fix: always select all available pods when running action

apis/apps/v1/componentdefinition_types.go

@@ -1536,6 +1536,9 @@ type ComponentLifecycleActions struct {
 	// during events such as planned maintenance or when performing stop, shutdown, restart, or upgrade operations.
 	// In a typical consensus system, this action is used to transfer leader role to another replica.
 	//
+	// When a pod is about to be updated, a switchover action will be triggered for it. So addon implementation must determine
+	// if the pod's current role needs to be transferred.
+	//
 	// The container executing this action has access to following variables:
 	//
 	// - KB_SWITCHOVER_CANDIDATE_NAME: The name of the pod of the new role's candidate, which may not be specified (empty).
@@ -1717,6 +1720,9 @@ type ComponentLifecycleActions struct {
 // designed to be invoked at predetermined points within the lifecycle of a Component instance.
 // It provides a modular and extensible way to customize a Component's behavior through the execution of defined actions.
 //
+// Action should be idempotent if possible. In some circumstances (for example, an UPDATE to an k8s object fails due to concurrent updates),
+// the action may be retried even after a success.
+//
 // Available Action triggers include:
 //
 //   - `postProvision`: Defines the hook to be executed after the creation of a Component,

config/crd/bases/apps.kubeblocks.io_componentdefinitions.yaml

@@ -9802,6 +9802,10 @@ spec:
                       In a typical consensus system, this action is used to transfer leader role to another replica.
 
 
+                      When a pod is about to be updated, a switchover action will be triggered for it. So addon implementation must determine
+                      if the pod's current role needs to be transferred.
+
+
                       The container executing this action has access to following variables:
 
 

controllers/apps/component/transformer_component_account_provision.go

@@ -142,7 +142,7 @@ func (t *componentAccountProvisionTransformer) lifecycleAction(transCtx *compone
 		return nil, err
 	}
 	lfa, err := lifecycle.New(synthesizedComp.Namespace, synthesizedComp.ClusterName, synthesizedComp.Name,
-		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods...)
+		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods)
 	if err != nil {
 		return nil, err
 	}

controllers/apps/component/transformer_component_post_provision.go

@@ -102,7 +102,7 @@ func (t *componentPostProvisionTransformer) lifecycleAction4Component(transCtx *
 		return nil, fmt.Errorf("has no pods to running the post-provision action")
 	}
 	return lifecycle.New(synthesizedComp.Namespace, synthesizedComp.ClusterName, synthesizedComp.Name,
-		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods...)
+		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods)
 }
 
 func checkPostProvisionDone(transCtx *componentTransformContext) bool {

controllers/apps/component/transformer_component_pre_terminate.go

@@ -167,7 +167,7 @@ func (t *componentPreTerminateTransformer) lifecycleAction4Component(transCtx *c
 		return nil, fmt.Errorf("has no pods to running the pre-terminate action")
 	}
 	return lifecycle.New(synthesizedComp.Namespace, synthesizedComp.ClusterName, synthesizedComp.Name,
-		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods...)
+		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, nil, pods)
 }
 
 func (t *componentPreTerminateTransformer) synthesizedComponent(transCtx *componentTransformContext, compDef *appsv1.ComponentDefinition) (*component.SynthesizedComponent, error) {

controllers/apps/component/transformer_component_workload_ops.go

@@ -222,7 +222,7 @@ func (r *componentWorkloadOps) leaveMemberForPod(pod *corev1.Pod, pods []*corev1
 	}
 
 	lfa, err := lifecycle.New(synthesizedComp.Namespace, synthesizedComp.ClusterName, synthesizedComp.Name,
-		lifecycleActions, synthesizedComp.TemplateVars, pod, pods...)
+		lifecycleActions, synthesizedComp.TemplateVars, pod, pods)
 	if err != nil {
 		return err
 	}
@@ -387,7 +387,7 @@ func (r *componentWorkloadOps) joinMember4ScaleOut() error {
 func (r *componentWorkloadOps) joinMemberForPod(pod *corev1.Pod, pods []*corev1.Pod) error {
 	synthesizedComp := r.synthesizeComp
 	lfa, err := lifecycle.New(synthesizedComp.Namespace, synthesizedComp.ClusterName, synthesizedComp.Name,
-		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, pod, pods...)
+		synthesizedComp.LifecycleActions, synthesizedComp.TemplateVars, pod, pods)
 	if err != nil {
 		return err
 	}

controllers/apps/componentdefinition_controller.go

@@ -28,7 +28,6 @@ import (
 	"slices"
 	"strings"
 
-	"github.com/pkg/errors"
 	k8sappsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -259,7 +258,7 @@ func (r *ComponentDefinitionReconciler) validateVars(cli client.Client, rctx int
 			continue
 		}
 		if err := component.ValidateDefNameRegexp(compDef); err != nil {
-			return errors.Wrapf(err, "invalid reference to component definition name pattern: %s", compDef)
+			return fmt.Errorf("invalid reference to component definition name pattern: %s: %w", compDef, err)
 		}
 	}
 	return nil

deploy/helm/crds/apps.kubeblocks.io_componentdefinitions.yaml

@@ -9802,6 +9802,10 @@ spec:
                       In a typical consensus system, this action is used to transfer leader role to another replica.
 
 
+                      When a pod is about to be updated, a switchover action will be triggered for it. So addon implementation must determine
+                      if the pod's current role needs to be transferred.
+
+
                       The container executing this action has access to following variables:
 
 

docs/developer_docs/api-reference/cluster.md

@@ -5867,6 +5867,8 @@ Action
 This approach aims to minimize downtime and maintain availability
 during events such as planned maintenance or when performing stop, shutdown, restart, or upgrade operations.
 In a typical consensus system, this action is used to transfer leader role to another replica.</p>
+<p>When a pod is about to be updated, a switchover action will be triggered for it. So addon implementation must determine
+if the pod&rsquo;s current role needs to be transferred.</p>
 <p>The container executing this action has access to following variables:</p>
 <ul>
 <li>KB_SWITCHOVER_CANDIDATE_NAME: The name of the pod of the new role&rsquo;s candidate, which may not be specified (empty).</li>

pkg/controller/instance/reconciler_update.go

@@ -195,7 +195,8 @@ func (r *updateReconciler) switchover(tree *kubebuilderx.ObjectTree, inst *workl
 		return nil
 	}
 
-	lfa, err := newLifecycleAction(inst, nil, pod)
+	// FIXME: select all available pods
+	lfa, err := newLifecycleAction(inst, []*corev1.Pod{pod}, pod)
 	if err != nil {
 		return err
 	}

pkg/controller/instance/utils.go

@@ -351,19 +351,15 @@ func copyAndMerge(oldObj, newObj client.Object) client.Object {
 	}
 }
 
-func newLifecycleAction(inst *workloads.Instance, objects []client.Object, pod *corev1.Pod) (lifecycle.Lifecycle, error) {
+func newLifecycleAction(inst *workloads.Instance, pods []*corev1.Pod, pod *corev1.Pod) (lifecycle.Lifecycle, error) {
 	var (
 		clusterName      = inst.Labels[constant.AppInstanceLabelKey]
 		compName         = inst.Labels[constant.KBAppComponentLabelKey]
 		lifecycleActions = &kbappsv1.ComponentLifecycleActions{
 			Switchover:  inst.Spec.LifecycleActions.Switchover,
 			Reconfigure: inst.Spec.LifecycleActions.Reconfigure,
 		}
-		pods []*corev1.Pod
 	)
-	for i := range objects {
-		pods = append(pods, objects[i].(*corev1.Pod))
-	}
 	return lifecycle.New(inst.Namespace, clusterName, compName,
-		lifecycleActions, inst.Spec.LifecycleActions.TemplateVars, pod, pods...)
+		lifecycleActions, inst.Spec.LifecycleActions.TemplateVars, pod, pods)
 }

pkg/controller/instanceset/reconciler_update.go

@@ -279,7 +279,7 @@ func (r *updateReconciler) switchover(tree *kubebuilderx.ObjectTree, its *worklo
 		return nil
 	}
 
-	lfa, err := newLifecycleAction(its, nil, pod)
+	lfa, err := newLifecycleAction(its, tree, pod)
 	if err != nil {
 		return err
 	}
@@ -318,7 +318,7 @@ func (r *updateReconciler) reconfigureConfig(tree *kubebuilderx.ObjectTree, its
 		itsCopy.Spec.LifecycleActions = &workloads.LifecycleActions{}
 	}
 	itsCopy.Spec.LifecycleActions.Reconfigure = config.Reconfigure
-	lfa, err := newLifecycleAction(itsCopy, nil, pod)
+	lfa, err := newLifecycleAction(itsCopy, tree, pod)
 	if err != nil {
 		return err
 	}

pkg/controller/instanceset/utils.go

@@ -30,11 +30,11 @@ import (
 	"k8s.io/apimachinery/pkg/util/intstr"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/utils/integer"
-	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	kbappsv1 "github.com/apecloud/kubeblocks/apis/apps/v1"
 	workloads "github.com/apecloud/kubeblocks/apis/workloads/v1"
 	"github.com/apecloud/kubeblocks/pkg/constant"
+	"github.com/apecloud/kubeblocks/pkg/controller/kubebuilderx"
 	"github.com/apecloud/kubeblocks/pkg/controller/lifecycle"
 )
 
@@ -203,19 +203,19 @@ func getMemberUpdateStrategy(its *workloads.InstanceSet) workloads.MemberUpdateS
 	return updateStrategy
 }
 
-func newLifecycleAction(its *workloads.InstanceSet, objects []client.Object, pod *corev1.Pod) (lifecycle.Lifecycle, error) {
+func newLifecycleAction(its *workloads.InstanceSet, tree *kubebuilderx.ObjectTree, pod *corev1.Pod) (lifecycle.Lifecycle, error) {
 	var (
 		clusterName      = its.Labels[constant.AppInstanceLabelKey]
 		compName         = its.Labels[constant.KBAppComponentLabelKey]
 		lifecycleActions = &kbappsv1.ComponentLifecycleActions{
 			Switchover:  its.Spec.LifecycleActions.Switchover,
 			Reconfigure: its.Spec.LifecycleActions.Reconfigure,
 		}
-		pods []*corev1.Pod
 	)
-	for i := range objects {
-		pods = append(pods, objects[i].(*corev1.Pod))
+	pods := make([]*corev1.Pod, 0)
+	for _, object := range tree.List(&corev1.Pod{}) {
+		pods = append(pods, object.(*corev1.Pod))
 	}
 	return lifecycle.New(its.Namespace, clusterName, compName,
-		lifecycleActions, its.Spec.LifecycleActions.TemplateVars, pod, pods...)
+		lifecycleActions, its.Spec.LifecycleActions.TemplateVars, pod, pods)
 }

pkg/controller/lifecycle/kbagent.go

@@ -50,8 +50,8 @@ type kbagent struct {
 	compName         string
 	lifecycleActions *appsv1.ComponentLifecycleActions
 	templateVars     map[string]string
-	pods             []*corev1.Pod
-	pod              *corev1.Pod
+	pods             []*corev1.Pod // available pods to execute action, typically all pods of the component
+	pod              *corev1.Pod   // target pod the action will be effective on. Some actions, like postprovision, does not care about this field.
 }
 
 var _ Lifecycle = &kbagent{}

pkg/controller/lifecycle/lifecycle.go

@@ -60,16 +60,13 @@ type Lifecycle interface {
 }
 
 func New(namespace, clusterName, compName string, lifecycleActions *appsv1.ComponentLifecycleActions,
-	templateVars map[string]string, pod *corev1.Pod, pods ...*corev1.Pod) (Lifecycle, error) {
-	if pod == nil && len(pods) == 0 {
-		return nil, fmt.Errorf("either pod or pods must be provided to call lifecycle actions")
+	templateVars map[string]string, pod *corev1.Pod, pods []*corev1.Pod) (Lifecycle, error) {
+	if len(pods) == 0 {
+		return nil, fmt.Errorf("pods must be provided to call lifecycle actions")
 	}
 	if pod == nil {
 		pod = pods[0]
 	}
-	if len(pods) == 0 {
-		pods = []*corev1.Pod{pod}
-	}
 	return &kbagent{
 		namespace:        namespace,
 		clusterName:      clusterName,