Merge pull request #3885 from aledbf/status

Refactor status update

internal/ingress/controller/controller.go

@@ -188,7 +188,11 @@ func (n *NGINXController) syncIngress(interface{}) error {
 		klog.Infof("Backend successfully reloaded.")
 		n.metricCollector.ConfigSuccess(hash, true)
 		n.metricCollector.IncReloadCount()
-		n.metricCollector.SetSSLExpireTime(servers)
+
+		if n.isLeader() {
+			klog.V(2).Infof("Updating ssl expiration metrics.")
+			n.metricCollector.SetSSLExpireTime(servers)
+		}
 	}
 
 	isFirstSync := n.runningConfig.Equal(&ingress.Configuration{})

internal/ingress/controller/nginx.go

@@ -31,6 +31,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"syscall"
 	"text/template"
 	"time"
@@ -115,6 +116,7 @@ func NewNGINXController(config *Configuration, mc metric.Collector, fs file.File
 	if err != nil {
 		klog.Fatalf("unexpected error obtaining pod information: %v", err)
 	}
+	n.podInfo = pod
 
 	n.store = store.New(
 		config.EnableSSLChainCompletion,
@@ -132,15 +134,13 @@ func NewNGINXController(config *Configuration, mc metric.Collector, fs file.File
 		config.DisableCatchAll)
 
 	n.syncQueue = task.NewTaskQueue(n.syncIngress)
+
 	if config.UpdateStatus {
-		n.syncStatus = status.NewStatusSyncer(status.Config{
+		n.syncStatus = status.NewStatusSyncer(pod, status.Config{
 			Client:                 config.Client,
 			PublishService:         config.PublishService,
 			PublishStatusAddress:   config.PublishStatusAddress,
 			IngressLister:          n.store,
-			ElectionID:             config.ElectionID,
-			IngressClass:           class.IngressClass,
-			DefaultIngressClass:    class.DefaultClass,
 			UpdateStatusOnShutdown: config.UpdateStatusOnShutdown,
 			UseNodeInternalIP:      config.UseNodeInternalIP,
 		})
@@ -215,13 +215,15 @@ Error loading new template: %v
 
 // NGINXController describes a NGINX Ingress controller.
 type NGINXController struct {
+	podInfo *k8s.PodInfo
+
 	cfg *Configuration
 
 	recorder record.EventRecorder
 
 	syncQueue *task.Queue
 
-	syncStatus status.Sync
+	syncStatus status.Syncer
 
 	syncRateLimiter flowcontrol.RateLimiter
 
@@ -254,6 +256,8 @@ type NGINXController struct {
 	fileSystem filesystem.Filesystem
 
 	metricCollector metric.Collector
+
+	currentLeader uint32
 }
 
 // Start starts a new NGINX master process running in the foreground.
@@ -262,10 +266,35 @@ func (n *NGINXController) Start() {
 
 	n.store.Run(n.stopCh)
 
-	if n.syncStatus != nil {
-		go n.syncStatus.Run()
+	// we need to use the defined ingress class to allow multiple leaders
+	// in order to update information about ingress status
+	electionID := fmt.Sprintf("%v-%v", n.cfg.ElectionID, class.DefaultClass)
+	if class.IngressClass != "" {
+		electionID = fmt.Sprintf("%v-%v", n.cfg.ElectionID, class.IngressClass)
 	}
 
+	setupLeaderElection(&leaderElectionConfig{
+		Client:     n.cfg.Client,
+		ElectionID: electionID,
+		OnStartedLeading: func(stopCh chan struct{}) {
+			if n.syncStatus != nil {
+				go n.syncStatus.Run(stopCh)
+			}
+
+			n.setLeader(true)
+			n.metricCollector.OnStartedLeading(electionID)
+			// manually update SSL expiration metrics
+			// (to not wait for a reload)
+			n.metricCollector.SetSSLExpireTime(n.runningConfig.Servers)
+		},
+		OnStoppedLeading: func() {
+			n.setLeader(false)
+			n.metricCollector.OnStoppedLeading(electionID)
+		},
+		PodName:      n.podInfo.Name,
+		PodNamespace: n.podInfo.Namespace,
+	})
+
 	cmd := nginxExecCommand()
 
 	// put NGINX in another process group to prevent it
@@ -1099,3 +1128,15 @@ func buildRedirects(servers []*ingress.Server) []*redirect {
 
 	return redirectServers
 }
+
+func (n *NGINXController) setLeader(leader bool) {
+	var i uint32
+	if leader {
+		i = 1
+	}
+	atomic.StoreUint32(&n.currentLeader, i)
+}
+
+func (n *NGINXController) isLeader() bool {
+	return atomic.LoadUint32(&n.currentLeader) != 0
+}

internal/ingress/controller/status.go

@@ -0,0 +1,123 @@
+/*
+Copyright 2019 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package controller
+
+import (
+	"context"
+	"os"
+	"time"
+
+	"k8s.io/klog"
+
+	apiv1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	clientset "k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/kubernetes/scheme"
+	"k8s.io/client-go/tools/leaderelection"
+	"k8s.io/client-go/tools/leaderelection/resourcelock"
+	"k8s.io/client-go/tools/record"
+)
+
+type leaderElectionConfig struct {
+	PodName      string
+	PodNamespace string
+
+	Client clientset.Interface
+
+	ElectionID string
+
+	OnStartedLeading func(chan struct{})
+	OnStoppedLeading func()
+}
+
+func setupLeaderElection(config *leaderElectionConfig) {
+	var elector *leaderelection.LeaderElector
+
+	// start a new context
+	ctx := context.Background()
+
+	var cancelContext context.CancelFunc
+
+	var newLeaderCtx = func(ctx context.Context) context.CancelFunc {
+		// allow to cancel the context in case we stop being the leader
+		leaderCtx, cancel := context.WithCancel(ctx)
+		go elector.Run(leaderCtx)
+		return cancel
+	}
+
+	var stopCh chan struct{}
+	callbacks := leaderelection.LeaderCallbacks{
+		OnStartedLeading: func(ctx context.Context) {
+			klog.V(2).Infof("I am the new leader")
+			stopCh = make(chan struct{})
+
+			if config.OnStartedLeading != nil {
+				config.OnStartedLeading(stopCh)
+			}
+		},
+		OnStoppedLeading: func() {
+			klog.V(2).Info("I am not leader anymore")
+			close(stopCh)
+
+			// cancel the context
+			cancelContext()
+
+			cancelContext = newLeaderCtx(ctx)
+
+			if config.OnStoppedLeading != nil {
+				config.OnStoppedLeading()
+			}
+		},
+		OnNewLeader: func(identity string) {
+			klog.Infof("new leader elected: %v", identity)
+		},
+	}
+
+	broadcaster := record.NewBroadcaster()
+	hostname, _ := os.Hostname()
+
+	recorder := broadcaster.NewRecorder(scheme.Scheme, apiv1.EventSource{
+		Component: "ingress-leader-elector",
+		Host:      hostname,
+	})
+
+	lock := resourcelock.ConfigMapLock{
+		ConfigMapMeta: metav1.ObjectMeta{Namespace: config.PodNamespace, Name: config.ElectionID},
+		Client:        config.Client.CoreV1(),
+		LockConfig: resourcelock.ResourceLockConfig{
+			Identity:      config.PodName,
+			EventRecorder: recorder,
+		},
+	}
+
+	ttl := 30 * time.Second
+	var err error
+
+	elector, err = leaderelection.NewLeaderElector(leaderelection.LeaderElectionConfig{
+		Lock:          &lock,
+		LeaseDuration: ttl,
+		RenewDeadline: ttl / 2,
+		RetryPeriod:   ttl / 4,
+
+		Callbacks: callbacks,
+	})
+	if err != nil {
+		klog.Fatalf("unexpected error starting leader election: %v", err)
+	}
+
+	cancelContext = newLeaderCtx(ctx)
+}