Make build operator leader durations configurable

cmd/manager/main.go

@@ -100,16 +100,19 @@ func main() {
 	}
 	defer r.Unset()
 
-	c := buildconfig.NewDefaultConfig()
-	if err := c.SetConfigFromEnv(); err != nil {
+	buildCfg := buildconfig.NewDefaultConfig()
+	if err := buildCfg.SetConfigFromEnv(); err != nil {
 		ctxlog.Error(ctx, err, "")
 		os.Exit(1)
 	}
 
-	mgr, err := controller.NewManager(ctx, c, cfg, manager.Options{
+	mgr, err := controller.NewManager(ctx, buildCfg, cfg, manager.Options{
 		LeaderElection:          true,
 		LeaderElectionID:        "build-operator-lock",
-		LeaderElectionNamespace: "default",
+		LeaderElectionNamespace: buildCfg.ManagerOptions.LeaderElectionNamespace,
+		LeaseDuration:           buildCfg.ManagerOptions.LeaseDuration,
+		RenewDeadline:           buildCfg.ManagerOptions.RenewDeadline,
+		RetryPeriod:             buildCfg.ManagerOptions.RetryPeriod,
 		Namespace:               "",
 		MetricsBindAddress:      fmt.Sprintf("%s:%d", metricsHost, metricsPort),
 	})
@@ -120,7 +123,7 @@ func main() {
 
 	// Add the Metrics Service
 	addMetrics(ctx, cfg, namespace)
-	buildMetrics.InitPrometheus(c)
+	buildMetrics.InitPrometheus(buildCfg)
 
 	ctxlog.Info(ctx, "Starting the Cmd.")
 

deploy/operator.yaml

@@ -25,6 +25,10 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
+            - name: BUILD_OPERATOR_LEADER_ELECTION_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: POD_NAME
               valueFrom:
                 fieldRef:

docs/configuration.md

@@ -0,0 +1,20 @@
+<!--
+Copyright The Shipwright Contributors
+
+SPDX-License-Identifier: Apache-2.0
+-->
+
+# Configuration
+
+The `build-operator` is installed into Kubernetes with reasonable defaults. However, there are some settings that can be overridden using environment variables in [`operator.yaml`](../deploy/operator.yaml).
+
+The following environment variables are available:
+
+| Environment Variable | Description |
+| --- | --- |
+| `CTX_TIMEOUT` | Override the default context timeout used for all Custom Resource Definition reconciliation operations. |
+| `KANIKO_CONTAINER_IMAGE` | Specify the Kaniko container image to be used instead of the default, for example `gcr.io/kaniko-project/executor:v1.0.1`. |
+| `BUILD_OPERATOR_LEADER_ELECTION_NAMESPACE` |  Set the namespace to be used to store the `build-operator` lock, by default it is in the same namespace as the operator itself. |
+| `BUILD_OPERATOR_LEASE_DURATION` |  Override the `LeaseDuration`, which is the duration that non-leader candidates will wait to force acquire leadership. |
+| `BUILD_OPERATOR_RENEW_DEADLINE` |  Override the `RenewDeadline`, which is the duration that the acting master will retry refreshing leadership before giving up. |
+| `BUILD_OPERATOR_RETRY_PERIOD` |  Override the `RetryPeriod`, which is the duration the LeaderElector clients should wait between tries of actions. |

pkg/config/config.go

@@ -31,6 +31,13 @@ const (
 
 	// environment variable to enable histogram labels
 	prometheusHistogramEnabledLabelsEnvVar = "PROMETHEUS_HISTOGRAM_ENABLED_LABELS"
+
+	leaderElectionNamespaceDefault = "default"
+	leaderElectionNamespaceEnvVar  = "BUILD_OPERATOR_LEADER_ELECTION_NAMESPACE"
+
+	leaseDurationEnvVar = "BUILD_OPERATOR_LEASE_DURATION"
+	renewDeadlineEnvVar = "BUILD_OPERATOR_RENEW_DEADLINE"
+	retryPeriodEnvVar   = "BUILD_OPERATOR_RETRY_PERIOD"
 )
 
 var (
@@ -46,6 +53,7 @@ type Config struct {
 	CtxTimeOut           time.Duration
 	KanikoContainerImage string
 	Prometheus           PrometheusConfig
+	ManagerOptions       ManagerOptions
 }
 
 // PrometheusConfig contains the specific configuration for the
@@ -56,6 +64,14 @@ type PrometheusConfig struct {
 	HistogramEnabledLabels            []string
 }
 
+// ManagerOptions contains configurable options for the build operator manager
+type ManagerOptions struct {
+	LeaderElectionNamespace string
+	LeaseDuration           *time.Duration
+	RenewDeadline           *time.Duration
+	RetryPeriod             *time.Duration
+}
+
 // NewDefaultConfig returns a new Config, with context timeout and default Kaniko image.
 func NewDefaultConfig() *Config {
 	return &Config{
@@ -66,6 +82,9 @@ func NewDefaultConfig() *Config {
 			BuildRunEstablishDurationBuckets:  metricBuildRunEstablishDurationBuckets,
 			BuildRunRampUpDurationBuckets:     metricBuildRunRampUpDurationBuckets,
 		},
+		ManagerOptions: ManagerOptions{
+			LeaderElectionNamespace: leaderElectionNamespaceDefault,
+		},
 	}
 }
 
@@ -97,6 +116,22 @@ func (c *Config) SetConfigFromEnv() error {
 
 	c.Prometheus.HistogramEnabledLabels = strings.Split(os.Getenv(prometheusHistogramEnabledLabelsEnvVar), ",")
 
+	if leaderElectionNamespace := os.Getenv(leaderElectionNamespaceEnvVar); leaderElectionNamespace != "" {
+		c.ManagerOptions.LeaderElectionNamespace = leaderElectionNamespace
+	}
+
+	if err := updateBuildOperatorDurationOption(&c.ManagerOptions.LeaseDuration, leaseDurationEnvVar); err != nil {
+		return err
+	}
+
+	if err := updateBuildOperatorDurationOption(&c.ManagerOptions.RenewDeadline, renewDeadlineEnvVar); err != nil {
+		return err
+	}
+
+	if err := updateBuildOperatorDurationOption(&c.ManagerOptions.RetryPeriod, retryPeriodEnvVar); err != nil {
+		return err
+	}
+
 	return nil
 }
 
@@ -126,3 +161,16 @@ func updateBucketsConfig(buckets *[]float64, envVarName string) error {
 
 	return nil
 }
+
+func updateBuildOperatorDurationOption(d **time.Duration, envVarName string) error {
+	if value := os.Getenv(envVarName); value != "" {
+		valueDuration, err := time.ParseDuration(value)
+		if err != nil {
+			return err
+		}
+
+		*d = &valueDuration
+	}
+
+	return nil
+}

pkg/config/config_test.go

@@ -55,6 +55,27 @@ var _ = Describe("Config", func() {
 				Expect(config.Prometheus.HistogramEnabledLabels).To(Equal([]string{"namespace", "strategy"}))
 			})
 		})
+
+		It("should allow for an override of the operator leader election namespace using an environment variable", func() {
+			var overrides = map[string]string{"BUILD_OPERATOR_LEADER_ELECTION_NAMESPACE": "build-operator"}
+			configWithEnvVariableOverrides(overrides, func(config *Config) {
+				Expect(config.ManagerOptions.LeaderElectionNamespace).To(Equal("build-operator"))
+			})
+		})
+
+		It("should allow for an override of the operator leader election times using environment variables", func() {
+			var overrides = map[string]string{
+				"BUILD_OPERATOR_LEASE_DURATION": "42s",
+				"BUILD_OPERATOR_RENEW_DEADLINE": "32s",
+				"BUILD_OPERATOR_RETRY_PERIOD":   "10s",
+			}
+
+			configWithEnvVariableOverrides(overrides, func(config *Config) {
+				Expect(*config.ManagerOptions.LeaseDuration).To(Equal(time.Duration(42 * time.Second)))
+				Expect(*config.ManagerOptions.RenewDeadline).To(Equal(time.Duration(32 * time.Second)))
+				Expect(*config.ManagerOptions.RetryPeriod).To(Equal(time.Duration(10 * time.Second)))
+			})
+		})
 	})
 })