Merge pull request #2294 from ArangoGutierrez/rebase_2001

Rebase PR 2001: add configurable pagination to nfd-gc

Author: k8s-ci-robot

cmd/nfd-gc/main.go

@@ -85,6 +85,8 @@ func initFlags(flagset *flag.FlagSet) *nfdgarbagecollector.Args {
 		"Kubeconfig to use")
 	flagset.IntVar(&args.Port, "port", 8080,
 		"Port which metrics and healthz endpoints are served on")
+	flagset.Int64Var(&args.ListSize, "list-size", 200,
+		"the pagination size used when listing node features")
 
 	klog.InitFlags(flagset)
 

docs/reference/gc-commandline-reference.md

@@ -31,6 +31,21 @@ Print usage and exit.
 
 Print version and exit.
 
+### -list-size
+
+The pagination size to use when calling api-server to list nodefeatures.
+Pagination is useful for controlling the load on api-server/etcd as the
+nodefeature resources can be large.
+A value of 0 will disable pagination.
+
+Default: 200
+
+Example:
+
+```bash
+nfd-gc -list-size=100
+```
+
 ### -gc-interval
 
 The `-gc-interval` specifies the interval between periodic garbage collector runs.

docs/usage/nfd-gc.md

@@ -27,3 +27,11 @@ default garbage collector interval is set to 1h which is the value when no
 In Helm deployments see
 [garbage collector parameters](../deployment/helm.md#garbage-collector-parameters)
 for altering the nfd-gc configuration.
+
+## List Pagination & Scalability
+
+When NFD GC starts up it lists nodefeatures from api-server.
+These resources can be large and in a large cluster this initial list call to
+sync the informer cache can be expensive and heavy on api-server/etcd.
+You can use the `list-size` argument to control pagination size
+to help control the load from this list.

pkg/nfd-gc/nfd-gc.go

@@ -52,6 +52,7 @@ type Args struct {
 	GCPeriod   time.Duration
 	Kubeconfig string
 	Port       int
+	ListSize   int64
 }
 
 type NfdGarbageCollector interface {
@@ -162,7 +163,7 @@ func (n *nfdGarbageCollector) garbageCollect() {
 
 	listAndHandle := func(gvr schema.GroupVersionResource, handler func(metav1.PartialObjectMetadata)) {
 		opts := metav1.ListOptions{
-			Limit: 200,
+			Limit: n.args.ListSize,
 		}
 		for {
 			rsp, err := n.client.Resource(gvr).List(context.TODO(), opts)