Answer A-472: a pipeline's end is said without waiting on another part

This commit is contained in:
2026-11-17 14:49:36 +01:00
parent bdb4efc160
commit 76c92e4c1e
@@ -1818,3 +1818,9 @@ the principle that settles it, cited as "design 47 G A-n".
mesh-delivery-sdk `buildstep` (the agreement every reader uses: indexer, builder, checker, a build node's step guard,
the composer); a fingerprint names the language version, so every module's fingerprint changes once. The first bugfix
through a working pipeline (the operator, 2026-11-17).
- **A-472** (met three times: the runner's ends 2026-11-16 22:16-22:52 and 2026-11-17 14:20 one per 30 s, and on
2026-11-17 14:4x one composition step-ended taken per ~80 s; principles 3 and 7) A pipeline's end and a step's end are
written and said without waiting on another part: the runner reads a catalog revision once and keeps it (a revision
never changes), every read `widens` makes is bounded to a few seconds and a read that fails or is cut says so in the
journal and in the widening's could-not-run, and `runner_cancel` reads a pipeline's state after an ask that did not
answer, so it never reports refused for a pipeline it cancelled.