Hello! I noticed that the Quickwit control_plane component explicitly uses the Recreate deployment strategy in the Helm chart (see here). It appears RollingUpdate was previously supported but was intentionally removed in this commit: fdd30be.
(I'm using a database as a metastore, so there shouldn't be some PVC problem)
I’d love to get more context on the reasoning behind this change and its impact on high availability:
- Does scaling the control_plane to 0 pods during a Recreate cycle risk downtime for indexing or search queries?
- What is the recommended best practice for updating Quickwit versions without service interruption?
- From my understanding, the control_plane manages the indexers/searchers. What happens when the control_plane is (temporary) absent? Do the indexers/searcher stop/pause their operations?
Thank you in advance!!
Hello! I noticed that the Quickwit control_plane component explicitly uses the
Recreatedeployment strategy in the Helm chart (see here). It appearsRollingUpdatewas previously supported but was intentionally removed in this commit: fdd30be.(I'm using a database as a metastore, so there shouldn't be some PVC problem)
I’d love to get more context on the reasoning behind this change and its impact on high availability:
Thank you in advance!!