Prevent MySQLCluster controller startup slowdown in large clusters

.github/instructions/focused-tests.instructions.md

@@ -0,0 +1,36 @@
+---
+name: Focused Test Execution
+description: "Use when running, debugging, or suggesting focused tests in MOCO. Covers Makefile targets, FOCUS, envtest, CI workflow/action patterns, and avoiding raw go test detours."
+applyTo: "**/*.go"
+---
+
+# Focused test execution
+
+When running a specific MOCO test case, first use the same entry points as CI and the Makefile. Do not start with raw `go test` unless there is a clear reason.
+
+## Preferred focused commands
+
+- Controller envtest: `make controller-envtest FOCUS='part of spec name'`
+- Clustering envtest: `make clustering-envtest FOCUS='part of spec name'`
+- API envtest: `make api-envtest FOCUS='part of spec name'`
+- Backup envtest: `make backup-envtest FOCUS='part of spec name'`
+- Full envtest set: `make envtest`
+- Small package tests/install/vet/gofmt: `make test`
+- MySQL integration tests: `make test-bkop MYSQL_VERSION='8.4.8'` or `make test-dbop MYSQL_VERSION='8.4.8'`
+- E2E and upgrade tests should follow `.github/actions/e2e/action.yaml` and `.github/actions/upgrade/action.yaml`; use the `e2e/Makefile` targets from inside `e2e/`.
+
+## Why Make targets first
+
+- `Makefile` exports `ENVTEST_KUBERNETES_VERSION` and `ENVTEST_ASSETS_DIR`; raw controller/envtest commands can fail with `could not parse "v" as version` if these are missing.
+- The `*-envtest` targets run with the repo's expected flags: `-race`, `-count 1`, `-ginkgo.randomize-all`, `-ginkgo.v`, and usually `-ginkgo.fail-fast`.
+- `controller-envtest` also sets `DEBUG_CONTROLLER=1`.
+- CI calls `make lint`, `make test`, `make check-generate`, and `make envtest` after `.github/actions/setup-aqua`, so local verification should mirror those paths when practical.
+
+## If raw `go test` is necessary
+
+Only use raw `go test` for a quick diagnostic. For envtest packages, carry over the Makefile environment explicitly:
+
+```sh
+ENVTEST_KUBERNETES_VERSION=1.35.0 ENVTEST_ASSETS_DIR="$PWD/bin" \
+  go test ./controllers -run TestAPIs -ginkgo.focus 'part of spec name'
+```

clustering/manager_test.go

@@ -137,10 +137,9 @@ var _ = Describe("manager", func() {
 		ctx, cancel := context.WithCancel(ctx)
 		stopFunc = cancel
 		go func() {
+			defer GinkgoRecover()
 			err := mgr.Start(ctx)
-			if err != nil {
-				panic(err)
-			}
+			Expect(err).NotTo(HaveOccurred())
 		}()
 		time.Sleep(10 * time.Millisecond)
 	})

cmd/moco-controller/cmd/run.go

@@ -102,6 +102,8 @@ func subMain(ns, addr string, port int) error {
 	clusterMgr := clustering.NewClusterManager(config.interval, mgr, opf, af, clusterLog)
 	defer clusterMgr.StopAll()
 
+	ctx := ctrl.SetupSignalHandler()
+
 	if err = (&controllers.MySQLClusterReconciler{
 		Client:                        mgr.GetClient(),
 		Scheme:                        mgr.GetScheme(),
@@ -117,7 +119,7 @@ func subMain(ns, addr string, port int) error {
 		MaxConcurrentReconciles:       config.maxConcurrentReconciles,
 		MySQLConfigMapHistoryLimit:    config.mySQLConfigMapHistoryLimit,
 		DisableDefaultSecurityContext: config.disableDefaultSecurityContext,
-	}).SetupWithManager(mgr); err != nil {
+	}).SetupWithManager(ctx, mgr); err != nil {
 		setupLog.Error(err, "unable to create controller", "controller", "MySQLCluster")
 		return err
 	}
@@ -169,7 +171,6 @@ func subMain(ns, addr string, port int) error {
 	metrics.Register(k8smetrics.Registry)
 
 	setupLog.Info("starting manager")
-	ctx := ctrl.SetupSignalHandler()
 	go reloader.Run(ctx, 1*time.Hour)
 	if err := mgr.Start(ctx); err != nil {
 		setupLog.Error(err, "problem running manager")

controllers/mysqlcluster_controller.go

@@ -1950,7 +1950,7 @@ func setControllerReferenceWithPVC(cluster *mocov1beta2.MySQLCluster, pvc *corev
 }
 
 // SetupWithManager sets up the controller with the Manager.
-func (r *MySQLClusterReconciler) SetupWithManager(mgr ctrl.Manager) error {
+func (r *MySQLClusterReconciler) SetupWithManager(ctx context.Context, mgr ctrl.Manager) error {
 	certHandler := handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, a client.Object) []reconcile.Request {
 		// the certificate name is formatted as "moco-agent-<cluster.Namespace>.<cluster.Name>"
 		if a.GetNamespace() != r.SystemNamespace {
@@ -1970,38 +1970,45 @@ func (r *MySQLClusterReconciler) SetupWithManager(mgr ctrl.Manager) error {
 		}
 	})
 
-	configMapHandler := handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, a client.Object) []reconcile.Request {
+	requestsForIndexedClusters := func(ctx context.Context, namespace, field, value string) []reconcile.Request {
 		clusters := &mocov1beta2.MySQLClusterList{}
-		if err := r.List(ctx, clusters, client.InNamespace(a.GetNamespace())); err != nil {
+		if err := r.List(ctx, clusters, client.InNamespace(namespace), client.MatchingFields{field: value}); err != nil {
 			return nil
 		}
-		var req []reconcile.Request
+
+		req := make([]reconcile.Request, 0, len(clusters.Items))
 		for _, c := range clusters.Items {
-			if c.Spec.MySQLConfigMapName == nil {
-				continue
-			}
-			if *c.Spec.MySQLConfigMapName == a.GetName() {
-				req = append(req, reconcile.Request{NamespacedName: client.ObjectKeyFromObject(&c)})
-			}
+			req = append(req, reconcile.Request{NamespacedName: client.ObjectKeyFromObject(&c)})
 		}
 		return req
-	})
+	}
 
-	backupPolicyHandler := handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, a client.Object) []reconcile.Request {
-		clusters := &mocov1beta2.MySQLClusterList{}
-		if err := r.List(ctx, clusters, client.InNamespace(a.GetNamespace())); err != nil {
+	if err := mgr.GetFieldIndexer().IndexField(ctx, &mocov1beta2.MySQLCluster{}, "spec.mysqlConfigMapName", func(rawObj client.Object) []string {
+		c := rawObj.(*mocov1beta2.MySQLCluster)
+		if c.Spec.MySQLConfigMapName == nil {
 			return nil
 		}
-		var req []reconcile.Request
-		for _, c := range clusters.Items {
-			if c.Spec.BackupPolicyName == nil {
-				continue
-			}
-			if *c.Spec.BackupPolicyName == a.GetName() {
-				req = append(req, reconcile.Request{NamespacedName: client.ObjectKeyFromObject(&c)})
-			}
+		return []string{*c.Spec.MySQLConfigMapName}
+	}); err != nil {
+		return fmt.Errorf("failed to index MySQLCluster by spec.mysqlConfigMapName: %w", err)
+	}
+
+	configMapHandler := handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, a client.Object) []reconcile.Request {
+		return requestsForIndexedClusters(ctx, a.GetNamespace(), "spec.mysqlConfigMapName", a.GetName())
+	})
+
+	if err := mgr.GetFieldIndexer().IndexField(ctx, &mocov1beta2.MySQLCluster{}, "spec.backupPolicyName", func(rawObj client.Object) []string {
+		c := rawObj.(*mocov1beta2.MySQLCluster)
+		if c.Spec.BackupPolicyName == nil {
+			return nil
 		}
-		return req
+		return []string{*c.Spec.BackupPolicyName}
+	}); err != nil {
+		return fmt.Errorf("failed to index MySQLCluster by spec.backupPolicyName: %w", err)
+	}
+
+	backupPolicyHandler := handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, a client.Object) []reconcile.Request {
+		return requestsForIndexedClusters(ctx, a.GetNamespace(), "spec.backupPolicyName", a.GetName())
 	})
 
 	return ctrl.NewControllerManagedBy(mgr).

controllers/mysqlcluster_controller_slowdown_test.go

@@ -0,0 +1,140 @@
+package controllers
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"time"
+
+	mocov1beta2 "github.com/cybozu-go/moco/api/v1beta2"
+	"github.com/cybozu-go/moco/pkg/constants"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/client-go/util/retry"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/config"
+	metricsserver "sigs.k8s.io/controller-runtime/pkg/metrics/server"
+)
+
+func testNewStoppedMySQLCluster(ns, name, configMapName string) *mocov1beta2.MySQLCluster {
+	cluster := testNewMySQLCluster(ns)
+	cluster.Name = name
+	cluster.Finalizers = nil
+	cluster.Annotations = map[string]string{constants.AnnReconciliationStopped: "true"}
+	cluster.Spec.MySQLConfigMapName = new(configMapName)
+	return cluster
+}
+
+func testDeleteAllMySQLClusters(ctx context.Context, ns string) {
+	clusters := &mocov1beta2.MySQLClusterList{}
+	err := k8sClient.List(ctx, clusters, client.InNamespace(ns))
+	ExpectWithOffset(1, err).NotTo(HaveOccurred())
+	for _, cluster := range clusters.Items {
+		if len(cluster.Finalizers) == 0 {
+			continue
+		}
+		err := retry.RetryOnConflict(retry.DefaultRetry, func() error {
+			latest := &mocov1beta2.MySQLCluster{}
+			if err := k8sClient.Get(ctx, client.ObjectKeyFromObject(&cluster), latest); err != nil {
+				return client.IgnoreNotFound(err)
+			}
+			latest.Finalizers = nil
+			return k8sClient.Update(ctx, latest)
+		})
+		ExpectWithOffset(1, err).NotTo(HaveOccurred())
+	}
+	err = k8sClient.DeleteAllOf(ctx, &mocov1beta2.MySQLCluster{}, client.InNamespace(ns))
+	ExpectWithOffset(1, err).NotTo(HaveOccurred())
+	EventuallyWithOffset(1, func() (int, error) {
+		clusters := &mocov1beta2.MySQLClusterList{}
+		if err := k8sClient.List(ctx, clusters, client.InNamespace(ns)); err != nil {
+			return 0, err
+		}
+		return len(clusters.Items), nil
+	}).Should(BeZero())
+}
+
+var _ = Describe("MySQLCluster reconciler startup", func() {
+	const largeScaleStartupObjectCount = 1000
+
+	BeforeEach(func(ctx SpecContext) {
+		testDeleteAllMySQLClusters(ctx, "test")
+		err := k8sClient.DeleteAllOf(ctx, &corev1.ConfigMap{}, client.InNamespace("test"))
+		Expect(err).NotTo(HaveOccurred())
+	})
+
+	AfterEach(func(ctx SpecContext) {
+		testDeleteAllMySQLClusters(ctx, "test")
+		err := k8sClient.DeleteAllOf(ctx, &corev1.ConfigMap{}, client.InNamespace("test"))
+		Expect(err).NotTo(HaveOccurred())
+	})
+
+	It("should start with many MySQLClusters and ConfigMaps", func(ctx SpecContext) {
+		By("creating many MySQLClusters and ConfigMaps before the manager starts")
+		for i := range largeScaleStartupObjectCount {
+			configMapName := fmt.Sprintf("startup-config-%04d", i)
+			cm := &corev1.ConfigMap{
+				ObjectMeta: metav1.ObjectMeta{Namespace: "test", Name: configMapName},
+				Data:       map[string]string{"my.cnf": "[mysqld]\n"},
+			}
+			err := k8sClient.Create(ctx, cm)
+			Expect(err).NotTo(HaveOccurred())
+
+			cluster := testNewStoppedMySQLCluster("test", fmt.Sprintf("startup-%04d", i), configMapName)
+			err = k8sClient.Create(ctx, cluster)
+			Expect(err).NotTo(HaveOccurred())
+		}
+
+		mgr, err := ctrl.NewManager(cfg, ctrl.Options{
+			Scheme:         scheme,
+			LeaderElection: false,
+			Metrics: metricsserver.Options{
+				BindAddress: "0",
+			},
+			Controller: config.Controller{
+				SkipNameValidation: new(true),
+			},
+		})
+		Expect(err).ToNot(HaveOccurred())
+
+		mockMgr := &mockManager{
+			clusters: make(map[string]struct{}),
+		}
+		mysqlr := &MySQLClusterReconciler{
+			Client:                     mgr.GetClient(),
+			Scheme:                     scheme,
+			Recorder:                   mgr.GetEventRecorderFor("moco-controller"),
+			SystemNamespace:            testMocoSystemNamespace,
+			ClusterManager:             mockMgr,
+			AgentImage:                 testAgentImage,
+			BackupImage:                testBackupImage,
+			FluentBitImage:             testFluentBitImage,
+			ExporterImage:              testExporterImage,
+			MySQLConfigMapHistoryLimit: 2,
+		}
+		err = mysqlr.SetupWithManager(ctx, mgr)
+		Expect(err).ToNot(HaveOccurred())
+
+		startCtx, cancel := context.WithCancel(ctx)
+		defer cancel()
+		go func() {
+			defer GinkgoRecover()
+			err := mgr.Start(startCtx)
+			Expect(err).NotTo(HaveOccurred())
+		}()
+
+		By("waiting for the controller to start and process an initial cluster event")
+		start := time.Now()
+		Eventually(func() error {
+			if !mockMgr.getKeys()["test/startup-0000"] {
+				GinkgoLogr.Info("controller has not processed the startup cluster yet")
+				return errors.New("controller has not processed the startup cluster yet")
+			}
+			return nil
+		}, 15*time.Second, 2*time.Second).Should(Succeed())
+		GinkgoLogr.Info("controller has processed the startup cluster", "elapsed", time.Since(start))
+	})
+})

controllers/mysqlcluster_controller_test.go

@@ -189,16 +189,15 @@ var _ = Describe("MySQLCluster reconciler", func() {
 			ExporterImage:              testExporterImage,
 			MySQLConfigMapHistoryLimit: 2,
 		}
-		err = mysqlr.SetupWithManager(mgr)
+		err = mysqlr.SetupWithManager(ctx, mgr)
 		Expect(err).ToNot(HaveOccurred())
 
 		ctx, cancel := context.WithCancel(ctx)
 		stopFunc = cancel
 		go func() {
+			defer GinkgoRecover()
 			err := mgr.Start(ctx)
-			if err != nil {
-				panic(err)
-			}
+			Expect(err).NotTo(HaveOccurred())
 		}()
 		time.Sleep(100 * time.Millisecond)
 	})

controllers/partition_controller_test.go

@@ -213,10 +213,9 @@ func setupNewManager(ctx context.Context, updateInterval time.Duration) context.
 
 	ctx, cancel := context.WithCancel(ctx)
 	go func() {
+		defer GinkgoRecover()
 		err := mgr.Start(ctx)
-		if err != nil {
-			panic(err)
-		}
+		Expect(err).NotTo(HaveOccurred())
 	}()
 	return cancel
 }

controllers/pod_watcher_test.go

@@ -83,10 +83,9 @@ var _ = Describe("PodWatcher", func() {
 		ctx, cancel := context.WithCancel(ctx)
 		stopFunc = cancel
 		go func() {
+			defer GinkgoRecover()
 			err := mgr.Start(ctx)
-			if err != nil {
-				panic(err)
-			}
+			Expect(err).NotTo(HaveOccurred())
 		}()
 		time.Sleep(100 * time.Millisecond)
 	})