Fixed
- Regression in
v0.2.0
- it was correct to use non-rate limited functions in certain scenarios, rate limit should only be used for retry after errors - Retry in case of errors was broken - it was using
Forget
instead ofDone
which was just resetting the rate limit counter but not actually removing previous item from the queue - No longer watch for SC - lister is already using efficient caching API, so it is cheap to get a SC by name from the Index during PV reconciliation; that removes some race condition when PV was reconciled and forgotten about before Releaser learned that SC is annotated
- Do not re-queue an object that has the same version
- There was race conditions when
claimRef
was already removed from the PV but Kubernetes didn't yet marked it asAvailable
, re-queuing such a PV was resulting in duplicativeReleased
events - now Released verifies that theclaimRef
is not alreadynil
before trying to release the volume - De-queue objects that are deleted from the cluster