|
|
def2bf4038
|
feat(automation): poisoned-run quarantine policy with durable failure signatures (SCHED-SOAK offline)
- poisoned_store.py: durable JSONL identical-failure counter (fcntl.flock + intra-process lock
+ thread-local reentrancy, append+fsync, torn-line tolerant, CAS quarantine epoch); no
Alembic/ORM changes (046 migrations frozen).
- poisoned.py: N=3 identical infra failures -> typed POISONED_RUN_QUARANTINE, matching
schedules disabled, exactly one blocked notification carrying the DLQ payload; success
resets the pair; operator unquarantine (CAS) re-enables schedules and resets counters.
- REST: GET /scenario-automation/quarantine and POST /quarantine/{scenario_id}/release
(scenario:automation MANAGE; 404 NOT_QUARANTINED / 409 STALE_QUARANTINE_VERSION).
- Tests: 13 policy + 2 API quarantine vectors; merged with the master ReadAcl/RetentionReceipts
classes (conflict resolved keeping both). Focused 31 passed; 046 T023 offline slice.
|
2026-09-22 12:35:39 +03:00 |
|
