Enable profiling for operator pod

Signed-off-by: alexander-demicev <[email protected]>

Author: alexander-demicev

cmd/main.go

@@ -104,11 +104,11 @@ func InitFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&watchNamespace, "namespace", "",
 		"Namespace that the controller watches to reconcile cluster-api objects. If unspecified, the controller watches for cluster-api objects across all namespaces.")
 
-	fs.StringVar(&profilerAddress, "profiler-address", "",
-		"Bind address to expose the pprof profiler (e.g. localhost:6060)")
+	fs.StringVar(&profilerAddress, "profiler-address", "localhost:6060",
+		"Bind address to expose the pprof profiler (e.g. localhost:6060). Set to empty string to disable.")
 
 	fs.BoolVar(&enableContentionProfiling, "contention-profiling", false,
-		"Enable block profiling")
+		"Enable block profiling when profiler is active")
 
 	fs.IntVar(&concurrencyNumber, "concurrency", 1,
 		"Number of core resources to process simultaneously")
@@ -148,7 +148,7 @@ func main() {
 		}
 	}
 
-	if enableContentionProfiling {
+	if enableContentionProfiling && profilerAddress != "" {
 		goruntime.SetBlockProfileRate(1)
 	}
 

config/manager/manager.yaml

@@ -25,6 +25,9 @@ spec:
         image: controller:latest
         name: manager
         ports:
+          - containerPort: 6060
+            name: profiler
+            protocol: TCP
           - containerPort: 8443
             name: metrics
             protocol: TCP

docs/book/src/04_developer/03_profiling.md

@@ -0,0 +1,32 @@
+# Profiling
+
+This section explains how to set up and use debugging endpoints like pprof for the Cluster API Operator.
+
+### Configuring Helm Values
+
+Profiling is enabled by default but some values can be customized. You can set the following values in your `values.yaml` file:
+
+```yaml
+profilerAddress: ":6060"
+contentionProfiling: true
+```
+
+Install with these custom values using [Helm chart installation methods](../02_installation/04_helm-chart-installation.md)
+
+### Enabling Port-Forwarding
+
+To access the pprof server on your local machine, run:
+
+```bash
+kubectl port-forward deployment/capi-operator -n <namespace> 6060
+```
+
+This will forward port 6060 from the container to your local machine.
+
+### Running pprof Commands
+
+With port-forwarding in place, you can run pprof commands like this:
+
+```bash
+go tool pprof http://localhost:6060/debug/pprof/profile?seconds=30
+```

hack/charts/cluster-api-operator/templates/deployment.yaml

@@ -85,6 +85,12 @@ spec:
         {{- if .retryPeriod }}
         - --leader-elect-retry-period={{ .retryPeriod }}
         {{- end }}
+        {{- if $.Values.profilerAddress }}
+        - --profiler-address=localhost{{ $.Values.profilerAddress }}
+        {{- end }}
+        {{- if $.Values.contentionProfiling }}
+        - --contention-profiling={{ $.Values.contentionProfiling }}
+        {{- end }}
         {{- end }}
         command:
         - /manager
@@ -94,18 +100,21 @@ spec:
         imagePullPolicy: {{ .Values.image.manager.pullPolicy }}
         name: manager
         ports:
+        {{- if $.Values.profilerAddress }}
+        {{- $profilerPort := $.Values.profilerAddress | toString | trimPrefix ":" | int }}
+        - containerPort: {{ $profilerPort }}
+          name: profiler
+          protocol: TCP
+        {{- end }}
         - containerPort: 9443
           name: webhook-server
           protocol: TCP
-          {{- if $.Values.diagnosticsAddress }}
-          {{- $diagnosticsPort := $.Values.diagnosticsAddress }}
-          {{- if contains ":" $diagnosticsPort -}}
-          {{ $diagnosticsPort = ( split ":" $.Values.diagnosticsAddress)._1 | int }}
-          {{- end }}
-        - containerPort: {{ $diagnosticsPort | int }}
-          name: metrics
+        {{- if .Values.diagnosticsAddress }}
+        {{- $diagnosticsPort := .Values.diagnosticsAddress | toString | trimPrefix ":" | int }}
+        - containerPort: {{ $diagnosticsPort }}
+          name: diagnostics
           protocol: TCP
-          {{- end }}
+        {{- end }}
         {{- with .Values.resources.manager }}
         resources:
         {{- toYaml . | nindent 12 }}
@@ -123,15 +132,15 @@ spec:
         {{- toYaml . | nindent 12 }}
         {{- end }}
         terminationMessagePolicy: FallbackToLogsOnError
-        {{- $healthAddr := $.Values.healthAddr }}
-        {{- if contains ":" $healthAddr -}}
-        {{ $healthAddr = ( split ":" $.Values.healthAddr)._1 | int }}
+        {{- $healthPort := 9440 }}
+        {{- if .Values.healthAddr }}
+        {{- $healthPort = .Values.healthAddr | toString | trimPrefix ":" | int }}
         {{- end }}
         livenessProbe:
           failureThreshold: 3
           httpGet:
             path: /healthz
-            port: {{ $healthAddr | default 9440 }}
+            port: {{ $healthPort }}
             scheme: HTTP
           initialDelaySeconds: 15
           periodSeconds: 20
@@ -141,7 +150,7 @@ spec:
           failureThreshold: 3
           httpGet:
             path: /readyz
-            port: {{ $healthAddr | default 9440 }}
+            port: {{ $healthPort }}
             scheme: HTTP
           initialDelaySeconds: 5
           periodSeconds: 10

hack/charts/cluster-api-operator/values.yaml

@@ -45,6 +45,8 @@ env:
   manager: []
 diagnosticsAddress: ":8443"
 healthAddr: ":9440"
+profilerAddress: ":6060"
+contentionProfiling: false
 insecureDiagnostics: false
 watchConfigSecret: false
 imagePullSecrets: {}

test/e2e/resources/full-chart-install.yaml

@@ -22077,17 +22077,21 @@ spec:
         - --health-addr=:9440
         - --diagnostics-address=:8443
         - --leader-elect=true
+        - --profiler-address=localhost:6060
         command:
         - /manager
         image: "gcr.io/k8s-staging-capi-operator/cluster-api-operator:dev"
         imagePullPolicy: IfNotPresent
         name: manager
         ports:
+        - containerPort: 6060
+          name: profiler
+          protocol: TCP
         - containerPort: 9443
           name: webhook-server
           protocol: TCP
         - containerPort: 8443
-          name: metrics
+          name: diagnostics
           protocol: TCP
         resources:
             limits: