Chapter 6. Bug fixes

Arbiter and flexible scaling can’t be enabled at the same time.

When arbiter and flexible scaling both are enabled, the storage cluster was shown in READY state even though there were logs or messages with the error arbiter and flexibleScaling both can’t be enabled. This was happening because of the incorrect specs of the storage cluster CR. With this update, the storage cluster is in "ERROR" state with the correct error message.

Buckets are always deleted when cleanup is required by the library

Previously, on OBC creation failure, lib-bucket-provisioner sent a delete request to the provisioner for cleanup purposes before retrying. Noobaa provisioners would look at the reclaim policy of the object bucket but in some cases did not delete the underlying bucket. With this update, on cleanup scenarios, the underlying bucket should have been deleted regardless of the reclaim policy.

Collect configuration of each OSD attached

Previously, there was no way to find out detailed configuration of each OSD. With this update, must-gather collects all the configuration of an OSD to further improve the debugging.

gRPC metrics are now disabled by default

Earlier, the cephcsi pods exposed the remote procedure call (gRPC) metrics for debugging purposes. The cephcsi node plugin pods used the host ports 9091 for CephFS and 9090 for RBD on the node where the cephcsi node plugin pods were running. This meant the cephcsi pods failed to come up. With this update, gRPC metrics are disabled by default and cephcsi pods do not use ports 9091 and 9090 on the node where the node plugin pods are running.

MDS report oversized cache

Rook has not previously applied mds_cache_memory_limit upon upgrades. This means OpenShift Container Storage 4.2 clusters that did not have that option applied were not updated with the correct value, which is typically half the size of the pod’s memory limit. Therefore, MDSs in standby-replay may report oversized cache.

Newly restored PVC can now be mounted on nodes

Previously, a bug in Ceph-CSI driver caused a wrong ‘RBD image not found’ error while mounting newly restored PVC with deleted parent snapshot on nodes with Red Hat Enterprise Linux version of less than 8.2 (with no deep flattening feature). This issue was fixed by flattening the newly restored PVC before mounting on nodes with Red Hat Enterprise Linux version of less than 8.2 (with no deep flattening feature).

Reliable mon quorum

Earlier, if the operator was restarted during a mon failover, the operator could erroneously remove the new mon. Hence, the mon quorum was at risk when the operator removed the new mon. With this update, the operator will restore the state when the mon failover is in progress and properly complete the mon failover after the operator is restarted. Now, the mon quorum is more reliable in the node drains and mon failover scenarios.