Repair kai-server disk-pressure watcher #618

New issue

Closed

opened 2026-07-24 04:20:11 +00:00 by coilysiren · 0 comments

coilysiren commented

2026-07-24 04:20:11 +00:00

Owner

The 2026-07-23 Forgejo outage exposed two failures in the kai-server disk-pressure detector.

df -B1P is parsed as one-petabyte blocks, so a 90% root filesystem appeared healthy.
The unprivileged classifier cannot traverse root-owned K3s storage, so its ownership report is incomplete.

Done condition: split the df flags, install a root-owned read-only classifier with exact sudoers commands, add regression coverage, and update the operator docs. Live convergence and disk cleanup remain attended operations.

The 2026-07-23 Forgejo outage exposed two failures in the kai-server disk-pressure detector. * df -B1P is parsed as one-petabyte blocks, so a 90% root filesystem appeared healthy. * The unprivileged classifier cannot traverse root-owned K3s storage, so its ownership report is incomplete. Done condition: split the df flags, install a root-owned read-only classifier with exact sudoers commands, add regression coverage, and update the operator docs. Live convergence and disk cleanup remain attended operations.

coilysiren referenced this issue from a commit

2026-07-24 04:20:28 +00:00

fix: repair disk pressure watcher measurements

coilysiren closed this issue

2026-07-24 04:20:28 +00:00