Rollback fails in case of only 1 Segment Store Pod

[Prabhaker24]

Description

When the upgrade of a pravega cluster containing a single segmentstore replica fails (due to upgrade to an invalid pravega version), the readiness check on the controller pods soon starts to fail.

NAME READY STATUS RESTARTS AGE
bookkeeper-bookie-0 1/1 Running 0 66m
bookkeeper-bookie-1 1/1 Running 0 66m
bookkeeper-bookie-2 1/1 Running 0 66m
bookkeeper-operator-cd5b76d74-frvp9 1/1 Running 0 8h
nfs-server-provisioner-1599135458-0 1/1 Running 0 8d
pravega-operator-5c5776fb84-bwhfb 1/1 Running 0 8h
pravega-pravega-controller-667b796659-ft85s 0/1 Running 0 64m
pravega-pravega-segment-store-0 0/1 ImagePullBackOff 0 60m
zookeeper-0 1/1 Running 0 67m
zookeeper-1 1/1 Running 0 67m
zookeeper-2 1/1 Running 0 66m
zookeeper-operator-55ccb88b87-r4l5j 1/1 Running 0 2d2h

Subsequently, even the rollback of the pravega cluster to its previous version fails.
This happens only in case there is 1 ss pod and upgrade for that ss pod has failed and as not even one ss pod is available get request for system scope fails and when rollback happens the controller pod isn't marked ready for the same reason.

Importance

should_have

Location

upgrade.go file

Suggestions for an improvement

In case of only 1 ss replica change the order of rollback, Rollback first the ss pod and then the controller pod.