Ansible: add leader election flags, use lease-based election (#3457)

* Ansible: add leader election flags, use lease-based election
* enable leader election by default

changelog/fragments/ansible-leader-election.yaml

@@ -0,0 +1,9 @@
+# entries is a list of entries to include in
+# release notes and/or the migration guide
+entries:
+  - description: >
+      In the Ansible operator, use controller-runtime's lease-based leader
+      election instead of SDK's leader-for-life mechnism.
+
+    kind: "change"
+    breaking: false

cmd/ansible-operator/main.go

@@ -15,7 +15,6 @@
 package main
 
 import (
-	"context"
 	"fmt"
 	"os"
 	"runtime"
@@ -40,7 +39,6 @@ import (
 	"github.com/operator-framework/operator-sdk/pkg/ansible/proxy/controllermap"
 	"github.com/operator-framework/operator-sdk/pkg/ansible/runner"
 	"github.com/operator-framework/operator-sdk/pkg/ansible/watches"
-	"github.com/operator-framework/operator-sdk/pkg/leader"
 	"github.com/operator-framework/operator-sdk/pkg/log/zap"
 	sdkVersion "github.com/operator-framework/operator-sdk/version"
 )
@@ -71,11 +69,26 @@ func main() {
 		os.Exit(1)
 	}
 
+	// Deprecated: OPERATOR_NAME environment variable is an artifact of the
+	// legacy operator-sdk project scaffolding. Flag `--leader-election-id`
+	// should be used instead.
+	if operatorName, found := os.LookupEnv("OPERATOR_NAME"); found {
+		log.Info("Environment variable OPERATOR_NAME has been deprecated, use --leader-election-id instead.")
+		if pflag.CommandLine.Lookup("leader-election-id").Changed {
+			log.Info("Ignoring OPERATOR_NAME environment variable since --leader-election-id is set")
+		} else {
+			f.LeaderElectionID = operatorName
+		}
+	}
+
 	// Set default manager options
 	// TODO: probably should expose the host & port as an environment variables
 	options := manager.Options{
-		HealthProbeBindAddress: fmt.Sprintf("%s:%d", metricsHost, healthProbePort),
-		MetricsBindAddress:     f.MetricsAddress,
+		HealthProbeBindAddress:  fmt.Sprintf("%s:%d", metricsHost, healthProbePort),
+		MetricsBindAddress:      f.MetricsAddress,
+		LeaderElection:          f.EnableLeaderElection,
+		LeaderElectionID:        f.LeaderElectionID,
+		LeaderElectionNamespace: f.LeaderElectionNamespace,
 		NewClient: func(cache cache.Cache, config *rest.Config, options client.Options) (client.Client, error) {
 			c, err := client.New(config, options)
 			if err != nil {
@@ -152,19 +165,6 @@ func main() {
 		}, w.Blacklist)
 	}
 
-	operatorName, err := k8sutil.GetOperatorName()
-	if err != nil {
-		log.Error(err, "Failed to get the operator name")
-		os.Exit(1)
-	}
-
-	// Become the leader before proceeding
-	err = leader.Become(context.TODO(), operatorName+"-lock")
-	if err != nil {
-		log.Error(err, "Failed to become leader.")
-		os.Exit(1)
-	}
-
 	err = mgr.AddHealthzCheck("ping", healthz.Ping)
 	if err != nil {
 		log.Error(err, "Failed to add Healthz check.")

internal/scaffold/ansible/deploy_operator.go

@@ -58,6 +58,9 @@ spec:
         - name: [[.ProjectName]]
           # Replace this with the built image name
           image: "REPLACE_IMAGE"
+          args:
+          - "--enable-leader-election"
+          - "--leader-election-id=[[.ProjectName]]"
           imagePullPolicy: "Always"
           volumeMounts:
           - mountPath: /tmp/ansible-operator/runner

pkg/ansible/flags/flag.go

@@ -28,11 +28,14 @@ type Flags struct {
 	ReconcilePeriod         time.Duration
 	WatchesFile             string
 	InjectOwnerRef          bool
+	EnableLeaderElection    bool
 	MaxConcurrentReconciles int
 	AnsibleVerbosity        int
 	AnsibleRolesPath        string
 	AnsibleCollectionsPath  string
 	MetricsAddress          string
+	LeaderElectionID        string
+	LeaderElectionNamespace string
 }
 
 const AnsibleRolesPathEnvVar = "ANSIBLE_ROLES_PATH"
@@ -81,5 +84,22 @@ func (f *Flags) AddTo(flagSet *pflag.FlagSet) {
 		":8080",
 		"The address the metric endpoint binds to",
 	)
-
+	flagSet.BoolVar(&f.EnableLeaderElection,
+		"enable-leader-election",
+		false,
+		"Enable leader election for controller manager. Enabling this will"+
+			" ensure there is only one active controller manager.",
+	)
+	flagSet.StringVar(&f.LeaderElectionID,
+		"leader-election-id",
+		"",
+		"Name of the configmap that is used for holding the leader lock.",
+	)
+	flagSet.StringVar(&f.LeaderElectionNamespace,
+		"leader-election-namespace",
+		"",
+		"Namespace in which to create the leader election configmap for"+
+			" holding the leader lock (required if running locally with leader"+
+			" election enabled).",
+	)
 }