Add leader election for nfd-master

cmd/nfd-master/main.go

@@ -128,6 +128,8 @@ func initFlags(flagset *flag.FlagSet) (*master.Args, *master.ConfigOverrideArgs)
 	flagset.StringVar(&args.Options, "options", "",
 		"Specify config options from command line. Config options are specified "+
 			"in the same format as in the config file (i.e. json or yaml). These options")
+	flagset.BoolVar(&args.EnableLeaderElection, "enable-leader-election", false,
+		"Enables a leader election. Enable this when running more than one replica on nfd master.")
 
 	overrides := &master.ConfigOverrideArgs{
 		LabelWhiteList: &utils.RegexpVal{},

deployment/base/rbac/master-clusterrole.yaml

@@ -22,3 +22,18 @@ rules:
   - get
   - list
   - watch
+- apiGroups:
+  - coordination.k8s.io
+  resources:
+  - leases
+  verbs:
+  - create
+- apiGroups:
+  - coordination.k8s.io
+  resources:
+  - leases
+  resourceNames:
+  - "nfd-master.nfd.kubernetes.io"
+  verbs:
+  - get
+  - update

deployment/components/master-config/nfd-master.conf.example

@@ -5,3 +5,9 @@
 # enableTaints: false
 # labelWhiteList: "foo"
 # resyncPeriod: "2h"
+# leaderElection:
+#   leaseDuration: 15s
+#   # this value has to be lower than leaseDuration and greater than retryPeriod*1.2
+#   renewDeadline: 10s
+#   # this value has to be greater than 0
+#   retryPeriod: 2s

deployment/helm/node-feature-discovery/templates/clusterrole.yaml

@@ -25,6 +25,21 @@ rules:
   - get
   - list
   - watch
+- apiGroups:
+  - coordination.k8s.io
+  resources:
+  - leases
+  verbs:
+  - create
+- apiGroups:
+  - coordination.k8s.io
+  resources:
+  - leases
+  resourceNames:
+  - "nfd-master.nfd.kubernetes.io"
+  verbs:
+  - get
+  - update
 {{- end }}
 
 ---

deployment/helm/node-feature-discovery/templates/master.yaml

@@ -82,6 +82,9 @@ spec:
             - "-port={{ .Values.master.port | default "8080" }}"
             {{- if .Values.enableNodeFeatureApi }}
             - "-enable-nodefeature-api"
+            {{- if  gt (int .Values.master.replicaCount) 1 }}
+            - "-enable-leader-election"
+            {{- end }}
             {{- end }}
             {{- if .Values.master.extraLabelNs | empty | not }}
             - "-extra-label-ns={{- join "," .Values.master.extraLabelNs }}"

deployment/helm/node-feature-discovery/values.yaml

@@ -21,6 +21,12 @@ master:
     # enableTaints: false
     # labelWhiteList: "foo"
     # resyncPeriod: "2h"
+    # leaderElection:
+    #   leaseDuration: 15s
+    #   # this value has to be lower than leaseDuration and greater than retryPeriod*1.2
+    #   renewDeadline: 10s
+    #   # this value has to be greater than 0
+    #   retryPeriod: 2s
   ### <NFD-MASTER-CONF-END-DO-NOT-REMOVE>
   # The TCP port that nfd-master listens for incoming requests. Default: 8080
   port: 8080

docs/reference/master-commandline-reference.md

@@ -151,6 +151,20 @@ Example:
 nfd-master -enable-nodefeature-api
 ```
 
+### -enable-leader-election
+
+The `-enable-leader-election` flag enables leader election for NFD-Master.
+It is advised to turn on this flag when running more than one instance of
+NFD-Master.
+
+This flag takes effect only when combined with `-enable-nodefeature-api` flag.
+
+Default: false
+
+```bash
+nfd-master -enable-nodefeature-api -enable-leader-election
+```
+
 ### -enable-taints
 
 The `-enable-taints` flag enables/disables node tainting feature of NFD.

docs/reference/master-configuration-reference.md

@@ -114,7 +114,7 @@ Example:
 labelWhiteList: "foo"
 ```
 
-### resyncPeriod
+## resyncPeriod
 
 The `resyncPeriod` option specifies the NFD API controller resync period.
 The resync means nfd-master replaying all NodeFeature and NodeFeatureRule objects,
@@ -128,5 +128,64 @@ Default: 1 hour.
 Example:
 
 ```yaml
-resyncPeriod=2h
+resyncPeriod: 2h
+```
+
+## leaderElection
+
+The `leaderElection` section exposes configuration to tweak leader election.
+
+### leaderElection.leaseDuration
+
+`leaderElection.leaseDuration` is the duration that non-leader candidates will
+wait to force acquire leadership. This is measured against time of
+last observed ack.
+
+A client needs to wait a full LeaseDuration without observing a change to
+the record before it can attempt to take over. When all clients are
+shutdown and a new set of clients are started with different names against
+the same leader record, they must wait the full LeaseDuration before
+attempting to acquire the lease. Thus LeaseDuration should be as short as
+possible (within your tolerance for clock skew rate) to avoid a possible
+long waits in the scenario.
+
+Default: 15 seconds.
+
+Example:
+
+```yaml
+leaderElection:
+  leaseDurtation: 15s
+```
+
+### leaderElection.renewDeadline
+
+`leaderElection.renewDeadline` is the duration that the acting master will retry
+refreshing leadership before giving up.
+
+This value has to be lower than leaseDuration and greater than retryPeriod*1.2.
+
+Default: 10 seconds.
+
+Example:
+
+```yaml
+leaderElection:
+  renewDeadline: 10s
+```
+
+### leaderElection.retryPeriod
+
+`leaderElection.retryPeriod` is the duration the LeaderElector clients should wait
+between tries of actions.
+
+It has to be greater than 0.
+
+Default: 2 seconds.
+
+Example:
+
+```yaml
+leaderElection:
+  retryPeriod: 2s
 ```

docs/usage/nfd-master.md

@@ -29,6 +29,11 @@ are created on the node (note the allowed
 > its [`-enable-nodefeature-api`](../reference/worker-commandline-reference.md#-enable-nodefeature-api)
 > flag.
 
+When `-enable-nodefeature-api` option is enabled and NFD-Master is intended to run
+with more than one replica, it is advised to use `-enable-leader-election` flag.
+This flag turns on leader election for NFD-Master and let only one replica
+to act on changes in NodeFeature and NodeFeatureRule objects.
+
 ## NodeFeatureRule controller
 
 NFD-Master acts as the controller for

pkg/nfd-master/nfd-master-internal_test.go

@@ -558,6 +558,10 @@ denyLabelNs: ["denied.ns.io","denied.kubernetes.io"]
 resourceLabels: ["vendor-1.com/feature-1","vendor-2.io/feature-2"]
 enableTaints: false
 labelWhiteList: "foo"
+leaderElection:
+  leaseDuration: 20s
+  renewDeadline: 4s
+  retryPeriod: 30s
 `)
 		f.Close()
 		So(err, ShouldBeNil)
@@ -573,6 +577,9 @@ labelWhiteList: "foo"
 				So(master.config.ResourceLabels, ShouldResemble, utils.StringSetVal{"vendor-1.com/feature-1": struct{}{}, "vendor-2.io/feature-2": struct{}{}}) // from cmdline
 				So(master.config.DenyLabelNs, ShouldResemble, utils.StringSetVal{"denied.ns.io": struct{}{}, "denied.kubernetes.io": struct{}{}})
 				So(master.config.LabelWhiteList.String(), ShouldEqual, "foo")
+				So(master.config.LeaderElection.LeaseDuration.Seconds(), ShouldEqual, float64(20))
+				So(master.config.LeaderElection.RenewDeadline.Seconds(), ShouldEqual, float64(4))
+				So(master.config.LeaderElection.RetryPeriod.Seconds(), ShouldEqual, float64(30))
 			})
 		})
 

pkg/nfd-master/nfd-master.go

@@ -30,6 +30,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/google/uuid"
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials"
@@ -38,9 +39,12 @@ import (
 	"google.golang.org/grpc/peer"
 	corev1 "k8s.io/api/core/v1"
 	k8sQuantity "k8s.io/apimachinery/pkg/api/resource"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	k8sLabels "k8s.io/apimachinery/pkg/labels"
 	"k8s.io/client-go/kubernetes"
 	restclient "k8s.io/client-go/rest"
+	"k8s.io/client-go/tools/leaderelection"
+	"k8s.io/client-go/tools/leaderelection/resourcelock"
 	"k8s.io/klog/v2"
 	controller "k8s.io/kubernetes/pkg/controller"
 	taintutils "k8s.io/kubernetes/pkg/util/taints"
@@ -71,6 +75,14 @@ type NFDConfig struct {
 	ResourceLabels utils.StringSetVal
 	EnableTaints   bool
 	ResyncPeriod   utils.DurationVal
+	LeaderElection LeaderElectionConfig
+}
+
+// LeaderElectionConfig contains the configuration for leader election
+type LeaderElectionConfig struct {
+	LeaseDuration utils.DurationVal
+	RenewDeadline utils.DurationVal
+	RetryPeriod   utils.DurationVal
 }
 
 // ConfigOverrideArgs are args that override config file options
@@ -98,6 +110,7 @@ type Args struct {
 	Prune                bool
 	VerifyNodeName       bool
 	Options              string
+	EnableLeaderElection bool
 
 	Overrides ConfigOverrideArgs
 }
@@ -174,6 +187,11 @@ func newDefaultConfig() *NFDConfig {
 		ResourceLabels: utils.StringSetVal{},
 		EnableTaints:   false,
 		ResyncPeriod:   utils.DurationVal{Duration: time.Duration(1) * time.Hour},
+		LeaderElection: LeaderElectionConfig{
+			LeaseDuration: utils.DurationVal{Duration: time.Duration(15) * time.Second},
+			RetryPeriod:   utils.DurationVal{Duration: time.Duration(2) * time.Second},
+			RenewDeadline: utils.DurationVal{Duration: time.Duration(10) * time.Second},
+		},
 	}
 }
 
@@ -221,7 +239,11 @@ func (m *nfdMaster) Run() error {
 
 	// Run updater that handles events from the nfd CRD API.
 	if m.nfdController != nil {
-		go m.nfdAPIUpdateHandler()
+		if m.args.EnableLeaderElection {
+			go m.nfdAPIUpdateHandlerWithLeaderElection()
+		} else {
+			go m.nfdAPIUpdateHandler()
+		}
 	}
 
 	// Notify that we're ready to accept connections
@@ -1230,3 +1252,47 @@ func (m *nfdMaster) startNfdApiController() error {
 	}
 	return nil
 }
+
+func (m *nfdMaster) nfdAPIUpdateHandlerWithLeaderElection() {
+	ctx := context.Background()
+	client, err := m.apihelper.GetClient()
+	if err != nil {
+		klog.ErrorS(err, "failed to get Kubernetes client")
+		m.Stop()
+	}
+	lock := &resourcelock.LeaseLock{
+		LeaseMeta: metav1.ObjectMeta{
+			Name:      "nfd-master.nfd.kubernetes.io",
+			Namespace: m.namespace,
+		},
+		Client: client.CoordinationV1(),
+		LockConfig: resourcelock.ResourceLockConfig{
+			// add uuid to prevent situation where 2 nfd-master nodes run on same node
+			Identity: m.nodeName + "_" + uuid.NewString(),
+		},
+	}
+	config := leaderelection.LeaderElectionConfig{
+		Lock: lock,
+		// make it configurable?
+		LeaseDuration: m.config.LeaderElection.LeaseDuration.Duration,
+		RetryPeriod:   m.config.LeaderElection.RetryPeriod.Duration,
+		RenewDeadline: m.config.LeaderElection.RenewDeadline.Duration,
+		Callbacks: leaderelection.LeaderCallbacks{
+			OnStartedLeading: func(_ context.Context) {
+				m.nfdAPIUpdateHandler()
+			},
+			OnStoppedLeading: func() {
+				// We lost the lock.
+				klog.ErrorS(err, "leaderelection lock was lost")
+				m.Stop()
+			},
+		},
+	}
+	leaderElector, err := leaderelection.NewLeaderElector(config)
+	if err != nil {
+		klog.ErrorS(err, "couldn't create leader elector")
+		m.Stop()
+	}
+
+	leaderElector.Run(ctx)
+}