Open
Description
This is at the moment mostly a placeholder issue to gather ideas for enhancement in the migration area.
Motivation
Current migration story is not ideal, especially with the Operator deployments.
- Migration can be interrupted by the Startup probe.
- Migration can be interrupted by a transaction timeout.
- Migration is not atomic. Interrupting it might lead to a broken state.
- Operator by default deploys a matching Server image. Updating the Operator consequently also updates the Server. This might be confusing and problematic, especially when we do not provide clear instructions for a rollback.
- Just rolling back even micro releases, without rolling back the database may not work - Liquibase Checksum Error #37641
- Docs are not sufficient
See the following for more context on the motivation:
- liveness probe /health/live not UP while DB migrations initialization #35261
- Operator: Implement ForceRolling upgrade strategy #37295 (comment)