kubevirt / containerized-data-importer / #6164

50%

main: 50%

Build Type

Pull #4225

travis-ci

Committed by mnecas

Commit Message

Avoid PVC writes during scheduler retry to prevent SchedulerError loop

The import controller watches both PVCs and owned pods. When the importer
pod is pending and the scheduler hits a VolumeBinding error, each reconcile
writes PVC annotations, bumping resourceVersion. This fires the PVC watch
causing an immediate re-reconcile (bypassing the 2s RequeueAfter), which
writes again, creating a tight loop at ~10 writes/sec.

The scheduler's VolumeBinding plugin assumes a PVC resourceVersion during
filtering and expects it unchanged at reserve time. The constant writes
make this race unwinnable, so the pod is never scheduled.

Fix: skip PVC writes and RequeueAfter while the pod has PodScheduled=False
with a transient reason (e.g. SchedulerError). Pod/PVC watches resume
reconciliation once scheduling succeeds.

Signed-off-by: Martin Necas <mnecas@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

Pull Request Pull Request #4225: Avoid PVC writes during scheduler retry to prevent SchedulerError loop

Coverage Stats

13 of 16 new or added lines in 1 file covered. (81.25%)

4 existing lines in 1 file now uncovered.

15151 of 30424 relevant lines covered (49.8%)

0.56 hits per line