fix: split pod controller from workflow controller (#14129)

Signed-off-by: Alan Clucas <[email protected]>

Author: Joibel

workflow/common/common.go

@@ -52,6 +52,9 @@ const (
 	// AnnotationKeyArtifactGCStrategy is listed as an annotation on the Artifact GC Pod to identify
 	// the strategy whose artifacts are being deleted
 	AnnotationKeyArtifactGCStrategy = workflow.WorkflowFullName + "/artifact-gc-strategy"
+	// AnnotationKeyPodGCStrategy is listed as an annotation on the Pod
+	// the strategy for the pod, in case the pod is orphaned from its workflow
+	AnnotationKeyPodGCStrategy = workflow.WorkflowFullName + "/pod-gc-strategy"
 
 	// LabelKeyControllerInstanceID is the label the controller will carry forward to workflows/pod labels
 	// for the purposes of workflow segregation

workflow/controller/agent.go

@@ -10,7 +10,6 @@ import (
 	apierr "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	"k8s.io/client-go/tools/cache"
 	"k8s.io/utils/ptr"
 
 	"github.com/argoproj/argo-workflows/v3/errors"
@@ -113,16 +112,12 @@ func (woc *wfOperationCtx) createAgentPod(ctx context.Context) (*apiv1.Pod, erro
 	podName := woc.getAgentPodName()
 	log := woc.log.WithField("podName", podName)
 
-	obj, exists, err := woc.controller.podInformer.GetStore().Get(cache.ExplicitKey(woc.wf.Namespace + "/" + podName))
+	pod, err := woc.controller.PodController.GetPod(woc.wf.Namespace, podName)
 	if err != nil {
 		return nil, fmt.Errorf("failed to get pod from informer store: %w", err)
 	}
-	if exists {
-		existing, ok := obj.(*apiv1.Pod)
-		if ok {
-			log.WithField("podPhase", existing.Status.Phase).Debug("Skipped pod creation: already exists")
-			return existing, nil
-		}
+	if pod != nil {
+		return pod, nil
 	}
 
 	certVolume, certVolumeMount, err := woc.getCertVolumeMount(ctx, common.CACertificatesVolumeMountName)
@@ -197,7 +192,7 @@ func (woc *wfOperationCtx) createAgentPod(ctx context.Context) (*apiv1.Pod, erro
 	agentMainCtr.Name = common.MainContainerName
 	agentMainCtr.Args = append([]string{"agent", "main"}, woc.getExecutorLogOpts()...)
 
-	pod := &apiv1.Pod{
+	pod = &apiv1.Pod{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      podName,
 			Namespace: woc.wf.ObjectMeta.Namespace,

workflow/controller/artifact_gc.go

@@ -503,7 +503,7 @@ func (woc *wfOperationCtx) createArtifactGCPod(ctx context.Context, strategy wfv
 // go through any GC pods that are already running and may have completed
 func (woc *wfOperationCtx) processArtifactGCCompletion(ctx context.Context) error {
 	// check if any previous Artifact GC Pods completed
-	pods, err := woc.controller.podInformer.GetIndexer().ByIndex(indexes.WorkflowIndex, woc.wf.GetNamespace()+"/"+woc.wf.GetName())
+	pods, err := woc.controller.PodController.GetPodsByIndex(indexes.WorkflowIndex, woc.wf.GetNamespace()+"/"+woc.wf.GetName())
 	if err != nil {
 		return fmt.Errorf("failed to get pods from informer: %w", err)
 	}