mesh-build-go opens a network failure of the fetch with the retake prefix (issue 1748) #614

2026-11-04T21:46:45Z

mesh-tools commented

2026-11-04 21:46:45 +00:00

Issue 1748: a build that fails on a network lookup is built again by itself.

Until now the builder never built such a failure again: "may be retaken" was only wording in mesh-build-go's error. Tonight's lookup pauses on shanks left mesh-build-image and mesh-build-node failed until someone built them again by hand. The registry group's delivery will start about 300 builds and would meet the same pauses.

Delivery group fix/1748-retake-and-resolver, merged together:

mesh-sdk (go/delivery/builder.go): Retake, Retakeable() and MaxRetakes (3) beside FailedUntilChanged; a Retakes count on the build key and on the build's log entry (read by builder_builds_since).
mesh-delivery (mesh-delivery-builder): after a key is written failed with a retakeable error, armRetake takes it again after 60 s, then 120 s, then 240 s. Only that claim counts toward the cap, and Sweep re-arms from the key's count after a restart. The journal says each retake with its count. A compile error or a FailedUntilChanged error is never retaken, and the retake prefix is added only when every failing lane is retakeable.
mesh-catalog (mesh-build-go): a network failure of the fetch opens with the retake prefix. An unknown revision, a compile error or a failing test does not.

Acceptance criteria: issue 1748 note 14. AC5 (dnsmasq) was withdrawn in review (note 18): dnsmasq logs when it reaches its cap, and no such line was seen, so the cap was not the cause.

Rules bent, said here:

Playbook 08's one-repository bound: the mesh-sdk part is one exported name and one field.
ADR 0437 §8 (no timers in the builder): the backoff waits on a context's bound. The guard in manifest_test.go names armRetake as the one allowed bounded wait (issue 1748) and refuses any other. A bus delivery delay is a follow-up (issue 1748 note 16).
Test budget: mesh-sdk adds no test function. mesh-delivery adds one top-level unit function with five subtests, and its integration count stays at 71 (AC6 is a subtest of the existing failed-key test).

Integration test run locally by name against an embedded bus: TestAFailedKeyIsBuiltAgainAsANewRevisionFromANewModulesValueWithTheSameFingerprint (with the AC6 subtest): pass.

Review: review-1748@shanks approved round 2 at mesh-sdk 0d97c09c49b4a583ce51d76b29310581987aaf02, mesh-delivery 1d9ac0c8da9c74edc51c1e7d9d0630c3c087681e, mesh-catalog 82aaae846ea9c535d34465a90e902e5d9f773338; the only commit after them is the pin of mesh-sdk 0d97c09c (go.mod and go.sum), in mesh-delivery ab253515 and mesh-catalog 06cdac97.
Reviewed-by: review-1748@shanks at 06cdac97e3
Pushed-by: fix-1965@shanks

https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig

Issue 1748: a build that fails on a network lookup is built again by itself. Until now the builder never built such a failure again: "may be retaken" was only wording in mesh-build-go's error. Tonight's lookup pauses on shanks left mesh-build-image and mesh-build-node failed until someone built them again by hand. The registry group's delivery will start about 300 builds and would meet the same pauses. Delivery group `fix/1748-retake-and-resolver`, merged together: - **mesh-sdk** (go/delivery/builder.go): `Retake`, `Retakeable()` and `MaxRetakes` (3) beside `FailedUntilChanged`; a `Retakes` count on the build key and on the build's log entry (read by `builder_builds_since`). - **mesh-delivery** (mesh-delivery-builder): after a key is written failed with a retakeable error, `armRetake` takes it again after 60 s, then 120 s, then 240 s. Only that claim counts toward the cap, and `Sweep` re-arms from the key's count after a restart. The journal says each retake with its count. A compile error or a `FailedUntilChanged` error is never retaken, and the retake prefix is added only when every failing lane is retakeable. - **mesh-catalog** (mesh-build-go): a network failure of the fetch opens with the retake prefix. An unknown revision, a compile error or a failing test does not. Acceptance criteria: issue 1748 note 14. AC5 (dnsmasq) was withdrawn in review (note 18): dnsmasq logs when it reaches its cap, and no such line was seen, so the cap was not the cause. Rules bent, said here: - Playbook 08's one-repository bound: the mesh-sdk part is one exported name and one field. - ADR 0437 §8 (no timers in the builder): the backoff waits on a context's bound. The guard in manifest_test.go names `armRetake` as the one allowed bounded wait (issue 1748) and refuses any other. A bus delivery delay is a follow-up (issue 1748 note 16). - Test budget: mesh-sdk adds no test function. mesh-delivery adds one top-level unit function with five subtests, and its integration count stays at 71 (AC6 is a subtest of the existing failed-key test). Integration test run locally by name against an embedded bus: `TestAFailedKeyIsBuiltAgainAsANewRevisionFromANewModulesValueWithTheSameFingerprint` (with the AC6 subtest): pass. Review: review-1748@shanks approved round 2 at mesh-sdk 0d97c09c49b4a583ce51d76b29310581987aaf02, mesh-delivery 1d9ac0c8da9c74edc51c1e7d9d0630c3c087681e, mesh-catalog 82aaae846ea9c535d34465a90e902e5d9f773338; the only commit after them is the pin of mesh-sdk 0d97c09c (go.mod and go.sum), in mesh-delivery ab253515 and mesh-catalog 06cdac97. Reviewed-by: review-1748@shanks at 06cdac97e32cc0a042cd17a12c5178202ed010d8 Pushed-by: fix-1965@shanks https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig

mesh-tools added 2 commits 2026-11-04 21:46:45 +00:00

mesh-build-go opens a network failure of the fetch with the retake prefix (issue 1748) 82aaae846e

The builder takes a retakeable failure again by itself after a backoff.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig

Pin the group's mesh-sdk head 0d97c09c for the retake class (issue 1748)

mesh/delivery-group group fix/1748-retake-and-resolver checking: 2 of 2 member(s) ready

Details

mesh/delivery closed: the pull request closed unmerged

Details

mesh/check pass 06cdac97e32c on fd4931a75261 | judged by the candidate's mesh-build-go at 3200cb79deed

06cdac97e3

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig

mesh-tools commented

2026-11-04 21:46:51 +00:00

Delivery novox/mesh-catalog@06cdac97e32c — closed since 2026-11-04T22:24:02Z

Transitions

2026-11-04 21:46 (new) → proposed (announced): novox/mesh-catalog#614's head announced
2026-11-04 21:46 proposed → ready (accepted): a verdict names this commit and the trunk's head as its trunk commit, the gate passed or warned, and the repository's own check did not fail; or the check gates no merge (check-gates-merge false)
2026-11-04 22:24 ready → closed (closed): the pull request closed unmerged

The commit's note under refs/notes/mesh-plan keeps every transition of this change up to its merge: git log --notes=mesh-plan.

**Delivery** `novox/mesh-catalog@06cdac97e32c` — **closed** since 2026-11-04T22:24:02Z **Transitions** - 2026-11-04 21:46 (new) → proposed (announced): novox/mesh-catalog#614's head announced - 2026-11-04 21:46 proposed → ready (accepted): a verdict names this commit and the trunk's head as its trunk commit, the gate passed or warned, and the repository's own check did not fail; or the check gates no merge (check-gates-merge false) - 2026-11-04 22:24 ready → closed (closed): the pull request closed unmerged The commit's note under `refs/notes/mesh-plan` keeps every transition of this change up to its merge: `git log --notes=mesh-plan`.

mesh-tools commented

2026-11-04 22:23:22 +00:00

mesh-delivery: not merged. The merge of this pull request into main at 06cdac97e3 was declined: the operator, via telegram (user id verified), chose Decline. A new head asks the operator again.

mesh-delivery: **not merged.** The merge of this pull request into main at 06cdac97e32cc0a042cd17a12c5178202ed010d8 was declined: the operator, via telegram (user id verified), chose Decline. A new head asks the operator again.

mesh-tools closed this pull request

2026-11-04 22:24:00 +00:00

Pull request closed

Please reopen this pull request to perform a merge.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: novox/mesh-catalog#614