Fixed the issue where node was not deleted when evicting pods timed out.

kubernetes · Sep 27, 2023 · 1e5918a · 1e5918a
1 parent cf8c507
commit 1e5918a
Showing 1 changed file with 12 additions and 2 deletions.
diff --git a/cluster-autoscaler/core/scaledown/actuation/group_deletion_scheduler.go b/cluster-autoscaler/core/scaledown/actuation/group_deletion_scheduler.go
@@ -89,8 +89,18 @@ func (ds *GroupDeletionScheduler) ScheduleDeletion(nodeInfo *framework.NodeInfo,
 
 	nodeDeleteResult := ds.prepareNodeForDeletion(nodeInfo, drain)
 	if nodeDeleteResult.Err != nil {
-		ds.AbortNodeDeletion(nodeInfo.Node(), nodeGroup.Id(), drain, "prepareNodeForDeletion failed", nodeDeleteResult)
-		return
+		isTimeout := true
+		for _, podEvictionResult := range nodeDeleteResult.PodEvictionResults {
+			// If there is any error message, it means it is not a pure timeout and node deletion is aborted.
+			if podEvictionResult.Err != nil {
+				isTimeout = false
+				break
+			}
+		}
+		if !isTimeout {
+			ds.AbortNodeDeletion(nodeInfo.Node(), nodeGroup.Id(), drain, "prepareNodeForDeletion failed", nodeDeleteResult)
+			return
+		}
 	}
 
 	ds.addToBatcher(nodeInfo, nodeGroup, batchSize, drain, opts.ZeroOrMaxNodeScaling)