Extend MaintenanceWindows parameter usage

[hughcapet]

Consider maintenance window when migrating master pods and replacing pods (rolling update)

[idanovinda]

👍

[FxKu]

If I recall correctly moving the primary is a once or never operation. Removing the node readiness label triggers an node event which the operator is listening on to perform the migration. If the migration does not work the operator will not try again, because no other node event is send. Hence the primary stays on the node and blocks further operators by the K8s admins. We have seen it in the past where we were asked to trigger a switchover to unblock K8s updates.

Therefore, I think we greatly increase the possibility that pod migrations will not happen at all if they are omitted due to maintenance window restrictions. However, we could instead mark the primary pod to be rotated so another sync will trigger a switchover so that the old instance can be removed by K8s drainage.

[FxKu]

We have to update reference docs and also sections that talk about rolling updates like in admin docs. About user docs I'm not sure. So far they do not mention maintenance windows at all.

[FxKu]

👍

[hughcapet]

👍