feat: revert appproject updates on peer clusters

Signed-off-by: Chetan Banavikalmutt <[email protected]>

Author: chetan-rns

agent/agent.go

@@ -28,6 +28,7 @@ import (
 	kubeappproject "github.com/argoproj-labs/argocd-agent/internal/backend/kubernetes/appproject"
 	kubenamespace "github.com/argoproj-labs/argocd-agent/internal/backend/kubernetes/namespace"
 	kuberepository "github.com/argoproj-labs/argocd-agent/internal/backend/kubernetes/repository"
+	"github.com/argoproj-labs/argocd-agent/internal/cache"
 	"github.com/argoproj-labs/argocd-agent/internal/config"
 	"github.com/argoproj-labs/argocd-agent/internal/event"
 	"github.com/argoproj-labs/argocd-agent/internal/informer"
@@ -49,7 +50,6 @@ import (
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/watch"
 
-	appCache "github.com/argoproj-labs/argocd-agent/internal/cache"
 	"github.com/argoproj/argo-cd/v3/pkg/apis/application/v1alpha1"
 	cacheutil "github.com/argoproj/argo-cd/v3/util/cache"
 	appstatecache "github.com/argoproj/argo-cd/v3/util/cache/appstate"
@@ -104,6 +104,9 @@ type Agent struct {
 
 	cacheRefreshInterval time.Duration
 	clusterCache         *appstatecache.Cache
+
+	// sourceCache is a cache of resources from the source. We use it to revert any changes made to the local resources.
+	sourceCache *cache.SourceCache
 }
 
 const defaultQueueName = "default"
@@ -310,6 +313,8 @@ func NewAgent(ctx context.Context, client *kube.KubernetesClient, namespace stri
 	}
 	a.clusterCache = clusterCache
 
+	a.sourceCache = cache.NewSourceCache()
+
 	return a, nil
 }
 
@@ -319,21 +324,10 @@ func (a *Agent) Start(ctx context.Context) error {
 	a.context = infCtx
 	a.cancelFn = cancelFn
 
-	// For managed-agent we need to maintain a cache to keep applications in sync with last known state of
-	// principal in case agent is disconnected with principal or application in managed-cluster is modified.
+	// For managed-agent we need to maintain a cache to keep resources in sync with last known state of
+	// principal in case agent is disconnected with principal or resources in managed-cluster are modified.
 	if a.mode == types.AgentModeManaged {
-		log().Infof("Recreating application spec cache from existing resources on cluster")
-		appList, err := a.appManager.List(ctx, backend.ApplicationSelector{Namespaces: []string{a.namespace}})
-		if err != nil {
-			log().Errorf("Error while fetching list of applications: %v", err)
-		}
-
-		for _, app := range appList {
-			sourceUID, exists := app.Annotations[manager.SourceUIDAnnotation]
-			if exists {
-				appCache.SetApplicationSpec(ty.UID(sourceUID), app.Spec, log())
-			}
-		}
+		a.populateSourceCache(ctx)
 	}
 
 	if a.options.metricsPort > 0 {
@@ -482,3 +476,46 @@ func (a *Agent) healthzHandler(w http.ResponseWriter, r *http.Request) {
 		w.WriteHeader(http.StatusServiceUnavailable)
 	}
 }
+
+func (a *Agent) populateSourceCache(ctx context.Context) {
+	log().Infof("Recreating application spec cache from existing resources on cluster")
+	appList, err := a.appManager.List(ctx, backend.ApplicationSelector{Namespaces: []string{a.namespace}})
+	if err != nil {
+		log().Errorf("Error while fetching list of applications: %v", err)
+	}
+
+	for _, app := range appList {
+		sourceUID, exists := app.Annotations[manager.SourceUIDAnnotation]
+		if exists {
+			a.sourceCache.Application.Set(ty.UID(sourceUID), app.Spec)
+		}
+	}
+
+	log().Infof("Recreating appProject spec cache from existing resources on cluster")
+	appProjectList, err := a.projectManager.List(ctx, backend.AppProjectSelector{Namespace: a.namespace})
+	if err != nil {
+		log().Errorf("Error while fetching list of appProjects: %v", err)
+	}
+
+	for _, appProject := range appProjectList {
+		sourceUID, exists := appProject.Annotations[manager.SourceUIDAnnotation]
+		if exists {
+			a.sourceCache.AppProject.Set(ty.UID(sourceUID), appProject.Spec)
+		}
+	}
+
+	log().Infof("Recreating repository spec cache from existing resources on cluster")
+	repoList, err := a.repoManager.List(ctx, backend.RepositorySelector{Namespace: a.namespace})
+	if err != nil {
+		log().Errorf("Error while fetching list of repositories: %v", err)
+	}
+
+	for _, repo := range repoList {
+		sourceUID, exists := repo.Annotations[manager.SourceUIDAnnotation]
+		if exists {
+			a.sourceCache.Repository.Set(ty.UID(sourceUID), repo.Data)
+		}
+	}
+
+	log().Infof("Source cache populated successfully")
+}

agent/inbound.go

@@ -19,7 +19,6 @@ import (
 	"time"
 
 	"github.com/argoproj-labs/argocd-agent/internal/backend"
-	appCache "github.com/argoproj-labs/argocd-agent/internal/cache"
 	"github.com/argoproj-labs/argocd-agent/internal/checkpoint"
 	"github.com/argoproj-labs/argocd-agent/internal/event"
 	"github.com/argoproj-labs/argocd-agent/internal/manager"
@@ -459,7 +458,7 @@ func (a *Agent) createApplication(incoming *v1alpha1.Application) (*v1alpha1.App
 
 	if a.mode == types.AgentModeManaged {
 		// Store app spec in cache
-		appCache.SetApplicationSpec(incoming.UID, incoming.Spec, logCtx)
+		a.sourceCache.Application.Set(incoming.UID, incoming.Spec)
 	}
 
 	created, err := a.appManager.Create(a.context, incoming)
@@ -500,7 +499,7 @@ func (a *Agent) updateApplication(incoming *v1alpha1.Application) (*v1alpha1.App
 
 		// Update app spec in cache
 		logCtx.Tracef("Calling update spec for this event")
-		appCache.SetApplicationSpec(incoming.UID, incoming.Spec, logCtx)
+		a.sourceCache.Application.Set(incoming.UID, incoming.Spec)
 
 		napp, err = a.appManager.UpdateManagedApp(a.context, incoming)
 	case types.AgentModeAutonomous:
@@ -535,15 +534,15 @@ func (a *Agent) deleteApplication(app *v1alpha1.Application) error {
 		if apierrors.IsNotFound(err) {
 			logCtx.Debug("application is not found, perhaps it is already deleted")
 			if a.mode == types.AgentModeManaged {
-				appCache.DeleteApplicationSpec(app.UID, logCtx)
+				a.sourceCache.Application.Delete(app.UID)
 			}
 			return nil
 		}
 		return err
 	}
 
 	if a.mode == types.AgentModeManaged {
-		appCache.DeleteApplicationSpec(app.UID, logCtx)
+		a.sourceCache.Application.Delete(app.UID)
 	}
 
 	err = a.appManager.Unmanage(app.QualifiedName())
@@ -577,6 +576,8 @@ func (a *Agent) createAppProject(incoming *v1alpha1.AppProject) (*v1alpha1.AppPr
 
 	logCtx.Infof("Creating a new AppProject on behalf of an incoming event")
 
+	a.sourceCache.AppProject.Set(incoming.UID, incoming.Spec)
+
 	// Get rid of some fields that we do not want to have on the AppProject
 	// as we start fresh.
 	if incoming.Annotations != nil {
@@ -611,6 +612,8 @@ func (a *Agent) updateAppProject(incoming *v1alpha1.AppProject) (*v1alpha1.AppPr
 
 	logCtx.Infof("Updating appProject")
 
+	a.sourceCache.AppProject.Set(incoming.UID, incoming.Spec)
+
 	logCtx.Tracef("Calling update spec for this event")
 	return a.projectManager.UpdateAppProject(a.context, incoming)
 
@@ -637,10 +640,14 @@ func (a *Agent) deleteAppProject(project *v1alpha1.AppProject) error {
 	if err != nil {
 		if apierrors.IsNotFound(err) {
 			logCtx.Debug("appProject not found, perhaps it is already deleted")
+			a.sourceCache.AppProject.Delete(project.UID)
 			return nil
 		}
 		return err
 	}
+
+	a.sourceCache.AppProject.Delete(project.UID)
+
 	err = a.projectManager.Unmanage(project.Name)
 	if err != nil {
 		log().Warnf("Could not unmanage appProject %s: %v", project.Name, err)

agent/outbound.go

@@ -81,7 +81,7 @@ func (a *Agent) addAppUpdateToQueue(old *v1alpha1.Application, new *v1alpha1.App
 
 	// Revert any direct modifications done in application on managed-cluster
 	// because for managed-agent all changes should be done through principal
-	if reverted := a.appManager.RevertManagedAppChanges(a.context, new); reverted {
+	if reverted := a.appManager.RevertManagedAppChanges(a.context, new, a.sourceCache.Application); reverted {
 		logCtx.Debugf("Modifications done in application: %s are reverted", new.Name)
 		return
 	}
@@ -215,6 +215,13 @@ func (a *Agent) addAppProjectUpdateToQueue(old *v1alpha1.AppProject, new *v1alph
 		return
 	}
 
+	// Revert any direct modifications done to appProject on managed-cluster
+	// because for managed-agent all changes should be done through principal
+	if reverted := a.projectManager.RevertAppProjectChanges(a.context, new, a.sourceCache.AppProject); reverted {
+		logCtx.Debugf("Modifications done to appProject are reverted")
+		return
+	}
+
 	// Only send the update event when we're in autonomous mode
 	if !a.mode.IsAutonomous() {
 		return