You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The core primitive of rolling upgrade is preparing a node for being shutdown gracefully. This is primarily achieved by reliably relinquishing leadership of raft groups. A per-node maintenance mode should be created that achieves this state.
A node should be able to be put into maintenance mode:
via admin api
via signal (subject to any configuration that controls signal handling behavior)
A node should be able to be brought out of maintenance mode. If triggered via signal (e.g. shutting down as part of a k8s replica set rolling upgrade) this may not be feasible.
The core primitive of rolling upgrade is preparing a node for being shutdown gracefully. This is primarily achieved by reliably relinquishing leadership of raft groups. A per-node maintenance mode should be created that achieves this state.
Fist new leadership should be rejected:
Next, leadership should be transfer to other nodes. This might require throttling in some cases:
A node should be able to be put into maintenance mode:
A node should be able to be brought out of maintenance mode. If triggered via signal (e.g. shutting down as part of a k8s replica set rolling upgrade) this may not be feasible.
Additional notes
The text was updated successfully, but these errors were encountered: