Background
Currently, when modifying configurations via the Milvus CRD, there are only two behaviors:
- When updateConfigMapOnly is true: No resources are recreated at all.
- When updateConfigMapOnly is false: The entire Milvus cluster is rolled/restarted (all component Deployments).
Proposal
It would be highly beneficial to introduce a third behavior:
- Only restart/recreate the affected components or their corresponding Deployments when a configuration change only impacts specific components.
The logic for determining affected components can be:
- Explicitly specified by the upstream or user for each configuration change.
- Or automatically inferred based on which Milvus components use the modified configuration.
This would avoid unnecessary full cluster restarts and reduce service disruption during runtime configuration updates.
Would the community consider supporting this fine-grained rolling update mechanism?
Background
Currently, when modifying configurations via the Milvus CRD, there are only two behaviors:
Proposal
It would be highly beneficial to introduce a third behavior:
The logic for determining affected components can be:
This would avoid unnecessary full cluster restarts and reduce service disruption during runtime configuration updates.
Would the community consider supporting this fine-grained rolling update mechanism?