Compare commits

...
Author SHA1 Message Date
jochen ee2160d00c Drop every mesh-tools token by id when the forge holds two, and stop re-minting on every call (hq issue 739)
mesh/merge-gate head ee2160d00c22 merged into trunk c71c513027e5 — fail: builds gitea → novox; no bus step; head ee2160d00c22 does not merge cleanly into t…
mesh/repo-check head ee2160d00c22 merged into trunk c71c513027e5 — fail: head ee2160d00c22 does not merge cleanly into trunk c71c513027e5: in conflict modu…
mesh/delivery stopped: the pull request closed unmerged
On 2026-10-14 from 22:28 local the forge held two tokens named mesh-tools.
gitea refuses a delete by such a name with 422 "multiple matches", so every
gitea tool call re-entered the mint, made two basic-auth calls that hash the
admin password, and failed: several times a minute, on a control-node that was
already overloaded, and mesh-delivery could read neither open pulls nor views.

The watcher and the tools run as separate bundles, so the in-process sharing
of one token source does not cover them; two mints racing leave two tokens by
the name, because the forge's check that a name is unused is not atomic.

A 422 on the delete by name now lists the account's tokens and drops each one
carrying the name by its id, then mints. A failed mint is not asked of the
forge again for 30 s: the same error is answered without a call.
2026-10-14 22:45:16 +02:00
mesh-admin 0e7d9e74d2 Merge pull request 'Say every forge event on the bus through a signed system webhook (hq issue 729)' (#304) from feat/729-the-forge-says-every-event-on-the-bus into main 2026-10-14 20:28:01 +00:00
jschoubben 1d8a244ba4 Say every forge event on the bus through a signed system webhook, and announce a head or merge at once (novox/hq issue 729)
mesh/merge-gate out of date: judged against trunk 2edef5a00ae0, the trunk is now 0e7d9e74d22a; judged again on a recheck or when it is merged
mesh/repo-check out of date: judged against trunk 2edef5a00ae0, the trunk is now 0e7d9e74d22a; judged again on a recheck or when it is merged
mesh/delivery covered after its stop: a later walk that contains it was delivered: plan-1792011830935225963 (novox/mesh-catalog at 0361c44a into main, do…
The gitea module keeps one system webhook for every event family, made or
corrected at start with the admin account (POST /admin/hooks, forge 1.27.3)
and a secret it mints and keeps. Its receiver, at the docker bridge's
gateway, refuses an unsigned request before reading it, a body over 5 MB,
and a signature that does not verify; it answers a verified delivery at once,
drops one already heard by its delivery id, and publishes each as
forge.<kind>.<action> on the module's events (its own stream, ADR 0363,
follows issue 725). A pull request opened, pushed to or merged is announced
at once. The polls look every five minutes only while deliveries arrive, and
every 30 s again, saying so loudly, when one finds what no event said or ten
minutes pass without a delivery. The announced records keep both paths from
announcing twice.
2026-10-14 22:21:22 +02:00
mesh-admin 2edef5a00a Merge pull request 'Build build-agent and route-proxy without a build ID (hq issue 721)' (#301) from fix/a-module-moves-only-when-its-program-changes into main 2026-10-14 20:21:12 +00:00
mesh-admin fadf5025cc Merge pull request 'Merge a pull request marked ready at its head once its checks pass (issue 722)' (#303) from feat/merge-on-ready into main 2026-10-14 19:57:00 +00:00
jschoubben baa5ef802e Name the keys gitea's mergeable config takes: none (ADR 0326)
mesh/merge-gate head baa5ef802e82 merged into trunk 7e3e62062bef — pass: builds gitea, mesh-delivery → novox; no bus step; every machine composes with the …
mesh/repo-check head baa5ef802e82 merged into trunk 7e3e62062bef — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
gitea now declares a setting, so every mergeable file of it must name its
keys; the runtime config reads no setting, and ready-authors reaches its
own file.
2026-10-14 21:50:05 +02:00
jschoubben 5b20e419ec Count a ready mark only from a listed account, after the last push and the label (issue 722)
mesh/merge-gate head 5b20e419ec1f merged into trunk 7e3e62062bef — fail: builds gitea, mesh-delivery → novox; no bus step; a manifest the change touches fa…
mesh/repo-check head 5b20e419ec1f merged into trunk 7e3e62062bef — fail: its merge-check.sh failed: files asking for a setting without saying whether it is…
mesh/delivery superseded: a newer head of the same pull request
Review of #303: anyone who can comment could turn a stale label into a mark,
and a long thread hid the newest comment. The forge now reads the pull
request's whole timeline and takes a "ready at" comment only from the
accounts in the trusted setting ready-authors (default the mesh's own forge
account and the operator's), written after the label was last added and after
the last push; any other author is ignored and said in the log.
2026-10-14 21:45:39 +02:00
mesh-admin 7e3e62062b Merge pull request 'mesh-catalog: hear the build seat that builds now (hq issue 712)' (#298) from fix-712-catalogue-hears-build-agent into main 2026-10-14 19:35:13 +00:00
jschoubben 7e0001e304 Merge a pull request marked ready at its head once its checks pass (issue 722)
mesh/merge-gate out of date: judged against trunk 482232e8c51c, the trunk is now 7e3e62062bef; judged again on a recheck or when it is merged
mesh/repo-check out of date: judged against trunk 482232e8c51c, the trunk is now 7e3e62062bef; judged again on a recheck or when it is merged
mesh/delivery superseded: a newer head of the same pull request
The operator wants a reviewed pull request merged without a session calling
the merge tool again after every recheck. The reviewer marks it ready (label
mesh/ready and a comment "ready at <full head sha>"); the forge's open-pulls
read carries the mark, and mesh-delivery merges through gitea_merge_pull_request
at that head, in a group's order once its composed check passed. A push after
the mark clears it; a failed check is said on the pull request and as a
stalled line.
2026-10-14 21:23:26 +02:00
jschoubben d94764f001 Build build-agent and route-proxy without a build ID (hq issue 721)
mesh/merge-gate head d94764f00168 merged into trunk fadf5025cc5a — pass: builds build-agent, route-proxy → ace, g14, novox, shanks; no bus step; every mach…
mesh/repo-check head d94764f00168 merged into trunk fadf5025cc5a — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery stopped: its group fix/a-module-moves-only-when-its-program-changes stopped: novox/mesh-controller@b34055f580d4 failed
mesh/delivery-group group fix/a-module-moves-only-when-its-program-changes delivering: 1 of 2 delivered; novox/mesh-catalog@d94764f00168, stopped for a member …
Go's build ID hashes every input package, so a change mesh-builder never reaches
still changed its bytes; without it, a rebuild of the same program is the same
file, and the check can say the module is unchanged.
2026-10-14 21:07:26 +02:00
mesh-admin 482232e8c5 Merge pull request 'Give the build agent a judge cache of its own (issue 704, ADR 0358)' (#297) from fix/704-release-branch-judge into main 2026-10-14 18:19:31 +00:00
mesh-admin e223262408 Merge pull request 'An answered ask keeps its question whole, its answer below it (issue 714)' (#299) from fix/an-answered-ask-keeps-its-text into main 2026-10-14 18:11:06 +00:00
jochen dbe2949f84 mesh-catalog: pass over a build off its repository's trunk, as the controller does (ADR 0238)
mesh/merge-gate head dbe2949f8456 merged into trunk 482232e8c51c — pass: builds mesh-catalog → novox; no bus step; every machine composes with the change a…
mesh/repo-check head dbe2949f8456 merged into trunk 482232e8c51c — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
A branch built through the build verb is a real build, and placing it would make it the catalogue's
current version though the controller never registers it. Say which of the controller's rules are
mirrored and which need its records.
2026-10-14 20:09:39 +02:00
jschoubben b7d9590fdc An answered ask keeps its question whole, its answer below it (issue 714)
mesh/merge-gate head b7d9590fdc95 merged into trunk e1c8965ddbf1 — pass: builds messenger, telegram → novox; no bus step; every machine composes with the c…
mesh/repo-check head b7d9590fdc95 merged into trunk e1c8965ddbf1 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
Issue 372's fix with issue 639's one line edited an answered ask down to
'You chose X at HH:MM · headline', so the operator could not read back what
they approved. The headline stays the title, the summary, points, link and
Details stay the body, and how it ended and who acts on it become the note.
Telegram cuts the body from its end past 4096 characters, never the title or
the note.
2026-10-14 20:06:24 +02:00
jochen 7ee2869d60 mesh-catalog: narrow the placing by a string, which the check's non-strict compile can follow
mesh/merge-gate head 7ee2869d60eb merged into trunk e1c8965ddbf1 — pass: builds mesh-catalog → novox; no bus step; every machine composes with the change a…
mesh/repo-check head 7ee2869d60eb merged into trunk e1c8965ddbf1 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery superseded: a newer head of the same pull request
2026-10-14 20:04:37 +02:00
jochen 0b74c04aa8 mesh-catalog: hear the build seat that builds now, so a merged build is registered when it is built (hq issue 712)
mesh/delivery superseded: a newer head of the same pull request
mesh/merge-gate head 0b74c04aa89f merged into trunk e1c8965ddbf1 — pass: builds mesh-catalog → novox; no bus step; every machine composes with the change a…
mesh/repo-check head 0b74c04aa89f merged into trunk e1c8965ddbf1 — fail: merge-check.sh (go) passed; its merge-check-typescript.sh failed: TypeScript modul…
Since the build role moved to node-build-agent (ADR 0190), the catalogue consumed only the retired
mesh-build-machine.built, so builds reached it only through the replay it asks for at its start: mailu
e1c8965d, built two minutes after the last replay, was never registered. Consume node-build-agent.built,
and pass over the seat's merge checks, tries, dry runs and failures, which the old seat never said.
2026-10-14 20:03:25 +02:00
mesh-admin e1c8965ddb Merge pull request 'mailu: declare its unhealthy urgent (hq ADR 0355 §5, issue 691)' (#296) from adr-355-mailu-urgent into main 2026-10-14 17:01:22 +00:00
mesh-adminandClaude Opus 5.5 47ef2644ac Give the build agent a judge cache of its own, and prove a release branch's move puts its verdicts out of date
mesh/merge-gate head 47ef2644aca1 merged into trunk e22326240833 — pass: builds build-agent, gitea → ace, g14, novox, shanks; no bus step; every machine co…
mesh/repo-check head 47ef2644aca1 merged into trunk e22326240833 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
mesh/delivery-group group fix/704-release-branch-judge delivered: every member is delivered
Every merge check's judge is built where no pull request can reach (hq ADR 0358 §6,
issue 704): the build agent needs a directory beside its workspace, mounted at the
same path, that only judge builds mount. ADR 0358 §5 already holds in the forge's
holder; this test keeps it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 13:48:30 +00:00
mesh-admin cb7bf94a77 Merge pull request 'mailu: ban clients that try to log in on port 25 (hq issue 708)' (#295) from fix-708-mailu-jail-port25 into main 2026-10-14 13:39:51 +00:00
jochen 8e821ef449 Say mail's failure urgently, on every channel
mesh/merge-gate head 8e821ef44992 merged into trunk cb7bf94a77ea — pass: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; every machine com…
mesh/repo-check head 8e821ef44992 merged into trunk cb7bf94a77ea — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
A broken mailu was a warning, said only on the desk; refused or stuck
mail must reach the operator wherever they are (ADR 0355 §5, issue 691).
2026-10-14 15:37:08 +02:00
jochen 07b9bd0ed4 Ban clients that try to log in on mail port 25
mesh/merge-gate head 07b9bd0ed44a merged into trunk 02427a60ef78 — pass: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; every machine com…
mesh/repo-check head 07b9bd0ed44a merged into trunk 02427a60ef78 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
Port 25 takes no login, so a client that tries is guessing passwords;
mailu's jail only matched Dovecot's failures and never banned them
(issue 708). Relay attempts stay unbanned: postfix already refuses them.
2026-10-14 15:34:55 +02:00
mesh-admin 02427a60ef Merge pull request 'mailu: judge the smtp container by whether mail flows (hq issue 690)' (#294) from fix-690-mail-flows into main 2026-10-14 13:29:25 +00:00
jochen 8661bcfa22 Declare mailu's tools, so the controller knows the one its check asks
mesh/merge-gate head 8661bcfa2275 merged into trunk 50861c74f8b4 — pass: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; every machine com…
mesh/repo-check head 8661bcfa2275 merged into trunk 50861c74f8b4 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
2026-10-14 15:21:39 +02:00
jochen 6049537655 Judge mailu's deferred queue by this side's faults, not one dead recipient
mesh/merge-gate head 60495376551d merged into trunk 50861c74f8b4 — fail: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; a manifest the ch…
mesh/repo-check head 60495376551d merged into trunk 50861c74f8b4 — fail: its merge-check.sh failed: files asking for a setting without saying whether it is…
mesh/delivery superseded: a newer head of the same pull request
A message to a far host that is down stays deferred for up to five days,
and mailu's unhealthy now pages the operator, so one dead recipient would
page for days. Overdue deferred mail is unhealthy only when a lookup failed
here (a name-service failure, as in issue 687) or when it reaches three or
more domains; one or two far domains are said in the answer, not alarmed.
2026-10-14 15:19:08 +02:00
jochen bf3e801f67 Judge mailu by whether mail flows, not only by its parts
mesh/merge-gate head bf3e801f6700 merged into trunk 50861c74f8b4 — fail: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; a manifest the ch…
mesh/repo-check head bf3e801f6700 merged into trunk 50861c74f8b4 — fail: its merge-check.sh failed: files asking for a setting without saying whether it is…
mesh/delivery superseded: a newer head of the same pull request
On 2026-10-14 every mailu container read healthy while postfix could not
resolve a name: inbound mail was refused with 450 and outbound mail sat in
the deferred queue for days (issue 690). The smtp container is now judged
by a tool check, mailu_health, asked inside it: postfix runs (the image's
own check, kept), no message waits deferred longer than 30 minutes, and a
public mail host resolves through the resolver postfix is given.
2026-10-14 15:14:51 +02:00
mesh-admin 50861c74f8 Merge pull request 'mailu: judge the front by SMTP and its certificate, not only its web health (hq issue 688)' (#285) from fix-688-mailu-health into main 2026-10-14 13:06:45 +00:00
mesh-admin b15846dd6e Merge pull request 'mailu: move the resolver to Mailu 2024.06.63, which follows the root key change (hq issue 687)' (#284) from fix-687-mailu-resolver into main 2026-10-14 12:32:36 +00:00
mesh-admin a44d490d97 Merge pull request 'Five postgres consumers say which provision their check exercises (hq issue 145, to-be 48 Phase E)' (#289) from fix/145-a-consumer-is-held-under-its-provider into main 2026-10-14 05:26:47 +00:00
mesh-admin af5cf53a34 Merge pull request 'lab: name the base image's age, refuse a stale one, and rebuild it (issue 698)' (#288) from fix/698-the-lab-names-its-base-age into main 2026-10-14 05:12:37 +00:00
jochen 2bb5d344f6 Hold umami's check to the share route's own 404, so an unknown route cannot pass it (hq issue 145, review 262)
mesh/merge-gate head 2bb5d344f6f1 merged into trunk af5cf53a3450 — pass: builds gitea, mesh-knowledge, n8n, nextcloud, umami → ace, novox; no bus step; eve…
mesh/repo-check head 2bb5d344f6f1 merged into trunk af5cf53a3450 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery covered: a later merge that contains it was delivered: novox/mesh-catalog@054dc9618ce3 (merged as b15846dd into main, walk plan-17919811644…
mesh/delivery-group group fix/145-a-consumer-is-held-under-its-provider delivered: every member is delivered
Any route umami does not serve answers 404 too, as an HTML page. The
share route answers JSON with code not-found only after the database
was asked; with the database down it answers 500.
2026-10-14 07:07:11 +02:00
mesh-admin e846c289c8 Merge pull request 'lab: a run whose suite proved nothing reads as failed, with its skips (issue 693)' (#287) from fix/693-a-run-that-proved-nothing-is-not-passed into main 2026-10-14 05:05:40 +00:00
jochen 0fcc560fec lab: one job at a time with no await between check and start, and no rebuild left holding anything (issue 698, review 256)
mesh/merge-gate head 0fcc560fecd6 merged into trunk e846c289c878 — pass: builds lab → shanks; no bus step; every machine composes with the change as it did…
mesh/repo-check head 0fcc560fecd6 merged into trunk e846c289c878 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
A run and a rebuild each asked the package registry between checking for
the other and writing their status, so both could start and a bed could
raise machines from an image being replaced. The registry is asked first
now. A hard-killed rebuild's credential is cleaned like a run's; a job that
never recorded its process is lost rather than blocking every run; and
lab_check reports a missing or unreadable base image.
2026-10-14 03:34:46 +02:00
jochen acf8a76646 lab: read the suite's verdict only after its recorded line, NOT REACHED in any case (issue 693, review 254)
mesh/merge-gate head acf8a76646de merged into trunk ebf7f3773dc7 — pass: builds lab → shanks; no bus step; every machine composes with the change as it did…
mesh/repo-check head acf8a76646de merged into trunk ebf7f3773dc7 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
mesh/delivery-group group fix/693-a-run-that-proved-nothing-is-not-passed delivered: every member is delivered
A test's own output, printed before the suite reports, could carry a line
shaped like the suite's NOT PASSED or skipped line and stand for its verdict.
2026-10-14 03:28:20 +02:00
jochen b62bb15839 Say which provision five consumers' checks exercise, so a database down is one condition (hq issue 145)
mesh/merge-gate out of date: judged against trunk ebf7f3773dc7, the trunk is now e846c289c878; judged again on a recheck or when it is merged
mesh/repo-check out of date: judged against trunk ebf7f3773dc7, the trunk is now e846c289c878; judged again on a recheck or when it is merged
mesh/delivery-group group fix/145-a-consumer-is-held-under-its-provider checking: composed check out of date: novox/mesh-catalog@b62bb158398d was judged agains…
mesh/delivery superseded: a newer head of the same pull request
gitea, nextcloud, umami and n8n gain an http check that reaches their
database through their own endpoint; mesh-knowledge's watch already
asks the database and now names it. While postgres is unhealthy, their
findings are held under its condition instead of raising one each.
2026-10-14 03:21:53 +02:00
jochen ff9d3ee6a0 lab: name the base image's age, refuse a stale one, and rebuild it (issue 698)
mesh/merge-gate head ff9d3ee6a06c merged into trunk ebf7f3773dc7 — pass: builds lab → shanks; no bus step; every machine composes with the change as it did…
mesh/repo-check head ff9d3ee6a06c merged into trunk ebf7f3773dc7 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery superseded: a newer head of the same pull request
A bed machine carries the package database its base image was built with,
and the mirrors stop serving it within weeks: genesis-keeps-its-bus failed
on a 404 from an image 44 days old, with nothing saying its age and no mesh
tool to rebuild it. lab_check and lab_run now say the age, lab_run refuses
an image older than 14 days and names the remedy, and lab_base reads the
image and rebuilds it with mesh-lab's base build when asked.
2026-10-14 03:21:23 +02:00
jochen c8255eb15d lab: a run whose suite proved nothing reads as failed, with its skips (issue 693)
mesh/merge-gate head c8255eb15d12 merged into trunk ebf7f3773dc7 — pass: builds lab → shanks; no bus step; every machine composes with the change as it did…
mesh/repo-check head c8255eb15d12 merged into trunk ebf7f3773dc7 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery-group group fix/693-a-run-that-proved-nothing-is-not-passed checking: 1 of 2 member(s) ready
mesh/delivery superseded: a newer head of the same pull request
lab_status took the suite's exit as the verdict, and the runner exits 0 when
every test skipped: a genesis bed that skipped its only test as NOT REACHED was
recorded as passed. A run that ended now has its outcome read from its log and
kept in its status - totals, each skip with its reason, and why it did not pass
- and one whose exit said 0 while no test passed reads as failed. Read from the
lab's own lines when it prints them, from the runner's summary for a lab of
before.
2026-10-14 03:14:06 +02:00
jochen 4ded07bc63 Judge mailu's front by what mail needs, not only its web health
mesh/merge-gate head 4ded07bc635f merged into trunk b15846dd6e78 — pass: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; every machine com…
mesh/repo-check head 4ded07bc635f merged into trunk b15846dd6e78 — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
The image's own check asks nginx's /health and that dovecot runs, so the
front stayed healthy while mail could not flow (issue 688). It now also
needs port 25 to greet over SMTP and the served certificate to hold for
14 more days; certbot renews at 30, so a renewal stuck for two weeks is
said long before the certificate expires.
2026-10-14 02:19:54 +02:00
jochen 054dc9618c Move mailu's resolver to Mailu 2024.06.63, which follows the root key change
mesh/merge-gate head 054dc9618ce3 merged into trunk a44d490d974c — pass: builds mailu, sent nowhere; no bus step; 1 wait(s) for a person; every machine com…
mesh/repo-check head 054dc9618ce3 merged into trunk a44d490d974c — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
mesh/delivery delivered
The root zone signs with KSK-2024 since 2026-10-11. The pinned resolver
image (2024.06.59) trusted only KSK-2017 and kept its anchor inside the
container, so every lookup failed with SERVFAIL and mail was refused or
stuck (issue 687). 2024.06.63 refreshes the anchor with unbound-anchor at
every start and fails its health check on SERVFAIL, which the old
'dig @127.0.0.1' never did (issue 688).
2026-10-14 02:17:25 +02:00
56 changed files with 3184 additions and 206 deletions
+1 -1
View File
@@ -1,4 +1,4 @@
66
62
The long-running resources of this catalogue that do not say how they are ready (`health`, novox/hq ADR 0240
rule 8), as `module check` counts them. It may only go down: merge-check.sh fails a change that raises it, and
one that lowers it writes the new number on the first line. From 2026-11-18, or once it is 0, a long-running
+1 -1
View File
@@ -14,7 +14,7 @@ WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
COPY . .
RUN CGO_ENABLED=0 GOOS=linux go build -trimpath -o /mesh-builder ./cmd/mesh-builder
RUN CGO_ENABLED=0 GOOS=linux go build -trimpath -buildvcs=false -ldflags=-buildid= -o /mesh-builder ./cmd/mesh-builder
# Unlike mesh-controller's own FROM scratch (ADR 0006: nothing to audit but one binary), the build
# machine's whole job is shelling out to git and docker — it needs a real userland to do that in,
+54
View File
@@ -115,3 +115,57 @@ func TestTheGenesisDocumentIsDeclaredAndIssuedOutside(t *testing.T) {
}
t.Fatal("the build agent has no environment file")
}
// **The judge cache is the agent's own, beside its workspace and never in it** (novox/hq ADR 0358 §6, issue 704): a
// directory resource of its own, kept as the agent's data, mounted into the agent at the same path as on the host —
// the container runtime resolves a judge build's mounts on the host — and handed to the agent as MESH_JUDGE_CACHE.
func TestTheJudgeCacheIsADirectoryOfItsOwnMountedAtTheSamePath(t *testing.T) {
raw, err := os.ReadFile("module.json")
if err != nil {
t.Fatal(err)
}
var m struct {
Data struct {
Own []struct {
ID string `json:"id"`
Path string `json:"path"`
Class string `json:"class"`
} `json:"own"`
} `json:"data"`
Resources []struct {
ID string `json:"id"`
Type string `json:"type"`
Content string `json:"content"`
Volumes []string `json:"volumes"`
} `json:"resources"`
}
if err := json.Unmarshal(raw, &m); err != nil {
t.Fatal(err)
}
dir, data, env, mounted := false, false, false, false
for _, d := range m.Data.Own {
if d.ID == "judge-cache" && d.Path == "${dir:judge-cache}" && d.Class == "cache" {
data = true
}
}
for _, r := range m.Resources {
switch {
case r.ID == "judge-cache" && r.Type == "directory":
dir = true
case r.ID == "agent-env":
env = strings.Contains(r.Content, "MESH_JUDGE_CACHE=${dir:judge-cache}\n")
case r.ID == "server":
for _, v := range r.Volumes {
if v == "${dir:judge-cache}:${dir:judge-cache}" {
mounted = true
}
if strings.HasPrefix(v, "${dir:workspace}") && strings.Contains(v, "judge-cache") {
t.Errorf("the judge cache is mounted inside the workspace: %s", v)
}
}
}
}
if !dir || !data || !env || !mounted {
t.Fatalf("the judge cache: directory %v, data %v, env %v, mounted at the same path %v", dir, data, env, mounted)
}
}
+13 -1
View File
@@ -39,6 +39,12 @@
"path": "${dir:workspace}",
"class": "cache",
"why": "a build's working copy, cloned again for every build"
},
{
"id": "judge-cache",
"path": "${dir:judge-cache}",
"class": "cache",
"why": "the Go cache every merge check's judge is built with and the release-branch judges kept per head; mounted into judge builds only, never into a check's containers (novox/hq ADR 0358 §6)"
}
]
},
@@ -54,12 +60,17 @@
"type": "directory",
"mode": "0700"
},
{
"id": "judge-cache",
"type": "directory",
"mode": "0700"
},
{
"id": "agent-env",
"type": "file",
"path": "${dir:mesh-state}/build-agent.env",
"mode": "0600",
"content": "MESH_BROKER_FILE=/run/mesh/broker\nMESH_NODE=${machine:name}\nMESH_REGISTRY=${bound:artifact-store:at}:${bound:artifact-store:port}\nMESH_PACKAGE_BINDING=/run/mesh/package-registry.json\nMESH_NPM_TOKEN_FILE=/run/mesh/package-registry.secret\nMESH_GENESIS_PACKAGE_REGISTRY=/run/mesh/genesis-package-registry.json\nMESH_WORKSPACE=${dir:workspace}\n"
"content": "MESH_BROKER_FILE=/run/mesh/broker\nMESH_NODE=${machine:name}\nMESH_REGISTRY=${bound:artifact-store:at}:${bound:artifact-store:port}\nMESH_PACKAGE_BINDING=/run/mesh/package-registry.json\nMESH_NPM_TOKEN_FILE=/run/mesh/package-registry.secret\nMESH_GENESIS_PACKAGE_REGISTRY=/run/mesh/genesis-package-registry.json\nMESH_WORKSPACE=${dir:workspace}\nMESH_JUDGE_CACHE=${dir:judge-cache}\n"
},
{
"id": "server",
@@ -72,6 +83,7 @@
"volumes": [
"${dir:mesh-state}:/run/mesh:ro",
"${dir:workspace}:${dir:workspace}",
"${dir:judge-cache}:${dir:judge-cache}",
"/var/run/docker.sock:/var/run/docker.sock"
],
"restart-on": [
+121 -5
View File
@@ -101,6 +101,92 @@ export interface OpenPull {
/** The head's statuses by context, the newest of each, with its state (novox/hq issue 438): what tells a check
* that started and never answered — pending — from one that answered. */
checks: Record<string, HeadCheck>;
/** The pull request's labels, by name. */
labels?: string[];
/** Its ready mark (novox/hq issue 722): the label `mesh/ready` and the newest comment "ready at <full head sha>", the
* head a reviewer approved. Absent without the label; `head` empty when the label names no head in a comment. */
ready?: ReadyMark;
}
/** The label that marks a pull request ready to merge at the head a comment names (novox/hq issue 722). */
export const READY_LABEL = "mesh/ready";
/** A comment that names the head a pull request is marked ready at: "ready at <full sha>", on a line of its own. */
const READY_AT = /^\s*ready at ([0-9a-f]{40}|[0-9a-f]{64})\b/im;
/** A pull request's ready mark: the head it names, who said it and when, and the label's id, to take it off. */
export interface ReadyMark {
head: string;
by?: string;
at?: string;
label_id: number;
}
/** **The accounts whose "ready at" comment counts** (novox/hq issue 722): the trusted setting ready-authors, which
* the mesh writes to the file MESH_GITEA_READY_AUTHORS_FILE names as a line `authors=<account> <account> …`. A file
* that cannot be read leaves the mesh's own forge account alone, said; never an empty list read as everyone. */
export function readyAuthors(file = process.env.MESH_GITEA_READY_AUTHORS_FILE, own = process.env.MESH_GITEA_ADMIN_USER,
read: (f: string) => string = (f) => readFileSync(f, "utf8")): string[] {
const fallback = own ? [own] : [];
if (!file) return fallback;
try {
const line = read(file).split("\n").find((l) => l.startsWith("authors="));
const authors = (line ?? "").slice("authors=".length).split(/[\s,]+/).filter(Boolean);
if (authors.length) return authors;
console.error(`[gitea] ${file} names no account that may mark a pull request ready; only ${fallback.join(", ") || "none"} may`);
} catch (err) {
console.error(`[gitea] ${file} cannot be read (${err instanceof Error ? err.message : err}); only ` +
`${fallback.join(", ") || "none"} may mark a pull request ready`);
}
return fallback;
}
/** One event of an issue's or pull request's timeline, as the forge keeps it, oldest first: a comment, a push to a
* pull request's branch (`pull_push`), a label added (body "1") or taken off (body ""), and the rest. */
export interface TimelineEvent {
id: number;
type: string;
body: string;
created_at?: string;
user?: string;
label?: { id: number; name: string };
}
/** The comments that wrote "ready at" without counting, by id: each said once in the log (novox/hq issue 722). */
const ignoredSaid = new Set<number>();
/** **The ready mark a pull request's timeline makes** (novox/hq issue 722), read in the forge's order — every event is
* one row of one table, numbered as it happened, so the order is exact where times share a second. A "ready at"
* comment counts only when its author is one of the accounts given, and only when it was written after the label
* mesh/ready was last added and after the last push to the pull request's branch: a push or the label taken off
* undoes every mark before it, so a mark older than the latest push never counts, even with the label still on. A
* comment by any other account is ignored, and said once. */
export function readyMarkOf(labelID: number, events: TimelineEvent[], authors: string[],
say: (line: string) => void = (l) => console.error(l)): ReadyMark {
const listed = new Set(authors.map((a) => a.toLowerCase()));
let labelled = false;
let mark: ReadyMark | undefined;
for (const e of [...events].sort((x, y) => x.id - y.id)) {
if (e.type === "pull_push") {
mark = undefined;
} else if (e.type === "label" && e.label?.name === READY_LABEL) {
labelled = e.body === "1";
mark = undefined;
} else if (e.type === "comment") {
const m = READY_AT.exec(e.body);
if (!m) continue;
if (!listed.has(String(e.user ?? "").toLowerCase())) {
if (!ignoredSaid.has(e.id)) {
ignoredSaid.add(e.id);
say(`[gitea] a "ready at ${m[1].slice(0, 12)}" comment (${e.id}) by ${e.user ?? "an unknown account"} is ignored: ` +
`only ${authors.join(", ") || "no account"} may mark a pull request ready (novox/hq issue 722)`);
}
continue;
}
if (labelled) mark = { head: m[1].toLowerCase(), by: e.user, at: e.created_at, label_id: labelID };
}
}
return mark ?? { head: "", label_id: labelID };
}
/** One status on a head, the newest of its context: its state, what it said, and when it was set. */
@@ -332,8 +418,9 @@ export class GiteaClient {
/** The open pull requests of every repository the mesh's account sees, updated since a moment when one is given,
* page by page (novox/hq issue 347): each with its repository and its own update time. */
async searchOpenPulls(since?: string): Promise<{ owner: string; name: string; number: number; updated_at?: string }[]> {
const out: { owner: string; name: string; number: number; updated_at?: string }[] = [];
async searchOpenPulls(since?: string): Promise<{ owner: string; name: string; number: number; updated_at?: string;
labels?: { id: number; name: string }[] }[]> {
const out: { owner: string; name: string; number: number; updated_at?: string; labels?: { id: number; name: string }[] }[] = [];
for (let page = 1; page <= 20; page++) {
const qs = new URLSearchParams({ type: "pulls", state: "open", limit: "50", page: String(page) });
if (since) qs.set("since", since);
@@ -343,7 +430,8 @@ export class GiteaClient {
const [fo, fn] = String(r.full_name ?? "").split("/");
const owner = String(r.owner ?? fo ?? "");
const name = String(r.name ?? fn ?? "");
if (owner && name && i?.number) out.push({ owner, name, number: Number(i.number), updated_at: i.updated_at });
const labels = (Array.isArray(i?.labels) ? i.labels : []).map((l: any) => ({ id: Number(l?.id ?? 0), name: String(l?.name ?? "") }));
if (owner && name && i?.number) out.push({ owner, name, number: Number(i.number), updated_at: i.updated_at, labels });
}
if (!found || found.length < 50) break;
}
@@ -354,7 +442,7 @@ export class GiteaClient {
* requires and the contexts its head has a status of, in any state. What mesh-delivery reads to say a pull
* request the mesh was never asked to check; asked of the forge, never of what the announcer remembers, so a
* fault in the announcer cannot hide itself. Reads only. */
async openPulls(): Promise<OpenPull[]> {
async openPulls(readyAuthors: string[] = []): Promise<OpenPull[]> {
const out: OpenPull[] = [];
const rules = new Map<string, BranchProtection[] | undefined>();
for (const found of await this.searchOpenPulls()) {
@@ -364,9 +452,14 @@ export class GiteaClient {
if (!rules.has(full)) rules.set(full, await this.branchProtections(found.owner, found.name).catch(() => undefined));
const rule = rules.get(full)?.find((p) => (p.rule_name ?? p.branch_name) === (pull.base || "main"));
const checks = await this.headChecks(found.owner, found.name, pull.head_sha);
const labels = found.labels ?? [];
// The ready mark (novox/hq issue 722): its timeline is read, whole, only for a pull request that carries the label.
const label = labels.find((l) => l.name === READY_LABEL);
const ready = label ? readyMarkOf(label.id, await this.timeline(found.owner, found.name, pull.number), readyAuthors)
: undefined;
out.push({ owner: found.owner, repo: found.name, number: pull.number, base: pull.base ?? "", head_sha: pull.head_sha,
html_url: pull.html_url, required: rule?.enable_status_check ? (rule.status_check_contexts ?? []) : [],
statuses: Object.keys(checks), checks });
statuses: Object.keys(checks), checks, labels: labels.map((l) => l.name), ...(ready ? { ready } : {}) });
}
return out;
}
@@ -639,6 +732,24 @@ export class GiteaClient {
return this.requestText(`${repoPath(owner, repo)}/pulls/${index}.diff`);
}
/** **Every event of an issue's or pull request's timeline**, oldest first, page by page (novox/hq issue 722): a long
* thread hides nothing. Bounded at 100 pages of 50; a longer one is said, and its first 5000 events read. */
async timeline(owner: string, repo: string, index: number): Promise<TimelineEvent[]> {
indexOf(index);
const out: TimelineEvent[] = [];
for (let page = 1; page <= 100; page++) {
const raw = await this.request<any[]>(`${repoPath(owner, repo)}/issues/${index}/timeline?page=${page}&limit=50`);
for (const e of raw ?? []) {
out.push({ id: Number(e?.id ?? 0), type: String(e?.type ?? ""), body: String(e?.body ?? ""),
created_at: e?.created_at, user: e?.user?.login,
...(e?.label ? { label: { id: Number(e.label.id ?? 0), name: String(e.label.name ?? "") } } : {}) });
}
if (!raw || raw.length < 50) return out;
}
console.error(`[gitea] ${owner}/${repo}#${index} has more than 5000 timeline events; only the first 5000 are read`);
return out;
}
/** Every comment on an issue or pull request, oldest first. */
async listComments(owner: string, repo: string, index: number): Promise<GiteaComment[]> {
indexOf(index);
@@ -803,6 +914,11 @@ export class GiteaAdmin {
return { status: res.status, body };
}
/** One admin-API call with a JSON body, answering the status and body (the system webhook, novox/hq issue 729). */
call(method: string, path: string, body?: unknown): Promise<AdminResponse> {
return this.request(path, { method, ...(body === undefined ? {} : { body: JSON.stringify(body) }) });
}
/** Fail with the forge's own message when a status the caller did not expect comes back. */
private static fail(path: string, res: AdminResponse): never {
const detail = typeof res.body === "string" ? res.body : JSON.stringify(res.body);
+152 -64
View File
@@ -40,10 +40,11 @@ import { emit, on } from "@novox/mesh-sdk/events";
import { registerModuleTools } from "@novox/mesh-sdk/tools";
import { commitStatusTool } from "./delivery.js";
import { mergeTool } from "./merge.js";
import { GiteaClient, movedSince } from "./client.js";
import { GiteaAdmin, GiteaClient, movedSince } from "./client.js";
import { CHECK_CONTEXT, REPO_CHECK_CONTEXT, commentFor, headsToAnnounce, statusesFor, type Announced, type Checked,
type RepoCheckFacts, reposToLook } from "./pulls.js";
import { GateWriter, clonedAs, type BuildStarted } from "./gate.js";
import { HOOK_PATH, HOOK_PORT, Pace, claimHead, ensureSystemHook, keptSecret, receiver, type ForgeEvent } from "./webhook.js";
import { announcementOf, keepTook, lateVerdictComment, outOfDateWords, sameCommit, tookFile, weighTrunks } from "./trunk.js";
// Without a way to a token — configured, or mintable with the admin account (token.ts) — there is
@@ -145,57 +146,9 @@ async function pollMerged(client: GiteaClient): Promise<void> {
// at once.
const fresh = !!pull.merged_at && !!since && pull.merged_at > since;
if (primedMerges && fresh) {
// What it changed, asked for only now: a module is rebuilt because a file inside its own
// directory moved, and without this every module built from a repository is rebuilt for a
// change to any of them (novox/hq 04-ISSUES/131).
const changed = await client.listPullFiles(repo.owner, repo.name, pull.number);
// Which of the directories they are in hold a module at the merge commit (novox/hq issue 278): a
// change inside one is that module's, held or not, and only a file in none is shared code. Not
// said when the list is cut or the forge could not be asked; the mesh then keeps its old rule.
let moduleDirs: string[] | null = null;
if (!changed.truncated) {
try {
moduleDirs = await client.moduleDirsAt(repo.owner, repo.name, pull.merge_commit_sha, changed.paths);
} catch (err) {
console.error(`[gitea] ${repo.full_name}#${pull.number}: which directories hold a module could not be read, ` +
`so the mesh reads its files by the old rule — ${err instanceof Error ? err.message : String(err)}`);
}
}
// The head it merged, and what its head was checked as (novox/hq ADR 0239): the delivery it was, made
// from the forge's word when its owner never heard the head.
let headChecks: Record<string, string> | null = null;
if (pull.head_sha) {
try {
headChecks = await client.commitStatuses(repo.owner, repo.name, pull.head_sha);
} catch (err) {
console.error(`[gitea] ${repo.full_name}#${pull.number}: its head's statuses could not be read — ${err instanceof Error ? err.message : err}`);
}
}
await emit("pull.merged", {
owner: repo.owner,
repo: repo.name,
number: pull.number,
title: pull.title,
body: pull.body,
head_sha: pull.head_sha,
...(headChecks ? { head_checks: headChecks } : {}),
head: pull.head,
base: pull.base,
merge_commit_sha: pull.merge_commit_sha,
merged_at: pull.merged_at,
clone_url: repo.clone_url,
html_url: pull.html_url,
paths: changed.paths,
paths_truncated: changed.truncated,
// Which of them the merge deleted (novox/hq ADR 0236): a module whose manifest went is forgotten,
// not built.
removed: changed.removed,
...(moduleDirs ? { module_dirs: moduleDirs, module_dirs_said: true } : {}),
});
// Said, because a trigger that fires silently is indistinguishable from one that did not
// fire (novox/hq 04-ISSUES/131) — this line is how an operator knows the mesh was told.
console.log(`[gitea] announced merge ${repo.full_name}#${pull.number} (${pull.merge_commit_sha.slice(0, 8)}) into ${pull.base}`);
if (await announceMerge(client, repo, pull)) pace.miss(`the merge of ${repo.full_name}#${pull.number}`);
}
else announced.add(pull.merge_commit_sha);
announced.add(pull.merge_commit_sha);
changed = true;
}
@@ -206,6 +159,73 @@ async function pollMerged(client: GiteaClient): Promise<void> {
lastLook = began;
}
type RepoOf = { owner: string; name: string; full_name: string; clone_url?: string };
type PullOf = Awaited<ReturnType<GiteaClient["listPullRequests"]>>[number];
/** **A merge, announced once, whichever path saw it first** (novox/hq issue 729): the poll or the forge's event. */
async function announceMerge(client: GiteaClient, repo: RepoOf, pull: PullOf): Promise<boolean> {
if (!pull.merged || !pull.merge_commit_sha || announced.has(pull.merge_commit_sha)) return false;
// Recorded before the first wait, so the other path finds it taken while this one is still asking the forge.
announced.add(pull.merge_commit_sha);
try {
// What it changed, asked for only now: a module is rebuilt because a file inside its own
// directory moved, and without this every module built from a repository is rebuilt for a
// change to any of them (novox/hq 04-ISSUES/131).
const changed = await client.listPullFiles(repo.owner, repo.name, pull.number);
// Which of the directories they are in hold a module at the merge commit (novox/hq issue 278): a
// change inside one is that module's, held or not, and only a file in none is shared code. Not
// said when the list is cut or the forge could not be asked; the mesh then keeps its old rule.
let moduleDirs: string[] | null = null;
if (!changed.truncated) {
try {
moduleDirs = await client.moduleDirsAt(repo.owner, repo.name, pull.merge_commit_sha, changed.paths);
} catch (err) {
console.error(`[gitea] ${repo.full_name}#${pull.number}: which directories hold a module could not be read, ` +
`so the mesh reads its files by the old rule — ${err instanceof Error ? err.message : String(err)}`);
}
}
// The head it merged, and what its head was checked as (novox/hq ADR 0239): the delivery it was, made
// from the forge's word when its owner never heard the head.
let headChecks: Record<string, string> | null = null;
if (pull.head_sha) {
try {
headChecks = await client.commitStatuses(repo.owner, repo.name, pull.head_sha);
} catch (err) {
console.error(`[gitea] ${repo.full_name}#${pull.number}: its head's statuses could not be read — ${err instanceof Error ? err.message : err}`);
}
}
await emit("pull.merged", {
owner: repo.owner,
repo: repo.name,
number: pull.number,
title: pull.title,
body: pull.body,
head_sha: pull.head_sha,
...(headChecks ? { head_checks: headChecks } : {}),
head: pull.head,
base: pull.base,
merge_commit_sha: pull.merge_commit_sha,
merged_at: pull.merged_at,
clone_url: repo.clone_url,
html_url: pull.html_url,
paths: changed.paths,
paths_truncated: changed.truncated,
// Which of them the merge deleted (novox/hq ADR 0236): a module whose manifest went is forgotten,
// not built.
removed: changed.removed,
...(moduleDirs ? { module_dirs: moduleDirs, module_dirs_said: true } : {}),
});
// Said, because a trigger that fires silently is indistinguishable from one that did not
// fire (novox/hq 04-ISSUES/131) — this line is how an operator knows the mesh was told.
console.log(`[gitea] announced merge ${repo.full_name}#${pull.number} (${pull.merge_commit_sha.slice(0, 8)}) into ${pull.base}`);
} catch (err) {
announced.delete(pull.merge_commit_sha);
throw err;
}
keepAnnounced();
return true;
}
// **Every new head of an open pull request is announced, once** (novox/hq to-be 45 §9): the mesh checks it
// against every machine of its facts before it merges. Kept beside the merges' record, so a restart
// announces nothing twice; the first look on a machine with no record announces only what moved in the
@@ -228,6 +248,31 @@ function keepHeads(): void {
writeFileSync(tmp, JSON.stringify({ heads }));
renameSync(tmp, pullsRecord);
}
/** **A head, announced once, whichever path saw it first** (novox/hq issue 729): the poll or the forge's event. */
async function announceHead(client: GiteaClient, repo: RepoOf, pull: OpenOf): Promise<boolean> {
const key = `${repo.full_name}#${pull.number}`;
const sha = String(pull.head_sha ?? "");
const before = heads[key];
// Claimed before the first wait, so the other path finds it taken while this one is still announcing.
if (!claimHead(heads, key, sha)) return false;
try {
// The announcement the merge tool makes too when it asks a check (trunk.ts, novox/hq ADR 0323 §5).
await emit("pull.updated", await announcementOf(client, repo.owner, repo.name, repo.clone_url ?? "", pull));
} catch (err) {
if (before === undefined) delete heads[key];
else heads[key] = before;
throw err;
}
keepHeads();
// Said, so an operator knows the mesh was asked to check it.
console.log(`[gitea] announced ${key} at ${sha.slice(0, 8)} to be checked before it merges`);
// Pending until the verdict comes, so the pull request says a check is running rather than nothing — and
// where it is: in the build queue from now, until a build machine takes it (novox/hq issue 446).
await gates.pending(repo.owner, repo.name, sha);
return true;
}
type OpenOf = Awaited<ReturnType<GiteaClient["listOpenPullsSince"]>>[number];
let lastPullLook = "";
// **A sweep now and then looks at every repository** (novox/hq issue 347), whatever moved: the floor is a
// heuristic, and a heuristic that misses is caught by the next sweep rather than never.
@@ -259,15 +304,8 @@ async function pollPulls(client: GiteaClient): Promise<void> {
const key = `${repo.full_name}#${pull.number}`;
const fresh = primedPulls || (!!pull.updated_at && pull.updated_at > dayAgo);
if (fresh) {
// The announcement the merge tool makes too when it asks a check (trunk.ts, novox/hq ADR 0323 §5).
await emit("pull.updated", await announcementOf(client, repo.owner, repo.name, repo.clone_url, pull));
// Said, so an operator knows the mesh was asked to check it.
console.log(`[gitea] announced ${key} at ${String(pull.head_sha).slice(0, 8)} to be checked before it merges`);
// Pending until the verdict comes, so the pull request says a check is running rather than nothing — and
// where it is: in the build queue from now, until a build machine takes it (novox/hq issue 446).
await gates.pending(repo.owner, repo.name, String(pull.head_sha));
}
heads[key] = String(pull.head_sha);
if (await announceHead(client, repo, pull)) pace.miss(`${key} at ${String(pull.head_sha).slice(0, 8)}`);
} else heads[key] = String(pull.head_sha);
changed = true;
}
if (await trunksMoved(client, repo.owner, repo.name)) changedTrunks = true;
@@ -413,12 +451,59 @@ async function buildStarted(event: { body: unknown }): Promise<void> {
}
}
/** How fast the polls look: 30 s until the forge's events are heard, five minutes while they are (webhook.ts). */
const pace = new Pace();
/** What the forge's event asks of the announcer at once (issue 729): a new head is announced, and a merge. */
async function actOn(client: GiteaClient, event: ForgeEvent): Promise<void> {
const b = event.body as { owner?: string; repo?: string; number?: number };
if (!b.owner || !b.repo || !b.number) return;
const head = ["forge.pull-request.opened", "forge.pull-request.synchronized", "forge.pull-request.reopened"].includes(event.name);
const merged = event.name === "forge.pull-request.merged";
if (!head && !merged) return;
const [repo, pull] = await Promise.all([client.getRepo(b.owner, b.repo), client.getPullRequest(b.owner, b.repo, b.number)]);
if (head && pull.state === "open") await announceHead(client, repo, pull);
if (merged) await announceMerge(client, repo, pull);
}
/**
* **The receiver, and the forge's one system webhook pointing at it** (issue 729). The secret is minted and kept beside
* the token; the webhook is made or corrected with the admin account (ADR 0266: the tools' token holds no admin scope).
* Answers whether events are heard; a forge that refuses is said, and the polls keep their pace.
*/
async function listenForForge(client: GiteaClient): Promise<boolean> {
const stateDir = process.env.MESH_GITEA_STATE_DIR;
const url = process.env.MESH_GITEA_HOOK_URL;
if (!stateDir || !url) {
console.log("[gitea] no webhook receiver: MESH_GITEA_STATE_DIR or MESH_GITEA_HOOK_URL unset, so the polls alone watch the forge");
return false;
}
try {
const secret = keptSecret(join(stateDir, "webhook.secret"));
const port = Number(process.env.MESH_GITEA_HOOK_PORT ?? HOOK_PORT);
const server = receiver({ secret, port, heard: () => pace.heard(),
publish: (event) => emit(event.name, event.body),
act: (event) => actOn(client, event) });
await new Promise<void>((resolve, reject) => {
server.once("error", reject);
server.listen(port, process.env.MESH_GITEA_HOOK_HOST ?? "0.0.0.0", () => resolve());
});
const did = await ensureSystemHook(GiteaAdmin.fromEnv(), url.replace(/\/+$/, "") + HOOK_PATH, secret);
console.log(`[gitea] listening for the forge's events on port ${port}; its system webhook ${did}, every event, signed`);
return true;
} catch (err) {
console.error(`[gitea] the forge's events are not heard, so the polls keep their pace — ${err instanceof Error ? err.message : err}`);
return false;
}
}
if (gitea) {
const client = gitea;
// A poll that fails says so once, not once a minute: the same reason repeating (the forge not up
// yet, the admin account refused on a restored forge) is one fact, and a recovery is worth a line.
let failing: string | null = null;
const tick = (fn: () => Promise<void>, everyMs: number): void => {
const tick = (fn: () => Promise<void>, every: number | (() => number)): void => {
const everyMs = (): number => (typeof every === "number" ? every : every());
// **One pass at a time** (novox/hq issue 250): the next pass is scheduled when this one has ended, so a
// pass that outlasts its interval delays the next instead of running beside it — two passes at once
// could each announce the same merge before either recorded it.
@@ -433,12 +518,15 @@ if (gitea) {
if (why !== failing) console.error(`[gitea] not watching until this clears — ${why}`);
failing = why;
})
.finally(() => setTimeout(run, everyMs));
.finally(() => setTimeout(run, everyMs()));
run();
};
// **The forge's events first, the polls as the safety net** (novox/hq issue 729): while verified deliveries arrive the
// polls slow to every five minutes and announce only what no event said (Pace); otherwise they keep 30 s.
await listenForForge(client);
tick(() => pollRepos(client), 60_000);
tick(() => pollMerged(client), 30_000);
tick(() => pollPulls(client), 30_000);
tick(() => pollMerged(client), () => pace.every());
tick(() => pollPulls(client), () => pace.every());
await on("mesh-controller.checked", (event) => {
const c = (event.body ?? {}) as Checked;
// A head's own verdict takes its turn after every write of that head's gate already asked for (gate.ts).
+79 -5
View File
@@ -42,7 +42,47 @@
"issue.opened",
"pull.merged",
"pull.updated",
"pull.closed"
"pull.closed",
"forge.push",
"forge.create",
"forge.delete",
"forge.fork",
"forge.status",
"forge.pull-request.opened",
"forge.pull-request.reopened",
"forge.pull-request.synchronized",
"forge.pull-request.edited",
"forge.pull-request.closed",
"forge.pull-request.merged",
"forge.pull-request.labeled",
"forge.pull-request.unlabeled",
"forge.pull-request.assigned",
"forge.pull-request.unassigned",
"forge.pull-request.reviewed",
"forge.pull-request.review-requested",
"forge.pull-request.commented",
"forge.pull-request.changed",
"forge.issue-comment.created",
"forge.issue-comment.edited",
"forge.issue-comment.deleted",
"forge.issues.opened",
"forge.issues.reopened",
"forge.issues.closed",
"forge.issues.edited",
"forge.issues.labeled",
"forge.issues.unlabeled",
"forge.issues.assigned",
"forge.issues.unassigned",
"forge.issues.changed",
"forge.release.published",
"forge.release.updated",
"forge.release.deleted",
"forge.release.changed",
"forge.repository.created",
"forge.repository.deleted",
"forge.repository.changed",
"forge.wiki.changed",
"forge.other"
],
"consumes": [
"mesh-controller.checked",
@@ -61,7 +101,14 @@
"port": 22,
"protocol": "tcp",
"from": "mesh",
"why": "git over ssh, gitea's own unmodified sshd. Published on the machine's own side at 222, the mesh's fixed public convention \u2014 not 22, which the machine's own daemon holds and a module does not take"
"why": "git over ssh, gitea's own unmodified sshd. Published on the machine's own side at 222, the mesh's fixed public convention — not 22, which the machine's own daemon holds and a module does not take"
},
{
"name": "forge-events",
"port": 3047,
"protocol": "tcp",
"from": "machine",
"why": "the forge's system webhook, every event signed with a secret the module keeps, from the forge's container on the docker bridge to the module's receiver at the bridge's gateway 172.17.0.1 (novox/hq issue 729)"
}
],
"serves": {
@@ -109,6 +156,13 @@
}
}
},
"settings": {
"ready-authors": {
"kind": "preference",
"default": "mesh-admin jschoubben",
"why": "the forge accounts whose comment \"ready at <full head sha>\" marks a pull request ready for mesh-delivery to merge, separated by spaces: the mesh's own forge account and the operator's (novox/hq issue 722). A comment by any other account is ignored and said in the forge's log. Trusted: who may have a change merged is the operator's word"
}
},
"resources": [
{
"id": "mesh-state",
@@ -122,6 +176,14 @@
"path": "${dir:mesh-state}/state",
"mode": "0700"
},
{
"id": "ready-authors",
"type": "file",
"path": "${dir:mesh-state}/ready-authors",
"mode": "0600",
"trusted": true,
"content": "# The accounts whose \"ready at <sha>\" comment marks a pull request ready (novox/hq issue 722), from the setting\n# ready-authors, written by the mesh. Replaced on every push; change the setting, never this file.\nauthors=${setting:ready-authors}\n"
},
{
"id": "state",
"type": "directory",
@@ -138,7 +200,7 @@
"type": "file",
"path": "${dir:state}/server.env",
"mode": "0600",
"content": "GITEA__security__INTERNAL_TOKEN=${secret:internal-token}\nGITEA__database__DB_TYPE=postgres\nGITEA__database__HOST=${bound:postgres-database:at}:${bound:postgres-database:port}\nGITEA__database__NAME=${bound:postgres-database:as}\nGITEA__database__USER=${bound:postgres-database:as}\nGITEA__database__PASSWD=${secret:postgres-database}\n"
"content": "GITEA__security__INTERNAL_TOKEN=${secret:internal-token}\nGITEA__database__DB_TYPE=postgres\nGITEA__database__HOST=${bound:postgres-database:at}:${bound:postgres-database:port}\nGITEA__database__NAME=${bound:postgres-database:as}\nGITEA__database__USER=${bound:postgres-database:as}\nGITEA__database__PASSWD=${secret:postgres-database}\nGITEA__webhook__ALLOWED_HOST_LIST=172.17.0.1\n"
},
{
"id": "data",
@@ -149,6 +211,13 @@
{
"id": "server",
"type": "container",
"health": {
"kind": "http",
"endpoint": "web",
"path": "/api/healthz",
"status": 200,
"needs": "postgres-database"
},
"name": "gitea",
"image": "gitea/gitea@sha256:87a67ee09d3ae0d1df5fda5dcda3e2a1f9236a45b0a59025d6e00e46adc43bef",
"env": {
@@ -200,7 +269,8 @@
"path": "${dir:mesh-state}/config.json",
"mode": "0600",
"content": "{}\n",
"merge": "json"
"merge": "json",
"takes-settings": []
}
],
"provides": [
@@ -239,7 +309,11 @@
"MESH_GITEA_ADMIN_USER": "mesh-admin",
"MESH_GITEA_ADMIN_PASSWORD_FILE": "${dir:state}/admin.secret",
"MESH_GITEA_STATE_DIR": "${dir:runtime-state}",
"MESH_RECEIVES": "${dir:grants}/npm.json"
"MESH_RECEIVES": "${dir:grants}/npm.json",
"MESH_GITEA_READY_AUTHORS_FILE": "${dir:mesh-state}/ready-authors",
"MESH_GITEA_HOOK_URL": "http://172.17.0.1:3047",
"MESH_GITEA_HOOK_PORT": "3047",
"MESH_GITEA_HOOK_HOST": "172.17.0.1"
}
},
{
+117
View File
@@ -137,3 +137,120 @@ test("a merged pull request whose branch was deleted is announced with its branc
assert.equal(headBranchOf({ ref: "refs/pull/7/head" }), "refs/pull/7/head");
assert.equal(headBranchOf(undefined), undefined);
});
// **A pull request marked ready says the head it is marked at** (novox/hq issue 722): the label mesh/ready and the
// newest comment "ready at <full head sha>" by a listed account, written after the label was added and after the last
// push. Read from the pull request's whole timeline, page by page; without the label nothing is read.
const head = "c".repeat(40), older = "d".repeat(40);
const ev = (id: number, type: string, body: string, user = "mesh-admin", label?: { id: number; name: string }) =>
({ id, type, body, user: { login: user }, created_at: `2026-10-14T10:${String(id % 60).padStart(2, "0")}:00Z`,
...(label ? { label } : {}) });
const readyLabel = { id: 31, name: "mesh/ready" };
function forgeWith(timeline: unknown[]) {
const asked: string[] = [];
const server = createServer((req, res) => {
const url = req.url ?? "";
asked.push(url);
const send = (v: unknown) => res.end(JSON.stringify(v));
if (url.startsWith("/api/v1/repos/issues/search")) {
return send(url.includes("page=1") ? [
{ number: 7, repository: { owner: "novox", name: "hq", full_name: "novox/hq" },
labels: [{ id: 31, name: "mesh/ready" }, { id: 2, name: "bug" }] },
{ number: 8, repository: { owner: "novox", name: "hq", full_name: "novox/hq" }, labels: [] },
] : []);
}
if (url.startsWith("/api/v1/repos/novox/hq/pulls/7")) {
return send({ number: 7, state: "open", base: { ref: "main" }, head: { ref: "x", sha: head }, html_url: "h" });
}
if (url.startsWith("/api/v1/repos/novox/hq/pulls/8")) {
return send({ number: 8, state: "open", base: { ref: "main" }, head: { ref: "y", sha: older }, html_url: "h" });
}
const t = /\/issues\/7\/timeline\?page=(\d+)&limit=50/.exec(url);
if (t) {
const page = Number(t[1]);
return send(timeline.slice((page - 1) * 50, page * 50));
}
if (url.includes("/branch_protections")) return send([]);
if (url.includes("/status")) return send({ statuses: [] });
res.statusCode = 404;
res.end("{}");
});
return { server, asked };
}
async function readOpen(timeline: unknown[], authors = ["mesh-admin", "jschoubben"]) {
const { GiteaClient } = await import("../client.ts");
const { server, asked } = forgeWith(timeline);
await new Promise<void>((r) => server.listen(0, r));
try {
const client = new GiteaClient(`http://127.0.0.1:${(server.address() as any).port}`, "t");
return { open: await client.openPulls(authors), asked };
} finally {
server.close();
}
}
test("a pull request's ready mark is read: the label and the head the newest listed ready comment names", async () => {
const { open, asked } = await readOpen([
ev(1, "pull_push", JSON.stringify({ commit_ids: [head] })),
ev(2, "label", "1", "mesh-admin", readyLabel),
ev(3, "comment", `ready at ${older}`),
ev(4, "comment", `Reviewed: approve.\nready at ${head}\n`),
ev(5, "comment", "thanks", "jschoubben"),
]);
assert.deepEqual(open[0].labels, ["mesh/ready", "bug"]);
assert.deepEqual(open[0].ready, { head, by: "mesh-admin", at: "2026-10-14T10:04:00Z", label_id: 31 });
assert.equal(open[1].ready, undefined, "no label, no mark");
assert.ok(!asked.some((u) => u.includes("/issues/8/")), "no timeline read without the label");
});
test("a ready comment by an account not listed is ignored, and said", async () => {
const { readyMarkOf } = await import("../client.ts");
const said: string[] = [];
const mark = readyMarkOf(31, [
ev(1, "pull_push", "{}") as any, ev(2, "label", "1", "mesh-admin", readyLabel) as any,
{ ...ev(3, "comment", `ready at ${head}`, "outsider"), user: "outsider" },
].map((e: any) => ({ ...e, user: typeof e.user === "string" ? e.user : e.user.login })), ["mesh-admin"], (l) => said.push(l));
assert.equal(mark.head, "", "an outsider's mark counts");
assert.equal(said.length, 1);
assert.match(said[0], /by outsider is ignored/);
});
test("a ready comment older than the latest push, or than the label, never counts", async () => {
const pushedAfter = await readOpen([
ev(1, "label", "1", "mesh-admin", readyLabel),
ev(2, "comment", `ready at ${head}`),
ev(3, "pull_push", JSON.stringify({ commit_ids: [head] })),
]);
assert.equal(pushedAfter.open[0].ready?.head, "", "a mark written before the latest push counts, the label still on");
const beforeLabel = await readOpen([
ev(1, "pull_push", "{}"),
ev(2, "comment", `ready at ${head}`),
ev(3, "label", "1", "mesh-admin", readyLabel),
]);
assert.equal(beforeLabel.open[0].ready?.head, "", "a mark written before the label was added counts");
const takenOff = await readOpen([
ev(1, "label", "1", "mesh-admin", readyLabel),
ev(2, "comment", `ready at ${head}`),
ev(3, "label", "", "mesh-admin", readyLabel),
]);
assert.equal(takenOff.open[0].ready?.head, "", "a mark whose label was taken off since counts");
});
test("a long thread hides nothing: every page of the timeline is read", async () => {
const events: unknown[] = [ev(1, "pull_push", "{}"), ev(2, "label", "1", "mesh-admin", readyLabel),
ev(3, "comment", `ready at ${older}`)];
for (let id = 4; id < 130; id++) events.push(ev(id, "comment", "a long discussion", "jschoubben"));
events.push(ev(130, "comment", `ready at ${head}`, "jschoubben"));
const { open, asked } = await readOpen(events);
assert.equal(open[0].ready?.head, head, "the newest mark, on the third page, is not read");
assert.equal(asked.filter((u) => u.includes("/issues/7/timeline")).length, 3);
});
test("the accounts that may mark are the setting's, and only the mesh's own when it cannot be read", async () => {
const { readyAuthors } = await import("../client.ts");
assert.deepEqual(readyAuthors("f", "mesh-admin", () => "# x\nauthors=mesh-admin jschoubben\n"), ["mesh-admin", "jschoubben"]);
assert.deepEqual(readyAuthors("f", "mesh-admin", () => { throw new Error("gone"); }), ["mesh-admin"]);
assert.deepEqual(readyAuthors("f", "mesh-admin", () => "authors=\n"), ["mesh-admin"]);
});
+77 -8
View File
@@ -15,7 +15,7 @@ import { tmpdir } from "node:os";
import { join } from "node:path";
import { collectTools } from "@novox/mesh-sdk/tools";
import { AdminRefused, MintedToken, TOKEN_SCOPES } from "../dist/token.js";
import { AdminRefused, MINT_RETRY_AFTER_MS, MintedToken, TOKEN_SCOPES } from "../dist/token.js";
import { GiteaClient } from "../dist/client.js";
import { getGiteaTools } from "../dist/tools/index.js";
@@ -29,6 +29,9 @@ interface Forge {
mints: number;
lastScopes: string[] | null;
tokens: Map<string, string>;
/** Tokens beyond the one per name `tokens` holds, by id: what two mints racing leave (novox/hq issue 739). */
twins: Map<number, { name: string; sha1: string }>;
tokenCalls: number;
scopesOf: Map<string, string[]>;
admins: Map<string, string>;
pullState: string;
@@ -42,6 +45,9 @@ function fakeForge(): Promise<Forge> {
mints: 0,
lastScopes: null as string[] | null,
tokens: new Map<string, string>(), // name -> value
twins: new Map<number, { name: string; sha1: string }>(),
idOf: new Map<string, number>(), // value -> id
tokenCalls: 0,
scopesOf: new Map<string, string[]>(), // value -> scopes, so a route can enforce them like gitea does
admins: new Map([[ADMIN, PASSWORD]]),
pullState: "open",
@@ -73,21 +79,41 @@ function fakeForge(): Promise<Forge> {
const tokens = url.pathname.match(/^\/api\/v1\/users\/([^/]+)\/tokens(?:\/([^/]+))?$/);
if (tokens) {
const user = basic(req);
forge.tokenCalls++;
if (user === null || user !== decodeURIComponent(tokens[1])) return json(res, 401, { message: "auth required" });
const named = (name: string): number =>
(forge.tokens.has(name) ? 1 : 0) + [...forge.twins.values()].filter((t) => t.name === name).length;
if (req.method === "POST") {
const { name, scopes } = await body(req);
if (forge.tokens.has(name)) return json(res, 400, { message: "token name has already been used" });
if (named(name) > 0) return json(res, 400, { message: "token name has already been used" });
forge.mints++;
forge.lastScopes = scopes;
const sha1 = `minted-${forge.mints}-${Math.random().toString(36).slice(2)}`;
forge.tokens.set(name, sha1);
forge.idOf.set(sha1, forge.mints);
forge.scopesOf.set(sha1, scopes);
return json(res, 201, { id: forge.mints, name, sha1, scopes, token_last_eight: sha1.slice(-8) });
}
if (req.method === "GET" && !tokens[2]) {
const page = Number(url.searchParams.get("page") ?? "1");
const all = [...[...forge.tokens].map(([name, v]) => ({ id: forge.idOf.get(v) ?? 0, name })),
...[...forge.twins].map(([id, t]) => ({ id, name: t.name }))];
return json(res, 200, page === 1 ? all : []);
}
if (req.method === "DELETE" && tokens[2]) {
const name = decodeURIComponent(tokens[2]);
if (!forge.tokens.has(name)) return json(res, 404, { message: "token not found" });
forge.tokens.delete(name);
const key = decodeURIComponent(tokens[2]);
// gitea: "identified by ID and if not available by name"
if (/^\d+$/.test(key)) {
const id = Number(key);
if (forge.twins.delete(id)) return json(res, 204, null);
const hit = [...forge.tokens].find(([, v]) => forge.idOf.get(v) === id);
if (!hit) return json(res, 404, { message: "token not found" });
forge.tokens.delete(hit[0]);
return json(res, 204, null);
}
if (named(key) > 1) return json(res, 422, { message: `multiple matches for token name '${key}'` });
if (!forge.tokens.has(key)) return json(res, 404, { message: "token not found" });
forge.tokens.delete(key);
return json(res, 204, null);
}
return json(res, 405, { message: "method not allowed" });
@@ -180,7 +206,9 @@ function fakeForge(): Promise<Forge> {
get pullState() { return forge.pullState; },
get pullTitle() { return forge.pullTitle; },
get branchDeleted() { return forge.branchDeleted; },
get tokenCalls() { return forge.tokenCalls; },
tokens: forge.tokens,
twins: forge.twins,
scopesOf: forge.scopesOf,
admins: forge.admins,
close: () => new Promise((r) => server.close(() => r())),
@@ -209,8 +237,9 @@ async function delivered(forge: Forge): Promise<{ env: NodeJS.ProcessEnv; file:
}
/** A client as a fresh process would build it: a new source over the kept file, its log captured. */
function minted(env: NodeJS.ProcessEnv, logs: string[]): GiteaClient {
function minted(env: NodeJS.ProcessEnv, logs: string[], now?: () => number): GiteaClient {
const source = new MintedToken({
now,
url: env.MESH_GITEA_URL!,
admin: env.MESH_GITEA_ADMIN_USER!,
passwordFile: env.MESH_GITEA_ADMIN_PASSWORD_FILE!,
@@ -325,6 +354,44 @@ test("the kept file is gone but the forge still holds a token by that name: repl
assert.ok(logs.some((l) => l.includes('already holds a token named "mesh-tools"')), logs.join("\n"));
});
test("two tokens carry the name (two mints raced): each is dropped by its id and one is minted, not a 422 for good (issue 739)", async () => {
const { env, logs } = await delivered(forge);
await minted(env, logs).listRepos();
// What two processes minting at once leave: a second token by the same name, whose value nobody kept.
// Two tokens by the name, neither of them the one kept.
forge.tokens.clear();
forge.twins.set(9001, { name: "mesh-tools", sha1: "raced-first" });
forge.twins.set(9002, { name: "mesh-tools", sha1: "raced-second" });
const repos = await minted(env, logs).listRepos();
assert.equal(repos.length, 1);
assert.equal(forge.twins.size, 0);
assert.equal([...forge.tokens.keys()].filter((n) => n === "mesh-tools").length, 1);
assert.ok(logs.some((l) => l.includes('holds 2 tokens named "mesh-tools"')), logs.join("\n"));
});
test("a mint that failed is not asked of the forge again within the retry window: the same error, no basic-auth call (issue 739)", async () => {
const { env, logs } = await delivered(forge);
forge.admins.delete(ADMIN);
try {
let clock = 5_000_000;
const client = minted(env, logs, () => clock);
await assert.rejects(client.listRepos(), AdminRefused);
const calls = forge.tokenCalls;
await assert.rejects(client.listRepos(), AdminRefused);
await assert.rejects(client.listRepos(), AdminRefused);
assert.equal(forge.tokenCalls, calls);
clock += MINT_RETRY_AFTER_MS;
await assert.rejects(client.listRepos(), AdminRefused);
assert.equal(forge.tokenCalls, calls + 1);
} finally {
forge.admins.set(ADMIN, PASSWORD);
}
});
test("concurrent first calls share one mint", async () => {
const { env, logs } = await delivered(forge);
const client = minted(env, logs);
@@ -339,7 +406,8 @@ test("the admin account is refused: said plainly, nothing kept, and the next cal
const { env, file, logs } = await delivered(forge);
forge.admins.delete(ADMIN); // the forge's data came from a predecessor; mesh-admin was never created there
try {
const client = minted(env, logs);
let clock = 1_000_000;
const client = minted(env, logs, () => clock);
const before = forge.mints;
await assert.rejects(client.listRepos(), (err: unknown) => {
@@ -354,8 +422,9 @@ test("the admin account is refused: said plainly, nothing kept, and the next cal
assert.equal(forge.mints, before);
await assert.rejects(stat(file), /ENOENT/);
// The account appears (the operator created it): the very next call mints and works.
// The account appears (the operator created it): the next call after the retry window mints and works.
forge.admins.set(ADMIN, PASSWORD);
clock += MINT_RETRY_AFTER_MS;
assert.equal((await client.listRepos()).length, 1);
assert.equal(forge.mints, before + 1);
} finally {
+26
View File
@@ -173,3 +173,29 @@ test("a verdict that landed out of date keeps its account on the pull request",
const bare = lateVerdictComment({ commit: HEAD, trunk: TRUNK.slice(0, 12), verdict: "pass" }, null, words);
assert.match(bare, /head dfb1415aaaaa merged into trunk a2c2c35bbbbb, pass/);
});
test("a release branch's move puts every open pull request into it out of date, as a trunk's does (ADR 0358 §5)", async () => {
// novox/hq ADR 0358 §5: for a pull request into a release branch, the release branch takes the trunk's place in ADR
// 0323 — the check merges the head onto the release branch's head and names it as the trunk commit, so the release
// branch's move puts its open verdicts out of date, and leaves those into main alone.
const release = "feat/57-config-and-provisioning";
const heads: Record<string, string> = { main: TRUNK, [release]: "eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee" };
const pulls = [{ number: 315, base: release, head: "3".repeat(40) }, { number: 1, base: "main", head: "1".repeat(40) }];
const { forge, statuses } = aForge(pulls, heads);
const gates = new GateWriter(forge);
const trunks: Record<string, string> = {};
await weighTrunks(forge, gates, trunks, "novox", "mesh-controller");
heads[release] = MOVED;
assert.equal(await weighTrunks(forge, gates, trunks, "novox", "mesh-controller"), true);
const s = statuses.get("3".repeat(40))!;
assert.equal(s["mesh/merge-gate"].state, "pending");
assert.match(s["mesh/merge-gate"].description!, /judged against trunk eeeeeeeeeeee, the trunk is now d28ca95ccccc/);
assert.equal(s["mesh/repo-check"].state, "pending");
assert.equal(statuses.get("1".repeat(40))!["mesh/merge-gate"].state, "success", "a pull request into main is left");
assert.equal(trunks[`novox/mesh-controller:${release}`], MOVED);
// The merge tool's rule reads the same: a verdict judged against the release branch's old head is not current.
assert.equal(verdictAtTrunk({ "mesh/merge-gate": { state: "success",
description: "head x merged into trunk eeeeeeeeeeee — pass" } }, MOVED).current, false);
});
+120
View File
@@ -0,0 +1,120 @@
import assert from "node:assert/strict";
import { test } from "node:test";
import { createHmac } from "node:crypto";
import type { AddressInfo } from "node:net";
// **The forge says every event on the bus** (novox/hq issue 729): a signed delivery is published once and acted on,
// an unsigned or unverified one is refused, and an event and the poll seeing the same head announce it once.
const secret = "s".repeat(64);
const sign = (body: string, key = secret) => createHmac("sha256", key).update(body).digest("hex");
const opened = JSON.stringify({ action: "opened", number: 7, pull_request: { number: 7, title: "t", user: { login: "jo" },
head: { ref: "feat/x", sha: "a".repeat(40) }, base: { ref: "main" }, merged: false },
repository: { name: "hq", full_name: "novox/hq", owner: { login: "novox" } }, sender: { login: "jo" } });
async function serve(published: { name: string }[], acted: { name: string }[]) {
const { receiver, HOOK_PATH } = await import("../webhook.ts");
const server = receiver({ secret, port: 0, log: () => {},
publish: async (e) => { published.push(e); }, act: async (e) => { acted.push(e); } });
await new Promise<void>((r) => server.listen(0, "127.0.0.1", () => r()));
const url = `http://127.0.0.1:${(server.address() as AddressInfo).port}${HOOK_PATH}`;
return { server, url };
}
test("a signed event is published on the bus and acted on once", async () => {
const published: { name: string }[] = [], acted: { name: string }[] = [];
const { server, url } = await serve(published, acted);
try {
const headers = { "X-Gitea-Event": "pull_request", "X-Gitea-Signature": sign(opened), "X-Gitea-Delivery": "d-1" };
const res = await fetch(url, { method: "POST", body: opened, headers });
assert.equal(res.status, 200);
// The same delivery sent again is answered and published nowhere.
const again = await fetch(url, { method: "POST", body: opened, headers });
assert.equal(again.status, 200);
await server.idle();
assert.deepEqual(published.map((e) => e.name), ["forge.pull-request.opened"]);
assert.deepEqual(acted.map((e) => e.name), ["forge.pull-request.opened"]);
} finally { server.close(); }
});
test("an unsigned event, or one whose signature does not verify, is refused and published nowhere", async () => {
const published: { name: string }[] = [], acted: { name: string }[] = [];
const { server, url } = await serve(published, acted);
try {
const unsigned = await fetch(url, { method: "POST", body: opened, headers: { "X-Gitea-Event": "pull_request" } });
assert.equal(unsigned.status, 401);
assert.match(await unsigned.text(), /unsigned/);
const forged = await fetch(url, { method: "POST", body: opened,
headers: { "X-Gitea-Event": "pull_request", "X-Gitea-Signature": sign(opened, "another secret") } });
assert.equal(forged.status, 401);
assert.match(await forged.text(), /does not verify/);
const big = "x".repeat(5 * 1024 * 1024 + 1);
const large = await fetch(url, { method: "POST", body: big,
headers: { "X-Gitea-Event": "push", "X-Gitea-Signature": sign(big) } });
assert.equal(large.status, 413);
await server.idle();
assert.deepEqual(published, []);
assert.deepEqual(acted, []);
} finally { server.close(); }
});
test("an event and the poll seeing the same head announce it once", async () => {
const { claimHead } = await import("../webhook.ts");
const heads: Record<string, string> = {};
const sha = "a".repeat(40);
assert.equal(claimHead(heads, "novox/hq#7", sha), true, "the event claims the head and announces it");
assert.equal(claimHead(heads, "novox/hq#7", sha), false, "the poll finds it taken");
assert.equal(claimHead(heads, "novox/hq#7", "b".repeat(40)), true, "a new head is announced again");
});
test("each forge event is named in the module's stream and trimmed", async () => {
const { eventOf } = await import("../webhook.ts");
const merged = eventOf("pull_request", { action: "closed", pull_request: { number: 3, merged: true, merge_commit_sha: "c".repeat(40),
head: { sha: "a".repeat(40) }, base: { ref: "main" } }, repository: { name: "hq", owner: { login: "novox" } } });
assert.equal(merged.name, "forge.pull-request.merged");
assert.equal(merged.body.merge_commit_sha, "c".repeat(40));
assert.equal(eventOf("pull_request", { action: "label_updated", pull_request: { number: 3 }, label: { name: "mesh/ready" } }).name,
"forge.pull-request.labeled");
const comment = eventOf("issue_comment", { action: "created", is_pull: true, issue: { number: 3 },
comment: { id: 9, body: "ready at abc", user: { login: "jo" } } });
assert.equal(comment.name, "forge.pull-request.commented");
assert.equal(comment.body.comment, "ready at abc");
assert.equal(eventOf("push", { ref: "refs/heads/main", after: "d".repeat(40) }).name, "forge.push");
assert.equal(eventOf("pull_request_review_approved", {}).name, "forge.other");
});
test("the system webhook is made when none points at the receiver, and corrected when one does", async () => {
const { ensureSystemHook } = await import("../webhook.ts");
const calls: string[] = [];
let hooks: unknown[] = [];
const admin = { call: async (method: string, path: string, body?: any) => {
calls.push(`${method} ${path}`);
if (method === "GET") return { status: 200, body: hooks };
if (method === "POST") { assert.ok(body.events.includes("pull_request") && body.events.includes("push")); assert.equal(body.config.secret, secret); return { status: 201, body: {} }; }
return { status: 200, body: {} };
} };
assert.equal(await ensureSystemHook(admin, "http://h:3047/forge-event", secret), "created");
hooks = [{ id: 4, config: { url: "http://h:3047/forge-event" } }];
assert.equal(await ensureSystemHook(admin, "http://h:3047/forge-event", secret), "corrected");
assert.ok(calls.includes("PATCH /admin/hooks/4"));
});
test("the polls look every 30 s until events are heard, every five minutes while they are, and 30 s again on a miss", async () => {
const { Pace } = await import("../webhook.ts");
let now = 1_000_000;
const said: string[] = [];
const pace = new Pace(() => now, (l) => said.push(l));
assert.equal(pace.every(), 30_000, "no delivery yet");
pace.heard();
assert.equal(pace.every(), 300_000, "a delivery heard");
now += 11 * 60_000;
assert.equal(pace.every(), 30_000, "ten minutes without a delivery");
pace.heard();
now += 1000;
pace.miss("novox/hq#7 at aaaaaaaa");
assert.equal(pace.every(), 30_000, "the poll found what no event said");
assert.match(said.join("\n"), /MISSED novox\/hq#7/);
now += 1000;
pace.heard();
assert.equal(pace.every(), 300_000, "the next delivery");
});
+60 -4
View File
@@ -97,13 +97,25 @@ export interface MintedTokenOptions {
readonly scopes?: readonly string[];
readonly log?: (line: string) => void;
readonly fetch?: typeof fetch;
/** The clock, in milliseconds; a test sets it. */
readonly now?: () => number;
}
/**
* **How long a failed mint is not tried again** (novox/hq issue 739). Every call that finds no token mints, and a
* mint is two or three calls over basic auth, each one hashing the admin password on the forge. A mint that keeps
* failing, once per tool call, is a loop the forge pays for in CPU: on 2026-10-14 it ran several times a minute
* on the control-node while it was already overloaded. Within this window the last failure is answered again,
* without asking the forge.
*/
export const MINT_RETRY_AFTER_MS = 30_000;
/** The token the module mints for itself, kept in its state and renewed when the forge rejects it. */
export class MintedToken implements TokenSource {
private held: string | null = null;
private readFile = false;
private inflight: Promise<string> | null = null;
private failed: { at: number; error: unknown } | null = null;
private readonly name: string;
private readonly scopes: readonly string[];
private readonly log: (line: string) => void;
@@ -114,8 +126,11 @@ export class MintedToken implements TokenSource {
this.scopes = opts.scopes ?? TOKEN_SCOPES;
this.log = opts.log ?? ((line) => console.log(`[gitea] ${line}`));
this.fetchImpl = opts.fetch ?? fetch;
this.now = opts.now ?? Date.now;
}
private readonly now: () => number;
/**
* Build from the runtime's environment: the forge's URL, the admin login and password file the
* manifest hands the runtime, and the module's state directory (MESH_GITEA_STATE_DIR, a directory
@@ -126,7 +141,10 @@ export class MintedToken implements TokenSource {
const opts = MintedToken.optionsFromEnv(url, env);
// One source per kept file in a process. The watcher and the tools entrypoint both build a
// client in the same runtime; two sources over one file would each renew on a 401 and drop the
// other's token by name, forever. Shared, a renewal is one renewal.
// other's token by name, forever. Shared, a renewal is one renewal. Across processes it is not shared: the
// runtime serves the watcher (index.js) and the tools (tools/index.js) as separate bundles, so two sources over
// one file do exist. renew() reads the kept file before it mints, and a race that still leaves two tokens by
// the name is undone by dropping each by its id (doMint; novox/hq issue 739).
const shared = MintedToken.shared.get(opts.file);
if (shared) return shared;
const source = new MintedToken(opts);
@@ -163,7 +181,7 @@ export class MintedToken implements TokenSource {
}
if (kept !== null) return this.mint("the kept token was minted with other scopes than the tools take now — minting one with them");
}
return this.mint("no token kept — minting one");
return this.mint(this.failed ? "no token held since the last mint failed — minting one" : "no token kept — minting one");
}
async renew(rejected: string): Promise<string> {
@@ -188,8 +206,19 @@ export class MintedToken implements TokenSource {
/** One mint at a time: concurrent first calls share it, rather than each minting its own. */
private mint(why: string): Promise<string> {
if (this.inflight === null) {
// A mint that failed a moment ago is not asked of the forge again yet (novox/hq issue 739): its error stands.
if (this.failed !== null && this.now() - this.failed.at < MINT_RETRY_AFTER_MS) return Promise.reject(this.failed.error);
this.log(why);
this.inflight = this.doMint().finally(() => {
this.inflight = this.doMint().then(
(token) => {
this.failed = null;
return token;
},
(err) => {
this.failed = { at: this.now(), error: err };
throw err;
},
).finally(() => {
this.inflight = null;
});
}
@@ -261,7 +290,12 @@ export class MintedToken implements TokenSource {
// went while the forge's data stayed. It is ours to replace: drop it by name and mint again.
this.log(`the forge already holds a token named "${this.name}" — replacing it`);
const dropped = await call("DELETE", `/${encodeURIComponent(this.name)}`);
if (dropped.status !== 204 && dropped.status !== 404) {
if (dropped.status === 422) {
// More than one token carries the name (novox/hq issue 739): two processes minted at once, and the forge's
// check that a name is unused is not atomic. The forge refuses a delete by such a name for good, so every
// one of them is dropped by its id — all are ours, and none of their values is kept.
await this.dropEveryNamed(call, dropped);
} else if (dropped.status !== 204 && dropped.status !== 404) {
throw new Error(`Gitea DELETE /users/${this.opts.admin}/tokens/${this.name}: ${dropped.status} ${detail(dropped.body)}`);
}
res = await call("POST", "", { name: this.name, scopes: this.scopes });
@@ -277,6 +311,28 @@ export class MintedToken implements TokenSource {
this.log(`minted a token for "${this.opts.admin}" (${this.scopes.join(", ")}), kept at ${this.opts.file}`);
return token;
}
/** Drop, by id, every token of the admin account that carries this name; the forge listed them, page by page. */
private async dropEveryNamed(call: (method: string, path?: string) => Promise<{ status: number; body: any }>,
refused: { status: number; body: any }): Promise<void> {
const ids: number[] = [];
for (let page = 1; page <= 20; page++) {
const listed = await call("GET", `?page=${page}&limit=50`);
if (listed.status !== 200 || !Array.isArray(listed.body)) {
throw new Error(`Gitea DELETE /users/${this.opts.admin}/tokens/${this.name}: ${refused.status} ${detail(refused.body)}; ` +
`and its tokens could not be listed to drop them one by one: ${listed.status} ${detail(listed.body)}`);
}
for (const t of listed.body) if (t?.name === this.name && Number.isInteger(t?.id)) ids.push(t.id);
if (listed.body.length < 50) break;
}
this.log(`the forge holds ${ids.length} tokens named "${this.name}" — dropping each by its id`);
for (const id of ids) {
const dropped = await call("DELETE", `/${id}`);
if (dropped.status !== 204 && dropped.status !== 404) {
throw new Error(`Gitea DELETE /users/${this.opts.admin}/tokens/${id}: ${dropped.status} ${detail(dropped.body)}`);
}
}
}
}
function detail(body: unknown): string {
+3 -3
View File
@@ -10,7 +10,7 @@
import { registerModuleTools, type ToolDefinition } from "@novox/mesh-sdk/tools";
import { emit } from "@novox/mesh-sdk/events";
import { GiteaClient, type BranchProtection } from "../client.js";
import { GiteaClient, readyAuthors, type BranchProtection } from "../client.js";
import { appendNote, run, viewBody, viewComment } from "../delivery.js";
import { containedFrom, isCommit, mergeableUnder, newestByContext } from "../checks.js";
import {
@@ -514,9 +514,9 @@ function allGiteaTools(gitea: GiteaClient): ToolDefinition[] {
},
{
name: "gitea_open_pulls",
description: "Every open pull request of every repository, from the forge itself: its head, its base, the statuses the base's protection requires and the ones its head has, each with its state, description and when it was set. What mesh-delivery reads to say a pull request the mesh was never asked to check (novox/hq issue 347), or whose check started and never answered (issue 438). Reads only.",
description: "Every open pull request of every repository, from the forge itself: its head, its base, the statuses the base's protection requires and the ones its head has, each with its state, description and when it was set. What mesh-delivery reads to say a pull request the mesh was never asked to check (novox/hq issue 347), or whose check started and never answered (issue 438), and each pull request's labels and ready mark — the label mesh/ready and the head named by the newest comment \"ready at <full head sha>\" that a listed account (the setting ready-authors) wrote after the label was added and after the last push — which mesh-delivery merges on (issue 722). Reads only.",
input: {},
run: async () => ({ pulls: await gitea.openPulls() }),
run: async () => ({ pulls: await gitea.openPulls(readyAuthors()) }),
},
{
name: "gitea_contains",
+314
View File
@@ -0,0 +1,314 @@
// **The forge says every event on the bus** (novox/hq issue 729; to-be 10: an event makes it fast, nothing makes it
// necessary). The module keeps one system webhook on its forge for every event, pointing at its own receiver, signed
// with a secret it mints and keeps beside its token; the receiver refuses whatever is unsigned or does not verify, and
// publishes each event, trimmed, on the module's events as forge.<kind>.<action> (forge.pull-request.opened, forge.push, …),
// on the shared EVENTS path its other events take; its own stream (ADR 0363) is a follow-up once issue 725 ships.
// The polls stay as the safety net, slowed; the announced-heads and announced-merges records keep anything from
// being announced twice, whichever path saw it first.
//
// The secret is never logged: lines say it was minted or reused and where it is kept, never what it is.
import { createHmac, randomBytes, timingSafeEqual } from "node:crypto";
import { chmodSync, mkdirSync, readFileSync, renameSync, writeFileSync } from "node:fs";
import { createServer, type Server } from "node:http";
import { dirname } from "node:path";
/** The name the receiver answers on, and the port it listens on unless the manifest says otherwise. */
export const HOOK_PATH = "/forge-event";
export const HOOK_PORT = 3047;
/** The secret the forge signs with: read from its 0600 file, or minted and kept there. */
export function keptSecret(file: string, log: (line: string) => void = (l) => console.log(`[gitea] ${l}`)): string {
try {
const kept = readFileSync(file, "utf8").trim();
if (kept.length >= 32) {
log(`reusing the webhook secret kept at ${file}`);
return kept;
}
} catch (err) {
if ((err as NodeJS.ErrnoException).code !== "ENOENT") throw err;
}
const secret = randomBytes(32).toString("hex");
mkdirSync(dirname(file), { recursive: true, mode: 0o700 });
const tmp = `${file}.tmp`;
writeFileSync(tmp, secret + "\n", { mode: 0o600 });
chmodSync(tmp, 0o600);
renameSync(tmp, file);
log(`minted a webhook secret, kept at ${file}`);
return secret;
}
/** Whether a body carries the forge's HMAC-SHA256 signature (X-Gitea-Signature, hex) under the secret. */
export function verified(secret: string, body: Buffer | string, signature: string | undefined): boolean {
if (!secret || !signature || !/^[0-9a-f]{64}$/i.test(signature.trim())) return false;
const want = createHmac("sha256", secret).update(body).digest();
const got = Buffer.from(signature.trim(), "hex");
return got.length === want.length && timingSafeEqual(got, want);
}
/** One forge event as the bus carries it: its name inside the stream, and what is needed of it — no secrets. */
export interface ForgeEvent {
name: string;
body: Record<string, unknown>;
}
/** Every event name the module emits for the forge, without its forge. prefix (module.json `emits`). */
export const EVENTS: readonly string[] = ["push", "create", "delete", "fork", "status", "pull-request.opened", "pull-request.reopened", "pull-request.synchronized", "pull-request.edited", "pull-request.closed", "pull-request.merged", "pull-request.labeled", "pull-request.unlabeled", "pull-request.assigned", "pull-request.unassigned", "pull-request.reviewed", "pull-request.review-requested", "pull-request.commented", "pull-request.changed", "issue-comment.created", "issue-comment.edited", "issue-comment.deleted", "issues.opened", "issues.reopened", "issues.closed", "issues.edited", "issues.labeled", "issues.unlabeled", "issues.assigned", "issues.unassigned", "issues.changed", "release.published", "release.updated", "release.deleted", "release.changed", "repository.created", "repository.deleted", "repository.changed", "wiki.changed", "other"];
/** The families the module's stream carries (module.json `streams`); any other kind is said as other.<kind>; underscores become hyphens. */
export const FAMILIES = ["push", "create", "delete", "fork", "status", "pull_request", "issue_comment", "issues",
"release", "repository", "wiki"] as const;
const ACTIONS: Record<string, string> = { label_updated: "labeled", label_cleared: "unlabeled", synchronized: "synchronized" };
function str(v: unknown): string | undefined {
return typeof v === "string" && v ? v : undefined;
}
function num(v: unknown): number | undefined {
return typeof v === "number" ? v : undefined;
}
function login(u: any): string | undefined {
return str(u?.login) ?? str(u?.username);
}
/**
* The event a forge delivery is: its kind (X-Gitea-Event) and the payload's action, trimmed to ids, repository, refs,
* shas, author, labels and a comment's body (a ready mark is one, ADR 0364). A merge is pull-request.merged, a comment on
* a pull request pull-request.commented.
*/
export function eventOf(kind: string, payload: any): ForgeEvent {
// The bus's event names take lower-case letters, digits and hyphens: the forge's underscores become hyphens.
const event = forgeEventOf(kind, payload);
const name = event.name.toLowerCase().replace(/_/g, "-").replace(/[^a-z0-9.-]/g, "-");
// Published on the module's events as it publishes its others, each name declared in its `emits`: one the manifest
// does not name is forge.other, saying what the forge called it. Its own stream (ADR 0363) follows issue 725.
if (EVENTS.includes(name)) return { name: `forge.${name}`, body: event.body };
return { name: "forge.other", body: { ...event.body, kind, said: name } };
}
function forgeEventOf(said: string, payload: any): ForgeEvent {
// The forge names some kinds finer than their family (pull_request_label, pull_request_sync): read by the payload.
let kind = said;
if (said.startsWith("pull_request") && payload?.pull_request) kind = "pull_request";
if ((said.startsWith("issue") || said === "pull_request_comment") && payload?.comment) kind = "issue_comment";
if (said.startsWith("issue") && !payload?.comment && payload?.issue) kind = "issues";
const repo = payload?.repository ?? {};
const base: Record<string, unknown> = {
owner: login(repo.owner), repo: str(repo.name), full_name: str(repo.full_name), clone_url: str(repo.clone_url),
sender: login(payload?.sender),
};
const family = (FAMILIES as readonly string[]).includes(kind) ? kind : `other`;
let action = str(payload?.action);
const pull = payload?.pull_request;
if (kind === "push") {
return { name: "push", body: { ...base, ref: str(payload.ref), before: str(payload.before), after: str(payload.after),
commits: Array.isArray(payload.commits) ? payload.commits.length : undefined, pusher: login(payload.pusher) } };
}
if (kind === "pull_request" && pull) {
if (action === "closed" && pull.merged) action = "merged";
action = ACTIONS[action ?? ""] ?? action ?? "changed";
return { name: `pull_request.${action}`, body: { ...base, number: num(pull.number ?? payload.number),
title: str(pull.title), author: login(pull.user), head: str(pull.head?.ref), head_sha: str(pull.head?.sha),
base: str(pull.base?.ref), merged: !!pull.merged, merge_commit_sha: str(pull.merge_commit_sha),
merged_at: str(pull.merged_at), html_url: str(pull.html_url),
labels: Array.isArray(pull.labels) ? pull.labels.map((l: any) => str(l?.name)).filter(Boolean) : undefined,
label: str(payload.label?.name) } };
}
if (kind === "issue_comment") {
const issue = payload.issue ?? {};
const onPull = !!payload.is_pull || !!issue.pull_request;
return { name: onPull ? "pull_request.commented" : `issue_comment.${action ?? "created"}`, body: { ...base,
number: num(issue.number), comment_id: num(payload.comment?.id), author: login(payload.comment?.user),
comment: str(payload.comment?.body), action } };
}
if (kind === "create" || kind === "delete") {
return { name: kind, body: { ...base, ref: str(payload.ref), ref_type: str(payload.ref_type), sha: str(payload.sha) } };
}
if (kind === "status") {
return { name: "status", body: { ...base, sha: str(payload.sha), context: str(payload.context), state: str(payload.state) } };
}
const name = family === "other" ? `other.${kind.replace(/[^a-z0-9_-]/gi, "_") || "unknown"}` :
["fork"].includes(family) ? family : `${family}.${action ?? "changed"}`;
return { name, body: { ...base, number: num(payload?.number ?? payload?.issue?.number), action } };
}
export interface ReceiverOptions {
secret: string;
port: number;
host?: string;
/** Publishes one event on the bus. */
publish: (event: ForgeEvent) => Promise<void>;
/** Acts on it at once, beside publishing it (the announcer). */
act?: (event: ForgeEvent) => Promise<void>;
/** Told of every verified delivery, so the polls know events are heard. */
heard?: () => void;
log?: (line: string) => void;
}
/** The most a delivery may carry: the forge's push of many commits stays well under it. */
export const MAX_BODY = 5 * 1024 * 1024;
/** How many deliveries' ids are remembered, so a delivery sent again is published once. */
export const REMEMBERED = 2000;
/**
* The receiver. A request without a signature is refused before its body is read, a body over MAX_BODY is refused, and
* one whose signature does not verify is refused, each with 401 or 413 and a line saying so. A verified delivery is
* answered 200 at once and queued: it is published and acted on after the reply, one at a time, so the forge's
* delivery timeout is never reached. A delivery whose X-Gitea-Delivery id was already heard is answered and dropped.
*/
export function receiver(opts: ReceiverOptions): Server & { idle(): Promise<void> } {
const log = opts.log ?? ((l) => console.log(`[gitea] ${l}`));
const seen = new Set<string>();
let queue: Promise<void> = Promise.resolve();
const server = createServer((req, res) => {
if (req.method !== "POST" || (req.url ?? "").split("?")[0] !== HOOK_PATH) {
res.statusCode = 404;
return res.end();
}
const from = req.socket.remoteAddress ?? "?";
const signature = String(req.headers["x-gitea-signature"] ?? "").trim();
if (!signature) {
log(`refused a forge event from ${from}: it is unsigned`);
res.writeHead(401, { Connection: "close" });
res.end("unsigned\n");
req.destroy();
return;
}
if (Number(req.headers["content-length"] ?? 0) > MAX_BODY) {
log(`refused a forge event from ${from}: its body is over ${MAX_BODY} bytes`);
res.writeHead(413, { Connection: "close" });
res.end("too large\n");
req.destroy();
return;
}
const chunks: Buffer[] = [];
let size = 0;
let over = false;
req.on("data", (c: Buffer) => {
size += c.length;
if (size > MAX_BODY) {
if (!over) {
over = true;
log(`refused a forge event from ${from}: its body is over ${MAX_BODY} bytes`);
res.writeHead(413, { Connection: "close" });
res.end("too large\n");
req.destroy();
}
} else chunks.push(c);
});
req.on("end", () => {
if (over) return;
const body = Buffer.concat(chunks);
if (!verified(opts.secret, body, signature)) {
log(`refused a forge event from ${from}: its signature does not verify`);
res.statusCode = 401;
return res.end("signature does not verify\n");
}
let payload: unknown;
try {
payload = JSON.parse(body.toString("utf8"));
} catch {
res.statusCode = 400;
return res.end("not JSON\n");
}
opts.heard?.();
const delivery = String(req.headers["x-gitea-delivery"] ?? "");
if (delivery && seen.has(delivery)) {
res.statusCode = 200;
return res.end("already heard\n");
}
if (delivery) {
seen.add(delivery);
if (seen.size > REMEMBERED) seen.delete(seen.values().next().value as string);
}
const event = eventOf(String(req.headers["x-gitea-event"] ?? "other"), payload);
res.statusCode = 200;
res.end("queued\n");
queue = queue.then(async () => {
try {
await opts.publish(event);
if (opts.act) await opts.act(event);
} catch (err) {
log(`a forge event ${event.name} could not be published or acted on; the polls catch it — ${err instanceof Error ? err.message : err}`);
}
});
});
});
return Object.assign(server, { idle: () => queue });
}
/** What ensuring the system webhook needs of the forge's administration (GiteaAdmin's login). */
export interface HookAdmin {
call(method: string, path: string, body?: unknown): Promise<{ status: number; body: any }>;
}
/**
* **How fast the polls look** (issue 729): every five minutes only while the forge's events are heard — a verified
* delivery within the last ten minutes, and none missed since. Until the first delivery, after ten minutes without one,
* or once the poll found something no event said, every thirty seconds.
*/
export class Pace {
private last = 0;
private missed = 0;
private readonly now: () => number;
private readonly log: (l: string) => void;
constructor(now: () => number = Date.now, log: (l: string) => void = (l) => console.error(`[gitea] ${l}`)) {
this.now = now;
this.log = log;
}
heard(): void {
this.last = this.now();
}
/** The poll found a head or merge no forge event announced. */
miss(what: string): void {
if (this.hearing()) this.log(`THE FORGE'S EVENTS MISSED ${what}: the poll found it, so the polls look every 30 s again until the next event`);
this.missed = this.now();
}
hearing(): boolean {
const t = this.now();
return this.last > 0 && t - this.last < 10 * 60_000 && this.last > this.missed;
}
every(): number {
return this.hearing() ? 5 * 60_000 : 30_000;
}
}
/** Every event family the forge (1.27) can send a webhook. */
export const HOOK_EVENTS = ["create", "delete", "fork", "push", "issues", "issue_comment", "pull_request",
"pull_request_review_request", "wiki", "repository", "release", "package", "status", "workflow_run", "workflow_job"];
/**
* **One system webhook, for every event, at the receiver** (issue 729): created when none points there, corrected
* when one does but differs (events, active, content type), and its secret set again each start — the forge never
* hands a secret back, so a corrected secret cannot be told from a kept one. Answers what it did.
*/
export async function ensureSystemHook(admin: HookAdmin, url: string, secret: string): Promise<"created" | "corrected"> {
// POST /admin/hooks makes a system webhook, and its options take no flag for it (the forge's 1.27.3 API schema). The
// forge has no "every event" word: each family is named, and a family name (issues, pull_request) takes all of it.
const want = { type: "gitea", active: true, events: HOOK_EVENTS, branch_filter: "*",
config: { url, content_type: "json", secret } };
const listed = await admin.call("GET", "/admin/hooks?type=system&limit=50");
if (listed.status !== 200) throw new Error(`the forge's system webhooks could not be listed: ${listed.status}`);
const hooks: any[] = Array.isArray(listed.body) ? listed.body : [];
const ours = hooks.filter((h) => h?.config?.url === url);
if (ours.length === 0) {
const made = await admin.call("POST", "/admin/hooks", want);
if (made.status !== 201 && made.status !== 200) throw new Error(`the forge's system webhook could not be made: ${made.status}`);
return "created";
}
for (const extra of ours.slice(1)) await admin.call("DELETE", `/admin/hooks/${extra.id}`);
const fixed = await admin.call("PATCH", `/admin/hooks/${ours[0].id}`, { active: true, events: HOOK_EVENTS, branch_filter: "*",
config: { url, content_type: "json", secret } });
if (fixed.status !== 200) throw new Error(`the forge's system webhook could not be corrected: ${fixed.status}`);
return "corrected";
}
/**
* **Claim a head for announcing, once** (issue 729): true when `sha` is not already the head recorded for `key`, which
* it then is. The poll and the forge's event both claim before they announce, so whichever comes second finds it taken.
*/
export function claimHead(heads: Record<string, string>, key: string, sha: string): boolean {
if (!sha || heads[key] === sha) return false;
heads[key] = sha;
return true;
}
+198
View File
@@ -0,0 +1,198 @@
import { test } from "node:test";
import assert from "node:assert/strict";
import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import { join } from "node:path";
import {
BASE_IMAGE_ALIAS, BASE_MAX_AGE_DAYS, baseImageFrom, baseProblem, baseScript, buildingBase, cleanCredentials, credentialPath,
describeBase, launchRebuild, launchRun, listBaseBuilds, listRuns, readStatus, runRefusal, running, staleBase,
type BaseImage,
} from "../tools/runs.ts";
// Defends novox/hq issue 698: the lab's base image was 44 days old, a bed machine raised from it carried a
// package database the mirrors no longer served, and genesis-keeps-its-bus stopped on a 404 for
// postgresql-libs. Nothing said how old the image was, nothing refused it, and no mesh tool rebuilt it.
const now = new Date("2026-10-14T01:00:00Z");
const listed = (uploaded: string, alias = BASE_IMAGE_ALIAS) => JSON.stringify([
{ aliases: [{ name: "images:other" }], uploaded_at: "2026-10-13T00:00:00Z" },
{ aliases: [{ name: alias }], uploaded_at: uploaded, created_at: "2026-08-01T00:00:00Z" },
]);
test("the base image's age is read from when this machine made it", () => {
const b = baseImageFrom(listed("2026-08-31T09:00:00Z"), now);
assert.equal(b.exists, true);
assert.equal(b.built, "2026-08-31T09:00:00.000Z");
assert.equal(b.ageDays, 43);
assert.match(describeBase(b), /built 2026-08-31T09:00:00.000Z \(43 day\(s\) ago; the bound is 14\)/);
});
test("an image older than the bound refuses a run, naming the tool that rebuilds it", () => {
const stale = baseImageFrom(listed("2026-08-31T09:00:00Z"), now);
const reason = runRefusal(stale);
assert.ok(reason, "a 43-day-old base image was not refused");
assert.match(reason, /older than the lab's bound of 14 days/);
assert.match(reason, /lab_base \{"rebuild": true\}/);
assert.equal(staleBase(stale), reason);
});
test("an image within the bound runs", () => {
const fresh = baseImageFrom(listed(new Date(now.getTime() - BASE_MAX_AGE_DAYS * 86_400_000).toISOString()), now);
assert.equal(fresh.ageDays, BASE_MAX_AGE_DAYS);
assert.equal(runRefusal(fresh), undefined);
});
test("a missing image, or one whose age cannot be read, is said and not refused", () => {
const none = baseImageFrom(listed("2026-10-01T00:00:00Z", "something-else"), now);
assert.equal(none.exists, false);
assert.equal(runRefusal(none), undefined);
assert.match(describeBase(none), /is not on this machine; rebuild it with lab_base/);
const unread = baseImageFrom(undefined, now);
assert.equal(runRefusal(unread), undefined);
assert.match(describeBase(unread), /its age is not known/);
});
test("no run starts while the base image is rebuilt", () => {
const fresh = baseImageFrom(listed("2026-10-13T00:00:00Z"), now);
const reason = runRefusal(fresh, { id: "base-x", started: "2026-10-14T00:59:00Z" });
assert.match(reason ?? "", /is being rebuilt \(base-x/);
});
test("a rebuild is mesh-lab's base build from the forge's main, with no secret in its script", () => {
const s = baseScript("/work/base-x", "https://forge.invalid", "/work/.base-x.git-credentials");
assert.match(s, /clone --quiet --depth 1 'https:\/\/forge.invalid\/novox\/mesh-lab.git'/);
assert.match(s, /src\/cli.ts base build/);
assert.ok(s.indexOf("rm -f '/work/.base-x.git-credentials'") < s.indexOf("npm ci"), "the credential outlives the clone");
});
test("a rebuild is known by its status, and is not a run", () => {
const work = mkdtempSync(join(tmpdir(), "lab-base-"));
try {
mkdirSync(join(work, "base-1"));
writeFileSync(join(work, "base-1", "status.json"), JSON.stringify({ id: "base-1", state: "built", started: "2026-10-13T00:00:00Z" }));
mkdirSync(join(work, "base-2"));
// A build whose process is gone while it says building was lost, and holds nothing up.
writeFileSync(join(work, "base-2", "status.json"), JSON.stringify({ id: "base-2", state: "building", started: "2026-10-14T00:00:00Z", pid: 2 ** 22 + 7 }));
assert.deepEqual(listBaseBuilds(work).map((b) => `${b.id} ${b.state}`), ["base-2 lost", "base-1 built"]);
assert.equal(buildingBase(work), undefined);
writeFileSync(join(work, "base-2", "status.json"), JSON.stringify({ id: "base-2", state: "building", started: "2026-10-14T00:00:00Z", pid: process.pid }));
assert.equal(buildingBase(work)?.id, "base-2");
assert.deepEqual(listRuns(work), []);
assert.ok(readFileSync(join(work, "base-1", "status.json"), "utf8"));
} finally {
rmSync(work, { recursive: true, force: true });
}
});
// **A run and a rebuild asked together: one starts** (review 256). Each asks the package registry, which
// waits, and each refused while the other was going; asked at once, both passed their checks before either
// wrote its status. The registry is asked first now, and the check and the start are one step.
function aLab(): string {
return mkdtempSync(join(tmpdir(), "lab-launch-"));
}
const fresh: BaseImage = { alias: BASE_IMAGE_ALIAS, exists: true, built: "2026-10-13T00:00:00.000Z", ageDays: 1 };
const slowRegistry = () => new Promise<{ registry: string }>((ok) => setTimeout(() => ok({ registry: "https://forge.invalid/r/" }), 20));
const launch = { registry: slowRegistry, base: () => fresh };
function beginRun(work: string, id: string) {
return () => {
mkdirSync(join(work, id));
const s = { id, state: "running" as const, started: new Date().toISOString(), tests: [], refs: {}, pid: process.pid };
writeFileSync(join(work, id, "status.json"), JSON.stringify(s));
return s;
};
}
function beginBuild(work: string, id: string) {
return () => {
mkdirSync(join(work, id));
const b = { id, state: "building" as const, started: new Date().toISOString(), pid: process.pid };
writeFileSync(join(work, id, "status.json"), JSON.stringify(b));
return b;
};
}
test("a run and a rebuild asked at once: exactly one starts", async () => {
for (const runFirst of [true, false]) {
const work = aLab();
try {
// Both asked before either answers, in one order and then the other.
const ask = { run: () => launchRun(work, launch, beginRun(work, "run-a")), build: () => launchRebuild(work, launch, beginBuild(work, "base-a")) };
const [r, b] = runFirst ? await Promise.all([ask.run(), ask.build()])
: (await Promise.all([ask.build(), ask.run()])).reverse() as [Awaited<ReturnType<typeof ask.run>>, Awaited<ReturnType<typeof ask.build>>];
assert.equal([r.started, b.started].filter(Boolean).length, 1, `both or neither started: run ${r.started}, rebuild ${b.started}`);
assert.equal(listRuns(work).length + listBaseBuilds(work).length, 1);
} finally {
rmSync(work, { recursive: true, force: true });
}
}
});
test("two runs, or two rebuilds, asked at once: one of each starts", async () => {
const work = aLab();
try {
const runs = await Promise.all([launchRun(work, launch, beginRun(work, "run-a")), launchRun(work, launch, beginRun(work, "run-b"))]);
assert.deepEqual(runs.map((r) => r.started).sort(), [false, true]);
const other = aLab();
const builds = await Promise.all([launchRebuild(other, launch, beginBuild(other, "base-a")), launchRebuild(other, launch, beginBuild(other, "base-b"))]);
assert.deepEqual(builds.map((b) => b.started).sort(), [false, true]);
rmSync(other, { recursive: true, force: true });
} finally {
rmSync(work, { recursive: true, force: true });
}
});
test("a run is refused on a stale base, and starts nothing", async () => {
const work = aLab();
try {
const stale: BaseImage = { alias: BASE_IMAGE_ALIAS, exists: true, built: "2026-08-31T09:00:00.000Z", ageDays: 43 };
const r = await launchRun(work, { registry: slowRegistry, base: () => stale }, beginRun(work, "run-a"));
assert.equal(r.started, false);
assert.match(r.started ? "" : r.reason, /lab_base \{"rebuild": true\}/);
assert.deepEqual(listRuns(work), []);
} finally {
rmSync(work, { recursive: true, force: true });
}
});
test("a rebuild or a run that never recorded its process is lost, and holds nothing up", () => {
const work = aLab();
try {
mkdirSync(join(work, "base-x"));
writeFileSync(join(work, "base-x", "status.json"), JSON.stringify({ id: "base-x", state: "building", started: "2026-10-14T00:00:00Z" }));
mkdirSync(join(work, "run-x"));
writeFileSync(join(work, "run-x", "status.json"), JSON.stringify({ id: "run-x", state: "running", started: "2026-10-14T00:00:00Z", tests: [], refs: {} }));
assert.equal(listBaseBuilds(work)[0]?.state, "lost");
assert.equal(buildingBase(work), undefined);
assert.equal(readStatus(work, "run-x")?.state, "lost");
assert.equal(running(work), undefined);
} finally {
rmSync(work, { recursive: true, force: true });
}
});
test("a rebuild's credential left behind is cleaned, and a running one's is kept", () => {
const work = aLab();
try {
for (const id of ["base-old", "base-now", "run-old", "run-now"]) writeFileSync(credentialPath(work, id), "x");
const removed = cleanCredentials(work, [credentialPath(work, "base-now"), credentialPath(work, "run-now")]).sort();
assert.deepEqual(removed, [credentialPath(work, "base-old"), credentialPath(work, "run-old")].sort());
} finally {
rmSync(work, { recursive: true, force: true });
}
});
test("lab_check's problem: a stale, missing or unread base image; a fresh one is none", () => {
assert.match(baseProblem({ alias: BASE_IMAGE_ALIAS, exists: false }) ?? "", /is not on this machine/);
assert.match(baseProblem({ alias: BASE_IMAGE_ALIAS, exists: false, problem: "incus could not list the lab's images" }) ?? "", /age is not known/);
assert.match(baseProblem({ alias: BASE_IMAGE_ALIAS, exists: true, built: "x", ageDays: 30 }) ?? "", /older than the lab's bound/);
assert.equal(baseProblem(fresh), undefined);
});
// The tools use the gate and the check above (index.ts registers with the SDK when imported, so it is read).
test("lab_run and lab_base start through the gate, and lab_check names what is wrong with the base", () => {
const index = readFileSync(new URL("../tools/index.ts", import.meta.url), "utf8");
assert.match(index, /await launchRun\(work,/);
assert.match(index, /await launchRebuild\(work,/);
assert.match(index, /const unfit = baseProblem\(base\);\s*if \(unfit\) problems\.push\(unfit\)/);
assert.match(index, /if \(args\.rebuild !== true\) return reading;/);
});
+106 -1
View File
@@ -5,7 +5,7 @@ import { spawnSync } from "node:child_process";
import { tmpdir } from "node:os";
import { join } from "node:path";
import { installPlan, installStep, NPM_REPOSITORIES, REPOSITORIES, script } from "../tools/runs.ts";
import { installPlan, installStep, NPM_REPOSITORIES, outcomeOf, readStatus, REPOSITORIES, script } from "../tools/runs.ts";
// Defends novox/hq issue 624: lab_run ran `npm ci` at the root of every checkout it installs, and
// mesh-tools keeps its package in node-tools/ with no lock file at its root, so every run failed at its
@@ -381,3 +381,108 @@ test("a forge setting that is not http(s) trusts no origin of its own", async ()
packument("ftp://novox.internal/api/packages/novox/npm/x.tgz")).get, root);
assert.match(viaRoot.problem ?? "", /neither/);
});
// Defends novox/hq issue 693: a run whose only test skipped as NOT REACHED read "0 passed, 0 failed", exit 0,
// and lab_status said passed. The log's tail, as run-2026-10-14T00-21-44-014Z on shanks wrote it.
const oldLog = [
"STOPPED at step 16 of 24, base: \t\t\t\tMake sure you have all the referenced content and try again.",
"\ufe63 genesis adopts its bus as the nats module in place: one bus, its streams kept, the node connected (588152.016874ms) # ADR 0333 §5 NOT REACHED: the first run stopped at step 16 of 24, base, before the broker step:",
" \t\t\t\tMake sure you have all the referenced content and try again.",
"NotFound: content digest sha256:e2de96513ba9: not found",
"ℹ tests 1", "ℹ suites 0", "ℹ pass 0", "ℹ fail 0", "ℹ cancelled 0", "ℹ skipped 1", "ℹ todo 0",
"ℹ duration_ms 590254.163993",
"",
"recorded: test/integration/genesis-keeps-its-bus.test.ts — 0 passed, 0 failed, against mesh-lab 7ed7da1",
].join("\n");
// The same run, as the lab's suite says it once it names its skips (mesh-lab, issue 693).
const newLog = [
"ℹ tests 1", "ℹ pass 0", "ℹ fail 0", "ℹ skipped 1",
"",
"recorded: test/integration/genesis-keeps-its-bus.test.ts — 0 passed, 0 failed, 1 skipped, against mesh-lab 7ed7da1",
"NOT PASSED: no test passed: 1 skipped, 1 of them as NOT REACHED",
"skipped: genesis adopts its bus — one bus (test/integration/genesis-keeps-its-bus.test.ts) # ADR 0333 §5 NOT REACHED: the first run stopped at step 16 of 24, base: NotFound: content digest",
].join("\n");
function anEndedRun(log: string, state: "passed" | "failed", exit: number): { work: string; id: string } {
const work = mkdtempSync(join(tmpdir(), "lab-runs-"));
const id = "run-2026-10-14T00-21-44-014Z";
mkdirSync(join(work, id));
writeFileSync(join(work, id, "run.log"), log);
writeFileSync(join(work, id, "status.json"), JSON.stringify({
id, state, exit, started: "2026-10-14T00:21:44.014Z", ended: "2026-10-14T00:33:11.179Z",
tests: ["test/integration/genesis-keeps-its-bus.test.ts"], refs: {},
}));
return { work, id };
}
test("a run whose only test skipped as NOT REACHED, from a lab of before, reads as failed with the reason", () => {
const { work, id } = anEndedRun(oldLog, "passed", 0);
const s = readStatus(work, id);
assert.equal(s?.state, "failed", JSON.stringify(s));
assert.equal(s?.exit, 0);
assert.equal(s?.outcome?.passed, 0);
assert.equal(s?.outcome?.skipped, 1);
assert.equal(s?.outcome?.notPassed, "no test passed: 1 skipped, 1 of them as NOT REACHED");
assert.match(s?.outcome?.skips[0]?.reason ?? "", /^ADR 0333 §5 NOT REACHED: the first run stopped at step 16 of 24, base, before the broker step:$/);
// Kept in the run's record, not only answered.
const kept = JSON.parse(readFileSync(join(work, id, "status.json"), "utf8"));
assert.equal(kept.state, "failed");
assert.equal(kept.outcome.notPassed, "no test passed: 1 skipped, 1 of them as NOT REACHED");
rmSync(work, { recursive: true, force: true });
});
test("the lab's own NOT PASSED and skipped lines are read, a test name's punctuation and all", () => {
const { work, id } = anEndedRun(newLog, "failed", 1);
const s = readStatus(work, id);
assert.equal(s?.state, "failed");
assert.equal(s?.outcome?.notPassed, "no test passed: 1 skipped, 1 of them as NOT REACHED");
assert.deepEqual(s?.outcome?.skips, [{
file: "test/integration/genesis-keeps-its-bus.test.ts",
test: "genesis adopts its bus — one bus",
reason: "ADR 0333 §5 NOT REACHED: the first run stopped at step 16 of 24, base: NotFound: content digest",
}]);
rmSync(work, { recursive: true, force: true });
});
test("a run that passed tests and skipped one for a declared reason stays passed, its skip listed", () => {
const log = [
"✔ a module lands (12ms)",
"﹣ a builder builds (0.1ms) # set MESH_LAB_BUILDER to a built mesh-builder",
"ℹ tests 2", "ℹ pass 1", "ℹ fail 0", "ℹ skipped 1",
].join("\n");
const { work, id } = anEndedRun(log, "passed", 0);
const s = readStatus(work, id);
assert.equal(s?.state, "passed");
assert.equal(s?.outcome?.notPassed, undefined);
assert.equal(s?.outcome?.skips[0]?.reason, "set MESH_LAB_BUILDER to a built mesh-builder");
rmSync(work, { recursive: true, force: true });
});
test("a run that ran no test at all is not passed", () => {
assert.equal(outcomeOf("ℹ tests 0\nℹ pass 0\nℹ fail 0\nℹ skipped 0\n").notPassed, "no test passed: none ran");
});
// Review 254 on issue 693: a test's own output, printed before the suite reports, is not the suite's verdict.
test("a line a test printed that looks like the suite's is not read as the suite's", () => {
const log = [
"NOT PASSED: a test printed this",
"skipped: a forged skip (test/integration/x.test.ts) # FORGED NOT REACHED",
"✔ a module lands (12ms)",
"ℹ tests 1", "ℹ pass 1", "ℹ fail 0", "ℹ skipped 0",
"",
"recorded: test/integration/x.test.ts — 1 passed, 0 failed, 0 skipped, against mesh-lab 7ed7da1",
].join("\n");
const o = outcomeOf(log);
assert.equal(o.notPassed, undefined, JSON.stringify(o));
assert.deepEqual(o.skips, []);
});
test("not reached is matched in any case", () => {
const log = [
"✔ a module lands (12ms)",
"﹣ the broker step (0.1ms) # the broker step was not reached",
"ℹ tests 2", "ℹ pass 1", "ℹ fail 0", "ℹ skipped 1",
].join("\n");
assert.equal(outcomeOf(log).notPassed, "1 test(s) skipped as NOT REACHED: their subject was not reached");
});
+51 -12
View File
@@ -9,7 +9,8 @@ import { tmpdir } from "node:os";
import { join } from "node:path";
import {
checkDir, cleanCredentials, clearChecks, credentialArgs, forgeCredential, installPlan, listRuns, npmRegistry, NPM_REPOSITORIES, publicRootOf, readStatus, REPOSITORIES, running, start,
stop, tail, writeCredential,
stop, tail, writeCredential, baseBuildLog, buildingBase, listBaseBuilds, startBaseBuild, credentialPath, launchRun,
launchRebuild, BASE_MAX_AGE_DAYS, describeBase, readBaseImage, baseProblem,
} from "./runs.js";
export function getLabTools(env: NodeJS.ProcessEnv): ToolDefinition[] {
@@ -59,12 +60,18 @@ export function getLabTools(env: NodeJS.ProcessEnv): ToolDefinition[] {
// **And what a run would check out and install, before a run finds out** (novox/hq issue 624): every
// repository it clones answers on the forge, and every one it installs has a lock file where the run
// installs from.
// A credential a stopped, crashed or rebooted run left behind is removed, and said; one still in use
// by a running run is not.
const busy = running(work);
const left = cleanCredentials(work, busy ? `${work}/.${busy.id}.git-credentials` : undefined);
const removed = left.length > 0 ? `removed ${left.length} forge credential(s) a run left behind: ${left.join(", ")}` : "";
const output = [`${check.stdout}${check.stderr}`.trim(), removed, ...leftovers, ...problems].filter(Boolean).join("\n");
// A credential a stopped, crashed or rebooted run or rebuild left behind is removed, and said; one still in
// use by a running run or rebuild is not.
const busy = [running(work)?.id, buildingBase(work)?.id].filter((id): id is string => !!id);
const left = cleanCredentials(work, busy.map((id) => credentialPath(work, id)));
const removed = left.length > 0 ? `removed ${left.length} forge credential(s) a run or rebuild left behind: ${left.join(", ")}` : "";
// **And how old the base image is** (novox/hq issue 698): past the bound, missing, or of an age that cannot
// be read is a problem here, though lab_run refuses only the first.
const base = readBaseImage();
const unfit = baseProblem(base);
if (unfit) problems.push(unfit);
const output = [`${check.stdout}${check.stderr}`.trim(), describeBase(base), removed, ...leftovers, ...problems]
.filter(Boolean).join("\n");
return { ok: check.status === 0 && problems.length === 0, output };
},
},
@@ -96,11 +103,43 @@ export function getLabTools(env: NodeJS.ProcessEnv): ToolDefinition[] {
}
const stranger = Object.keys(refs).filter((r) => !REPOSITORIES.includes(r));
if (stranger.length > 0) return { started: false, reason: `the lab does not build ${stranger.join(", ")}` };
const busy = running(work);
if (busy) return { started: false, reason: `${busy.id} is still ${busy.state}; one run at a time`, running: busy };
const registry = await registryOf(forge);
return { started: true, run: start(work, forge, tests, refs, credentialOf(forge), registry.registry),
registry: registry.registry, ...(registry.problem ? { problem: registry.problem } : {}) };
// **One job at a time, not on a base image past its bound, nor while it is rebuilt** (novox/hq issue 698,
// review 256): the registry asked first, then the check and the start with nothing awaited between.
const launched = await launchRun(work, { registry: () => registryOf(forge), base: () => readBaseImage() },
(registry) => start(work, forge, tests, refs, credentialOf(forge), registry));
if (!launched.started) return launched;
return { started: true, run: launched.job, base: describeBase(launched.base), registry: launched.registry,
...(launched.problem ? { problem: launched.problem } : {}) };
},
},
{
name: "lab_base",
description:
"The lab's base image, which every bed machine with a container runtime is raised from: when it was built, its " +
`age, the bound lab_run refuses past (${BASE_MAX_AGE_DAYS} days), and the last rebuilds with the newest one's log. ` +
"Reads only, unless rebuild is true: then it rebuilds the image with mesh-lab's `base build` from the forge's " +
"main, in the background, and answers at once; ask again for the build's state and the image's new age.",
input: {
rebuild: { type: "boolean", description: "rebuild the image (default false: only read)" },
},
run: async (args) => {
const before = readBaseImage();
const builds = listBaseBuilds(work).slice(0, 3);
const newest = builds[0];
const reading = {
base: before, said: describeBase(before), bound: BASE_MAX_AGE_DAYS,
...(baseProblem(before) ? { problem: baseProblem(before) } : {}),
builds, ...(newest ? { log: baseBuildLog(work, newest.id) } : {}),
};
if (args.rebuild !== true) return reading;
const forge = forgeOf();
if (!forge) return { ...reading, started: false, reason: "the lab's forge is not set: settings for lab, {\"forge\": \"<url>\"}" };
// Not under a run, nor beside another rebuild: the registry asked first, then the check and the start with
// nothing awaited between (review 256).
const launched = await launchRebuild(work, { registry: () => registryOf(forge), base: () => before },
(registry) => startBaseBuild(work, forge, credentialOf(forge), registry));
if (launched.started === false) return { ...reading, started: false, reason: launched.reason };
return { ...reading, started: true, build: launched.job };
},
},
{
+364 -9
View File
@@ -6,7 +6,7 @@
// and that file is the whole of what the tools read back: a runtime that restarts mid-run still answers
// for it, and says it was lost rather than pretending it is still going.
import { spawn } from "node:child_process";
import { spawn, spawnSync } from "node:child_process";
import { existsSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, rmSync, writeFileSync } from "node:fs";
import { join } from "node:path";
@@ -223,16 +223,18 @@ trap ${shellQuote(`${rm}; exit 129`)} HUP`;
}
/**
* cleanCredentials removes every run's credential file left in the work directory but the one named — what
* a crash or a reboot left behind, which no trap could remove. Answers what it removed.
* cleanCredentials removes every run's and every base rebuild's credential file left in the work directory
* but the ones named — what a crash or a reboot left behind, which no trap could remove. Answers what it
* removed.
*/
export function cleanCredentials(work: string, keep?: string): string[] {
export function cleanCredentials(work: string, keep?: string | string[]): string[] {
if (!existsSync(work)) return [];
const kept = new Set([keep ?? []].flat());
const removed: string[] = [];
for (const name of readdirSync(work)) {
if (!/^\.run-.*\.git-credentials$/.test(name)) continue;
if (!/^\.(run|base)-.*\.git-credentials$/.test(name)) continue;
const path = join(work, name);
if (path === keep) continue;
if (kept.has(path)) continue;
rmSync(path, { force: true });
removed.push(path);
}
@@ -286,6 +288,117 @@ export interface RunStatus {
commits?: Record<string, string>;
exit?: number;
pid?: number;
/** What the suite said of its tests, read from the run's log once it ended (novox/hq issue 693). */
outcome?: Outcome;
}
/** One skipped test, its file and the reason it gave. */
export interface Skip {
file: string;
test: string;
reason: string;
}
/**
* What a run's suite said of its tests: its totals, each skip with its reason, and — when the run did not
* pass although its exit said 0 — why (`notPassed`). Totals the runner did not print are absent.
*/
export interface Outcome {
passed?: number;
failed?: number;
skipped?: number;
skips: Skip[];
notPassed?: string;
}
/**
* outcomeOf reads what a run's log says of its tests (novox/hq issue 693).
*
* **A run in which no test passed has proven nothing**, and the runner exits 0 when every test skipped:
* the genesis bed skipped its only test as "NOT REACHED", the suite printed "0 passed, 0 failed", and the
* lab recorded the run as passed. The lab's suite now says it itself — a "NOT PASSED: <why>" line and a
* "skipped: <test> (<file>) # <reason>" line per skip after its "recorded:" line, and an exit that is not
* 0 — and this reads those lines. For a run of a lab from before (a branch, or an old log), the same rule is
* read from the runner's own output: its summary totals, and its skip lines' first line of reason.
*
* The rule, the lab suite's (mesh-lab src/outcomes.ts): **not passed when no test passed**, or when a test
* skipped as NOT REACHED (its subject was not reached); a skip for another reason beside tests that passed
* leaves the run passed, its reason listed.
*/
export function outcomeOf(log: string): Outcome {
const plain = log.replace(/\u001b\[[0-9;]*m/g, "");
const total = (word: string) => {
// The last summary in the log: the runner's own, once per run.
const all = [...plain.matchAll(new RegExp(`^\\s*(?:\u2139|#)\\s*${word}\\s+(\\d+)\\s*$`, "gm"))];
const last = all[all.length - 1];
return last ? Number(last[1]) : undefined;
};
const outcome: Outcome = { skips: [] };
const passed = total("pass");
const failed = total("fail");
const skipped = total("skipped");
if (passed !== undefined) outcome.passed = passed;
if (failed !== undefined) outcome.failed = failed;
if (skipped !== undefined) outcome.skipped = skipped;
// **The suite's own lines are read only after its last "recorded:" line** (novox/hq issue 693, review
// 254): a test prints what it likes before the suite reports, and a line of its that looked like the
// suite's — "NOT PASSED: …", "skipped: … # NOT REACHED" — would otherwise stand for the suite's verdict.
const recorded = [...plain.matchAll(/^recorded: .*$/gm)];
const lastRecorded = recorded[recorded.length - 1];
const suiteSaid = lastRecorded ? plain.slice((lastRecorded.index ?? 0) + lastRecorded[0].length) : "";
// The suite's own line, "skipped: <test> (<file>) # <reason>": the file in brackets, a .ts, parts it.
for (const m of suiteSaid.matchAll(/^skipped: (.*?) \(([^()]+\.ts)\) # (.*)$/gm)) {
outcome.skips.push({ file: m[2] ?? "", test: m[1] ?? "", reason: m[3] === "no reason given" ? "" : m[3] ?? "" });
}
if (outcome.skips.length === 0) {
// A lab from before names no skip itself; its runner's spec line does: "﹣ <test> (<ms>ms) # <reason>".
for (const m of plain.matchAll(/^\s*\ufe63 (.*?) \([\d.]+m?s\) # (.*)$/gm)) {
const reason = (m[2] ?? "").trim();
outcome.skips.push({ file: "", test: m[1] ?? "", reason: reason === "SKIP" ? "" : reason });
}
}
const said = suiteSaid.match(/^NOT PASSED: (.*)$/m);
if (said) {
outcome.notPassed = (said[1] ?? "").trim();
} else if ((failed ?? 0) === 0) {
// In any case: "not reached" says the same as "NOT REACHED".
const unreached = outcome.skips.filter((s) => s.reason.toUpperCase().includes("NOT REACHED")).length;
if (passed === 0) {
outcome.notPassed = outcome.skips.length > 0 || (skipped ?? 0) > 0
? `no test passed: ${skipped ?? outcome.skips.length} skipped${unreached > 0 ? `, ${unreached} of them as NOT REACHED` : ""}`
: "no test passed: none ran";
} else if (unreached > 0) {
outcome.notPassed = `${unreached} test(s) skipped as NOT REACHED: their subject was not reached`;
}
}
return outcome;
}
/**
* settled is a run that ended, with its outcome read from its log: a run whose exit said 0 and whose suite
* proved nothing reads as failed, with why (novox/hq issue 693). Read once and kept in its status, so the
* record says it too; a run still going, stopped or lost is left as it is.
*/
export function settled(work: string, s: RunStatus): RunStatus {
if (s.outcome || !s.ended || (s.state !== "passed" && s.state !== "failed")) return s;
let log: string;
try {
log = readFileSync(join(runDir(work, s.id), "run.log"), "utf8");
} catch {
return s;
}
const outcome = outcomeOf(log);
const next: RunStatus = { ...s, outcome };
if (next.state === "passed" && outcome.notPassed) next.state = "failed";
try {
writeFileSync(statusPath(work, s.id), JSON.stringify(next, null, 2));
} catch {
// Read again next time: what lab_status answers is the same either way.
}
return next;
}
export function runDir(work: string, id: string): string {
@@ -300,16 +413,23 @@ export function readStatus(work: string, id: string): RunStatus | undefined {
try {
const s = JSON.parse(readFileSync(statusPath(work, id), "utf8")) as RunStatus;
// A run whose process is gone while its status still says it is going was lost — the runtime or
// the machine restarted under it. Said, rather than left reading as running for ever.
if (!["passed", "failed", "stopped", "lost"].includes(s.state) && s.pid && !alive(s.pid)) {
// the machine restarted under it. Said, rather than left reading as running for ever. One that never
// recorded a process (the runtime died between writing its status and starting it) is lost too: it
// would otherwise hold up every run after it (review 256).
if (!["passed", "failed", "stopped", "lost"].includes(s.state) && gone(s.pid)) {
s.state = "lost";
}
return s;
return settled(work, s);
} catch {
return undefined;
}
}
/** gone is whether a job's process is not running: none recorded, or none by that id. */
function gone(pid: number | undefined): boolean {
return !pid || !alive(pid);
}
function alive(pid: number): boolean {
try {
process.kill(pid, 0);
@@ -446,3 +566,238 @@ export function tail(work: string, id: string, lines: number): string {
return "";
}
}
// ---- the base image (novox/hq issue 698)
//
// Every bed machine that needs a container runtime is a clone of one base image, `mesh-lab/base`, which
// mesh-lab's `base build` makes from the distribution's current image. A machine raised from it carries the
// package database the image was built with. On a rolling distribution the mirrors stop serving what an old
// database names within weeks, and a bed that installs a package then fails on a 404 that reads as a mirror
// outage (novox/hq issue 205; seen 2026-10-14 when genesis-keeps-its-bus installed postgresql-libs from a
// database six weeks old). So the lab names the image's age wherever it is asked, refuses a run on an image
// older than the bound, and rebuilds the image when asked to.
//
// **The bound is 14 days** (BASE_MAX_AGE_DAYS). Arch's mirrors keep a package version for as long as it is
// current and drop it soon after it is replaced; an image a fortnight old has reliably installed what beds
// install, and the one that failed was 44 days old. A rebuild takes minutes, so a short bound costs little;
// a run refused is cheaper than a bed that fails an hour in for a reason that is not the change under test.
/** The image every bed machine with a runtime is raised from, as mesh-lab names it (src/lifecycle/place.ts). */
export const BASE_IMAGE_ALIAS = "mesh-lab/base";
/** How old the base image may be, in days, before a run is refused (see the head of this file). */
export const BASE_MAX_AGE_DAYS = 14;
const DAY = 24 * 60 * 60 * 1000;
export interface BaseImage {
alias: string;
/** Whether the image is there at all. */
exists: boolean;
/** When this machine made it, ISO. */
built?: string;
/** Its age in whole days. */
ageDays?: number;
/** Why its age could not be read, when it could not. */
problem?: string;
}
/**
* baseImageFrom reads `incus image list <alias> --format json`: the image whose aliases hold the base's
* name, and when it was made here (its upload, which a publish sets).
*/
export function baseImageFrom(listed: string | undefined, now: Date): BaseImage {
const alias = BASE_IMAGE_ALIAS;
if (listed === undefined) return { alias, exists: false, problem: "incus could not list the lab's images" };
let images: Array<{ aliases?: Array<{ name?: string }>; uploaded_at?: string; created_at?: string }>;
try {
images = JSON.parse(listed);
} catch {
return { alias, exists: false, problem: "incus listed the lab's images as something that is not JSON" };
}
const image = (Array.isArray(images) ? images : []).find((i) => (i.aliases ?? []).some((a) => a.name === alias));
if (!image) return { alias, exists: false };
const made = [image.uploaded_at, image.created_at].map((t) => (t ? Date.parse(t) : NaN))
.find((t) => Number.isFinite(t) && t > 0);
if (made === undefined) return { alias, exists: true, problem: "incus gives no time the image was made" };
return { alias, exists: true, built: new Date(made).toISOString(), ageDays: Math.floor((now.getTime() - made) / DAY) };
}
/** readBaseImage asks incus, on this machine, for the base image. */
export function readBaseImage(now = new Date()): BaseImage {
const listed = spawnSync("incus", ["image", "list", BASE_IMAGE_ALIAS, "--format", "json"], { encoding: "utf8" });
return baseImageFrom(listed.status === 0 ? listed.stdout : undefined, now);
}
/** The base image's age in one sentence, as lab_check and lab_run say it. */
export function describeBase(b: BaseImage): string {
if (b.problem) return `the base image ${b.alias}: its age is not known: ${b.problem}`;
if (!b.exists) return `the base image ${b.alias} is not on this machine; rebuild it with lab_base {"rebuild": true}`;
return `the base image ${b.alias} was built ${b.built} (${b.ageDays} day(s) ago; the bound is ${BASE_MAX_AGE_DAYS})`;
}
/**
* staleBase is why a run is refused for its base image, or undefined when it is not: an image older than the
* bound. An image that is missing or whose age cannot be read is not refused here — a bed that needs it says
* so itself, and a bed that does not need it runs.
*/
export function staleBase(b: BaseImage): string | undefined {
if (!b.exists || b.ageDays === undefined || b.ageDays <= BASE_MAX_AGE_DAYS) return undefined;
return `the base image ${b.alias} was built ${b.built}, ${b.ageDays} days ago, older than the lab's bound of ` +
`${BASE_MAX_AGE_DAYS} days: a bed machine raised from it carries a package database the mirrors no longer serve ` +
`(novox/hq issues 205, 698). Rebuild it with lab_base {"rebuild": true}, then run again`;
}
/**
* baseProblem is what lab_check and lab_base say is wrong with the base image, or undefined: past the bound, not
* on this machine, or of an age that cannot be read (review 256). Only the first refuses a run.
*/
export function baseProblem(b: BaseImage): string | undefined {
if (b.problem || !b.exists) return describeBase(b);
return staleBase(b);
}
/**
* runRefusal is why lab_run starts no run for the base image, or undefined: a rebuild still going, which
* replaces the image a bed raises its machines from, or an image older than the bound.
*/
export function runRefusal(base: BaseImage, rebuilding?: { id: string; started: string }): string | undefined {
if (rebuilding) return `the base image is being rebuilt (${rebuilding.id}, since ${rebuilding.started}); run once lab_base says it is built`;
return staleBase(base);
}
// ---- its rebuild: mesh-lab's `base build`, in the background, as a run is.
export interface BaseBuild {
id: string;
state: "building" | "built" | "failed" | "lost";
started: string;
ended?: string;
exit?: number;
pid?: number;
}
function buildStatusPath(work: string, id: string): string {
return join(work, id, "status.json");
}
/** The base builds, newest first. A build whose process is gone while it says building was lost. */
export function listBaseBuilds(work: string): BaseBuild[] {
if (!existsSync(work)) return [];
const builds: BaseBuild[] = [];
for (const id of readdirSync(work).filter((d) => d.startsWith("base-"))) {
try {
const b = JSON.parse(readFileSync(buildStatusPath(work, id), "utf8")) as BaseBuild;
if (b.state === "building" && gone(b.pid)) b.state = "lost";
builds.push(b);
} catch {
// A directory without a status is not a build.
}
}
return builds.sort((a, b) => b.started.localeCompare(a.started));
}
/** The base build still going, if any. */
export function buildingBase(work: string): BaseBuild | undefined {
return listBaseBuilds(work).find((b) => b.state === "building");
}
/** The last lines of a base build's log. */
export function baseBuildLog(work: string, id: string, lines = 40): string {
try {
return readFileSync(join(work, id, "build.log"), "utf8").split("\n").slice(-Math.max(1, lines)).join("\n");
} catch {
return "";
}
}
/**
* baseScript is what one rebuild runs: mesh-lab's main from the forge, installed the way a run installs it,
* then its `base build`, which replaces the image and reads back that it works before publishing it.
*/
export function baseScript(dir: string, forge: string, credential?: string, registry = `${forge}${NPM_PATH}`): string {
const lab = join(dir, "mesh-lab");
return `set -euo pipefail
export GIT_TERMINAL_PROMPT=0
${credentialGuard(credential)}
cd ${shellQuote(dir)}
git ${credentialArgs(credential)}clone --quiet --depth 1 ${shellQuote(`${forge}/novox/mesh-lab.git`)} ${shellQuote(lab)}
${credential ? `rm -f ${shellQuote(credential)}` : ""}
echo "mesh-lab $(git -C ${shellQuote(lab)} rev-parse HEAD)"
printf '%s\\n' ${shellQuote(`@novox:registry=${registry}`)} > ${shellQuote(join(dir, ".npmrc"))}
export NPM_CONFIG_USERCONFIG=${shellQuote(join(dir, ".npmrc"))}
${installStep(lab, false)}
cd ${shellQuote(lab)}
node --experimental-strip-types src/cli.ts base build
`;
}
/** startBaseBuild rebuilds the base image in the background and answers at once with the build. */
export function startBaseBuild(work: string, forge: string, credential?: string, registry?: string): BaseBuild {
const id = `base-${new Date().toISOString().replace(/[:.]/g, "-")}`;
const dir = join(work, id);
mkdirSync(dir, { recursive: true });
const build: BaseBuild = { id, state: "building", started: new Date().toISOString() };
writeFileSync(buildStatusPath(work, id), JSON.stringify(build, null, 2));
// What a crashed or rebooted run or rebuild left behind goes before this one writes its own.
cleanCredentials(work);
const credentials = writeCredential(credentialPath(work, id), credential);
writeFileSync(join(dir, "build.sh"), baseScript(dir, forge, credentials, registry), { mode: 0o700 });
const wrapper = `bash ${shellQuote(join(dir, "build.sh"))} > ${shellQuote(join(dir, "build.log"))} 2>&1; code=$?
node -e ${shellQuote(
`const f=${JSON.stringify(buildStatusPath(work, id))};const s=JSON.parse(require("fs").readFileSync(f,"utf8"));s.state=process.argv[1]==="0"?"built":"failed";s.exit=Number(process.argv[1]);s.ended=new Date().toISOString();require("fs").writeFileSync(f,JSON.stringify(s,null,2))`,
)} "$code"
rm -f ${shellQuote(credentialPath(work, id))}
rm -rf ${shellQuote(join(dir, "mesh-lab"))}`;
const child = spawn("bash", ["-c", wrapper], { detached: true, stdio: "ignore" });
child.unref();
build.pid = child.pid;
writeFileSync(buildStatusPath(work, id), JSON.stringify(build, null, 2));
return build;
}
// ---- starting a run or a rebuild: one job on the lab's machine at a time (review 256).
//
// **Everything that waits is done first; the check and the start are one step.** A run and a rebuild each
// refused while the other was going, but each asked the package registry — an await — between its check and
// writing its status, so a run and a rebuild asked together could both pass their checks and both start: a
// bed raising machines from an image being replaced under it. Here the registry is asked first, and nothing
// is awaited between reading what is going and writing the new job's status, so in the one runtime that
// serves the lab's tools no other call runs in between.
export interface Launch {
/** The address a run or a rebuild names the package registry by, asked of the forge (npmRegistry). */
registry: () => Promise<{ registry: string; problem?: string }>;
/** The base image as it is now. */
base: () => BaseImage;
}
export type Launched<T> = { started: true; registry: string; problem?: string; base: BaseImage; job: T }
| { started: false; reason: string; base: BaseImage };
/** launchRun starts a run unless a run or a rebuild is going, or the base image is past its bound. */
export async function launchRun(work: string, launch: Launch, begin: (registry: string) => RunStatus):
Promise<Launched<RunStatus>> {
const registry = await launch.registry();
// From here to begin(): nothing awaited.
const base = launch.base();
const busy = running(work);
if (busy) return { started: false, reason: `${busy.id} is still ${busy.state}; one run at a time`, base };
const refused = runRefusal(base, buildingBase(work));
if (refused) return { started: false, reason: refused, base };
return { started: true, ...registry, base, job: begin(registry.registry) };
}
/** launchRebuild starts a rebuild of the base image unless a run or another rebuild is going. */
export async function launchRebuild(work: string, launch: Launch, begin: (registry: string) => BaseBuild):
Promise<Launched<BaseBuild>> {
const registry = await launch.registry();
// From here to begin(): nothing awaited.
const base = launch.base();
const rebuilding = buildingBase(work);
if (rebuilding) return { started: false, reason: `${rebuilding.id} is still rebuilding it`, base };
// A run raises its machines from the image; replacing it under one would change what the run tests.
const busy = running(work);
if (busy) return { started: false, reason: `${busy.id} is still ${busy.state}, and raises its machines from this image`, base };
return { started: true, ...registry, base, job: begin(registry.registry) };
}
+216
View File
@@ -0,0 +1,216 @@
// Whether mail flows through mailu, as the node-engine asks it every interval (novox/hq ADR 0240, kind
// `tool`; issue 690). Every container read healthy on 2026-10-14 while postfix could not resolve a name:
// inbound mail was refused with 450 and outbound mail sat in the deferred queue for days. The parts'
// checks judge parts; this judges the flow, from inside the postfix container, the way postfix sees it:
//
// 1. postfix runs (`postfix status`, standing in for the image's own check, `! /usr/libexec/postfix/master
// -t`, which this one replaces);
// 2. the deferred queue (`postqueue -j`) shows no fault of this side: a recipient deferred longer than the
// bound because a lookup failed here, or overdue mail to three or more domains. One or two far domains
// that do not answer are theirs, and can stay deferred for days: said, never unhealthy, so one dead
// recipient does not page the operator for days;
// 3. a public mail host's name resolves through the resolver postfix is given: a resolver that
// SERVFAILs refuses inbound mail before anything reaches the queue.
//
// The answer is the health answer every health tool gives (mesh-sdk go/health): `state`, `healthy` beside
// it for an older reader, and `why` in words — counts, the oldest message's age and its first reason —
// never a bare false.
import { execFile } from "node:child_process";
import { promisify } from "node:util";
const run = promisify(execFile);
/** How long a message may wait in the deferred queue before mail is said not to flow. */
export const DEFERRED_BOUND_SECONDS = 30 * 60;
/** The public name resolved through postfix's resolver: gmail's mail exchanger, which mail needs. */
export const PUBLIC_NAME = "gmail-smtp-in.l.google.com";
// The lookup postfix's own would make, through the resolver the container is given; a failure is said on one
// line ("gaierror: [Errno -3] Try again" is what a SERVFAIL reads as), never as a traceback.
const RESOLVE = [
"import socket, sys",
"try:",
` socket.getaddrinfo(${JSON.stringify(PUBLIC_NAME)}, 25)`,
"except Exception as e:",
" sys.exit(type(e).__name__ + ': ' + str(e))",
].join("\n");
/** Overdue mail to this many recipient domains or more is our side's fault, not one far server's. */
export const SPREAD_DOMAINS = 3;
/**
* What postfix writes when its own lookups fail, as on 2026-10-14 (issue 687): "Host or domain name not
* found. Name service error for name=gmail.com type=MX: Host not found, try again", or "Temporary lookup
* failure". A far server that is down reads otherwise ("Connection timed out", "Connection refused").
*/
export const LOCAL_LOOKUP_FAILURE =
/Name service error|Temporary lookup failure|Host or domain name not found.*try again/is;
/** The domain of an address, lower-cased; "(no recipient)" when there is none. */
function domainOf(address?: string): string {
const at = address?.lastIndexOf("@") ?? -1;
return address && at >= 0 ? address.slice(at + 1).toLowerCase() : "(no recipient)";
}
/** Runs a command in a container: its stdout, or a rejection carrying stderr. */
export type Exec = (container: string, command: string[]) => Promise<string>;
/** The health answer (mesh-sdk go/health): healthy, or unhealthy with why. */
export interface HealthAnswer {
state: "healthy" | "unhealthy";
healthy: boolean;
why: string;
queue?: {
messages: number;
deferred: number;
overdue: number;
oldestDeferredSeconds?: number;
overdueDomains?: string[];
nameServiceFailures?: number;
};
}
/** One message as `postqueue -j` prints it, one JSON object a line (postfix 3.1 and later). */
interface Queued {
queue_name?: string;
queue_id?: string;
arrival_time?: number;
recipients?: Array<{ address?: string; delay_reason?: string }>;
}
/** docker exec, bounded: a hung container is an answer too. */
export function dockerExec(timeoutMs = 10_000): Exec {
return async (container, command) =>
(await run("docker", ["exec", container, ...command], { timeout: timeoutMs })).stdout;
}
/** What a failed exec said, on one line: its stderr when it has one, else its message. */
function said(e: unknown): string {
const err = e as { stderr?: string; message?: string; killed?: boolean; signal?: string };
if (err?.killed || err?.signal === "SIGTERM") return "it did not answer in time";
const text = (err?.stderr ?? "").trim() || (err?.message ?? String(e)).trim();
return text.split("\n").map((l) => l.trim()).filter(Boolean)[0] ?? "it failed and said nothing";
}
/** An age in words: "2 days 3 hours", "45 minutes". */
export function age(seconds: number): string {
const s = Math.max(0, Math.floor(seconds));
const d = Math.floor(s / 86400), h = Math.floor((s % 86400) / 3600), m = Math.floor((s % 3600) / 60);
const unit = (n: number, w: string) => `${n} ${w}${n === 1 ? "" : "s"}`;
if (d > 0) return h > 0 ? `${unit(d, "day")} ${unit(h, "hour")}` : unit(d, "day");
if (h > 0) return m > 0 ? `${unit(h, "hour")} ${unit(m, "minute")}` : unit(h, "hour");
return unit(m, "minute");
}
function parseQueue(output: string): Queued[] {
const queued: Queued[] = [];
for (const line of output.split("\n")) {
const text = line.trim();
if (!text.startsWith("{")) continue; // "Mail queue is empty" and the like
queued.push(JSON.parse(text) as Queued);
}
return queued;
}
/**
* Judge whether mail flows. `now` is in seconds since the epoch. Every part is asked even when one
* fails, so the answer says everything that is wrong at once.
*/
export async function mailFlows(
exec: Exec,
container: string,
now: number = Date.now() / 1000,
boundSeconds: number = DEFERRED_BOUND_SECONDS,
): Promise<HealthAnswer> {
const faults: string[] = [];
const [status, queue, resolve] = await Promise.allSettled([
exec(container, ["postfix", "status"]),
exec(container, ["postqueue", "-j"]),
exec(container, ["python3", "-c", RESOLVE]),
]);
if (status.status === "rejected") {
faults.push(`postfix in ${container} does not run: ${said(status.reason)}.`);
}
let summary: HealthAnswer["queue"];
const notes: string[] = [];
if (queue.status === "rejected") {
faults.push(`postfix's queue in ${container} could not be read: ${said(queue.reason)}.`);
} else {
let queued: Queued[] = [];
try {
queued = parseQueue(queue.value);
} catch (e) {
faults.push(`postfix's queue in ${container} printed a line that is not JSON: ${said(e)}.`);
}
const deferred = queued.filter((q) => q.queue_name === "deferred");
const overdue = deferred
.filter((q) => typeof q.arrival_time === "number" && now - q.arrival_time > boundSeconds)
.sort((a, b) => (a.arrival_time ?? 0) - (b.arrival_time ?? 0));
const oldest = deferred.reduce<number | undefined>(
(o, q) => (typeof q.arrival_time === "number" && (o === undefined || q.arrival_time < o) ? q.arrival_time : o),
undefined,
);
// Each overdue recipient, oldest message first: what our side is judged by, not one far host.
const late = overdue.flatMap((q) =>
(q.recipients?.length ? q.recipients : [{}]).map((r) => ({
to: r.address,
domain: domainOf(r.address),
reason: r.delay_reason?.trim() || "postfix gave no reason",
waited: now - (q.arrival_time ?? now),
})),
);
const lookups = late.filter((r) => LOCAL_LOOKUP_FAILURE.test(r.reason));
const domains = [...new Set(late.map((r) => r.domain))];
summary = {
messages: queued.length,
deferred: deferred.length,
overdue: overdue.length,
...(oldest !== undefined ? { oldestDeferredSeconds: Math.max(0, Math.floor(now - oldest)) } : {}),
...(domains.length > 0 ? { overdueDomains: domains } : {}),
...(lookups.length > 0 ? { nameServiceFailures: lookups.length } : {}),
};
const one = (r: (typeof late)[number]) =>
`the oldest, ${age(r.waited)} old${r.to ? ` to ${r.to}` : ""}, waits because: ${r.reason}`;
const bound = age(boundSeconds);
if (lookups.length > 0) {
faults.push(
`${lookups.length} deferred ${lookups.length === 1 ? "recipient has" : "recipients have"} waited longer ` +
`than ${bound} on a name-service failure, which is this server's resolver; ${one(lookups[0])}.`,
);
} else if (domains.length >= SPREAD_DOMAINS) {
faults.push(
`${overdue.length} of ${deferred.length} deferred messages have waited longer than ${bound}, to ` +
`${domains.length} domains (${domains.join(", ")}): more than one far server's fault; ${one(late[0])}.`,
);
} else if (late.length > 0) {
notes.push(
`${overdue.length} deferred ${overdue.length === 1 ? "message has" : "messages have"} waited longer than ` +
`${bound}, only to ${domains.join(" and ")}, which reads as the far side's: ${one(late[0])}.`,
);
}
}
if (resolve.status === "rejected") {
faults.push(`${container} cannot resolve ${PUBLIC_NAME} through its resolver: ${said(resolve.reason)}.`);
}
if (faults.length > 0) {
const why = [`Mail does not flow.`, ...faults, ...notes].join(" ");
return { state: "unhealthy", healthy: false, why, queue: summary };
}
const waiting = !summary || summary.deferred === 0
? " The deferred queue is empty."
: notes.length > 0
? ` ${notes.join(" ")}`
: ` ${summary.deferred} deferred, the oldest ${age(summary.oldestDeferredSeconds ?? 0)} old, within ${age(boundSeconds)}.`;
return {
state: "healthy",
healthy: true,
why: `Mail flows: postfix runs, ${PUBLIC_NAME} resolves.${waiting}`,
queue: summary,
};
}
+32 -4
View File
@@ -5,6 +5,10 @@
"policy": "record",
"why": "people's mail: any change to how its containers are declared recreates them together in one send, and the mail is down for everyone until they are up again — a person chooses the moment (hq ADR 0242, issue 295)"
},
"unhealthy": {
"severity": "urgent",
"why": "while mailu is broken, mail is refused or stuck for everyone with a mailbox here (hq ADR 0355, issue 687)"
},
"capabilities": [
"container-runtime"
],
@@ -56,6 +60,19 @@
"api-token": "${dir:state}/api-token.secret"
}
},
"tools": [
"mailu_list_users",
"mailu_create_user",
"mailu_change_password",
"mailu_delete_user",
"mailu_list_aliases",
"mailu_create_alias",
"mailu_delete_alias",
"mailu_list_domains",
"mailu_read_mail",
"mailu_search_mail",
"mailu_health"
],
"emits": [
"user.created",
"user.deleted",
@@ -371,7 +388,7 @@
"id": "resolver",
"type": "container",
"name": "mailu-resolver",
"image": "ghcr.io/mailu/unbound@sha256:3a0fdfb364a63f4f9259526e013c1ef40f5f14de3621ce1560804b3a5909584a",
"image": "ghcr.io/mailu/unbound@sha256:2c037bc5c988b5c0ab7e1b6e6ffa762e14ba1e45c96281ac87f8b9be8eee8b6e",
"health": {
"kind": "runtime"
},
@@ -443,7 +460,12 @@
"name": "mailu-smtp",
"image": "ghcr.io/mailu/postfix@sha256:e2e49f39e53b80eac9e7a2f18d9df11edeb4914fd62dbba89b3155e8e034f62e",
"health": {
"kind": "runtime"
"kind": "tool",
"tool": "mailu_health",
"interval": "60s",
"timeout": "30s",
"looks": 3,
"grace": "120s"
},
"network": "mailu",
"env-file": [
@@ -560,7 +582,12 @@
"name": "mailu-front",
"image": "ghcr.io/mailu/nginx@sha256:36f98897cd1bc9d27628bbb4e04bdf60147af2ec7507d6da77f002c4f256896d",
"health": {
"kind": "runtime"
"kind": "exec",
"command": "curl -m3 -skfLo /dev/null http://127.0.0.1:10204/health && kill -0 $(cat /run/dovecot/master.pid) && python3 -c \"import smtplib; s = smtplib.SMTP('127.0.0.1', 25, timeout=5); s.ehlo('health'); s.quit()\" && openssl x509 -checkend 1209600 -noout -in /certs/letsencrypt/live/mailu/nginx-chain.pem",
"interval": "60s",
"timeout": "15s",
"looks": 3,
"grace": "120s"
},
"network": "mailu",
"env-file": [
@@ -637,6 +664,7 @@
"MESH_MAILU_URL": "http://127.0.0.1:${port:8080}/api/v1",
"MESH_MAILU_API_KEY_FILE": "${dir:state}/api-token.secret",
"MESH_MAILU_IMAP_CONTAINER": "mailu-imap",
"MESH_MAILU_SMTP_CONTAINER": "mailu-smtp",
"MESH_MAILU_CONFIG_FILE": "${dir:mesh-state}/config.json",
"MESH_RECEIVES": "${dir:grants}/mesh.json"
}
@@ -673,7 +701,7 @@
"jails": [
{
"name": "mailu-front",
"failregex": "^.*(?:imap|pop3|submission|managesieve)-login: .*\\(auth failed, \\d+ attempts(?: in \\d+ secs)?\\):.*rip=<HOST>(?:,|$)",
"failregex": "^.*(?:imap|pop3|submission|managesieve)-login: .*\\(auth failed, \\d+ attempts(?: in \\d+ secs)?\\):.*rip=<HOST>(?:,|$)\n ^.*client login failed: \"AUTH not supported\" while in http auth state, client: <HOST>, server: \\S+:25,",
"jail": "backend = systemd\njournalmatch = CONTAINER_NAME=mailu-front\nport = smtp,submission,submissions,imap,imaps,pop3,pop3s\nmaxretry = 3\nfindtime = 1d\nbantime = 1d"
}
]
+140
View File
@@ -0,0 +1,140 @@
// Whether mail flows (novox/hq issue 690): the health tool is unhealthy, in words, when a message waits in
// postfix's deferred queue longer than the bound, when postfix does not run, when a public name does not
// resolve, and when the container cannot be reached — and healthy otherwise. Run against the compiled
// module, the way the runtime loads it.
import { test } from "node:test";
import assert from "node:assert/strict";
import { mailFlows, DEFERRED_BOUND_SECONDS } from "../dist/health.js";
const NOW = 1_800_000_000;
type Answers = { status?: string | Error; queue?: string | Error; resolve?: string | Error };
/** A container that answers each command as told; an Error is a failed exec, with its stderr. */
function container(a: Answers) {
const asked: string[][] = [];
const exec = async (_name: string, command: string[]) => {
asked.push(command);
const answer = command[0] === "postfix" ? a.status ?? "" : command[0] === "postqueue" ? a.queue ?? "" : a.resolve ?? "";
if (answer instanceof Error) throw answer;
return answer;
};
return { exec, asked };
}
function failed(stderr: string): Error {
return Object.assign(new Error("Command failed: docker exec"), { stderr });
}
function deferred(id: string, ageSeconds: number, reason: string, to = "friend@example.com"): string {
return JSON.stringify({
queue_name: "deferred",
queue_id: id,
arrival_time: NOW - ageSeconds,
message_size: 1234,
sender: "someone@example.org",
recipients: [{ address: to, delay_reason: reason }],
});
}
test("an empty queue is healthy, and says so", async () => {
const c = container({ queue: "" });
const a = await mailFlows(c.exec, "mailu-smtp", NOW);
assert.equal(a.state, "healthy");
assert.equal(a.healthy, true);
assert.match(a.why, /deferred queue is empty/);
assert.deepEqual(a.queue, { messages: 0, deferred: 0, overdue: 0 });
assert.equal(c.asked.length, 3, "postfix, its queue and a lookup were each asked");
});
test("a recipient overdue on a name-service failure is unhealthy, with counts, its age and its reason", async () => {
const reason = "Host or domain name not found. Name service error for name=gmail.com type=MX: Host not found, try again";
const queue = [
deferred("AAA111", 3 * 86400 + 2 * 3600, reason),
deferred("BBB222", 5 * 60, "connect to mx.example.net[192.0.2.1]:25: Connection timed out"),
JSON.stringify({ queue_name: "active", queue_id: "CCC333", arrival_time: NOW - 10, recipients: [] }),
].join("\n");
const a = await mailFlows(container({ queue }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "unhealthy");
assert.equal(a.healthy, false);
assert.match(a.why, /^Mail does not flow\./);
assert.match(a.why, /1 deferred recipient has waited longer than 30 minutes on a name-service failure/);
assert.match(a.why, /3 days 2 hours old to friend@example\.com/);
assert.ok(a.why.includes(reason), "the reason is said");
assert.deepEqual(a.queue, {
messages: 3, deferred: 2, overdue: 1, oldestDeferredSeconds: 3 * 86400 + 2 * 3600,
overdueDomains: ["example.com"], nameServiceFailures: 1,
});
});
test("one dead far domain overdue is healthy, and said", async () => {
const queue = [
deferred("AAA111", 2 * 86400, "connect to mx.dead.example[192.0.2.9]:25: Connection timed out", "a@dead.example"),
deferred("BBB222", 3600, "connect to mx.dead.example[192.0.2.9]:25: Connection timed out", "b@dead.example"),
].join("\n");
const a = await mailFlows(container({ queue }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "healthy");
assert.equal(a.healthy, true);
assert.match(a.why, /2 deferred messages have waited longer than 30 minutes, only to dead\.example, which reads as the far side's/);
assert.match(a.why, /2 days old to a@dead\.example, waits because: connect to mx\.dead\.example/);
assert.deepEqual(a.queue?.overdueDomains, ["dead.example"]);
});
test("overdue mail to three domains is unhealthy, whatever the reason", async () => {
const timedOut = "connect to mx[192.0.2.1]:25: Connection timed out";
const queue = [
deferred("AAA111", 7200, timedOut, "a@one.example"),
deferred("BBB222", 3600, timedOut, "b@two.example"),
deferred("CCC333", 3600, timedOut, "c@three.example"),
].join("\n");
const a = await mailFlows(container({ queue }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "unhealthy");
assert.match(a.why, /3 of 3 deferred messages have waited longer than 30 minutes, to 3 domains \(one\.example, two\.example, three\.example\)/);
assert.match(a.why, /the oldest, 2 hours old to a@one\.example/);
});
test("a deferred message younger than the bound is healthy, and is said", async () => {
const queue = deferred("BBB222", DEFERRED_BOUND_SECONDS - 60, "connect to mx.example.net: Connection timed out");
const a = await mailFlows(container({ queue }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "healthy");
assert.equal(a.healthy, true);
assert.match(a.why, /1 deferred, the oldest 29 minutes old, within 30 minutes/);
});
test("a container that cannot be reached is unhealthy, and says what docker said", async () => {
const gone = failed("Error response from daemon: No such container: mailu-smtp\n");
const a = await mailFlows(container({ status: gone, queue: gone, resolve: gone }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "unhealthy");
assert.equal(a.healthy, false);
assert.match(a.why, /postfix in mailu-smtp does not run: Error response from daemon: No such container: mailu-smtp\./);
assert.match(a.why, /postfix's queue in mailu-smtp could not be read: Error response from daemon: No such container/);
assert.equal(a.queue, undefined);
});
test("a name that does not resolve is unhealthy, though the queue is empty", async () => {
const a = await mailFlows(
container({ queue: "", resolve: failed("gaierror: [Errno -3] Try again\n") }).exec,
"mailu-smtp",
NOW,
);
assert.equal(a.state, "unhealthy");
assert.match(a.why, /mailu-smtp cannot resolve gmail-smtp-in\.l\.google\.com through its resolver: gaierror: \[Errno -3\] Try again\./);
});
test("a container that does not answer in time is said so", async () => {
const hung = Object.assign(new Error("Command failed"), { killed: true, signal: "SIGTERM" });
const a = await mailFlows(container({ queue: hung }).exec, "mailu-smtp", NOW);
assert.equal(a.state, "unhealthy");
assert.match(a.why, /postfix's queue in mailu-smtp could not be read: it did not answer in time\./);
});
test("the health tool is served without the admin API, on the smtp container it is told", async () => {
const { collectTools } = await import("@novox/mesh-sdk/tools");
await import("../dist/tools/index.js");
const mailu = collectTools({ MESH_MAILU_SMTP_CONTAINER: "mailu-smtp" }).find((m) => m.module === "mailu");
const names = (mailu?.tools ?? []).map((t) => t.name);
assert.deepEqual(names, ["mailu_health"]);
assert.match(mailu!.tools[0].description, /inside mailu-smtp/);
});
+27 -4
View File
@@ -4,6 +4,27 @@
import { registerModuleTools, type ToolDefinition } from "@novox/mesh-sdk/tools";
import { MailuClient } from "../client.js";
import { dockerExec, mailFlows, DEFERRED_BOUND_SECONDS, PUBLIC_NAME, SPREAD_DOMAINS } from "../health.js";
/**
* Whether mail flows (novox/hq ADR 0240, issue 690): the tool the node-engine asks as the smtp container's
* check. It needs no admin API, only the postfix container, so it is served even when the API is not
* configured — a check that vanished with the API would say unhealthy for the wrong reason.
*/
export function getHealthTool(env: NodeJS.ProcessEnv): ToolDefinition {
const container = env.MESH_MAILU_SMTP_CONTAINER || "mailu-smtp";
return {
name: "mailu_health",
description:
`Whether mail flows through mailu, asked inside ${container}: postfix runs; no recipient has waited ` +
`deferred longer than ${DEFERRED_BOUND_SECONDS / 60} minutes on a name-service failure here, and overdue ` +
`deferred mail does not reach ${SPREAD_DOMAINS} or more domains (one or two far domains are said, not ` +
`unhealthy); and ${PUBLIC_NAME} resolves through the resolver postfix uses. Says counts, domains, the ` +
"oldest message's age and its reason (novox/hq ADR 0240, issue 690).",
input: {},
run: async () => mailFlows(dockerExec(), container),
};
}
export function getMailuTools(mailu: MailuClient): ToolDefinition[] {
return [
@@ -143,12 +164,14 @@ export function getMailuTools(mailu: MailuClient): ToolDefinition[] {
];
}
// The tools exist only when the admin API is configured; without it, mailu contributes none rather
// than failing the whole runtime.
// The admin tools exist only when the admin API is configured; without it, mailu contributes only its
// health tool rather than failing the whole runtime.
registerModuleTools("mailu", (env) => {
let tools: ToolDefinition[] = [];
try {
return getMailuTools(MailuClient.fromEnv(env));
tools = getMailuTools(MailuClient.fromEnv(env));
} catch {
return [];
tools = [];
}
return [...tools, getHealthTool(env)];
});
+1 -1
View File
@@ -8,5 +8,5 @@
"skipLibCheck": true,
"noEmit": true
},
"include": ["client.ts", "index.ts", "tools/index.ts"]
"include": ["client.ts", "health.ts", "index.ts", "tools/index.ts"]
}
+96
View File
@@ -0,0 +1,96 @@
// What the catalogue makes of a build outcome it hears (novox/hq issue 712).
//
// Kept apart from index.ts, which subscribes at its top level, so the decision can be tested without a bus.
import type { Made } from "./store.js";
/** What the build seat says when it has built something, and what the control plane replays. */
export interface Built {
module?: string;
commit?: string;
repository?: string;
path?: string;
ref?: string;
manifest?: unknown;
/** What this build published, each pinned as anything else would name it. */
made?: Made[];
/**
* Every artifact this was built on top of, so the edge is derived rather than declared.
*
* References, not module names: the builder sees a pinned image and cannot see which module
* produced it. Turning that into an edge between module-versions is this module's job.
*/
against?: string[];
/**
* This is history, not news — the mesh re-announcing a build this catalogue was not there for.
*
* Registered exactly as any other, and announced as nothing. A module that moved months ago is
* not something anything should act on now: emitting `upgraded` would have the control plane
* decide about a rollout, and `rebuild-needed` would ask for builds of things that are already
* current.
*/
replay?: boolean;
/** Why the build failed; a failed build made no module-version. */
failed?: string;
/** A dry run is looked at, never taken in (novox/hq issue 240). */
"dry-run"?: boolean;
/** A pull request's merge check or a branch tried before one (to-be 45 §9, issue 686): nothing merged. */
"check-outcome"?: unknown;
check?: unknown;
/** The repository's default branch as the build seat read the forge at the build (ADR 0238); empty from a seat older than the rule. */
trunk?: string;
/** Whether the commit is on `trunk`. */
"on-trunk"?: boolean;
/** The branches the commit is on. */
branches?: string[];
}
/** The events a build outcome reaches this module on: the build seat's, as it happens, and the control plane's replay. */
export const BUILD_EVENTS = [
// **The build seat that builds now** (novox/hq ADR 0190). Until issue 712 the catalogue heard only the
// retired seat below, so every build after the handover reached the graph only through the replay this
// module asks for when it starts: a merge built and pushed stayed out of the catalogue until its next start.
"node-build-agent.built",
// The retired build role, while a machine still serves it (ADR 0190's handover). Goes with the seat row.
"mesh-build-machine.built",
// What the mesh already held, when this module asks what it missed (novox/hq ADR 0134).
"mesh-controller.built-before",
] as const;
/**
* Whether a build outcome is a module-version to place in the graph, and why not when it is not.
*
* It mirrors two of the rules the controller keeps before it registers a build (mesh-controller
* cmd/mesh-controller/main.go `builds.Built`, build.go `publishable`): a merge check, a try and a dry run are
* never published, and **only a commit on its repository's trunk is** (ADR 0238) — a branch built through
* the `build` verb is checked, never a module's version. Where the seat names no trunk (one older than the
* rule) it is placed, as the controller lets it through. A failure made nothing and is passed over too.
*
* Not mirrored, because the catalogue cannot see what they read: the branch a module follows when it is not
* the default one (the controller's `follows`; such a module's builds off the default branch are passed over
* here and reach the graph by the replay), ADR 0266's rule that a module registers only from the repository
* it is built from, a superseded request, a build in the scratch namespace, one that failed its gate, and a
* build whose source is unchanged. Each needs the controller's records. This is safe for what the mesh
* sends, which is chosen from the controller's own module table and never from this graph; that the graph
* can still hold a build the controller refused under those rules is a follow-up (hq issue 712).
*/
export function placeable(body: Built):
| { placed: "yes"; module: string; commit: string }
| { placed: "no"; why: string; fault: boolean } {
if (body["check-outcome"] != null || body.check != null) {
return { placed: "no", why: "a merge check or a try: nothing of it merged", fault: false };
}
if (body["dry-run"]) {
return { placed: "no", why: "a dry run is looked at, never taken in", fault: false };
}
if (body.failed) {
return { placed: "no", why: "a failed build made no module-version", fault: false };
}
if (body.trunk && !body["on-trunk"] && !(body.branches ?? []).includes(body.trunk)) {
return { placed: "no", why: `the commit is not on its repository's trunk ${body.trunk}: checked, never published (ADR 0238)`, fault: false };
}
if (!body.module || !body.commit) {
return { placed: "no", why: "it named no module or no commit", fault: true };
}
return { placed: "yes", module: body.module, commit: body.commit };
}
+17 -41
View File
@@ -10,7 +10,8 @@
// diamond need no special handling, and nothing holds a plan.
import { on, emit } from "@novox/mesh-sdk/events";
import { Graph, type Made } from "./store.js";
import { Graph } from "./store.js";
import { BUILD_EVENTS, placeable, type Built } from "./heard.js";
const graph = Graph.fromEnv();
@@ -21,54 +22,29 @@ const graph = Graph.fromEnv();
// would block the very apply that brings the overlay up — stopped being true when a step's failure
// became this module's business and not the machine's (ADR 0136).
/** What the builder says when it has built something. */
interface Built {
module?: string;
commit?: string;
repository?: string;
path?: string;
ref?: string;
manifest?: unknown;
/** What this build published, each pinned as anything else would name it. */
made?: Made[];
/**
* Every artifact this was built on top of, so the edge is derived rather than declared.
*
* References, not module names: the builder sees a pinned image and cannot see which module
* produced it. Turning that into an edge between module-versions is this module's job.
*/
against?: string[];
/**
* This is history, not news — the mesh re-announcing a build this catalogue was not there for.
*
* Registered exactly as any other, and announced as nothing. A module that moved months ago is
* not something anything should act on now: emitting `upgraded` would have the control plane
* decide about a rollout, and `rebuild-needed` would ask for builds of things that are already
* current.
*/
replay?: boolean;
}
/**
* What a build means for the graph, wherever it came from.
*
* Two emitters say the same thing and neither is a mistake: the build machine says it as it happens,
* and the control plane says what it already held when this module asks what it missed
* (novox/hq ADR 0134). A replay is marked as one in its body, so nothing acts on a module that moved
* months ago — see `replay` above.
* months ago — see `replay` in heard.ts.
*/
const placeTheBuild = async (event: { body: unknown }): Promise<void> => {
const body = (event.body ?? {}) as Built;
if (!body.module || !body.commit) {
// Said rather than dropped: a build that announced itself without saying what it built is a
// fault in the builder, and a silent discard here would make it look like a missing event.
console.error("mesh-catalog: a build event named no module or no commit; ignored", body);
const placing = placeable(body);
if (placing.placed === "no") {
// A check, a try, a dry run and a failure are the build seat's ordinary outcomes: passed over. One
// that names no module or commit is a fault in the builder, said rather than dropped, so it does
// not look like a missing event.
if (placing.fault) console.error(`mesh-catalog: a build event was not placed: ${placing.why}`, body);
return;
}
const { module, commit } = placing;
const { upgraded, previous } = await graph.register({
module: body.module,
commit: body.commit,
module,
commit,
repository: body.repository ?? "",
path: body.path ?? "",
ref: body.ref ?? "",
@@ -80,7 +56,7 @@ const placeTheBuild = async (event: { body: unknown }): Promise<void> => {
if (body.replay) return;
await emit("registered", {
module: body.module, commit: body.commit, upgraded,
module, commit, upgraded,
});
// **A rebuild that changed nothing is not an upgrade.** Announcing it would ripple outward
@@ -88,7 +64,7 @@ const placeTheBuild = async (event: { body: unknown }): Promise<void> => {
if (!upgraded) return;
await emit("upgraded", {
module: body.module, commit: body.commit, previous,
module, commit, previous,
});
// What can be built now — stale, and waiting on nothing that is itself stale.
@@ -101,9 +77,9 @@ const placeTheBuild = async (event: { body: unknown }): Promise<void> => {
}
};
// As it happens, and what the mesh already held when this module asked what it missed.
await on("mesh-build-machine.built", placeTheBuild);
await on("mesh-controller.built-before", placeTheBuild);
// As it happens — from the build seat that builds now, and the retired one while it still answers — and
// what the mesh already held when this module asked what it missed (novox/hq issue 712).
for (const heard of BUILD_EVENTS) await on(heard, placeTheBuild);
// **And ask for what was built before this catalogue existed** (novox/hq 04-ISSUES/050).
//
+1
View File
@@ -26,6 +26,7 @@
"postgres-database": "${dir:state}/database.secret"
},
"consumes": [
"node-build-agent.built",
"mesh-build-machine.built",
"mesh-controller.built-before"
],
+72
View File
@@ -0,0 +1,72 @@
import { test } from "node:test";
import assert from "node:assert/strict";
import { readFile } from "node:fs/promises";
import { BUILD_EVENTS, placeable } from "../heard.js";
// A walk's build as the build seat says it (mesh-controller's link.BuildResult), here mailu's of novox/hq issue 712.
const walkBuild = {
id: "build-1791997333800041716",
repository: "http://novox.internal:20000/novox/mesh-catalog.git",
path: "modules/mailu",
ref: "main",
on: "shanks",
trunk: "main",
"on-trunk": true,
branches: ["main"],
module: "mailu",
commit: "e1c8965ddbf1cf9af9cf1eed7a72ae9ffa6d5a87",
manifest: { module: "mailu", unhealthy: { severity: "urgent", why: "mail stops" } },
made: [{ name: "code", kind: "bundle", reference: "artifact-store://mailu/code/blobs/sha256:d041" }],
};
test("the catalogue hears the build seat that builds now, not only the retired one (issue 712)", async () => {
assert.ok(BUILD_EVENTS.includes("node-build-agent.built"));
assert.ok(BUILD_EVENTS.includes("mesh-controller.built-before"));
// Subscribing is not enough: the mesh grants and routes only what the manifest consumes.
const manifest = JSON.parse(await readFile(new URL("../module.json", import.meta.url), "utf8"));
for (const heard of BUILD_EVENTS) assert.ok(manifest.consumes.includes(heard), `${heard} is not consumed`);
});
test("a walk's build is placed, under its module and commit", () => {
assert.deepEqual(placeable(walkBuild), { placed: "yes", module: "mailu", commit: walkBuild.commit });
// A walk that took the check's build of the same key says so, and is placed all the same (ADR 0350 §2).
assert.equal(placeable({ ...walkBuild, reused: "the check's build" } as never).placed, "yes");
});
test("a merge check, a try, a dry run and a failure are passed over without a fault", () => {
for (const [body, why] of [
[{ ...walkBuild, "check-outcome": { verdict: "passed" } }, "merge check"],
[{ ...walkBuild, check: { try: true } }, "merge check or a try"],
[{ ...walkBuild, "dry-run": true }, "dry run"],
[{ ...walkBuild, failed: "exit status 1" }, "failed"],
] as const) {
const placing = placeable(body);
assert.equal(placing.placed, "no");
if (placing.placed === "no") {
assert.match(placing.why, new RegExp(why));
assert.equal(placing.fault, false);
}
}
});
test("an outcome that names no module or commit is said as the builder's fault", () => {
const placing = placeable({ repository: walkBuild.repository, commit: walkBuild.commit });
assert.equal(placing.placed, "no");
if (placing.placed === "no") assert.equal(placing.fault, true);
});
test("a branch built through the build verb is passed over; a commit on the trunk, or from a seat that names none, is placed (ADR 0238)", () => {
const branch = { ...walkBuild, ref: "feature", "on-trunk": false, branches: ["feature"] };
const placing = placeable(branch);
assert.equal(placing.placed, "no");
if (placing.placed === "no") {
assert.match(placing.why, /not on its repository's trunk main/);
assert.equal(placing.fault, false);
}
// On the trunk by its branches although the seat did not set on-trunk.
assert.equal(placeable({ ...branch, branches: ["feature", "main"] }).placed, "yes");
// A seat older than the rule says no trunk: placed, as the controller lets it through.
const { trunk, "on-trunk": _on, branches, ...older } = walkBuild;
assert.equal(placeable(older).placed, "yes");
});
+1
View File
@@ -11,6 +11,7 @@
"include": [
"pg.d.ts",
"store.ts",
"heard.ts",
"index.ts",
"tools/index.ts",
"prepare/index.ts"
@@ -396,6 +396,9 @@ type fakeForge struct {
// open is what OpenPulls answers; openDown a forge that cannot say.
open []OpenPull
openDown bool
// merges, comments and unlabels are what was asked of the forge for ready marks (novox/hq issue 722).
merges, comments, unlabels []string
refuse error
}
func (f *fakeForge) OpenPulls() ([]OpenPull, error) {
@@ -649,3 +652,25 @@ func aWalk(id, repo, commit string, waits bool) Walk {
}
return w
}
// Merge, Comment and Unlabel record what was asked (novox/hq issue 722); Merge answers refuse when set.
func (f *fakeForge) Merge(owner, repo string, number int, head string) error {
f.mu.Lock()
defer f.mu.Unlock()
f.merges = append(f.merges, fmt.Sprintf("%s/%s#%d@%s", owner, repo, number, head))
return f.refuse
}
func (f *fakeForge) Comment(owner, repo string, number int, body string) error {
f.mu.Lock()
defer f.mu.Unlock()
f.comments = append(f.comments, fmt.Sprintf("%s/%s#%d: %s", owner, repo, number, body))
return nil
}
func (f *fakeForge) Unlabel(owner, repo string, number, label int) error {
f.mu.Lock()
defer f.mu.Unlock()
f.unlabels = append(f.unlabels, fmt.Sprintf("%s/%s#%d-%d", owner, repo, number, label))
return nil
}
@@ -69,6 +69,8 @@ type Holder struct {
leaveRefused map[string]string
// tried counts the owed effects Flush tried since this owner started, by how each ended (novox/hq issue 407).
tried owedTries
// marks are the open pull requests read with a ready mark, by owner/repo#number (novox/hq issue 722, ready.go).
marks map[string]*marked
}
// owedTries is how many owed effects were tried since this owner started, and when the last was: three numbers
@@ -120,6 +122,7 @@ func (h *Holder) init() {
h.askTried = map[string]time.Time{}
h.walkAsked = map[string]time.Time{}
h.unchecked = map[string]unchecked{}
h.marks = map[string]*marked{}
h.walks = map[string]seenWalk{}
h.orphans = map[string]time.Time{}
h.asking = map[string]bool{}
@@ -996,6 +999,12 @@ func (h *Holder) Tick() {
for _, ask := range h.leaveStopped() {
ask()
}
h.mu.Lock()
asks := h.readyAsks(h.Now())
h.mu.Unlock()
for _, ask := range asks {
ask()
}
}
// plan works out, under the lock, what to ask; each ask applies its own answer.
@@ -76,8 +76,9 @@ func run(h *Holder, l *listening) {
logf("[mesh-delivery] the controller's walks cannot be read: %v", err)
}
}
// Every open pull request, from the forge itself, every five minutes (novox/hq issue 347).
if now.Sub(looked) >= 5*time.Minute {
// Every open pull request, from the forge itself, every minute (novox/hq issues 347, 722): what says a
// head nobody checks, and the ready marks a merge is asked on.
if now.Sub(looked) >= time.Minute {
looked = now
h.SeeOpenPulls()
}
@@ -65,7 +65,9 @@ func TestItClaimsTheDeliverySeatAndFollowsWhatItHandles(t *testing.T) {
want = append(want, "seat:"+ControllerSeat+"."+v)
}
want = append(want, "gitea.gitea_note_append", "gitea.gitea_delivery_view", "gitea.gitea_commit_status",
"gitea.gitea_commit_statuses", "gitea.gitea_contains", "gitea.gitea_open_pulls")
"gitea.gitea_commit_statuses", "gitea.gitea_contains", "gitea.gitea_open_pulls",
// The merge on a ready mark (novox/hq issue 722): the merge tool, a comment, a label taken off.
"gitea.gitea_merge_pull_request", "gitea.gitea_add_comment", "gitea.gitea_api")
if !reflect.DeepEqual(m.Invokes, want) {
t.Fatalf("invokes %v, wanted %v", m.Invokes, want)
}
@@ -110,6 +110,9 @@ type Delivery struct {
LetGo time.Time `json:"let_go"`
LetGoWhy string `json:"let_go_why,omitempty"`
// Ready is what this owner did with its pull request's ready mark (novox/hq issue 722, ready.go).
Ready *ReadyRun `json:"ready,omitempty"`
Transitions []Transition `json:"transitions,omitempty"`
// Owed are the forge's and the bus's effects of its transitions not yet done: said only after the
// delivery is kept, retried until done.
@@ -91,6 +91,13 @@ type Forge interface {
// OpenPulls is every open pull request, with what its base requires and what its head has (novox/hq issue
// 347). Reads only.
OpenPulls() ([]OpenPull, error)
// Merge asks the forge's merge tool to merge a pull request at the head named (gitea_merge_pull_request, novox/hq
// issue 722): the one path every merge takes. Its refusal is the error, in its words.
Merge(owner, repo string, number int, head string) error
// Comment adds a comment to a pull request.
Comment(owner, repo string, number int, body string) error
// Unlabel takes a label off a pull request, by the label's id.
Unlabel(owner, repo string, number, label int) error
}
// ForgeChecks is what the forge says of a commit's checks (the gitea module's gitea_commit_statuses).
@@ -392,6 +399,19 @@ func (f toolForge) Statuses(owner, repo string, number int, commit, base string)
return &c, nil
}
func (f toolForge) Merge(owner, repo string, number int, head string) error {
return f.tool("gitea_merge_pull_request", map[string]any{"owner": owner, "repo": repo, "number": number, "head": head})
}
func (f toolForge) Comment(owner, repo string, number int, body string) error {
return f.tool("gitea_add_comment", map[string]any{"owner": owner, "repo": repo, "number": number, "body": body})
}
func (f toolForge) Unlabel(owner, repo string, number, label int) error {
return f.tool("gitea_api", map[string]any{"method": "DELETE",
"path": fmt.Sprintf("/repos/%s/%s/issues/%d/labels/%d", owner, repo, number, label)})
}
func (f toolForge) OpenPulls() ([]OpenPull, error) {
raw, err := f.ask("gitea.gitea_open_pulls", map[string]any{})
if err != nil {
@@ -0,0 +1,323 @@
package main
import (
"fmt"
"strings"
"time"
)
// Merge on ready (novox/hq issue 722).
//
// The operator, 2026-10-14: "the reviewer task should merge the request or mark the MR as ready to merge in any case".
// Before, a reviewed pull request with green checks waited until a session called the forge's merge tool, and again
// after every check the tool asked because the trunk had moved. Now the independent reviewer marks the pull request
// ready at the head it approved — the forge label `mesh/ready` and a comment "ready at <full head sha>" (approving
// through the forge is the operator's) — and this owner merges it, through the forge's own merge tool
// (gitea_merge_pull_request), the one path every merge takes:
//
// - only at the head the mark names, and only while that is the pull request's head: a push after the mark clears
// it — the label taken off, and said on the pull request — and the new head needs a mark of its own;
// - only when its delivery is ready here: its check passed against its trunk's head. A verdict out of date is judged
// again by the merge tool, which asks the check against the trunk's head itself (novox/hq ADR 0323 §5); the merge
// is asked again once a minute, and lands on the passing verdict;
// - a group member only at its turn in the group's order, and only once the group's composed check passed for its
// members' heads as they are;
// - a check that failed merges nothing: it is said once on the pull request, and as a stalled line, which the
// controller raises as the condition `delivery.<id>.stalled`. So is a mark the merge tool refused for an hour.
//
// What this owner does with a mark is kept on the delivery (Ready), said in its log, and the merge's transitions name
// it as their author.
// ReadyMark is a pull request's ready mark as the forge reads it (the gitea module's gitea_open_pulls): the head the
// newest "ready at" comment names, who wrote it and when, and the label's id. Head is empty when the label is on and
// no comment names a head.
type ReadyMark struct {
Head string `json:"head"`
By string `json:"by,omitempty"`
At string `json:"at,omitempty"`
LabelID int `json:"label_id"`
}
// ReadyLabel is the label that marks a pull request ready.
const ReadyLabel = "mesh/ready"
// readyTryEvery is how often a marked head's merge is asked while the merge tool refuses it: a check it asked takes
// minutes, and a refusal costs the forge a few reads.
const readyTryEvery = time.Minute
// readyRefusedFor is how long the merge tool may refuse a marked head before it is said as stalled: longer than a
// check takes (the controller's checkTimeout is 45 minutes), so a check on its way is never said.
const readyRefusedFor = time.Hour
// ReadyRun is what this owner did with a delivery's ready mark: kept on the delivery and shown whole by `show`.
type ReadyRun struct {
// Head, By and MarkedAt are the mark: the head it names, who marked it, when.
Head string `json:"head"`
By string `json:"by,omitempty"`
MarkedAt string `json:"marked_at,omitempty"`
// Seen is when this owner first read the mark.
Seen time.Time `json:"seen"`
// Waits is what the merge waits on, when it waits: a verdict, the group's turn or its composed check.
Waits string `json:"waits,omitempty"`
// Tries is how many times the merge was asked; LastTry when, and Answer what the merge tool last answered.
Tries int `json:"tries,omitempty"`
LastTry time.Time `json:"last_try"`
Answer string `json:"answer,omitempty"`
// RefusedSince is when the merge tool began refusing it without a break.
RefusedSince *time.Time `json:"refused_since,omitempty"`
// Merged is when the merge tool merged it.
Merged *time.Time `json:"merged,omitempty"`
// Failed is why it is not merged and will not be on this head: its check failed. Said once on the pull request.
Failed string `json:"failed,omitempty"`
}
// mergedBy is who a merge this owner made on a ready mark is said to be by, in its transitions.
func (r *ReadyRun) mergedBy() string {
if r == nil || r.Merged == nil {
return ""
}
by := byOwner + ", on the ready mark at " + shortOf(r.Head, 12)
if r.By != "" {
by += " by " + r.By
}
return by
}
// marked is a pull request read with a ready mark.
type marked struct {
pull OpenPull
// cleared is when the mark was cleared, the label taken off: not cleared again within readyTryEvery.
cleared time.Time
}
// pullKey is a pull request as a key: owner/repo#number.
func pullKey(p OpenPull) string {
return strings.ToLower(p.Owner+"/"+p.Repo) + fmt.Sprintf("#%d", p.Number)
}
// seeMarks keeps the open pull requests that carry a ready mark; under the lock. A pull request no longer open or
// no longer marked drops out.
func (h *Holder) seeMarks(open []OpenPull) {
seen := map[string]*marked{}
for _, p := range open {
if p.Ready == nil || p.HeadSHA == "" {
continue
}
k := pullKey(p)
m := &marked{pull: p}
if was := h.marks[k]; was != nil && was.pull.Ready != nil && was.pull.Ready.Head == p.Ready.Head &&
was.pull.HeadSHA == p.HeadSHA {
m.cleared = was.cleared
}
seen[k] = m
}
h.marks = seen
}
// sameFull is whether a mark names a head: the whole sha, or — read from an older forge module — at least twelve
// characters of it.
func sameFull(mark, head string) bool {
mark, head = strings.ToLower(mark), strings.ToLower(head)
return len(mark) >= 12 && strings.HasPrefix(head, mark)
}
// readyAsks works out, under the lock, what each ready mark needs: a mark cleared, a merge asked, a failure said.
// The asking happens outside the lock.
func (h *Holder) readyAsks(now time.Time) []func() {
var asks []func()
for k, m := range h.marks {
p, mark := m.pull, *m.pull.Ready
repository := strings.ToLower(p.Owner + "/" + p.Repo)
if !sameFull(mark.Head, p.HeadSHA) {
if !m.cleared.IsZero() && now.Sub(m.cleared) < readyTryEvery {
continue
}
m.cleared = now
asks = append(asks, h.clearMark(k, p, mark))
continue
}
d := h.deliveries[IDOf(repository, p.HeadSHA)]
if d == nil || d.State.Final() || d.MergedAs != "" {
continue // not announced yet, or past its merge
}
if d.Ready == nil || d.Ready.Head != mark.Head {
d.Ready = &ReadyRun{Head: mark.Head, By: mark.By, MarkedAt: mark.At, Seen: now}
h.Logf("[mesh-delivery] %s is marked ready at %s by %s: merged once its check passes against its trunk's head",
d.ID, shortOf(mark.Head, 12), orNone(mark.By))
h.keep(d)
}
r := d.Ready
if r.Merged != nil {
continue
}
if d.State == Rejected {
if r.Failed == "" {
r.Failed = "its check failed"
if d.Check != nil {
r.Failed += ": " + oneLine(d.Check.Summary)
if j := d.Check.Judged(); j != "" {
r.Failed += " (" + j + ")"
}
}
r.Waits = ""
h.Logf("[mesh-delivery] %s is marked ready and not merged: %s", d.ID, r.Failed)
h.keep(d)
asks = append(asks, h.sayOn(d, fmt.Sprintf("mesh-delivery: **not merged.** This pull request is marked "+
"ready at %s, and %s. A new head, marked ready once reviewed, or a recheck that passes, merges it.",
mark.Head, r.Failed)))
}
continue
}
r.Failed = ""
waits := h.readyWaits(d)
if waits != r.Waits {
r.Waits = waits
if waits != "" {
h.Logf("[mesh-delivery] %s is marked ready and waits: %s", d.ID, waits)
}
h.keep(d)
}
if waits != "" || (!r.LastTry.IsZero() && now.Sub(r.LastTry) < readyTryEvery) {
continue
}
r.Tries++
r.LastTry = now
h.keep(d)
asks = append(asks, h.askMerge(d.ID, p, mark.Head))
}
return asks
}
// readyWaits is what a marked delivery's merge waits on; empty when it is asked now. A ready delivery is asked; so is
// a proposed one whose verdict went out of date and whose check nobody asked again — the merge tool asks it.
func (h *Holder) readyWaits(d *Delivery) string {
switch {
case d.State == Ready:
case d.State == Proposed && d.outOfDate() != "" && d.AskedAgain == nil:
case d.State == Proposed && d.AskedAgain != nil:
return "its check against the trunk's head is asked and has not answered"
default:
return "its check has not passed yet (" + string(d.State) + ")"
}
if d.Group == "" {
return ""
}
return h.groupTurnToMerge(d)
}
// groupTurnToMerge is why a group member may not merge yet; empty at its turn: the group's composed check passed for
// its members' heads as they are and is not out of date, and every member before it in the group's order merged.
func (h *Holder) groupTurnToMerge(d *Delivery) string {
g := h.groups[d.Group]
if g == nil {
return ""
}
members := h.membersOf(g)
if state, why := GroupState(g, members); state != "ready" && state != "delivering" {
return fmt.Sprintf("its group %s is %s: %s", g.ID, state, why)
}
if c := g.Check; c == nil || (c.Verdict != "pass" && c.Verdict != "warning") || c.OutOfDate != "" ||
!sameMembers(c.Commits, commitsOf(members)) {
return "its group " + g.ID + "'s composed check has not passed for its members' heads"
}
if !containsString(g.Order, d.ID) {
return "its group " + g.ID + "'s order is not worked out for it yet"
}
for _, id := range g.Order {
if id == d.ID {
return ""
}
if o := h.deliveries[id]; o != nil && o.MergedAs == "" && !o.State.Final() {
return "its group " + g.ID + "'s order merges " + id + " first"
}
}
return ""
}
// askMerge asks the forge's merge tool to merge a marked head, and keeps its answer.
func (h *Holder) askMerge(id string, p OpenPull, head string) func() {
return func() {
err := h.Forge.Merge(p.Owner, p.Repo, p.Number, head)
h.mu.Lock()
defer h.mu.Unlock()
d := h.deliveries[id]
if d == nil || d.Ready == nil {
return
}
r, now := d.Ready, h.Now()
if err == nil {
r.Merged, r.Answer, r.RefusedSince = &now, "merged", nil
h.Logf("[mesh-delivery] %s merged on its ready mark at %s, its check passed against its trunk's head", id,
shortOf(head, 12))
h.keep(d)
return
}
answer := oneLine(err.Error())
if r.RefusedSince == nil {
r.RefusedSince = &now
}
if answer != r.Answer {
h.Logf("[mesh-delivery] %s is marked ready, and the merge tool did not merge it (asked again in %s): %s", id,
readyTryEvery, answer)
}
r.Answer = answer
h.keep(d)
}
}
// clearMark takes a mark off a pull request whose head moved after it, and says so on it.
func (h *Holder) clearMark(k string, p OpenPull, mark ReadyMark) func() {
return func() {
err := h.Forge.Unlabel(p.Owner, p.Repo, p.Number, mark.LabelID)
if err == nil {
err = h.Forge.Comment(p.Owner, p.Repo, p.Number, fmt.Sprintf("mesh-delivery: the ready mark at %s is cleared: "+
"the head is now %s, pushed after it. Mark the new head ready once it is reviewed.", mark.Head, p.HeadSHA))
}
if err != nil {
h.Logf("[mesh-delivery] %s's ready mark at %s names an older head than %s, and could not be cleared yet: %v",
k, shortOf(mark.Head, 12), shortOf(p.HeadSHA, 12), err)
return
}
h.Logf("[mesh-delivery] %s's ready mark at %s cleared: its head is now %s", k, shortOf(mark.Head, 12),
shortOf(p.HeadSHA, 12))
h.mu.Lock()
delete(h.marks, k)
h.mu.Unlock()
}
}
// sayOn comments on a delivery's pull request.
func (h *Holder) sayOn(d *Delivery, body string) func() {
owner, repo, number, id := d.Owner(), d.Repo(), d.Number, d.ID
return func() {
if err := h.Forge.Comment(owner, repo, number, body); err != nil {
h.Logf("[mesh-delivery] %s: could not say on its pull request: %v", id, err)
}
}
}
// readyLines are the marked heads that will not merge: a check that failed at once, a merge refused past its bound;
// as stalled lines, which the controller raises as conditions. Under the lock.
func (h *Holder) readyLines(now time.Time) []StalledLine {
var out []StalledLine
for _, d := range h.deliveries {
r := d.Ready
if r == nil || r.Merged != nil || d.MergedAs != "" || d.State.Final() {
continue
}
switch {
case r.Failed != "" && d.State == Rejected:
out = append(out, StalledLine{ID: d.ID, Number: d.Number, State: d.State, For: now.Sub(r.Seen).Round(time.Second).String(),
Bound: "0s", H2: "none: a new head marked ready, or a recheck that passes — the operator's",
Says: fmt.Sprintf("%s#%d is marked ready at %s and not merged: %s", d.Repository, d.Number,
shortOf(r.Head, 12), r.Failed)})
case r.RefusedSince != nil && now.Sub(*r.RefusedSince) >= readyRefusedFor:
out = append(out, StalledLine{ID: d.ID, Number: d.Number, State: d.State,
For: now.Sub(*r.RefusedSince).Round(time.Second).String(), Bound: readyRefusedFor.String(),
H2: "none: the merge is asked again each minute — the operator's",
Says: fmt.Sprintf("%s#%d is marked ready at %s, and the merge tool has refused it since %s: %s", d.Repository,
d.Number, shortOf(r.Head, 12), r.RefusedSince.UTC().Format(time.RFC3339), r.Answer)})
}
}
return out
}
@@ -0,0 +1,199 @@
package main
import (
"errors"
"strings"
"testing"
"time"
)
// **A pull request marked ready is merged by this owner** (novox/hq issue 722): before, a reviewed pull request with
// green checks waited until a session called the merge tool, and again after every check the tool asked because the
// trunk had moved. Each test below fails on a holder without the ready mark: nothing merged it.
const (
readyHead = "1111111111111111111111111111111111111111"
pushed = "2222222222222222222222222222222222222222"
)
// markedPull is an open pull request as gitea_open_pulls reads it, labelled mesh/ready at the head named.
func markedPull(repo string, number int, head, markedAt string) OpenPull {
owner, name, _ := strings.Cut(repo, "/")
return OpenPull{Owner: owner, Repo: name, Number: number, Base: "main", HeadSHA: head, Labels: []string{ReadyLabel},
Ready: &ReadyMark{Head: markedAt, By: "reviewer", At: "2026-10-14T19:00:00Z", LabelID: 31}}
}
func judgedAt(repo string, number int, sha, gate, trunk string) CheckedEvent {
c := verdict(repo, number, sha, gate)
c.Trunk = trunk
return c
}
func TestAReadyPullRequestMarkedAtItsHeadIsMergedThroughTheMergeTool(t *testing.T) {
w := newWorld(t)
id := IDOf("novox/app", readyHead)
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.h.Checked(judgedAt("novox/app", 7, readyHead, "pass", "t1t1t1t1t1t1"))
w.settleAll()
if len(w.forge.merges) != 0 {
t.Fatalf("an unmarked pull request was merged: %v", w.forge.merges)
}
w.forge.open = []OpenPull{markedPull("novox/app", 7, readyHead, readyHead)}
w.h.SeeOpenPulls()
w.settleAll()
if len(w.forge.merges) != 1 || w.forge.merges[0] != "novox/app#7@"+readyHead {
t.Fatalf("a ready, marked pull request was merged %v; wanted once, at the marked head", w.forge.merges)
}
d := w.delivery(id)
if d.Ready == nil || d.Ready.Merged == nil || d.Ready.By != "reviewer" || d.Ready.Tries != 1 {
t.Fatalf("what was done with the mark is not kept: %+v", d.Ready)
}
// Its merge's transition names who merged it.
w.h.PullMerged(merged("novox/app", 7, readyHead, "m7m7m7m7m7m7"))
w.h.WalkMoved(w.ctl.put(aWalk("plan-7", "novox/app", "m7m7m7m7m7m7", true)))
w.settleAll()
d = w.delivery(id)
var last Transition
for _, tr := range d.Transitions {
if tr.Event == EvMerged {
last = tr
}
}
if !strings.Contains(last.By, "ready mark at "+readyHead[:12]+" by reviewer") {
t.Fatalf("the merge's transition does not name the mark: %+v", last)
}
if len(w.forge.merges) != 1 {
t.Fatalf("merged again: %v", w.forge.merges)
}
}
func TestAPushAfterTheMarkClearsItAndNothingIsMerged(t *testing.T) {
w := newWorld(t)
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.h.PullUpdated(pr("novox/app", 7, pushed, "feat/x", "src/a.go"))
w.h.Checked(judgedAt("novox/app", 7, pushed, "pass", "t1t1t1t1t1t1"))
w.forge.open = []OpenPull{markedPull("novox/app", 7, pushed, readyHead)}
w.h.SeeOpenPulls()
w.settleAll()
if len(w.forge.merges) != 0 {
t.Fatalf("a head pushed after the mark was merged: %v", w.forge.merges)
}
if len(w.forge.unlabels) != 1 || w.forge.unlabels[0] != "novox/app#7-31" {
t.Fatalf("the stale mark's label was not taken off: %v", w.forge.unlabels)
}
if len(w.forge.comments) != 1 || !strings.Contains(w.forge.comments[0], "cleared") ||
!strings.Contains(w.forge.comments[0], pushed) {
t.Fatalf("the cleared mark is not said on the pull request: %v", w.forge.comments)
}
}
func TestAMovedTrunkIsJudgedAgainByTheMergeToolAndMergedOnItsVerdict(t *testing.T) {
w := newWorld(t)
id := IDOf("novox/app", readyHead)
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.h.Checked(judgedAt("novox/app", 7, readyHead, "pass", "t1t1t1t1t1t1"))
// Another merge moves the trunk: its verdict is out of date, and it goes back to proposed.
w.h.PullMerged(merged("novox/app", 9, "9999999999999999", "t2t2t2t2t2t2"))
if w.state(id) != Proposed {
t.Fatalf("out of date, it is %s", w.state(id))
}
// The merge tool asks the check against the trunk's head and refuses: nothing merged.
w.forge.refuse = errors.New("the forge refused gitea_merge_pull_request: The check of head merged into trunk was asked")
w.forge.open = []OpenPull{markedPull("novox/app", 7, readyHead, readyHead)}
w.h.SeeOpenPulls()
w.settleAll()
if len(w.forge.merges) != 1 {
t.Fatalf("the merge tool was not asked once to judge it again: %v", w.forge.merges)
}
// The tool's ask is the head announced again; it waits for that verdict, asking nothing.
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.later(5 * time.Minute)
w.settleAll()
if len(w.forge.merges) != 1 || !strings.Contains(w.delivery(id).Ready.Waits, "asked") {
t.Fatalf("asked again while its check was on its way: %v (%+v)", w.forge.merges, w.delivery(id).Ready)
}
// It passes against the trunk's head: merged, without anyone asking.
w.forge.refuse = nil
w.h.Checked(judgedAt("novox/app", 7, readyHead, "pass", "t2t2t2t2t2t2"))
w.settleAll()
if w.state(id) != Ready || len(w.forge.merges) != 2 || w.delivery(id).Ready.Merged == nil {
t.Fatalf("on its passing verdict it was not merged: %s %v %+v", w.state(id), w.forge.merges, w.delivery(id).Ready)
}
}
func TestAMarkedHeadWhoseCheckFailedIsNotMergedAndIsSaid(t *testing.T) {
w := newWorld(t)
id := IDOf("novox/app", readyHead)
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.h.Checked(judgedAt("novox/app", 7, readyHead, "fail", "t1t1t1t1t1t1"))
w.forge.open = []OpenPull{markedPull("novox/app", 7, readyHead, readyHead)}
w.h.SeeOpenPulls()
w.settleAll()
w.later(10 * time.Minute)
w.h.SeeOpenPulls()
w.settleAll()
if len(w.forge.merges) != 0 {
t.Fatalf("a head whose check failed was merged: %v", w.forge.merges)
}
if len(w.forge.comments) != 1 || !strings.Contains(w.forge.comments[0], "not merged") {
t.Fatalf("the failure is not said once on the pull request: %v", w.forge.comments)
}
lines := w.h.Stalled()
if len(lines) != 1 || lines[0].ID != id || lines[0].State != Rejected || !strings.Contains(lines[0].Says, "marked ready") {
t.Fatalf("the failure is not a stalled line, which the controller raises as a condition: %+v", lines)
}
}
func TestAMarkRefusedForAnHourIsSaid(t *testing.T) {
w := newWorld(t)
w.h.PullUpdated(pr("novox/app", 7, readyHead, "feat/x", "src/a.go"))
w.h.Checked(judgedAt("novox/app", 7, readyHead, "pass", "t1t1t1t1t1t1"))
w.forge.refuse = errors.New("the forge refused gitea_merge_pull_request: mesh/repo-check is pending")
w.forge.open = []OpenPull{markedPull("novox/app", 7, readyHead, readyHead)}
w.h.SeeOpenPulls()
w.settleAll()
w.later(30 * time.Second)
w.settleAll()
if len(w.forge.merges) != 1 {
t.Fatalf("asked again within the minute: %v", w.forge.merges)
}
w.later(readyRefusedFor)
w.settleAll()
if len(w.forge.merges) != 2 {
t.Fatalf("not asked again after a minute: %v", w.forge.merges)
}
if lines := w.h.Stalled(); len(lines) != 1 || !strings.Contains(lines[0].Says, "refused it since") {
t.Fatalf("a mark refused for an hour is not said: %+v", lines)
}
}
func TestAGroupMergesItsMarkedMembersInOrderOnlyOnceComposed(t *testing.T) {
w := newWorld(t)
const ctlHead, catHead = "c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1c1", "d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2d2"
ctl, cat := IDOf("novox/mesh-controller", ctlHead), IDOf("novox/mesh-catalog", catHead)
w.ctl.order = func(ms []Member) Order { return Order{Order: []string{ctl, cat}} }
w.h.PullUpdated(pr("novox/mesh-controller", 1, ctlHead, "feat/x", "internal/a.go"))
w.h.PullUpdated(pr("novox/mesh-catalog", 2, catHead, "feat/x", "modules/app/module.json"))
w.h.Checked(judgedAt("novox/mesh-controller", 1, ctlHead, "pass", "t1t1t1t1t1t1"))
w.h.Checked(judgedAt("novox/mesh-catalog", 2, catHead, "pass", "t3t3t3t3t3t3"))
w.forge.open = []OpenPull{markedPull("novox/mesh-controller", 1, ctlHead, ctlHead),
markedPull("novox/mesh-catalog", 2, catHead, catHead)}
w.h.SeeOpenPulls()
w.settleAll()
if len(w.forge.merges) != 0 {
t.Fatalf("a member merged before its group's composed check passed: %v", w.forge.merges)
}
w.h.Checked(CheckedEvent{Group: "feat/x", ID: "check-1", Verdict: "pass", Summary: "both compose"})
w.settleAll()
if len(w.forge.merges) != 1 || w.forge.merges[0] != "novox/mesh-controller#1@"+ctlHead {
t.Fatalf("composed, the first member in the group's order was not merged alone: %v", w.forge.merges)
}
if waits := w.delivery(cat).Ready.Waits; !strings.Contains(waits, "merges "+ctl+" first") {
t.Fatalf("the second member does not say it waits its turn: %q", waits)
}
w.h.PullMerged(merged("novox/mesh-controller", 1, ctlHead, "m1m1m1m1m1m1"))
w.settleAll()
if len(w.forge.merges) != 2 || w.forge.merges[1] != "novox/mesh-catalog#2@"+catHead {
t.Fatalf("after the first member merged, the next was not: %v", w.forge.merges)
}
}
@@ -498,7 +498,12 @@ func Apply(d *Delivery, ev Event, act bool, f Facts) (Transition, error) {
if why == "" {
why = r.Guard
}
t := Transition{At: f.Now, From: d.State, To: r.To, Event: ev, Why: why, By: f.By}
by := f.By
if by == "" && (ev == EvMerged || ev == EvDone) && d.MergedAs != "" {
// A merge this owner made on a ready mark names it as its author (novox/hq issue 722).
by = d.Ready.mergedBy()
}
t := Transition{At: f.Now, From: d.State, To: r.To, Event: ev, Why: why, By: by}
d.State, d.Since = r.To, f.Now
d.Transitions = append(d.Transitions, t)
if len(d.Transitions) > keptTransitions {
@@ -72,6 +72,9 @@ type OpenPull struct {
// than that says none: the head is then read by Statuses alone, as before, and a pending check is not told
// from an answered one.
Checks map[string]HeadCheck `json:"checks,omitempty"`
// Labels are its labels by name, and Ready its ready mark, when it carries one (novox/hq issue 722).
Labels []string `json:"labels,omitempty"`
Ready *ReadyMark `json:"ready,omitempty"`
}
// HeadCheck is one status on a head as the forge keeps it: pending, success, failure, error or warning.
@@ -201,6 +204,7 @@ func (h *Holder) SeeOpenPulls() {
seen[id] = u
}
h.unchecked = seen
h.seeMarks(open)
}
// started is whether the head's check started, rather than was never asked: it has any merge check, or a delivery
@@ -249,6 +249,7 @@ func (h *Holder) Stalled() []StalledLine {
}
out = append(out, h.withChecks(h.uncheckedLines(now))...)
out = append(out, h.orphanLines(now)...)
out = append(out, h.readyLines(now)...)
sort.Slice(out, func(i, j int) bool { return out[i].ID < out[j].ID })
return out
}
+4 -1
View File
@@ -47,7 +47,10 @@
"gitea.gitea_commit_status",
"gitea.gitea_commit_statuses",
"gitea.gitea_contains",
"gitea.gitea_open_pulls"
"gitea.gitea_open_pulls",
"gitea.gitea_merge_pull_request",
"gitea.gitea_add_comment",
"gitea.gitea_api"
],
"state": [
"deliveries",
@@ -121,6 +121,12 @@ func TestItsProcessesAreThePreparationTheActAndTheJudge(t *testing.T) {
!reflect.DeepEqual(m.Tools, []string{HealthTool, HighestTool}) {
t.Fatalf("watch %v, tools %v", watch, m.Tools)
}
// The watch's verdict is whether the database serves, so its check names the provision it exercises: while the
// database's provider is unhealthy, this finding is held under the provider's condition (hq ADR 0240 rule 5,
// issue 145).
if h["needs"] != "postgres-database" {
t.Fatalf("watch's check needs %v; it exercises postgres-database", h["needs"])
}
if watch["env"].(map[string]any)["MESH_KNOWLEDGE_STATUS"] != m.Build.Artifacts[0].Env["MESH_KNOWLEDGE_STATUS"] {
t.Fatal("watch writes another status file than the health tool reads")
}
+1
View File
@@ -156,6 +156,7 @@
"health": {
"kind": "tool",
"tool": "knowledge_health",
"needs": "postgres-database",
"interval": "60s",
"timeout": "10s",
"looks": 2,
+6 -4
View File
@@ -49,8 +49,9 @@ channel's client and no channel's secret.
warrant published to the asker alone. A warrant the bus did not take is published again every minute.
- **What the operator sees when they answer** (issue 372): first, in reply to the answer where it was given,
that it was recorded and whether its asker was told ("You chose Approve: recorded, and claude-code was
told.") — said only once the ask's end is recorded — or, refused, why; then the ask's message edited to how
it ended, as its title ("You chose Approve at 13:30 on Telegram · <headline>"), with no answers left on it.
told.") — said only once the ask's end is recorded — or, refused, why; then the ask's message edited with no
answers left on it: its question whole (headline, summary, points, link, Details), and below it how it ended
("You chose Approve at 13:30 on Telegram.") and who acts on it (issue 714).
What the asker then did is not said: the asks protocol carries no result back to the router, so the
message says only that the asker was told and acts on it itself.
- **An ask never defaults**: at its expiry it ends `expired` and its asker does what it said it would.
@@ -138,8 +139,9 @@ and said in the log). Everything else is the message's note (`asks.Message.Note`
smaller: who asked and when, with what happens if nobody answers; what each answer does; where else it is answered;
and the asker's Details (who proposed it, a fingerprint, how to read it whole) — on a kind that proves who answers
and was given the ask's answers, its Details whole (`asks.Ask.WholeDetails`): every value an approval sets, held to
the secret rule alone, as the whole words are. Answered, the message is one line —
"You chose Approve at 16:06 on Telegram · <headline>" — and the router's word on the asker is its note.
the secret rule alone, as the whole words are. Answered, the message keeps its headline, body and Details, so the
operator can read back what they answered (issue 714); its note becomes "You chose Approve at 16:06 on Telegram."
with the router's word on the asker.
**An ask's values are shown whole where the sender is proven** (hq issue 383). An ask may carry its explanation
twice: as it is held to the whole content rule, and whole (`asks.Ask.Whole`), its paths and addresses as they
@@ -365,10 +365,12 @@ func TestAnApproveFromTheLinkedAccountIsAWarrantForTheAskerAlone(t *testing.T) {
for _, m := range ch.edits {
edited = m
}
// The question's title is how it ended, read at a glance; then what the router knows of the asker (issue 372).
if len(edited.Actions) != 0 || edited.Title != "You chose Release at 12:00 on Telegram · Delivery of hq held for 36 hours" ||
edited.Body != "" || edited.Note != "Told to mesh-delivery, which acts on it itself." {
t.Errorf("%s still offers %q, or does not say how it ended in one line (issue 639): %q / %q / %q", ch.name, labelsOf(edited), edited.Title, edited.Body, edited.Note)
// The question stays whole (issue 714); below it, how it ended and what the router knows of the asker
// (issue 372).
if len(edited.Actions) != 0 || edited.Title != "Delivery of hq held for 36 hours" ||
edited.Body != heldAsk(c, "d-1").Explanation ||
edited.Note != "You chose Release at 12:00 on Telegram.\nTold to mesh-delivery, which acts on it itself." {
t.Errorf("%s still offers %q, or does not keep the question and say how it ended (issue 714): %q / %q / %q", ch.name, labelsOf(edited), edited.Title, edited.Body, edited.Note)
}
}
// The tap is confirmed where it was made, in reply to it, once the warrant stood — and before the question
@@ -624,8 +626,8 @@ func TestAnAskExpiresAndNeverDefaults(t *testing.T) {
for _, m := range tg.edits {
edited = m
}
if len(edited.Actions) != 0 || !strings.HasPrefix(edited.Title, "No longer answerable: nobody answered in time · Delivery of hq") ||
edited.Body != "" || edited.Note != "The delivery stays held." {
if len(edited.Actions) != 0 || !strings.HasPrefix(edited.Title, "Delivery of hq") ||
edited.Body == "" || edited.Note != "No longer answerable: nobody answered in time.\nThe delivery stays held." {
t.Errorf("the copy still offers answers, or does not say how it ended: %+v", edited)
}
}
@@ -51,7 +51,7 @@ func TestAnApprovalIsConfirmedWhereItWasGivenAndThenTheQuestionSaysHowItEnded(t
}
want := []string{
"send You chose Release: recorded, and mesh-delivery was told.",
"edit You chose Release at 12:00 on Telegram · Delivery of hq held for 36 hours",
"edit Delivery of hq held for 36 hours",
}
if strings.Join(log, "\n") != strings.Join(want, "\n") {
t.Fatalf("the operator saw, in order:\n%s\nwant:\n%s", strings.Join(log, "\n"), strings.Join(want, "\n"))
@@ -122,7 +122,7 @@ func TestTheConfirmationNeverClaimsWhatDidNotHappen(t *testing.T) {
for _, m := range tg.edits {
edited = m
}
if !strings.HasPrefix(edited.Note, "Told to mesh-delivery as soon as the bus takes it") {
if !strings.Contains(edited.Note, "\nTold to mesh-delivery as soon as the bus takes it") {
t.Errorf("the question claims its asker was told: %q", edited.Note)
}
})
@@ -9,8 +9,8 @@ import (
)
// novox/hq issue 639: an ask's message reads at a glance on the phone. Its body is the asker's summary and points
// alone; who asked and when, what each answer does, the expiry and the asker's Details are its note; once answered
// it is one line with the headline.
// alone; who asked and when, what each answer does, the expiry and the asker's Details are its note. Once answered
// it keeps the question whole, so the operator can read back what they answered (issue 714).
// decisionAsk is the ask for ADR 0343 as the knowledge base composes it now (its glance form).
func decisionAsk(c *clock) asks.Ask {
@@ -59,7 +59,8 @@ func TestAnAsksMessageIsItsSummaryWithTheRestInItsNote(t *testing.T) {
if n := len(strings.Split(m.Text(), "\n")); n > 9 || len(m.Actions) != 2 {
t.Errorf("the message is %d lines offering %s:\n%s", n, labelsOf(m), m.Text())
}
// Answered, it is one line with the headline; the body is not said again.
// Answered, it keeps its headline, summary, points and Details; below them, how it ended and who acts on it;
// its buttons are gone (issue 714: the one line issue 639 made left nothing to read back).
if out, err := h.Chosen("telegram", choiceOf(ticketFor(m, "Approve"), "42", c.now())); err != nil || out != "chosen" {
t.Fatalf("Approve: %q %v", out, err)
}
@@ -67,8 +68,9 @@ func TestAnAsksMessageIsItsSummaryWithTheRestInItsNote(t *testing.T) {
for _, e := range tg.edits {
edited = e
}
if edited.Title != "You chose Approve at "+asked+" on Telegram · "+a.Headline || edited.Body != "" ||
strings.Contains(edited.Note, "What an agent learns") || len(edited.Actions) != 0 {
if edited.Title != a.Headline || edited.Body != a.Explanation+"\n"+want[2]+"\n"+want[3] ||
edited.Note != "You chose Approve at "+asked+" on Telegram.\nTold to mesh-knowledge, which acts on it itself." ||
len(edited.Actions) != 0 {
t.Errorf("the answered copy: %q\n%q\n%q", edited.Title, edited.Body, edited.Note)
}
}
+43 -10
View File
@@ -856,7 +856,7 @@ func (h *Holder) sayOn(r *Record, what, kind string) Message {
}
}
if r.Answered != "" {
m = h.answeredWords(r, m)
m = h.answeredWords(r, m, kind)
}
if kind == "" {
return m
@@ -910,18 +910,24 @@ func (h *Holder) sayOn(r *Record, what, kind string) Message {
return m
}
// answeredWords is a message that says what came of its question. An ask's own message is the question
// itself, so it becomes how it ended and what it was about — "You chose Approve at 13:30 on Telegram ·
// <headline>" — one line, read at a glance on the phone, where the edit rings nothing (issue 372); the body
// is not said again (issue 639), and what the router knows of the asker after it is the note. A condition's
// message keeps its own title, and says it first.
func (h *Holder) answeredWords(r *Record, m Message) Message {
// answeredWords is a message that says what came of its question. An ask's own message keeps the question
// whole — its headline as the title, the asker's summary, points and link as the body, its Details after them —
// so the operator can read back what they answered (issue 714, which undid issue 639's one line). How it ended
// ("You chose Approve at 13:30 on Telegram.") and what the router knows of the asker after it are the note,
// below the question, and its answers are gone (issue 372). A channel that must shorten it cuts the body from
// its end, the Details first; never the headline or the note. A condition's message keeps its own title, and
// says how it ended first.
func (h *Holder) answeredWords(r *Record, m Message, kind string) Message {
if r.Origin == "ask" {
first, rest, _ := strings.Cut(r.Answered, "\n")
headline, _ := wordsOf(r)
out := Message{Title: strings.TrimSuffix(first, ".") + " · " + headline, Note: strings.TrimSpace(rest),
body := []string{}
if b := strings.TrimSpace(m.Body); b != "" {
body = append(body, b)
}
body = append(body, h.answeredDetails(r, kind)...)
out := Message{Title: headline, Body: strings.Join(body, "\n"), Note: strings.TrimSpace(r.Answered),
Urgent: m.Urgent, Quiet: m.Quiet}
if _, ok := Check(out.Text()); ok {
if _, ok := Check(out.Title + "\n" + out.Note); ok {
return out
}
// Its headline refused (it is the asker's): the outcome is still said, before what may leave.
@@ -930,6 +936,33 @@ func (h *Holder) answeredWords(r *Record, m Message) Message {
return m
}
// answeredDetails are an answered ask's Details, line by line, as its open message noted them (issue 714): whole
// on a kind that proves who answers, a line that may not leave withheld and said so.
func (h *Holder) answeredDetails(r *Record, kind string) []string {
if r.Ask == "" {
return nil
}
e := h.askOf(r.Ask)
if e == nil {
return nil
}
details, passes := e.rec.Ask.Details, Check
if wd := h.wholeDetailsOf(r, kind); wd != "" {
details, passes = wd, CheckSecret
}
var lines []string
for _, line := range strings.Split(strings.TrimSpace(details), "\n") {
if line = strings.TrimSpace(line); line == "" {
continue
}
if refusal, ok := passes(line); !ok {
line = withheldLine(line, refusal, passes)
}
lines = append(lines, line)
}
return lines
}
// wholeWords is a record's message with the ask's whole words in place of its explanation, when they carry no
// secret's shape once composed; else false, and the message under the whole content rule stands.
func (h *Holder) wholeWords(r *Record, what, whole string) (Message, bool) {
+1 -1
View File
@@ -91,7 +91,7 @@ func (h *Holder) Preview(key, what string) (map[string]any, error) {
asking := h.openAsk(&r) != nil // askOf takes the same lock, so this is asked outside it
m, line, refusal, ownWords := h.allowedWords(&r, what, now)
if r.Answered != "" {
m = h.answeredWords(&r, m)
m = h.answeredWords(&r, m, "")
}
out := map[string]any{
"key": key,
@@ -50,7 +50,7 @@ func TestAReplacedAsksMessageEndsWithIt(t *testing.T) {
m, edited = r.m, true
}
}
if !edited || len(m.Actions) != 0 || !strings.Contains(m.Title, "Replaced by a newer question") {
if !edited || len(m.Actions) != 0 || !strings.HasPrefix(m.Note, "Replaced by a newer question") {
t.Errorf("%s: the replaced ask's message still stands: edited %v, %q %v", ch.name, edited, m.Text(), labelsOf(m))
}
}
@@ -86,7 +86,7 @@ func TestAnAsksMessageLeftOpenAfterItsAskEndedIsEndedOnTheNextTick(t *testing.T)
if got := openAskRecords(h); len(got) != 0 {
t.Fatalf("the message of an ended ask is still open: %v", got)
}
if m, ok := tg.edits["1"]; !ok || len(m.Actions) != 0 || !strings.Contains(m.Title, "Replaced by a newer question") {
if m, ok := tg.edits["1"]; !ok || len(m.Actions) != 0 || !strings.HasPrefix(m.Note, "Replaced by a newer question") {
t.Errorf("the leftover's copy was not ended: %v %q %v", ok, m.Text(), labelsOf(m))
}
}
@@ -61,8 +61,7 @@ func TestAnAsksWholeWordsAreShownWhereTheSenderIsProvenAndMaskedElsewhere(t *tes
if got := w.of("mesh-controller"); len(got) != 0 || store.asks["mesh-controller.s-1"].State != StateOpen {
t.Errorf("the ask was answered: %+v / %s", got, store.asks["mesh-controller.s-1"].State)
}
// Approved, the message is edited to one line that says so, with the headline; the body is not said again
// (issue 639).
// Approved, the message keeps its headline and words, and says below them how it ended (issue 714).
if out, err := h.Chosen("telegram", choiceOf(ticketFor(phone, "Release"), "42", c.now())); err != nil || out != "chosen" {
t.Fatalf("Release: %q %v", out, err)
}
@@ -70,8 +69,8 @@ func TestAnAsksWholeWordsAreShownWhereTheSenderIsProvenAndMaskedElsewhere(t *tes
if !ok {
t.Fatalf("the phone's message was not edited: %+v", tg.edits)
}
if !strings.HasPrefix(edited.Title, "You chose Release at ") || !strings.HasSuffix(edited.Title, " · Set mounts on shanks?") ||
edited.Body != "" || len(edited.Actions) != 0 {
if edited.Title != "Set mounts on shanks?" || edited.Body == "" ||
!strings.HasPrefix(edited.Note, "You chose Release at ") || len(edited.Actions) != 0 {
t.Errorf("the edited message: %q\n%s\nactions %+v", edited.Title, edited.Body, edited.Actions)
}
}
+7
View File
@@ -97,6 +97,13 @@
{
"id": "server",
"type": "container",
"health": {
"kind": "http",
"endpoint": "web",
"path": "/healthz/readiness",
"status": 200,
"needs": "postgres-database"
},
"name": "n8n",
"artifact": "server",
"network": "n8n",
+8
View File
@@ -83,6 +83,14 @@
{
"id": "server",
"type": "container",
"health": {
"kind": "http",
"endpoint": "web",
"path": "/status.php",
"status": 200,
"body": "\"installed\":true",
"needs": "postgres-database"
},
"name": "nextcloud",
"image": "nextcloud@sha256:fb966733647ea03f0446b0c22eac9733c8eb616d37b960caca9d4c3010e14a08",
"env-file": [
+1 -1
View File
@@ -20,7 +20,7 @@ WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
COPY . .
RUN CGO_ENABLED=0 GOOS=linux go build -trimpath -o /mesh-route-proxy ./examples/route-proxy
RUN CGO_ENABLED=0 GOOS=linux go build -trimpath -buildvcs=false -ldflags=-buildid= -o /mesh-route-proxy ./examples/route-proxy
# A small runtime with the public CA roots the ACME client needs to reach a real authority, and run
# as root so it can bind :80 and :443 — the two privileged ports a public front door listens on.
+3 -2
View File
@@ -4,10 +4,11 @@ package main
// formatting", "too much text, no formatting, no summary"). The router's words are shown as given —
// `exact-render` — and formatted, never reworded:
//
// - the title in bold: the headline, or how an answered ask ended;
// - the title in bold: the headline, an answered ask's too (issue 714);
// - the body as it is, a line that is a key with its value (as a settings proposal lists them, "+ key: value")
// in monospace;
// - the note — who asked, what each answer does, the expiry, the asker's details — apart, after a blank line,
// - the note — who asked, what each answer does, the expiry, the asker's details; once answered, how it ended
// and who acts on it — apart, after a blank line,
// in a blockquote Telegram shows collapsed to its first lines, expanded on a tap.
//
// Every character Telegram's HTML would read as markup is escaped first, so a message carrying '<' neither
@@ -84,3 +84,20 @@ func TestTheNoteIsEscaped(t *testing.T) {
t.Errorf("%q", got)
}
}
// novox/hq issue 714: an answered ask keeps its question whole, with how it ended below it. Over Telegram's bound,
// the end of the body (the asker's Details) is cut; the headline and the answer are never.
func TestAnAnsweredAskOverTheBoundKeepsItsHeadlineAndItsAnswer(t *testing.T) {
answer := "You chose Approve at 20:03 on Telegram.\nTold to mesh-knowledge, which acts on it itself."
m := asks.Message{Handle: "h-714", To: "42", Title: "Fixes reach every machine at once, unchecked",
Body: "A fix goes to every machine in one step.\n• No machine goes first\n" + strings.Repeat("Details line. ", 400),
Note: answer}
s := partsOf(m)
if n := len([]rune(s.plain())); n > MaxLength {
t.Fatalf("the answered ask is %d characters, over %d", n, MaxLength)
}
if s.title != m.Title || s.note != answer || !strings.HasPrefix(s.body, "A fix goes to every machine in one step.\n• No machine goes first") ||
!strings.HasSuffix(s.body, cutNote) {
t.Errorf("the headline, the summary or the answer was cut:\n%s", s.plain())
}
}
+9
View File
@@ -89,6 +89,15 @@
{
"id": "server",
"type": "container",
"health": {
"kind": "http",
"endpoint": "web",
"path": "/api/share/mesh-health",
"status": 404,
"body": "\"code\":\"not-found\"",
"timeout": "10s",
"needs": "postgres-database"
},
"name": "umami",
"image": "ghcr.io/umami-software/umami@sha256:85909afc45bdcda1917394594a087421fdbb05610fded0fa9f6fb861abb2f367",
"network": "umami",