feat/Broker status - version/image extraction improvements

charts/kafka-operator/templates/operator-deployment-with-webhook.yaml

@@ -117,6 +117,8 @@ spec:
       app.kubernetes.io/name: {{ include "kafka-operator.name" . }}
       app.kubernetes.io/instance: {{ .Release.Name }}
       app.kubernetes.io/component: operator
+      app: kafka-operator
+      component: operator
   replicas: {{ .Values.replicaCount }}
   template:
     metadata:
@@ -133,8 +135,8 @@ spec:
         app.kubernetes.io/name: {{ include "kafka-operator.name" . }}
         app.kubernetes.io/instance: {{ .Release.Name }}
         app.kubernetes.io/component: operator
-        app: prometheus
-        component: alertmanager
+        app: kafka-operator
+        component: operator
     spec:
     {{- with .Values.imagePullSecrets }}
       imagePullSecrets:

charts/kafka-operator/templates/podmonitor.yaml

@@ -0,0 +1,32 @@
+{{- if .Values.prometheusMetrics.podMonitor.enabled }}
+kind: PodMonitor
+apiVersion: monitoring.coreos.com/v1
+metadata:
+  name: {{ include "kafka-operator.fullname" . }}
+  namespace: {{ .Release.Namespace | quote }}
+  labels:
+    helm.sh/chart: {{ include "kafka-operator.chart" . }}
+    app.kubernetes.io/name: {{ include "kafka-operator.name" . }}
+    app.kubernetes.io/instance: {{ .Release.Name }}
+    app.kubernetes.io/managed-by: {{ .Release.Service }}
+    app.kubernetes.io/version: {{ .Chart.AppVersion }}
+    app.kubernetes.io/component: operator
+    app: kafka-operator
+    component: operator
+  {{- with .Values.operator.annotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+spec:
+  namespaceSelector:
+    matchNames:
+    - {{ .Release.Namespace }}
+  selector:
+    matchLabels:
+      app: kafka-operator
+      component: operator
+  endpoints:
+  - interval: {{ .Values.prometheusMetrics.interval }}
+    port: metrics
+    path: /metrics
+{{- end }}

charts/kafka-operator/values.yaml

@@ -103,6 +103,10 @@ prometheusMetrics:
       create: true
       # -- ServiceAccount used by prometheus auth proxy
       name: kafka-operator-authproxy
+  podMonitor:
+    # -- If true, create a PodMonitor for Prometheus metrics
+    enabled: false
+    interval: 30s
 
 # -- Health probes configuration
 healthProbes: {}

pkg/resources/kafka/kafka.go

@@ -409,6 +409,7 @@ func (r *Reconciler) Reconcile(log logr.Logger) error {
 	reorderedBrokers := reorderBrokers(runningBrokers, boundPersistentVolumeClaims, r.KafkaCluster.Spec.Brokers, r.KafkaCluster.Status.BrokersState, controllerID, log)
 
 	allBrokerDynamicConfigSucceeded := true
+	brokerStatus := make(map[int32]*banzaiv1beta1.BrokerConfig)
 	for _, broker := range reorderedBrokers {
 		brokerConfig, err := broker.GetBrokerConfig(r.KafkaCluster.Spec)
 		if err != nil {
@@ -449,9 +450,8 @@ func (r *Reconciler) Reconcile(log logr.Logger) error {
 		if err != nil {
 			return err
 		}
-		if err = r.updateStatusWithDockerImageAndVersion(broker.Id, brokerConfig, log); err != nil {
-			return err
-		}
+		brokerStatus[broker.Id] = brokerConfig
+
 		// If dynamic configs can not be set then let the loop continue to the next broker,
 		// after the loop we return error. This solves that case when other brokers could get healthy,
 		// but the loop exits too soon because dynamic configs can not be set.
@@ -474,6 +474,10 @@ func (r *Reconciler) Reconcile(log logr.Logger) error {
 		return err
 	}
 
+	if err := r.updateStatusWithDockerImageAndVersion(brokerStatus, log); err != nil {
+		return err
+	}
+
 	// in case HeadlessServiceEnabled is changed, delete the service that was created by the previous
 	// reconcile flow. The services must be deleted at the end of the reconcile flow after the new services
 	// were created and broker configurations reflecting the new services otherwise the Kafka brokers
@@ -917,21 +921,38 @@ func (r *Reconciler) reconcileKafkaPod(log logr.Logger, desiredPod *corev1.Pod,
 	return nil
 }
 
-func (r *Reconciler) updateStatusWithDockerImageAndVersion(brokerId int32, brokerConfig *banzaiv1beta1.BrokerConfig,
-	log logr.Logger) error {
-	jmxExp := jmxextractor.NewJMXExtractor(r.KafkaCluster.GetNamespace(),
-		r.KafkaCluster.Spec.GetKubernetesClusterDomain(), r.KafkaCluster.GetName(), log)
+type brokerVersionResult struct {
+	brokerID     int32
+	kafkaVersion *banzaiv1beta1.KafkaVersion
+	err          error
+}
 
-	kafkaVersion, err := jmxExp.ExtractDockerImageAndVersion(brokerId, brokerConfig,
-		r.KafkaCluster.Spec.GetClusterImage(), r.KafkaCluster.Spec.HeadlessServiceEnabled)
-	if err != nil {
-		return err
+func (r *Reconciler) updateStatusWithDockerImageAndVersion(brokers map[int32]*banzaiv1beta1.BrokerConfig, log logr.Logger) error {
+	ch := make(chan brokerVersionResult, len(brokers))
+	for brokerID, brokerConfig := range brokers {
+		go func(id int32, cfg *banzaiv1beta1.BrokerConfig) {
+			jmxExp := jmxextractor.NewJMXExtractor(r.KafkaCluster.GetNamespace(),
+				r.KafkaCluster.Spec.GetKubernetesClusterDomain(), r.KafkaCluster.GetName(), log)
+			kv, err := jmxExp.ExtractDockerImageAndVersion(id, cfg,
+				r.KafkaCluster.Spec.GetClusterImage(), r.KafkaCluster.Spec.HeadlessServiceEnabled)
+			if err != nil {
+				ch <- brokerVersionResult{brokerID: id, err: err}
+				return
+			}
+			ch <- brokerVersionResult{brokerID: id, kafkaVersion: kv}
+		}(brokerID, brokerConfig)
 	}
-	err = k8sutil.UpdateBrokerStatus(r.Client, []string{strconv.Itoa(int(brokerId))}, r.KafkaCluster,
-		*kafkaVersion, log)
-	if err != nil {
-		return err
+
+	for range brokers {
+		result := <-ch
+		if result.err != nil {
+			return result.err
+		}
+		if err := k8sutil.UpdateBrokerStatus(r.Client, []string{strconv.Itoa(int(result.brokerID))}, r.KafkaCluster, *result.kafkaVersion, log); err != nil {
+			return err
+		}
 	}
+
 	return nil
 }