feat: support reference model for workloads (argoproj#676)

Signed-off-by: Alexander Matyushentsev <[email protected]>

cmd/rollouts-controller/main.go

@@ -11,6 +11,7 @@ import (
 	log "github.com/sirupsen/logrus"
 	"github.com/spf13/cobra"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/client-go/discovery"
 	"k8s.io/client-go/dynamic"
 	"k8s.io/client-go/dynamic/dynamicinformer"
 	kubeinformers "k8s.io/client-go/informers"
@@ -97,6 +98,8 @@ func newCommand() *cobra.Command {
 			checkError(err)
 			dynamicClient, err := dynamic.NewForConfig(config)
 			checkError(err)
+			discoveryClient, err := discovery.NewDiscoveryClientForConfig(config)
+			checkError(err)
 			smiClient, err := smiclientset.NewForConfig(config)
 			resyncDuration := time.Duration(rolloutResyncPeriod) * time.Second
 			kubeInformerFactory := kubeinformers.NewSharedInformerFactoryWithOptions(
@@ -130,6 +133,7 @@ func newCommand() *cobra.Command {
 				rolloutClient,
 				dynamicClient,
 				smiClient,
+				discoveryClient,
 				kubeInformerFactory.Apps().V1().ReplicaSets(),
 				kubeInformerFactory.Core().V1().Services(),
 				kubeInformerFactory.Extensions().V1beta1().Ingresses(),

controller/controller.go

@@ -12,6 +12,7 @@ import (
 	"k8s.io/apimachinery/pkg/util/runtime"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
 	"k8s.io/apimachinery/pkg/util/wait"
+	"k8s.io/client-go/discovery"
 	"k8s.io/client-go/dynamic"
 	appsinformers "k8s.io/client-go/informers/apps/v1"
 	batchinformers "k8s.io/client-go/informers/batch/v1"
@@ -85,6 +86,8 @@ type Manager struct {
 	experimentWorkqueue  workqueue.RateLimitingInterface
 	analysisRunWorkqueue workqueue.RateLimitingInterface
 
+	refResolver rollout.TemplateRefResolver
+
 	dynamicClientSet dynamic.Interface
 
 	namespace string
@@ -97,6 +100,7 @@ func NewManager(
 	argoprojclientset clientset.Interface,
 	dynamicclientset dynamic.Interface,
 	smiclientset smiclientset.Interface,
+	discoveryClient discovery.DiscoveryInterface,
 	replicaSetInformer appsinformers.ReplicaSetInformer,
 	servicesInformer coreinformers.ServiceInformer,
 	ingressesInformer extensionsinformers.IngressInformer,
@@ -141,11 +145,14 @@ func NewManager(
 	serviceWorkqueue := workqueue.NewNamedRateLimitingQueue(workqueue.DefaultControllerRateLimiter(), "Services")
 	ingressWorkqueue := workqueue.NewNamedRateLimitingQueue(workqueue.DefaultControllerRateLimiter(), "Ingresses")
 
+	refResolver := rollout.NewInformerBasedWorkloadRefResolver(namespace, dynamicclientset, discoveryClient, rolloutWorkqueue, rolloutsInformer.Informer())
+
 	rolloutController := rollout.NewController(rollout.ControllerConfig{
 		Namespace:                       namespace,
 		KubeClientSet:                   kubeclientset,
 		ArgoProjClientset:               argoprojclientset,
 		DynamicClientSet:                dynamicclientset,
+		RefResolver:                     refResolver,
 		SmiClientSet:                    smiclientset,
 		ExperimentInformer:              experimentsInformer,
 		AnalysisRunInformer:             analysisRunInformer,
@@ -238,6 +245,7 @@ func NewManager(
 		experimentController:          experimentController,
 		analysisController:            analysisController,
 		dynamicClientSet:              dynamicclientset,
+		refResolver:                   refResolver,
 		namespace:                     namespace,
 	}
 
@@ -255,6 +263,7 @@ func (c *Manager) Run(rolloutThreadiness, serviceThreadiness, ingressThreadiness
 	defer c.rolloutWorkqueue.ShutDown()
 	defer c.experimentWorkqueue.ShutDown()
 	defer c.analysisRunWorkqueue.ShutDown()
+
 	// Wait for the caches to be synced before starting workers
 	log.Info("Waiting for controller's informer caches to sync")
 	if ok := cache.WaitForCacheSync(stopCh, c.serviceSynced, c.ingressSynced, c.jobSynced, c.rolloutSynced, c.experimentSynced, c.analysisRunSynced, c.analysisTemplateSynced, c.replicasSetSynced); !ok {

docs/migrating.md

@@ -1,6 +1,11 @@
 # Migrating to Rollouts
 
-Migrating to Argo Rollouts involves converting an existing Deployment resource, to a Rollout resource.
+There are ways to migrate to Rollout:
+
+* Convert an existing Deployment resource to a Rollout resource.
+* Reference an existing Deployment from a Rollout using `workloadRef` field.
+
+## Convert Deployment to Rollout
 
 When converting a Deployment to a Rollout, it involves changing three fields:
 
@@ -36,8 +41,79 @@ spec:
       - pause: {}
 ```
 
-## Other Considerations
+!!! warning
+    When migrating a Deployment which is already serving live production traffic, a Rollout should
+    run next to the Deployment before deleting the Deployment or scaling down the Deployment.
+    **Not following this approach might result in downtime**. It also allows for the Rollout to be
+    tested before deleting the original Deployment.
+
+
+## Reference Deployment From Rollout
+
+Instead of removing Deployment you can scale-down in to zero and reference from the Rollout resource:
+
+1. Create a Rollout resource.
+1. Reference an existing Deployment using `workloadRef` field.
+1. Scale-down existing Deployment by changing `replicas` field of an existing Deployment to zero.
+
+Below is an example of a Rollout resource referencing a Deployment.
+
+```yaml
+apiVersion: argoproj.io/v1alpha1               # Create a rollout resource
+kind: Rollout
+metadata:
+  name: rollout-ref-deployment
+spec:
+  replicas: 5
+  workloadRef:                                 # Reference an existing Deployment using workloadRef field
+    apiVersion: apps/v1
+    kind: Deployment
+    name: rollout-ref-deployment
+  strategy:
+    canary:
+      steps:
+        - setWeight: 20
+        - pause: {duration: 10s}
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  labels:
+    app.kubernetes.io/instance: rollout-canary
+  name: rollout-ref-deployment
+spec:
+  replicas: 0                                  # Scale down existing deployment
+  selector:
+    matchLabels:
+      app: rollout-ref-deployment
+  template:
+    metadata:
+      labels:
+        app: rollout-ref-deployment
+    spec:
+      containers:
+        - name: rollouts-demo
+          image: argoproj/rollouts-demo:blue
+          imagePullPolicy: Always
+          ports:
+            - containerPort: 8080
+```
+
+Consider following if your Deployment runs in production:
+
+**Running Rollout and Deployment side-by-side**
+
+After creation Rollout will spinup required number of Pods side-by-side with the Deployment Pods.
+Rollout won't try to manage existing Deployment Pods. That means you can safely update add Rollout
+to the production environment without any interruption but you are going to run twice more Pods during migration.
+
+**Traffic Management During Migration**
+
+The Rollout offers traffic management functionality that manages routing rules and flows the traffic to different
+versions of an application. For example [Blue-Green](../docs/features/bluegreen.md) deployment strategy manipulates
+Kubernetes Service selector and direct production traffic to "green" instances only.
 
-When migrating a Deployment which is already serving live production traffic, a Rollout should
-run next to the Deployment before deleting the Deployment. **Not following this approach might result in 
-downtime**. It also allows for the Rollout to be tested before deleting the original Deployment.
+If you are using this feature then Rollout switches production traffic to Pods that it manages. The switch happens
+only when the required number of Pod is running and healthy so it is safe in production as well. However, if you
+want to be extra careful then consider creating a temporal Service or Ingress object to validate Rollout behavior.
+Once testing is done delete temporal Service/Ingress and switch rollout to production one.

examples/rollout-template-ref.yaml

@@ -0,0 +1,39 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  labels:
+    app.kubernetes.io/instance: rollout-canary
+  name: rollout-ref-deployment
+spec:
+  replicas: 0
+  selector:
+    matchLabels:
+      app: rollout-ref-deployment
+  template:
+    metadata:
+      labels:
+        app: rollout-ref-deployment
+    spec:
+      containers:
+        - name: rollouts-demo
+          image: argoproj/rollouts-demo:blue
+          imagePullPolicy: Always
+          ports:
+            - containerPort: 8080
+---
+apiVersion: argoproj.io/v1alpha1
+kind: Rollout
+metadata:
+  name: rollout-ref-deployment
+spec:
+  replicas: 5
+  revisionHistoryLimit: 2
+  workloadRef:
+    apiVersion: apps/v1
+    kind: Deployment
+    name: rollout-ref-deployment
+  strategy:
+    canary:
+      steps:
+        - setWeight: 20
+        - pause: {duration: 10s}