[task-controller] parallel reconciliation settable from cli

Author: justonedev1

control-operator/cmd/task-manager/main.go

@@ -57,11 +57,14 @@ func main() {
 	var metricsAddr string
 	var enableLeaderElection bool
 	var probeAddr string
+	var maxConcurrentReconciles int
 	flag.StringVar(&metricsAddr, "metrics-bind-address", ":9082", "The address the metric endpoint binds to.")
 	flag.StringVar(&probeAddr, "health-probe-bind-address", ":9083", "The address the probe endpoint binds to.")
 	flag.BoolVar(&enableLeaderElection, "leader-elect", false,
 		"Enable leader election for controller manager. "+
 			"Enabling this will ensure there is only one active controller manager.")
+	flag.IntVar(&maxConcurrentReconciles, "max-concurrent-reconciles", 1,
+		"Number of Tasks that are reconciled in parallel. Reconciles of a single Task are always serialized.")
 	opts := zap.Options{
 		Development: true,
 	}
@@ -90,10 +93,11 @@ func main() {
 	}
 
 	if err = (&controller.TaskReconciler{
-		Client:   mgr.GetClient(),
-		Scheme:   mgr.GetScheme(),
-		Recorder: mgr.GetEventRecorderFor("task-controller"),
-		NodeName: nodeName,
+		Client:                  mgr.GetClient(),
+		Scheme:                  mgr.GetScheme(),
+		Recorder:                mgr.GetEventRecorderFor("task-controller"),
+		NodeName:                nodeName,
+		MaxConcurrentReconciles: maxConcurrentReconciles,
 	}).SetupWithManager(mgr); err != nil {
 		setupLog.Error(err, "unable to create controller", "controller", "Task")
 		os.Exit(1)

control-operator/config/manager/task-manager/task-manager.yaml

@@ -33,6 +33,7 @@ spec:
           args:
             - --health-probe-bind-address=:9082
             - --metrics-bind-address=:9083
+            - --max-concurrent-reconciles=5
             - --zap-encoder=json
           env:
             - name: NODE_NAME

control-operator/internal/controller/task_controller.go

@@ -29,6 +29,7 @@ import (
 	"fmt"
 	"reflect"
 	"strings"
+	"sync"
 	"time"
 
 	v1 "k8s.io/api/core/v1"
@@ -53,12 +54,13 @@ import (
 // TaskReconciler reconciles a Task object
 type TaskReconciler struct {
 	client.Client
-	Scheme   *runtime.Scheme
-	Recorder record.EventRecorder
-	NodeName string
+	Scheme                  *runtime.Scheme
+	Recorder                record.EventRecorder
+	NodeName                string
+	MaxConcurrentReconciles int
 }
 
-var clientsForContainers map[string]*OccClient = make(map[string]*OccClient)
+var clientsForContainers sync.Map
 
 const taskFinalizer string = "aliecs.alice.cern/finalizer"
 
@@ -135,7 +137,7 @@ func (r *TaskReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.
 		return ctrl.Result{}, nil
 	}
 
-	if _, exists := clientsForContainers[t.Name]; !exists {
+	if _, exists := clientsForContainers.Load(t.Name); !exists {
 		if existingPod.Status.PodIP == "" {
 			log.Info("pod doesn't have IP yet, we wait for different event")
 			return ctrl.Result{}, nil
@@ -158,12 +160,12 @@ func (r *TaskReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.
 	// on them being implemented
 	if t.Status.State == "" {
 		log.V(1).Info("Status.State is empty, querying container")
-		client, exists := clientsForContainers[t.Name]
+		client, exists := clientsForContainers.Load(t.Name)
 		if !exists {
 			return ctrl.Result{Requeue: true}, nil
 		}
 
-		stateReply, err := client.GetState(ctx)
+		stateReply, err := client.(*OccClient).GetState(ctx)
 		if err != nil {
 			log.Error(err, "Failed to GetState")
 			return ctrl.Result{RequeueAfter: 5 * time.Second}, nil
@@ -187,16 +189,18 @@ func (r *TaskReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.
 
 	// Handle Spec -> gRPC State Sync
 	if t.Status.State != t.Spec.State {
-		client, exists := clientsForContainers[t.Name]
+		clientAny, exists := clientsForContainers.Load(t.Name)
 		if !exists {
 			return ctrl.Result{Requeue: true}, nil
 		}
 
+		client := clientAny.(*OccClient)
+
 		stateReply, err := client.GetState(ctx)
 		if err != nil {
 			log.Info("Failed to get state for sync, retrying in 5s", "error", err.Error())
 			client.Close()
-			delete(clientsForContainers, t.Name)
+			clientsForContainers.Delete(t.Name)
 			return ctrl.Result{RequeueAfter: 5 * time.Second}, nil
 		}
 
@@ -267,7 +271,7 @@ func (r *TaskReconciler) createGRPCConsumer(ctx context.Context, t *aliecsv1alph
 		return ctrl.Result{RequeueAfter: 5 * time.Second}, nil
 	}
 
-	clientsForContainers[t.Name] = client
+	clientsForContainers.Store(t.Name, client)
 
 	if err := r.recordCondition(ctx, t, aliecsv1alpha1.ConditionGRPCConnected, metav1.ConditionTrue, "Connected", fmt.Sprintf("gRPC connection established to %s", addr)); err != nil {
 		return ctrl.Result{}, err
@@ -276,14 +280,14 @@ func (r *TaskReconciler) createGRPCConsumer(ctx context.Context, t *aliecsv1alph
 }
 
 func (r *TaskReconciler) consumeGRPCConsumerIfReady(ctx context.Context, t *aliecsv1alpha1.Task, log logr.Logger) ctrl.Result {
-	client, exists := clientsForContainers[t.Name]
+	client, exists := clientsForContainers.Load(t.Name)
 
 	if !exists {
 		log.Info("didn't found existing client, retrying ", "task", t.Name)
 		return ctrl.Result{RequeueAfter: time.Second}
 	}
 
-	if !client.ConsumeIfReady(ctx) {
+	if !client.(*OccClient).ConsumeIfReady(ctx) {
 		log.Info("gRPC client is not ready, retrying in 5 seconds", "name", t.Name)
 		return ctrl.Result{RequeueAfter: 5 * time.Second}
 	}
@@ -341,12 +345,12 @@ func (r *TaskReconciler) deletePod(ctx context.Context, t *aliecsv1alpha1.Task,
 }
 
 func (*TaskReconciler) cleargRPC(t *aliecsv1alpha1.Task, log logr.Logger) {
-	if client, exists := clientsForContainers[t.Name]; exists {
+	if client, exists := clientsForContainers.Load(t.Name); exists {
 		log.Info("Cleaning up gRPC connection")
-		if err := client.Close(); err != nil {
+		if err := client.(*OccClient).Close(); err != nil {
 			log.Error(err, "Failed to close gRPC client during deletion")
 		}
-		delete(clientsForContainers, t.Name)
+		clientsForContainers.Delete(t.Name)
 		log.Info("gRPC cleaned")
 	}
 }
@@ -431,7 +435,7 @@ func (r *TaskReconciler) SetupWithManager(mgr ctrl.Manager) error {
 			}),
 		)).
 		Owns(&v1.Pod{}).
-		WithOptions(controller.Options{MaxConcurrentReconciles: 1}).
+		WithOptions(controller.Options{MaxConcurrentReconciles: r.MaxConcurrentReconciles}).
 		Complete(r)
 }