wake/reconcile (pilot #7, alarm-hygiene): re-enumerates CONSUMED detector states → dup wake + spurious rc=1 CRITICAL → store records last-consumed observed_hash as 3rd accounted check #932
Assignees
jason.woltje
Clear assignees
jason.woltje
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: mosaicstack/stack#932
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Pilot finding #7 (dragon-lin), A-lane, NORMAL cadence — joins the residual queue AFTER #917/#925. Safe-but-noisy alarm-hygiene defect.
Defect. The reconciler RE-ENUMERATES already-CONSUMED detector-observed states. After consume-truncation, a consumed state matches NO accounting record — the inbox is truncated, the seen-ledger only covers the reconciler's OWN enumerations, and the detector hash-file is (correctly) distrusted — so the reconciler treats it as unaccounted and re-enumerates it. Result: one DUPLICATE orientation wake + one SPURIOUS
rc=1CRITICAL alarm per detector-active window per cycle. Functionally safe (no lost obligation), but NOISY — the real cost is G2a ALARM HYGIENE at fleet scale: cry-wolf erosion kills real alarms.RULED SCALPEL (Mos). At consume-truncation, the STORE records the last-consumed
observed_hashper(kind,id); the reconciler adds it as its THIRD accounted check (alongside the inbox/observed_seq and its own seen-ledger).Trust argument (load-bearing — verbatim). A store-side record is trustworthy for the reconciler BY CONSTRUCTION — its existence implies the state WAS durably enqueued; the §5 swallow hole's signature is hash-advance-WITHOUT-enqueue, which this record structurally cannot exhibit. The rejected alternative (reconcile trusting detector hash-files) STAYS REJECTED — it would reopen §5.
Red-first: a CONSUMED state does NOT re-enumerate (no dup wake, no spurious CRITICAL); AND a genuinely-unaccounted state STILL re-enumerates (the anti-swallow catch keeps its teeth — do NOT weaken G3).
Pilot evidence: seqs 20/21 byte-matched to consumed 18/19 (re-enumerated the just-consumed states); the follow-up cycle was clean.
Refs: #892, #908 (store allocator / seen-ledger), §5 swallow hole (why detector-hash-files stay distrusted), #920/#924 (co-feed + alarm hygiene). Priority: normal cadence, after #917.