Commit Graph
16 Commits
Author SHA1 Message Date
jschoubben 8eafc0c5ce The placement reader names the seat mesh-placement's state, held by the assigner, as L2 moved it (issue 774, ADR 0567)
mesh/delivery-group group feat/774-vault-mints-l2 checking: the checker's verdict is out of date: novox/mesh-delivery@f6df07605dd0 was judged on trunk 5b2e5bc5…
mesh/delivery merged: source.main moved to novox/mesh-sdk@d747c1128e42
mesh/check checking 8eafc0c5ce3c on d747c1128e42
2026-10-28 21:54:04 +01:00
jschoubben 18a47ac3c7 Placed: a module in the first revision heard of a node's key stands since before it, so the seed's placement does not leave out what already ran (issue 774) 2026-10-28 21:54:04 +01:00
jschoubben 1386219d4d One reader of the catalogue's placement and the toolchain gate over it, in place of the three copies over the controller's assignments (issue 774, ADR 0549 §3, issue 1940); the assigner may read the lease (ADR 0549 §5) 2026-10-28 21:54:04 +01:00
jschoubben f91340fe82 buildrun: open a steps watch again that closed having heard nothing, up to 3 closes in a row
mesh/check checking f91340fe82ef on 6b43afa2ab38 | novox/mesh-sdk#121 in group fix/lost-check
mesh/delivery merged: source.main moved to novox/mesh-sdk@af08be36a27a
mesh/delivery-group group fix/lost-check merged: every member is merged
A watch that closed at once was left for good, so an ask whose own watch
went deaf was read only at a reconnect (review of fix/lost-check).
2026-10-28 20:35:02 +01:00
jschoubben 6aeb401095 buildrun: read an ask's step at every step end of its build module, so a deaf watch loses no end
mesh-catalog #553's verdict stayed running after its step passed on shanks
at 19:09:31Z on 2026-10-28: the watch of that one step key, opened at
18:46:05Z, heard neither the start nor the end and never closed. HeardEnds
stands between Asker and each ask's watch: nudged at each step end of the
same build module (HearEnds: one watch per steps state) and at a reconnect,
every waited ask reads its keys now; a watch that closes is opened again, at
most 3 times. The checker and the builder both wait through it.

Tests: a section of the asker's test (deaf, nudged by another module and by
its own, a watch that closes); counts unchanged.
2026-10-28 20:29:30 +01:00
jschoubben fed64292f0 Read build capacity as an integer and build hours as one window, default 0: no repository names a node (ADR 0567 §12, issue 1126) 2026-10-28 18:13:18 +01:00
jschoubben 02f50bd9d6 A build seat's capacity follows its node's build CPU, and a running ask says its phases
Since the build CPU cap (issue 1806) shanks' mesh-build-go kept capacity 3 inside a 12-core slice:
three full checks got about 4 cores each and ran past the 30-minute step bound (issue 1890).
The puller now judges by min(configured, floor(slice quota cores / cores-per-step)), the quota
read live from its cgroup, says it on change and writes capacity.<node> to the pauses state.
A run's phases (fetch and merge, digest, tests) are logged with the ask id and written to its
running step key; Asker.Wait hands every step value to a WithStepSeen hook.
2026-10-26 19:52:54 +01:00
jschoubben d515c71e1c process: a bucket missing at a part's start is bound when the bus says it was created, keeping what was written meanwhile; buscompose grants each principal the creation of the buckets it may bind (issue 1790)
Since the grants part creates a manifest's buckets after a delivery (ADR 0547), a module restarted by that
delivery can open a new state before it exists; OptionalBucket answered nil and the part never looked again
(mesh-delivery's checker and builder, state hearing, 2026-10-26). AwaitBucket binds on the server's
STREAM.CREATED advisory, or at the next use, and writes the latest value kept per key. The copied composer's
agreement test leaves that one row out: the controller's list is no longer loaded since the take-over.
2026-10-26 15:11:00 +01:00
jschoubben 3203358c4c buildrun: an asker keeps whether it hears DEAD_LETTERS, and its health tool answers from it (issue 734, review round 1)
mesh/delivery merged: source.main moved to novox/mesh-sdk@545a0c9cb43b
mesh/delivery-group group fix/734-lost-ask merged: every member is merged
mesh/check being carried to 3203358c4cb5 on 2aae6f5580f2 | from mesh-delivery c8b26735c8cf | novox/mesh-sdk#92 in group fix/734-lost-ask
A deaf asker must be loud: SayHearing keeps each Hearing in the asker's state hearing, and HearingHealth answers the
module's health tool from it, unhealthy with why while DEAD_LETTERS is not heard.
2026-10-26 14:18:41 +01:00
jschoubben 9b01c94a9e buildrun: the asker hears DEAD_LETTERS again on its next reconnect or ask, in the background, and says whether it hears; buscompose grants only what a pulled ordered consumer needs (issue 734, review round 1)
An asker started before its grant was composed was refused once and stayed deaf until it started again, silently.
DeadLetterHearing tries in the background and again on an event (Again: a reconnect, an ask waited for while deaf), and
tells each change of Hearing so the module's health tool can say it. DeadLetterReads drops the flow control a pull
consumer does not use and deletes one consumer by name.
2026-10-26 14:16:58 +01:00
jschoubben d96b9fd660 buildrun: an ask the build seat gave up on ends its asker's wait as failed with why, for every asker of the build seats (issue 734)
The checker's try and the builder's lanes waited for ever on an ask given up during a bus reload: no build node writes a
step for it. LostAskWaiter hands the asker a failed step on the node bus when DEAD_LETTERS keeps the ask's copy
(HearDeadLetters, granted by DeadLetterReads), and under NotTakenBound when no node took the ask within the bound.
2026-10-26 14:00:46 +01:00
jschoubben fc4f57c4b3 Restart an ask's bound at each run's start, also on the same node, so a step redelivered after its build module was replaced is not cut by the killed run's time (issue 1887, review: issue 734 path; ADR 0525 §6) 2026-10-23 16:36:46 +02:00
jschoubben 88fa74ce5e Bound an ask's step from its start on the node that took it, not from its put, so an ask waiting in its stream during a large rebuild is never failed and asked twice (issue 1887, review round 2 item 1; ADR 0525 §3, §6) 2026-10-23 16:32:21 +02:00
jschoubben 6003da0a58 Put an ask on its build seat's stream and read its step, and name a building key's asks, so the builder and the checker ask no node and a restarted builder never asks twice (issue 1887 A4a, A4b) 2026-10-23 15:43:59 +02:00
jschoubben 0730ef6092 Keep the unit level cheap: the merge's real-git tests start processes, so they are integration tests; the places tests wait 10 ms, not 50 (issue 895) 2026-10-21 19:26:02 +02:00
jschoubben 4375ed8b2c Move the builder's shared build-run parts into go/delivery/buildrun, so the checker shares them and no module keeps a copy (issue 1671) 2026-10-21 19:18:11 +02:00