We've got some tests that bring up a cluster and immediately terminate everything. Problem with this is you can get into a state where Couchbase has half initialized a bucket, and when the pods are recovered, it'll just hang in warmup forever. The fix, for us, is to wait until the bucket is reported as up and ready before termination.
In all fairness to CBS, it would look like the bucket is missing a bunch of vbuckets and something potentially bad could have happened. I guess a "hey I've not finished creating this" flag could solve the issue, but it sounds very niche.