fix: the pod of the custom ops is always pending (#8456)

apecloud · Nov 14, 2024 · 7fa68e4 · 7fa68e4
1 parent a16b2eb
commit 7fa68e4
Show file tree

Hide file tree

Showing 4 changed files with 34 additions and 2 deletions.
diff --git a/controllers/apps/operations/custom/action_exec.go b/controllers/apps/operations/custom/action_exec.go
@@ -168,10 +168,14 @@ func (e *ExecAction) buildExecPodSpec(actionCtx ActionContext,
 		}, execAction.Command...),
 	}
 	intctrlutil.InjectZeroResourcesLimitsIfEmpty(container)
+	toleration, err := getTolerations(e.Cluster, e.Comp)
+	if err != nil {
+		return nil, err
+	}
 	return &corev1.PodSpec{
 		Containers: []corev1.Container{*container},
 		// tolerate all taints
-		Tolerations:      e.Comp.Tolerations,
+		Tolerations:      toleration,
 		ImagePullSecrets: intctrlutil.BuildImagePullSecrets(),
 	}, nil
 }
diff --git a/controllers/apps/operations/custom/action_workload.go b/controllers/apps/operations/custom/action_workload.go
@@ -182,7 +182,11 @@ func (w *WorkloadAction) buildPodSpec(actionCtx ActionContext,
 		podSpec.RestartPolicy = corev1.RestartPolicyNever
 	}
 	if len(podSpec.Tolerations) == 0 {
-		podSpec.Tolerations = w.Comp.Tolerations
+		toleration, err := getTolerations(w.Cluster, w.Comp)
+		if err != nil {
+			return nil, err
+		}
+		podSpec.Tolerations = toleration
 	}
 	switch {
 	case w.OpsRequest.Spec.CustomOps.ServiceAccountName != nil:

diff --git a/controllers/apps/operations/custom/utils.go b/controllers/apps/operations/custom/utils.go
@@ -35,6 +35,7 @@ import (
 	"github.com/apecloud/kubeblocks/pkg/common"
 	"github.com/apecloud/kubeblocks/pkg/constant"
 	"github.com/apecloud/kubeblocks/pkg/controller/component"
+	"github.com/apecloud/kubeblocks/pkg/controller/scheduling"
 	intctrlutil "github.com/apecloud/kubeblocks/pkg/controllerutil"
 )
 
@@ -383,3 +384,14 @@ func getNameFromObjectKey(objectKey string) string {
 	}
 	return objectKey
 }
+
+func getTolerations(cluster *appsv1alpha1.Cluster, compSpec *appsv1alpha1.ClusterComponentSpec) ([]corev1.Toleration, error) {
+	schedulePolicy, err := scheduling.BuildSchedulingPolicy(cluster, compSpec)
+	if err != nil {
+		return nil, err
+	}
+	if schedulePolicy == nil {
+		return nil, nil
+	}
+	return schedulePolicy.Tolerations, nil
+}
diff --git a/controllers/apps/operations/custom_test.go b/controllers/apps/operations/custom_test.go
@@ -108,6 +108,11 @@ var _ = Describe("CustomOps", func() {
 				GetObject()
 
 			cluster = testapps.NewClusterFactory(testCtx.DefaultNamespace, clusterName, clusterDefinitionName, clusterVersionName).
+				SetSchedulingPolicy(appsv1alpha1.SchedulingPolicy{
+					Tolerations: []corev1.Toleration{
+						{Operator: corev1.TolerationOpExists, Key: "test"},
+					},
+				}).
 				WithRandomName().AddComponentV2(consensusComp, componentDefObj.Name).SetReplicas(1).Create(&testCtx).GetObject()
 
 			fullCompName := constant.GenerateClusterComponentName(cluster.Name, consensusComp)
@@ -201,6 +206,8 @@ var _ = Describe("CustomOps", func() {
 
 			By("mock job is completed, expect for ops phase is Succeed")
 			job := &jobList.Items[0]
+			Expect(job.Spec.Template.Spec.Tolerations).Should(HaveLen(1))
+			Expect(job.Spec.Template.Spec.Tolerations[0].Key).Should(Equal("test"))
 			patchJobPhase(job, batchv1.JobComplete)
 			By("reconcile once and make the action succeed")
 			_, err = GetOpsManager().Reconcile(reqCtx, k8sClient, opsResource)
@@ -235,6 +242,11 @@ var _ = Describe("CustomOps", func() {
 		It("Test custom ops with sharding cluster", func() {
 			By("init environment for sharding cluster")
 			cluster = testapps.NewClusterFactory(testCtx.DefaultNamespace, "", "", "").
+				SetSchedulingPolicy(appsv1alpha1.SchedulingPolicy{
+					Tolerations: []corev1.Toleration{
+						{Operator: corev1.TolerationOpExists, Key: "test"},
+					},
+				}).
 				WithRandomName().AddShardingSpecV2(consensusComp, compDefName).Create(&testCtx).GetObject()
 			opsResource.Cluster = cluster