feat: Implement watch for Istio resources

cmd/rollouts-controller/main.go

@@ -97,7 +97,11 @@ func newCommand() *cobra.Command {
 				kubeinformers.WithTweakListOptions(func(options *metav1.ListOptions) {
 					options.LabelSelector = jobprovider.AnalysisRunUIDLabelKey
 				}))
-			cm := controller.NewManager(kubeClient, rolloutClient, dynamicClient,
+			cm := controller.NewManager(
+				namespace,
+				kubeClient,
+				rolloutClient,
+				dynamicClient,
 				kubeInformerFactory.Apps().V1().ReplicaSets(),
 				kubeInformerFactory.Core().V1().Services(),
 				jobInformerFactory.Batch().V1().Jobs(),

controller/controller.go

@@ -73,10 +73,13 @@ type Manager struct {
 	serviceWorkqueue     workqueue.RateLimitingInterface
 	experimentWorkqueue  workqueue.RateLimitingInterface
 	analysisRunWorkqueue workqueue.RateLimitingInterface
+
+	defaultIstioVersion string
 }
 
 // NewManager returns a new manager to manage all the controllers
 func NewManager(
+	namespace string,
 	kubeclientset kubernetes.Interface,
 	argoprojclientset clientset.Interface,
 	dynamicclientset dynamic.Interface,
@@ -113,6 +116,7 @@ func NewManager(
 	metricsServer := metrics.NewMetricsServer(metricsAddr, rolloutsInformer.Lister())
 
 	rolloutController := rollout.NewRolloutController(
+		namespace,
 		kubeclientset,
 		argoprojclientset,
 		dynamicclientset,
@@ -178,6 +182,7 @@ func NewManager(
 		serviceController:      serviceController,
 		experimentController:   experimentController,
 		analysisController:     analysisController,
+		defaultIstioVersion:    defaultIstioVersion,
 	}
 
 	return cm

rollout/bluegreen.go

@@ -33,6 +33,7 @@ func (c *RolloutController) rolloutBlueGreen(r *v1alpha1.Rollout, rsList []*apps
 	allRSs := roCtx.AllRSs()
 	if reconcileBlueGreenTemplateChange(roCtx) {
 		roCtx.PauseContext().ClearPauseConditions()
+		roCtx.PauseContext().RemoveAbort()
 		logCtx.Infof("New pod template or template change detected")
 		return c.syncRolloutStatusBlueGreen(previewSvc, activeSvc, roCtx)
 	}

rollout/controller.go

@@ -9,6 +9,7 @@ import (
 	log "github.com/sirupsen/logrus"
 	appsv1 "k8s.io/api/apps/v1"
 	k8serrors "k8s.io/apimachinery/pkg/api/errors"
+	"k8s.io/apimachinery/pkg/runtime/schema"
 	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/client-go/dynamic"
 	appsinformers "k8s.io/client-go/informers/apps/v1"
@@ -19,6 +20,7 @@ import (
 	"k8s.io/client-go/tools/cache"
 	"k8s.io/client-go/tools/record"
 	"k8s.io/client-go/util/workqueue"
+	"k8s.io/kubernetes/cmd/kubeadm/app/util"
 	"k8s.io/kubernetes/pkg/controller"
 	"k8s.io/utils/pointer"
 
@@ -28,15 +30,22 @@ import (
 	clientset "github.com/argoproj/argo-rollouts/pkg/client/clientset/versioned"
 	informers "github.com/argoproj/argo-rollouts/pkg/client/informers/externalversions/rollouts/v1alpha1"
 	listers "github.com/argoproj/argo-rollouts/pkg/client/listers/rollouts/v1alpha1"
+	"github.com/argoproj/argo-rollouts/rollout/trafficrouting/istio"
 	"github.com/argoproj/argo-rollouts/utils/conditions"
 	controllerutil "github.com/argoproj/argo-rollouts/utils/controller"
 	"github.com/argoproj/argo-rollouts/utils/defaults"
 	logutil "github.com/argoproj/argo-rollouts/utils/log"
 	serviceutil "github.com/argoproj/argo-rollouts/utils/service"
 )
 
+const (
+	virtualServiceIndexName = "byVirtualService"
+)
+
 // RolloutController is the controller implementation for Rollout resources
 type RolloutController struct {
+	// namespace which namespace(s) operates on
+	namespace string
 	// rsControl is used for adopting/releasing replica sets.
 	replicaSetControl controller.RSControlInterface
 
@@ -80,6 +89,7 @@ type RolloutController struct {
 
 // NewRolloutController returns a new rollout controller
 func NewRolloutController(
+	namespace string,
 	kubeclientset kubernetes.Interface,
 	argoprojclientset clientset.Interface,
 	dynamicclientset dynamic.Interface,
@@ -102,6 +112,7 @@ func NewRolloutController(
 	}
 
 	controller := &RolloutController{
+		namespace:              namespace,
 		kubeclientset:          kubeclientset,
 		argoprojclientset:      argoprojclientset,
 		dynamicclientset:       dynamicclientset,
@@ -146,6 +157,15 @@ func NewRolloutController(
 		},
 	})
 
+	util.CheckErr(rolloutsInformer.Informer().AddIndexers(cache.Indexers{
+		virtualServiceIndexName: func(obj interface{}) (strings []string, e error) {
+			if rollout, ok := obj.(*v1alpha1.Rollout); ok {
+				return istio.GetRolloutVirtualServiceKeys(rollout), nil
+			}
+			return
+		},
+	}))
+
 	replicaSetInformer.Informer().AddEventHandler(cache.ResourceEventHandlerFuncs{
 		AddFunc: func(obj interface{}) {
 			controllerutil.EnqueueParentObject(obj, register.RolloutKind, controller.enqueueRollout)
@@ -197,6 +217,10 @@ func (c *RolloutController) Run(threadiness int, stopCh <-chan struct{}) error {
 		}, time.Second, stopCh)
 	}
 	log.Info("Started Rollout workers")
+
+	gvk := schema.ParseGroupResource("virtualservices.networking.istio.io").WithVersion(c.defaultIstioVersion)
+	go controllerutil.WatchResourceWithExponentialBackoff(stopCh, c.dynamicclientset, c.namespace, gvk, c.rolloutWorkqueue, c.rolloutsIndexer, virtualServiceIndexName)
+
 	<-stopCh
 	log.Info("Shutting down workers")
 

rollout/controller_test.go

@@ -372,7 +372,11 @@ func (f *fixture) newController(resync resyncFunc) (*RolloutController, informer
 	rolloutWorkqueue := workqueue.NewNamedRateLimitingQueue(workqueue.DefaultControllerRateLimiter(), "Rollouts")
 	serviceWorkqueue := workqueue.NewNamedRateLimitingQueue(workqueue.DefaultControllerRateLimiter(), "Services")
 
-	c := NewRolloutController(f.kubeclient, f.client, nil,
+	c := NewRolloutController(
+		metav1.NamespaceAll,
+		f.kubeclient,
+		f.client,
+		nil,
 		i.Argoproj().V1alpha1().Experiments(),
 		i.Argoproj().V1alpha1().AnalysisRuns(),
 		i.Argoproj().V1alpha1().AnalysisTemplates(),

rollout/trafficrouting/istio/istio.go

@@ -31,6 +31,23 @@ func NewReconciler(r *v1alpha1.Rollout, client dynamic.Interface, recorder recor
 	}
 }
 
+// GetRolloutVirtualServiceKeys gets the virtual service and its namespace from a rollout
+func GetRolloutVirtualServiceKeys(rollout *v1alpha1.Rollout) []string {
+	if rollout.Spec.Strategy.Canary == nil {
+		return []string{}
+	}
+	if rollout.Spec.Strategy.Canary.TrafficRouting == nil {
+		return []string{}
+	}
+	if rollout.Spec.Strategy.Canary.TrafficRouting.Istio == nil {
+		return []string{}
+	}
+	if rollout.Spec.Strategy.Canary.TrafficRouting.Istio.VirtualService.Name == "" {
+		return []string{}
+	}
+	return []string{fmt.Sprintf("%s/%s", rollout.Namespace, rollout.Spec.Strategy.Canary.TrafficRouting.Istio.VirtualService.Name)}
+}
+
 // Reconciler holds required fields to reconcile Istio resources
 type Reconciler struct {
 	rollout           *v1alpha1.Rollout

rollout/trafficrouting/istio/istio_test.go

@@ -291,3 +291,28 @@ func TestValidateHosts(t *testing.T) {
 	err = validateHosts(hr, "stable", "not-found-canary")
 	assert.Equal(t, fmt.Errorf("Canary Service 'not-found-canary' not found in route"), err)
 }
+
+func TestGetRolloutVirtualServiceKeys(t *testing.T) {
+	ro := &v1alpha1.Rollout{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "foo",
+			Namespace: metav1.NamespaceDefault,
+		},
+		Spec: v1alpha1.RolloutSpec{
+			Strategy: v1alpha1.RolloutStrategy{},
+		},
+	}
+	assert.Len(t, GetRolloutVirtualServiceKeys(ro), 0)
+	ro.Spec.Strategy.Canary = &v1alpha1.CanaryStrategy{}
+	assert.Len(t, GetRolloutVirtualServiceKeys(ro), 0)
+	ro.Spec.Strategy.Canary.TrafficRouting = &v1alpha1.RolloutTrafficRouting{}
+	assert.Len(t, GetRolloutVirtualServiceKeys(ro), 0)
+	ro.Spec.Strategy.Canary.TrafficRouting.Istio = &v1alpha1.IstioTrafficRouting{
+		VirtualService: v1alpha1.IstioVirtualService{},
+	}
+	assert.Len(t, GetRolloutVirtualServiceKeys(ro), 0)
+	ro.Spec.Strategy.Canary.TrafficRouting.Istio.VirtualService.Name = "test"
+	keys := GetRolloutVirtualServiceKeys(ro)
+	assert.Len(t, keys, 1)
+	assert.Equal(t, keys[0], "default/test")
+}

utils/controller/controller.go

@@ -6,10 +6,15 @@ import (
 	"time"
 
 	log "github.com/sirupsen/logrus"
+	"k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/labels"
+	"k8s.io/apimachinery/pkg/runtime/schema"
 	"k8s.io/apimachinery/pkg/selection"
 	"k8s.io/apimachinery/pkg/util/runtime"
+	"k8s.io/apimachinery/pkg/util/wait"
+	"k8s.io/apimachinery/pkg/watch"
+	"k8s.io/client-go/dynamic"
 	"k8s.io/client-go/tools/cache"
 	"k8s.io/client-go/util/workqueue"
 
@@ -18,6 +23,72 @@ import (
 	logutil "github.com/argoproj/argo-rollouts/utils/log"
 )
 
+// processNextWatchObj will process a single object from the watch by seeing if
+// that object is in an index and enqueueing the value object from the indexer
+func processNextWatchObj(watchEvent watch.Event, queue workqueue.RateLimitingInterface, indexer cache.Indexer, index string) {
+	obj := watchEvent.Object
+	acc, err := meta.Accessor(obj)
+	if err != nil {
+		log.Errorf("Error processing object from watch: %v", err)
+		return
+	}
+	objsToEnqueue, err := indexer.ByIndex(index, fmt.Sprintf("%s/%s", acc.GetNamespace(), acc.GetName()))
+	if err != nil {
+		log.Errorf("Cannot process indexer: %s", err.Error())
+		return
+	}
+	for i := range objsToEnqueue {
+		Enqueue(objsToEnqueue[i], queue)
+	}
+}
+
+// WatchResource is a long-running function that will continually call the
+// processNextWatchObj function in order to watch changes on a resource kind
+// and enqueue a different resources kind that interact with them
+func WatchResource(client dynamic.Interface, namespace string, gvk schema.GroupVersionResource, queue workqueue.RateLimitingInterface, indexer cache.Indexer, index string) error {
+	log.Infof("Starting watch on resource '%s'", gvk.Resource)
+	var watchI watch.Interface
+	var err error
+	if namespace == metav1.NamespaceAll {
+		watchI, err = client.Resource(gvk).Watch(metav1.ListOptions{})
+	} else {
+		watchI, err = client.Resource(gvk).Namespace(namespace).Watch(metav1.ListOptions{})
+	}
+
+	if err != nil {
+		log.Errorf("Error with watch: %v", err)
+		return err
+	}
+	for watchEvent := range watchI.ResultChan() {
+		processNextWatchObj(watchEvent, queue, indexer, index)
+	}
+	return nil
+}
+
+// WatchResourceWithExponentialBackoff creates a watch for the gvk provided. If there are any error,
+// the function will rety again using exponetial backoff. It starts at 1 second wait, and wait afterwards
+// increases by a factor of 2 and caps at 5 minutes.
+func WatchResourceWithExponentialBackoff(stopCh <-chan struct{}, client dynamic.Interface, namespace string, gvk schema.GroupVersionResource, queue workqueue.RateLimitingInterface, indexer cache.Indexer, index string) {
+	backoff := wait.Backoff{
+		Duration: 1 * time.Second,
+		Cap:      5 * time.Minute,
+		Factor:   float64(2),
+		Steps:    10,
+	}
+	for {
+		select {
+		case <-stopCh:
+			return
+		default:
+		}
+		err := WatchResource(client, namespace, gvk, queue, indexer, index)
+		if err == nil {
+			continue
+		}
+		time.Sleep(backoff.Step())
+	}
+}
+
 // RunWorker is a long-running function that will continually call the
 // processNextWorkItem function in order to read and process a message on the
 // workqueue.