Store scale downs by node group id

bskiba · Jul 24, 2024 · bd40d70 · bd40d70
1 parent dee3865
commit bd40d70
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 14 deletions.
diff --git a/cluster-autoscaler/clusterstate/clusterstate.go b/cluster-autoscaler/clusterstate/clusterstate.go
@@ -121,7 +121,7 @@ type ClusterStateRegistry struct {
 	sync.Mutex
 	config                             ClusterStateRegistryConfig
 	scaleUpRequests                    map[string]*ScaleUpRequest // nodeGroupName -> ScaleUpRequest
-	scaleDownRequests                  []*ScaleDownRequest
+	scaleDownRequests                  map[string][]*ScaleDownRequest
 	nodes                              []*apiv1.Node
 	nodeInfosForGroups                 map[string]*schedulerframework.NodeInfo
 	cloudProvider                      cloudprovider.CloudProvider
@@ -158,7 +158,7 @@ type NodeGroupScalingSafety struct {
 func NewClusterStateRegistry(cloudProvider cloudprovider.CloudProvider, config ClusterStateRegistryConfig, logRecorder *utils.LogEventRecorder, backoff backoff.Backoff, nodeGroupConfigProcessor nodegroupconfig.NodeGroupConfigProcessor) *ClusterStateRegistry {
 	return &ClusterStateRegistry{
 		scaleUpRequests:                 make(map[string]*ScaleUpRequest),
-		scaleDownRequests:               make([]*ScaleDownRequest, 0),
+		scaleDownRequests:               make(map[string][]*ScaleDownRequest),
 		nodes:                           make([]*apiv1.Node, 0),
 		cloudProvider:                   cloudProvider,
 		config:                          config,
@@ -251,7 +251,7 @@ func (csr *ClusterStateRegistry) RegisterScaleDown(nodeGroup cloudprovider.NodeG
 	}
 	csr.Lock()
 	defer csr.Unlock()
-	csr.scaleDownRequests = append(csr.scaleDownRequests, request)
+	csr.scaleDownRequests[nodeGroup.Id()] = append(csr.scaleDownRequests[nodeGroup.Id()], request)
 }
 
 // To be executed under a lock.
@@ -291,10 +291,12 @@ func (csr *ClusterStateRegistry) updateScaleRequests(currentTime time.Time) {
 		}
 	}
 
-	newScaleDownRequests := make([]*ScaleDownRequest, 0)
-	for _, scaleDownRequest := range csr.scaleDownRequests {
-		if scaleDownRequest.ExpectedDeleteTime.After(currentTime) {
-			newScaleDownRequests = append(newScaleDownRequests, scaleDownRequest)
+	newScaleDownRequests := make(map[string][]*ScaleDownRequest)
+	for nodeGroupId, scaleDownRequests := range csr.scaleDownRequests {
+		for _, scaleDownRequest := range scaleDownRequests {
+			if scaleDownRequest.ExpectedDeleteTime.After(currentTime) {
+				newScaleDownRequests[nodeGroupId] = append(newScaleDownRequests[nodeGroupId], scaleDownRequest)
+			}
 		}
 	}
 	csr.scaleDownRequests = newScaleDownRequests
@@ -566,10 +568,10 @@ func (csr *ClusterStateRegistry) updateAcceptableRanges(targetSize map[string]in
 		acceptableRange.MinNodes -= scaleUpRequest.Increase
 		result[nodeGroupName] = acceptableRange
 	}
-	for _, scaleDownRequest := range csr.scaleDownRequests {
-		acceptableRange := result[scaleDownRequest.NodeGroup.Id()]
-		acceptableRange.MaxNodes++
-		result[scaleDownRequest.NodeGroup.Id()] = acceptableRange
+	for nodeGroupName, scaleDownRequests := range csr.scaleDownRequests {
+		acceptableRange := result[nodeGroupName]
+		acceptableRange.MaxNodes += len(scaleDownRequests)
+		result[nodeGroupName] = acceptableRange
 	}
 	csr.acceptableRanges = result
 }

diff --git a/cluster-autoscaler/clusterstate/clusterstate_test.go b/cluster-autoscaler/clusterstate/clusterstate_test.go
@@ -478,7 +478,7 @@ func TestRegisterScaleDown(t *testing.T) {
 	}, fakeLogRecorder, newBackoff(), nodegroupconfig.NewDefaultNodeGroupConfigProcessor(config.NodeGroupAutoscalingOptions{MaxNodeProvisionTime: 15 * time.Minute}))
 	now := time.Now()
 	clusterstate.RegisterScaleDown(provider.GetNodeGroup("ng1"), "ng1-1", now.Add(time.Minute), now)
-	assert.Equal(t, 1, len(clusterstate.scaleDownRequests))
+	assert.Equal(t, 1, len(clusterstate.scaleDownRequests["ng1"]))
 	clusterstate.updateScaleRequests(now.Add(5 * time.Minute))
 	assert.Equal(t, 0, len(clusterstate.scaleDownRequests))
 	assert.Empty(t, clusterstate.GetScaleUpFailures())
@@ -1143,9 +1143,9 @@ func TestUpdateAcceptableRanges(t *testing.T) {
 			for nodeGroupName, targetSize := range tc.targetSizes {
 				provider.AddNodeGroup(nodeGroupName, 0, 1000, targetSize)
 			}
-			var scaleDownRequests []*ScaleDownRequest
+			scaleDownRequests := make(map[string][]*ScaleDownRequest)
 			for _, nodeGroupName := range tc.scaledDownGroups {
-				scaleDownRequests = append(scaleDownRequests, &ScaleDownRequest{
+				scaleDownRequests[nodeGroupName] = append(scaleDownRequests[nodeGroupName], &ScaleDownRequest{
 					NodeGroup: provider.GetNodeGroup(nodeGroupName),
 				})
 			}