jmclong / azuredisk-csi-driver / 27848737582

76%

main_v2: 70%

Build Type

push

github

Committed by jmclong

Commit Message

feat(azuredisk): skip detach for already-unattached disk on ControllerUnpublishVolume

During mass detach (e.g. when many spot/preemptible nodes are deleted at
once), ControllerUnpublishVolume calls DetachDisk per volume, which
serializes on the global VMManagementType mutex and can stall the
controller.

Add a fast-path that does a cheap disk read first: if the disk no longer
exists (HTTP 404) or exists but has no owning VM (ManagedBy and
ManagedByExtended both empty), the volume is already detached from every
node, so we return success without taking the node-centric DetachDisk
path. checkDiskExists honors GetDisk throttling and a timeout, so any read
error simply falls through to the normal detach.

The behavior is gated by a new CLI flag --skip-detach-for-unattached-disk
(default true). Adds unit tests for diskHasNoOwner and isDiskNotFoundError.

Coverage Stats

11 of 34 new or added lines in 3 files covered. (32.35%)

8 existing lines in 2 files now uncovered.

4688 of 6206 relevant lines covered (75.54%)

11.28 hits per line