feat(minReadySeconds): support minReadySeconds for PD (#5827)

pkg/manager/member/pd_upgrader.go

@@ -15,6 +15,7 @@ package member
 
 import (
 	"fmt"
+	"strconv"
 
 	"github.com/pingcap/tidb-operator/pkg/apis/pingcap/v1alpha1"
 	"github.com/pingcap/tidb-operator/pkg/controller"
@@ -23,13 +24,19 @@ import (
 
 	"github.com/pingcap/advanced-statefulset/client/apis/apps/v1/helper"
 	apps "k8s.io/api/apps/v1"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/klog/v2"
 	podutil "k8s.io/kubernetes/pkg/api/v1/pod"
 )
 
 const (
 	// set this PD clustre annotation to true to fail cluster upgrade if PD loose the quorum during one pod restart
 	annoKeyPDPeersCheck = "tidb.pingcap.com/pd-check-quorum-before-upgrade"
+
+	// TODO: change to use minReadySeconds in sts spec
+	// See https://kubernetes.io/blog/2021/08/27/minreadyseconds-statefulsets/
+	annoKeyPDMinReadySeconds = "tidb.pingcap.com/pd-min-ready-seconds"
 )
 
 type pdUpgrader struct {
@@ -79,6 +86,17 @@ func (u *pdUpgrader) gracefulUpgrade(tc *v1alpha1.TidbCluster, oldSet *apps.Stat
 		return nil
 	}
 
+	minReadySeconds := 0
+	s, ok := tc.Annotations[annoKeyPDMinReadySeconds]
+	if ok {
+		i, err := strconv.Atoi(s)
+		if err != nil {
+			klog.Warningf("tidbcluster: [%s/%s] annotation %s should be an integer: %v", ns, tcName, annoKeyPDMinReadySeconds, err)
+		} else {
+			minReadySeconds = i
+		}
+	}
+
 	mngerutils.SetUpgradePartition(newSet, *oldSet.Spec.UpdateStrategy.RollingUpdate.Partition)
 	podOrdinals := helper.GetPodOrdinals(*oldSet.Spec.Replicas, oldSet).List()
 	for _i := len(podOrdinals) - 1; _i >= 0; _i-- {
@@ -95,8 +113,13 @@ func (u *pdUpgrader) gracefulUpgrade(tc *v1alpha1.TidbCluster, oldSet *apps.Stat
 		}
 
 		if revision == tc.Status.PD.StatefulSet.UpdateRevision {
-			if !podutil.IsPodReady(pod) {
-				return controller.RequeueErrorf("tidbcluster: [%s/%s]'s upgraded pd pod: [%s] is not ready", ns, tcName, podName)
+			if !podutil.IsPodAvailable(pod, int32(minReadySeconds), metav1.Now()) {
+				readyCond := podutil.GetPodReadyCondition(pod.Status)
+				if readyCond == nil || readyCond.Status != corev1.ConditionTrue {
+					return controller.RequeueErrorf("tidbcluster: [%s/%s]'s upgraded pd pod: [%s] is not ready", ns, tcName, podName)
+
+				}
+				return controller.RequeueErrorf("tidbcluster: [%s/%s]'s upgraded pd pod: [%s] is not available, last transition time is %v", ns, tcName, podName, readyCond.LastTransitionTime)
 			}
 			if member, exist := tc.Status.PD.Members[PdName(tc.Name, i, tc.Namespace, tc.Spec.ClusterDomain, tc.Spec.AcrossK8s)]; !exist || !member.Health {
 				return controller.RequeueErrorf("tidbcluster: [%s/%s]'s pd upgraded pod: [%s] is not health", ns, tcName, podName)

pkg/manager/member/pd_upgrader_test.go

@@ -314,6 +314,30 @@ func TestPDUpgraderUpgrade(t *testing.T) {
 				g.Expect(newSet.Spec.UpdateStrategy.RollingUpdate.Partition).To(Equal(pointer.Int32Ptr(1)))
 			},
 		},
+		{
+			name: "upgraded pod is ready but not available",
+			changeFn: func(tc *v1alpha1.TidbCluster) {
+				tc.Status.PD.Synced = true
+				if tc.Annotations == nil {
+					tc.Annotations = map[string]string{}
+				}
+				// 5min is enough for unit test
+				tc.Annotations[annoKeyTiDBMinReadySeconds] = "300"
+			},
+			changePods: func(pods []*corev1.Pod) {
+				pods[1].Status.Conditions[0].LastTransitionTime = metav1.Now()
+			},
+			changeOldSet:       nil,
+			transferLeaderErr:  false,
+			pdPeersAreUnstable: true,
+			errExpectFn: func(g *GomegaWithT, err error) {
+				g.Expect(err).NotTo(HaveOccurred())
+			},
+			expectFn: func(g *GomegaWithT, tc *v1alpha1.TidbCluster, newSet *apps.StatefulSet) {
+				g.Expect(tc.Status.PD.Phase).To(Equal(v1alpha1.UpgradePhase))
+				g.Expect(newSet.Spec.UpdateStrategy.RollingUpdate.Partition).To(Equal(pointer.Int32Ptr(1)))
+			},
+		},
 	}
 
 	for i := range tests {