Skip to content

Commit

Permalink
Simplify description of logic for cancelling replica evictions
Browse files Browse the repository at this point in the history
Longhorn 2238

Signed-off-by: Eric Weber <eric.weber@suse.com>
  • Loading branch information
ejweber committed Nov 1, 2023
1 parent 6f2362f commit dd5e324
Showing 1 changed file with 3 additions and 7 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -191,13 +191,9 @@ Make changes so that:
`replica.status.evictionRequested` to influence replica scheduling/deletion behavior (e.g. rebuild an extra replica
to replace the evicting one or delete the evicting one once rebuilding is complete).
- The node controller sets `replica.spec.evictionRequested = false` when:
- `spec.evictionRequested == false` on the replica's node, AND
- `spec.evictionRequested == false` on the replica's disk, AND
- `status.Unschedulable == false` on the associated Kubernetes node object, OR
- `status.Unschedulable == true` on the associated Kubernetes node object and the node drain policy is not
`block-for-eviction` or `block-for-eviction-if-contains-last-replica`, OR
- `status.Unschedulable == true` on the associated Kubernetes node object, the node drain policy is
`block-for-eviction-if-contains-last-replica`, and there is another PDB-protected replica.
- The user is not requesting eviction with `node.spec.evictionRequested == true` or
`disk.spec.evictionRequested == true`, AND
- The conditions aren't right for auto-eviction based on the node status and drain policy.
- The node controller sets `status.autoEvicting = true` when a node has evicting replicas because of the new drain
policies and `status.autoEvicting == false` when it does not. This provides a clue to the user (and the UI) while auto
eviction is ongoing.
Expand Down

0 comments on commit dd5e324

Please sign in to comment.