Record overloaded Indee observation and eventual native recovery

2026-10-08 07:12:25 -04:00
parent 83632c2439
commit 214cebfac2
1 changed files with 30 additions and 0 deletions
@@ -717,3 +717,33 @@ Actual success qualification `f39bd824-1d3e-49dc-a0ad-d2a90523bdb0` is in
 progress; no success or deployment is claimed by this checkpoint. Historical
 published catalog bytes remain unchanged. The separately prepared unsigned
 candidate catalog already has the correct hooks (see worker runtime receipt).
 ### Corrected-plan attempt: overloaded observation, then native recovery
 Operation `f39bd824-1d3e-49dc-a0ad-d2a90523bdb0` did not reach target startup.
 During acquire, reading the exact original worker's persisted died event with
 `podman events` exceeded its unchanged 30-second bound. The initial recovery
 call then exceeded the unchanged 90-second systemd-start observation bound and
 reported unresolved Restoring at 11:08:12 UTC. This was an accurate caller-time
 failure, not the eventual terminal outcome.
 Subsequent read-only evidence found original worker
 `2ddac1b1c3e1454e1c1e61686e134940fd62fa6ce5a009920d960eede6240b04`
 died with exit137 at 11:05:13 UTC. Its recovered unit entered active state at
 11:08:13, one second after the caller's timeout. Supported native recovery then
 completed autonomously: **Restored, cleanup_done=true, maintenance Released**.
 No manual journal edit, ownership adoption, resume command or second RPC was
 used. Independent checks passed all seven running original/recovery image pins,
 exact saved units, frontend/API writable-layer sentinels and absent holds/fence.
 No backup or successful target-cutover acceptance is claimed for this attempt.
 Host pressure during the failure reached memory-full avg10 56% and I/O-full
 avg10 59%; QEMU resident memory fell to about 880 MiB. An unrelated compiler
 was running in another project's systemd unit. This supports overloaded
 observation as a contributor; it does not prove a new worker shutdown defect.
 No unrelated process, service, scheduler priority or production timeout changed.
 After verified safe recovery, the same QEMU was paused with memory retained to
 serialize queued work. Success qualification remains open and must resume only
 when the host can support the unchanged deadlines. The corrected hook itself
 was not reached in this operation.