ObservedGeneration and update block #1298
Conversation
burmanm
force-pushed
the
upgrade_process
branch
2 times, most recently
from
5801d6a
to
2881e5f
Compare
|
@burmanm , is this ready for another round of review? |
|
Yes |
…the end of the reconciliation
…also, not just Stargate
…tricted values for AutomatedUpdate, "always" and "once" and verify them in the webhook. Remove tee outputting to stdout in the helm prepare script. UpdateStatus should only delete and reset the ClusterRequiresUpdate if it was allowed to update in the first place. Also, we should listen to changes and reconcile if annotations change. Add additional checks to the GenerationCheck test to ensure we do not touch the CassandraDatacenter unless given permission.
Since ObservedGeneration is being added by this PR, it will be 0 when the operator gets upgraded to this version. We want to interpret that as "the generation didn't change". This was agreed upon previously but got lost in a force-push.
There was a problem hiding this comment.
I pushed commits to fix a few remaining issues. This is good to go as far as k8ssandra-operator is concerned.
However I'm noticing an issue with the refresh CassandraTask: it doesn't seem to ever be marked as finished (and as a consequence neither does the K8ssandraTask). It doesn't impact the new behavior we're introducing here, the conditions and annotations are correctly reset.
I'll test on cass-operator alone and file an issue if necessary.
|
OK, this might be a side-effect of my testing method: I was artificially setting the |
|
Given we found some bugs (or side effects) in cass-operator's implementation of this feature, I think we should first solve those and re-evaluate how it's done in this PR. |
What this PR does:
Adds new status field: ObservedGeneration and ensure it is updated at the end of the reconciliation
Which issue(s) this PR fixes:
Fixes #1274
Checklist