kubernetes-sigs / karpenter / 29748337182

82%

main: 82%

Pull #3159

github

fix: tolerate missing PersistentVolume in driverFromVolume to prevent cluster sync paralysis

When a PV is deleted out-of-band while a PVC remains in Lost phase,
driverFromVolume returns a raw NotFound error that bubbles up through
updateForPod, causing node state reconciliation to fail repeatedly.
Upon a Karpenter controller restart, the affected node is excluded from
the internal cluster map, preventing Cluster.Synced() from succeeding
and pinning karpenter_cluster_state_synced at 0 indefinitely — halting
all provisioning, consolidation, and disruption operations.

This adds apierrors.IsNotFound handling to driverFromVolume, aligning it
with the existing tolerant behavior for missing PVCs in GetVolumes()
(line 88-94) which already has a comment warning about this exact
failure mode.

Signed-off-by: Omer Dahan <omerd2@users.noreply.github.com>

4 of 4 new or added lines in 1 file covered. (100.0%)

16576 of 20100 relevant lines covered (82.47%)

97684.6 hits per line