scylla-operator-bot[bot]
commented
3 months ago The Scylla Operator project currently lacks enough contributors to adequately respond to all issues.
This bot triages un-triaged issues according to the following rules:
- After 30d of inactivity,
lifecycle/staleis applied - After 30d of inactivity since
lifecycle/stalewas applied,lifecycle/rottenis applied - After 30d of inactivity since
lifecycle/rottenwas applied, the issue is closed
You can:
- Mark this issue as fresh with
/remove-lifecycle stale - Close this issue with
/close - Offer to help out
/lifecycle stale
tnozicka
We advise users to disable GKE upgrades to avoid the scenario where GKE breaks the PDB contract after 1 hour to force the upgrade while scylla may still be streaming to a new node. Unfortunately that also goes for the manual cluster upgrades.
One ideas is to create a completely separate node pool with the same labels and drain the old one manually, gradually migrating scylla nodes to the new node pool and then decommissioning the old node pool. We need to test it and document it. QA needs to add a case where it artificially breaks the upgrade and makes sure the nodes are not forced removed after 1h.