Dedup adversarial repair and GC deadlines leave crash coverage incomplete #1287

Open
opened 2026-10-07 00:49:24 +00:00 by kayg · 0 comments
Owner

The adv31 production-release campaign reached the dedup repair and crash/GC scenarios on the retained large Home. The repair status was still scanning after the probe deadline, with 512 files and 128,712,635 bytes checked. It did not yet report the intentionally damaged blob as repaired. Its problem list included an earlier untouched-byte fixture; this evidence does not establish user-data loss.

The publish crash barriers (two intentional SIGKILL restarts) ran. The GC watcher then waited 120 seconds without seeing its 512 artificial garbage entries deleted. Thus the GC crash barrier was never requested, and the probe's additional 120-second restart wait and 180-second completion wait expired. The server remained alive. Orphan collection and repair were not proved within this round; no assertions were changed.

The raw bounded-probe output is attached to kayg/calternal#1202. Separate scheduler/fixture deadline investigation is required before treating this as a production correctness finding. Deletion requests preceding GC were slow (16.7s and 8.5s), which alone does not block the merge on the shared host.

The adv31 production-release campaign reached the dedup repair and crash/GC scenarios on the retained large Home. The repair status was still `scanning` after the probe deadline, with 512 files and 128,712,635 bytes checked. It did not yet report the intentionally damaged blob as repaired. Its problem list included an earlier untouched-byte fixture; this evidence does not establish user-data loss. The publish crash barriers (two intentional SIGKILL restarts) ran. The GC watcher then waited 120 seconds without seeing its 512 artificial garbage entries deleted. Thus the GC crash barrier was never requested, and the probe's additional 120-second restart wait and 180-second completion wait expired. The server remained alive. Orphan collection and repair were not proved within this round; no assertions were changed. The raw bounded-probe output is attached to kayg/calternal#1202. Separate scheduler/fixture deadline investigation is required before treating this as a production correctness finding. Deletion requests preceding GC were slow (16.7s and 8.5s), which alone does not block the merge on the shared host.
Commenting is not possible because the repository is archived.
No labels
No milestone
No project
No assignees
1 participant
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal-v0#1287
No description provided.