fix: reconstruct the cache for failed replica restarting

If the upper layer tries to launch a previously failed replica in a newly launched instance manager pod. The pod may have no info then directly treat it as a new replica and skip the cache reconstruction. Finally, this replica will be failed by r.validateAndUpdate() in the next moment. Longhorn 9488 Signed-off-by: Shuo Wu <[email protected]>
longhorn · Dec 31, 2024 · ed1b3cb · ed1b3cb
1 parent b3271d0
commit ed1b3cb
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/pkg/spdk/replica.go b/pkg/spdk/replica.go
@@ -803,8 +803,8 @@ func (r *Replica) Create(spdkClient *spdkclient.Client, portCount int32, superio
 		return nil, err
 	}
 
-	// In case of failed replica reuse being failed by r.validateAndUpdate(), we should make sure the caches are correct.
-	if r.reconstructRequired {
+	// In case of failed replica reuse/restart being errored by r.validateAndUpdate(), we should make sure the caches are correct.
+	if r.State == types.InstanceStatePending && r.reconstructRequired {
 		bdevLvolMap, err := GetBdevLvolMapWithFilter(spdkClient, r.replicaLvolFilter)
 		if err != nil {
 			return nil, err