1.9 KiB
| id | status | status_updated | status_note | component | severity | timestamp | last_seen | occurrences | files | tags | related | fix_commits | legacy | gitea_issue | gitea_synced_status | gitea_synced_hash | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| bug-1860 | fixed | 2026-09-27 | c69705ae: controller is ground truth on rejoin; controller refuses non-PEND dispatch and re-sends state; scheduler adopts controller allocation on RUN update |
controller | unknown | 2026-09-24T14:38:03Z | 2026-09-24T14:38:03Z | 1 |
|
|
|
true | 279 | fixed | 92556ffb1968 |
bug-1860: Scheduler keeps a job RUN (with allocation) forever after the controller requeued it to PEND while…
Description
Scheduler keeps a job RUN (with allocation) forever after the controller requeued it to PEND while the scheduler was disconnected: scheduler.log 'JOBS_PUSH: refusing resync state 0x1 for jobid ... -- existing job is RUN with a real allocation on record' on rejoin; job never redispatched (seen in bug-1859's chaotic soak)
Environment
Not recorded (migrated from buglog.json).
Reproduction
Not recorded (migrated from buglog.json).
Root cause
core_events.c JOBS_PUSH bug-1812 guard refuses any non-terminal resync state that would drop a real allocation, to protect against a committed snapshot racing the scheduler's own fresh dispatch. It cannot tell that race apart from a genuine, committed controller-side requeue (e.g. resync 'unrecognized' -> record_job_reconcile_pending_fact) that happened while the scheduler was disconnected, and JOBS_PUSH carries no per-job freshness (commit LSN/version) to decide. Needs a design decision (e.g. per-job last-modified LSN in the push).
Fix
Not recorded (migrated from buglog.json).
Verification
Not recorded (migrated from buglog.json).
References
- Gitea: #279