Language string `json:"language"`
System string `json:"system"`
From string `json:"from"`
Binary string `json:"binary"`
Name string `json:"name"`
Kind string `json:"kind"`
Language string `json:"language"`
System string `json:"system"`
From string `json:"from"`
Binary string `json:"binary"`
Loads []string `json:"loads"`
}
type manifest struct {
Module string `json:"module"`
Version string `json:"version"`
Slug string `json:"slug"`
Tools []string `json:"tools"`
Emits []string `json:"emits"`
Consumes []string `json:"consumes"`
State []state `json:"state"`
Reads []string `json:"reads"`
@@ -78,7 +81,11 @@ func TestManifestDeclarations(t *testing.T) { // 1238/AC1, 1444 (reads the build
if m.Module != "mesh-delivery-indexer" {
t.Errorf("module = %q", m.Module)
}
if want := []state{{Name: "candidates", History: 10}, {Name: "manifests", History: 1}, {Name: "modules", History: 20}}; !reflect.DeepEqual(m.State, want) {
// 1126 (ADR 0549 §2): the indexer's placement state, and the event it says when a node's placement changed.
if want := []string{"placement-changed"}; !reflect.DeepEqual(m.Emits, want) {
t.Errorf("emits = %v, want %v", m.Emits, want)
}
if want := []state{{Name: "candidates", History: 10}, {Name: "manifests", History: 1}, {Name: "modules", History: 20}, {Name: "placement", History: 10}, {Name: "taken", History: 1}, {Name: "takeins", History: 1}}; !reflect.DeepEqual(m.State, want) {
t.Errorf("state = %+v, want %+v", m.State, want)
}
if want := []string{"mesh-build-go.toolchains", "mesh-build-node.toolchains", "mesh-controller.assignments", "mesh-controller.builds", "mesh-delivery-checker.previews", "mesh-delivery-gate.source"}; !reflect.DeepEqual(sorted(m.Reads), want) {
@@ -95,14 +102,18 @@ func TestManifestDeclarations(t *testing.T) { // 1238/AC1, 1444 (reads the build
if m.SecretsOwner != "mesh-delivery-indexer" {
t.Errorf("secrets-owner = %q", m.SecretsOwner)
}
// Two invokes: the forge's trees (the named paths' objects, since the indexer is the one reader of a manifest's paths:
// Three invokes: the file of a registered module's manifest and lock files (registered.go), the forge's trees (the named paths' objects, since the indexer is the one reader of a manifest's paths:
// ADR 0499 clauses 2 and 3, ADR 0495 clause 1; and the module directories of the trunk, issue 1900) and whether a
// commit is in another (issue 1900).
if want := []string{"gitea.gitea_commit_trees", "gitea.gitea_contains"}; !reflect.DeepEqual(m.Invokes, want) {
if want := []string{"gitea.gitea_commit_trees", "gitea.gitea_contains", "gitea.gitea_get_file"}; !reflect.DeepEqual(m.Invokes, want) {
t.Errorf("invokes = %v, want %v", m.Invokes, want)
}
// No emits, no streams, no bus, no seat, no tool (ADR 0437 §4).
for _, k := range []string{"streams", "logs", "emits", "claims", "provides", "requires", "uses", "defines-seats", "tools", "seat", "seats", "bus"} {
// 1945: one tool, indexer_take_in, served by its tools bundle through the takeins state; the process serves none.
if want := []string{"indexer_take_in"}; !reflect.DeepEqual(m.Tools, want) {
t.Errorf("tools = %v, want %v", m.Tools, want)
}
// No streams, no bus, no seat (ADR 0437 §4); its one event is placement-changed (ADR 0549 §2), checked above.
for _, k := range []string{"streams", "logs", "claims", "provides", "requires", "uses", "defines-seats", "seat", "seats", "bus"} {
if _, ok := top[k]; ok {
t.Errorf("the manifest declares %q: the indexer has none", k)
}
@@ -134,8 +145,13 @@ func TestManifestProcess(t *testing.T) { // 1238/AC1
if _, bus := p["bus"]; bus {
t.Errorf("the process declares a bus: it runs on its own broker credential")
}
if len(m.Build.Artifacts) != 1 {
t.Fatalf("artifacts = %+v", m.Build.Artifacts)
if len(m.Build.Artifacts) != 2 {
t.Fatalf("artifacts = %+v, want the program and the tools", m.Build.Artifacts)
}
if a := m.Build.Artifacts[1]; a.Name != "tools" || a.Kind != "bundle" || a.Binary != "mesh-delivery-indexer-tools" || !reflect.DeepEqual(a.Loads, []string{"mesh-delivery-indexer-tools"}) {
t.Errorf("tools artifact = %+v", a) // 1945
} else if st, err := os.Stat(a.From); err != nil || !st.IsDir() {
t.Errorf("the tools' source %s: %v", a.From, err)
}
a := m.Build.Artifacts[0]
if st, err := os.Stat(a.From); err != nil || !st.IsDir() || a.Binary != "mesh-delivery-indexer" || !strings.HasSuffix(a.From, a.Binary) {
+23 -2
View File
@@ -2,6 +2,12 @@
"module": "mesh-delivery-indexer",
"version": "1",
"slug": "indexer",
"tools": [
"indexer_take_in"
],
"emits": [
"placement-changed"
],
"consumes": [
"mesh-build-go.toolchain.reported",
"mesh-build-node.toolchain.reported",
@@ -11,7 +17,10 @@
"state": [
{ "name": "candidates", "history": 10 },
{ "name": "manifests", "history": 1 },
{ "name": "modules", "history": 20 }
{ "name": "modules", "history": 20 },
{ "name": "placement", "history": 10 },
{ "name": "taken", "history": 1 },
{ "name": "takeins", "history": 1 }
],
"reads": [
"mesh-build-go.toolchains",
@@ -23,7 +32,8 @@
],
"invokes": [
"gitea.gitea_commit_trees",
"gitea.gitea_contains"
"gitea.gitea_contains",
"gitea.gitea_get_file"
],
"own-secrets": {
"broker": "${dir:state}/broker"
@@ -71,6 +81,17 @@
"system": "arch",
"from": "cmd/mesh-delivery-indexer",
"binary": "mesh-delivery-indexer"
},
{
"name": "tools",
"kind": "bundle",
"language": "go",
"system": "arch",
"from": "cmd/mesh-delivery-indexer-tools",
"binary": "mesh-delivery-indexer-tools",
"loads": [
"mesh-delivery-indexer-tools"
]
}
]
}
@@ -286,16 +286,17 @@ func (t *Teller) catchUp(id string) {
t.said(h, *w, err)
}
// mergeAct is what the approve option of the gate's ask binds: this ask, the answer, and the heads it covers. It is
// spelled as the gate composes it (mesh-catalog modules/mesh-delivery-gate approval.go mergeAct, from
// mergeAct is what the approve option of the gate's ask binds: the ask's verb (delivery.AskVerb: mesh-delivery.merge, or
// mesh-delivery.settle for a person's settle ask, novox/hq ADR 0556 clause 5), this ask, the answer, and the heads it covers.
// It is spelled as the gate composes it (mesh-catalog modules/mesh-delivery-gate approval.go mergeAct, from
// modules/mesh-delivery approval.go:63-65) and as mesh-controller verifies it (cmd/mesh-controller/moved.go:98-106).
func mergeAct(ask, option string, heads map[string]string) asks.Act {
func mergeAct(verb, ask, option string, heads map[string]string) asks.Act {
var out []string
for id, sha := range heads {
out = append(out, id+"="+sha)
}
sort.Strings(out)
return asks.Act{"verb": "mesh-delivery.merge", "ask": ask, "option": option, "heads": strings.Join(out, ",")}
return asks.Act{"verb": verb, "ask": ask, "option": option, "heads": strings.Join(out, ",")}
}
// headsOf is the heads an ask covers as an act names them: a delivery's id, `owner/repository@<twelve characters of its
@@ -363,7 +364,7 @@ func (t *Teller) Answer(o asks.Option, w asks.Warrant) error {
return fmt.Errorf("the gate's ask %s is no longer the ask the operator answered (its digest is %s, the warrant's %s)",
s.Key, ga.Ask.Digest(), w.AskDigest)
}
if err := o.Performs(mergeAct(w.Ask, o.ID, headsOf(ga.Heads))); err != nil {
if err := o.Performs(mergeAct(delivery.AskVerb(ga.Group), w.Ask, o.ID, headsOf(ga.Heads))); err != nil {
return err
}
body, err := delivery.EncodeAnswer(delivery.Answer{Option: o.ID, Warrant: warrantID(w), Words: w.Says()})

Some files were not shown because too many files have changed in this diff Show More