opt: skip transfer leader if only one PD; update some CRD comments (#…

…6050)
pingcap · Jan 23, 2025 · 3d2a1b5 · 3d2a1b5
1 parent c75ea80
commit 3d2a1b5
Show file tree

Hide file tree

Showing 7 changed files with 101 additions and 13 deletions.
diff --git a/apis/core/v1alpha1/cluster_types.go b/apis/core/v1alpha1/cluster_types.go
@@ -113,10 +113,10 @@ type TLSCluster struct {
 	//        - use the K8s built-in certificate signing system signed certificates: https://kubernetes.io/docs/tasks/tls/managing-tls-in-a-cluster/
 	//        - or use cert-manager signed certificates: https://cert-manager.io/
 	//   2. Create one secret object for one component group which contains the certificates created above.
-	//      The name of this Secret must be: <clusterName>-<groupName>-cluster-secret.
-	//        For PD: kubectl create secret generic <clusterName>-<pd-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
-	//        For TiKV: kubectl create secret generic <clusterName>-<tikv-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
-	//        For TiDB: kubectl create secret generic <clusterName>-<tidb-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+	//      The name of this Secret must be: <groupName>-<componentName>-cluster-secret.
+	//        For PD: kubectl create secret generic <groupName>-pd-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+	//        For TiKV: kubectl create secret generic <groupName>-tikv-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+	//        For TiDB: kubectl create secret generic <groupName>-tidb-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
 	//        For Client: kubectl create secret generic <clusterName>-cluster-client-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
 	//        Same for other components.
 	// +optional

diff --git a/apis/core/v1alpha1/pd_types.go b/apis/core/v1alpha1/pd_types.go
@@ -278,8 +278,9 @@ type PDGroupSpec struct {
 	Replicas *int32           `json:"replicas"`
 	Version  string           `json:"version"`
 
-	// Bootstrapped means that pd cluster has been bootstrapped
-	// It's no need to initialize a new cluster
+	// Bootstrapped means that pd cluster has been bootstrapped,
+	// and there is no need to initialize a new cluster.
+	// In other words, this PD group will just join an existing cluster.
 	// Normally, this field is automatically changed by operator.
 	// If it's true, it cannot be set to false for security
 	Bootstrapped bool `json:"bootstrapped,omitempty"`

diff --git a/manifests/crd/core.pingcap.com_clusters.yaml b/manifests/crd/core.pingcap.com_clusters.yaml
@@ -113,10 +113,10 @@ spec:
                              - use the K8s built-in certificate signing system signed certificates: https://kubernetes.io/docs/tasks/tls/managing-tls-in-a-cluster/
                              - or use cert-manager signed certificates: https://cert-manager.io/
                         2. Create one secret object for one component group which contains the certificates created above.
-                           The name of this Secret must be: <clusterName>-<groupName>-cluster-secret.
-                             For PD: kubectl create secret generic <clusterName>-<pd-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
-                             For TiKV: kubectl create secret generic <clusterName>-<tikv-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
-                             For TiDB: kubectl create secret generic <clusterName>-<tidb-groupName>-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+                           The name of this Secret must be: <groupName>-<componentName>-cluster-secret.
+                             For PD: kubectl create secret generic <groupName>-pd-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+                             For TiKV: kubectl create secret generic <groupName>-tikv-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
+                             For TiDB: kubectl create secret generic <groupName>-tidb-cluster-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
                              For Client: kubectl create secret generic <clusterName>-cluster-client-secret --namespace=<namespace> --from-file=tls.crt=<path/to/tls.crt> --from-file=tls.key=<path/to/tls.key> --from-file=ca.crt=<path/to/ca.crt>
                              Same for other components.
                     type: boolean

diff --git a/manifests/crd/core.pingcap.com_pdgroups.yaml b/manifests/crd/core.pingcap.com_pdgroups.yaml
@@ -53,8 +53,9 @@ spec:
             properties:
               bootstrapped:
                 description: |-
-                  Bootstrapped means that pd cluster has been bootstrapped
-                  It's no need to initialize a new cluster
+                  Bootstrapped means that pd cluster has been bootstrapped,
+                  and there is no need to initialize a new cluster.
+                  In other words, this PD group will just join an existing cluster.
                   Normally, this field is automatically changed by operator.
                   If it's true, it cannot be set to false for security
                 type: boolean

diff --git a/pkg/controllers/pd/tasks/pod.go b/pkg/controllers/pd/tasks/pod.go
@@ -112,6 +112,11 @@ func preDeleteCheck(
 ) (bool, error) {
 	// TODO: add quorum check. After stopping this pd, quorum should not be lost
 
+	if len(peers) == 1 {
+		logger.Info("no need to transfer leader because there is only one pd")
+		return false, nil
+	}
+
 	if isLeader {
 		peer := LongestHealthPeer(pd, peers)
 		if peer == "" {

diff --git a/pkg/controllers/pd/tasks/pod_test.go b/pkg/controllers/pd/tasks/pod_test.go
@@ -216,6 +216,17 @@ func TestTaskPod(t *testing.T) {
 							}
 							return obj
 						}),
+						fake.FakeObj("aaa-yyy", func(obj *v1alpha1.PD) *v1alpha1.PD {
+							obj.Spec.Version = fakeVersion
+							obj.Status.Conditions = []metav1.Condition{
+								{
+									Type:               v1alpha1.CondHealth,
+									Status:             metav1.ConditionFalse,
+									LastTransitionTime: metav1.Now(),
+								},
+							}
+							return obj
+						}),
 					},
 				},
 				Healthy:  true,
@@ -224,6 +235,46 @@ func TestTaskPod(t *testing.T) {
 
 			expectedStatus: task.SFail,
 		},
+		{
+			desc: "pod spec changed, pod is healthy, pod is leader, only one pd",
+			state: &ReconcileContext{
+				State: &state{
+					pd: fake.FakeObj("aaa-xxx", func(obj *v1alpha1.PD) *v1alpha1.PD {
+						obj.Spec.Version = fakeVersion
+						obj.Status.Conditions = []metav1.Condition{
+							{
+								Type:   v1alpha1.CondHealth,
+								Status: metav1.ConditionTrue,
+							},
+						}
+						return obj
+					}),
+					cluster: fake.FakeObj[v1alpha1.Cluster]("aaa"),
+					pod: fake.FakeObj("aaa-pd-xxx", func(obj *corev1.Pod) *corev1.Pod {
+						return obj
+					}),
+					pds: []*v1alpha1.PD{
+						fake.FakeObj("aaa-xxx", func(obj *v1alpha1.PD) *v1alpha1.PD {
+							obj.Spec.Version = fakeVersion
+							obj.Status.Conditions = []metav1.Condition{
+								{
+									Type:               v1alpha1.CondHealth,
+									Status:             metav1.ConditionTrue,
+									LastTransitionTime: metav1.Now(),
+								},
+							}
+							return obj
+						}),
+					},
+				},
+				Healthy:  true,
+				IsLeader: true,
+			},
+
+			expectUpdatedPod:         false,
+			expectedPodIsTerminating: true,
+			expectedStatus:           task.SWait,
+		},
 		{
 			desc: "pod spec changed, pod is healthy, pod is not leader",
 			state: &ReconcileContext{

diff --git a/tests/e2e/pd/pd.go b/tests/e2e/pd/pd.go
@@ -99,7 +99,7 @@ var _ = ginkgo.Describe("PD", label.PD, func() {
 	})
 
 	ginkgo.Context("Update", label.P0, label.Update, func() {
-		ginkgo.It("support rolling update PD by change config file", func(ctx context.Context) {
+		ginkgo.It("support rolling update PD by change config file with 3 replicas", func(ctx context.Context) {
 			pdg := data.NewPDGroup(
 				f.Namespace.Name,
 				data.WithReplicas[*runtime.PDGroup](3),
@@ -128,6 +128,36 @@ var _ = ginkgo.Describe("PD", label.PD, func() {
 			cancel()
 			<-ch
 		})
+
+		ginkgo.It("support update PD by change config file with 1 replica", func(ctx context.Context) {
+			pdg := data.NewPDGroup(
+				f.Namespace.Name,
+				data.WithReplicas[*runtime.PDGroup](1),
+			)
+
+			ginkgo.By("Create PDGroup")
+			f.Must(f.Client.Create(ctx, pdg))
+			f.WaitForPDGroupReady(ctx, pdg)
+
+			patch := client.MergeFrom(pdg.DeepCopy())
+			pdg.Spec.Template.Spec.Config = `log.level = 'warn'`
+
+			nctx, cancel := context.WithCancel(ctx)
+			ch := make(chan struct{})
+			go func() {
+				defer close(ch)
+				defer ginkgo.GinkgoRecover()
+				f.Must(waiter.WaitPodsRollingUpdateOnce(nctx, f.Client, runtime.FromPDGroup(pdg), waiter.LongTaskTimeout))
+			}()
+
+			changeTime := time.Now()
+			ginkgo.By("Change config of the PDGroup")
+			f.Must(f.Client.Patch(ctx, pdg, patch))
+			f.Must(waiter.WaitForPodsRecreated(ctx, f.Client, runtime.FromPDGroup(pdg), changeTime, waiter.LongTaskTimeout))
+			f.WaitForPDGroupReady(ctx, pdg)
+			cancel()
+			<-ch
+		})
 	})
 
 	ginkgo.Context("Suspend", label.P0, label.Suspend, func() {