Add preview endpoint for ArgoRollouts using BlueGreen (#199)

* Changes to support preview endpoint for bluegreen rollout Signed-off-by: nbn01 <nandan_bn@intuit.com> * Updated integration test for preview endpoint Signed-off-by: nbn01 <nandan_bn@intuit.com> * Minor check fixes Signed-off-by: nbn01 <nandan_bn@intuit.com> * Reverted integration test changes Signed-off-by: nbn01 <nandan_bn@intuit.com> * Update unit tests Signed-off-by: nbn01 <nandan_bn@intuit.com> * Add integration test for preview endpoint of bluegreen rollout Signed-off-by: nbn01 <nandan_bn@intuit.com> * Update makefile Signed-off-by: nbn01 <nandan_bn@intuit.com> * Modified permissions Signed-off-by: nbn01 <nandan_bn@intuit.com> * Reverting argo_rollouts default to false Signed-off-by: nbn01 <nandan_bn@intuit.com> * Updated unit tests to check when preview service not defined Signed-off-by: nbn01 <nandan_bn@intuit.com> * Improved integration test for bluegreen rollout Signed-off-by: nbn01 <nandan_bn@intuit.com> * Continue with normal flow if preview condition fails Signed-off-by: nbn01 <nandan_bn@intuit.com> * Uncomment clean up Signed-off-by: nbn01 <nandan_bn@intuit.com> * make "preview" standard prefix name for bluegreen Signed-off-by: nbn01 <nandan_bn@intuit.com> * abstract update preview endpoint to method Signed-off-by: nbn01 <nandan_bn@intuit.com> * Add unit test Signed-off-by: nbn01 <nandan_bn@intuit.com> * fix golintci lint to use go1.17.7 Signed-off-by: nbn01 <nandan_bn@intuit.com> Co-authored-by: nbn01 <nandan_bn@intuit.com>
istio-ecosystem · Apr 6, 2022 · 845e7da · 845e7da
1 parent 4c9ef81
commit 845e7da
Show file tree

Hide file tree

Showing 11 changed files with 520 additions and 73 deletions.
diff --git a/.github/workflows/golang-ci-lint.yml b/.github/workflows/golang-ci-lint.yml
@@ -13,11 +13,15 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v2
+      - uses: actions/setup-go@v3
+        with:
+          go-version: '1.17.7'
       - name: golangci-lint
         uses: golangci/golangci-lint-action@v2
         with:
           # Required: the version of golangci-lint is required and must be specified without patch version: we always use the latest patch version.
           version: v1.43.0
+          skip-go-installation: true
 
           # Optional: working directory, useful for monorepos
           # working-directory: somedir

diff --git a/Makefile b/Makefile
@@ -131,6 +131,7 @@ gen-yaml:
 	kustomize build ./install/sample/overlays/rollout-bluegreen > ./out/yaml/sample-greeting-rollout-bluegreen.yaml
 	kustomize build ./install/sample/overlays/remote > ./out/yaml/remotecluster_sample.yaml
 	cp ./install/sample/sample_dep.yaml ./out/yaml/sample_dep.yaml
+	cp ./install/sample/greeting_preview.yaml ./out/yaml/greeting_preview.yaml
 	cp ./install/sample/gtp.yaml ./out/yaml/gtp.yaml
 	cp ./install/sample/gtp_failover.yaml ./out/yaml/gtp_failover.yaml
 	cp ./install/sample/gtp_topology.yaml ./out/yaml/gtp_topology.yaml

diff --git a/admiral/pkg/clusters/handler.go b/admiral/pkg/clusters/handler.go
@@ -747,7 +747,7 @@ func copyEndpoint(e *v1alpha32.ServiceEntry_Endpoint) *v1alpha32.ServiceEntry_En
 
 // A rollout can use one of 2 stratergies :-
 // 1. Canary strategy - which can use a virtual service to manage the weights associated with a stable and canary service. Admiral created endpoints in service entries will use the weights assigned in the Virtual Service
-// 2. Blue green strategy- this contains 2 service instances in a namespace, an active service and a preview service. Admiral will always use the active service
+// 2. Blue green strategy- this contains 2 service instances in a namespace, an active service and a preview service. Admiral will use repective service to create active and preview endpoints
 func getServiceForRollout(rc *RemoteController, rollout *argo.Rollout) map[string]*WeightedService {
 
 	if rollout == nil {
@@ -769,9 +769,12 @@ func getServiceForRollout(rc *RemoteController, rollout *argo.Rollout) map[strin
 	var istioCanaryWeights = make(map[string]int32)
 
 	var blueGreenActiveService string
+	var blueGreenPreviewService string
+
 	if rolloutStrategy.BlueGreen != nil {
-		// If rollout uses blue green strategy, use the active service
+		// If rollout uses blue green strategy
 		blueGreenActiveService = rolloutStrategy.BlueGreen.ActiveService
+		blueGreenPreviewService = rolloutStrategy.BlueGreen.PreviewService
 	} else if rolloutStrategy.Canary != nil && rolloutStrategy.Canary.TrafficRouting != nil && rolloutStrategy.Canary.TrafficRouting.Istio != nil {
 		canaryService = rolloutStrategy.Canary.CanaryService
 		stableService = rolloutStrategy.Canary.StableService
@@ -844,7 +847,7 @@ func getServiceForRollout(rc *RemoteController, rollout *argo.Rollout) map[strin
 		var service = servicesInNamespace[s]
 		var match = true
 		//skip services that are not referenced in the rollout
-		if len(blueGreenActiveService) > 0 && service.ObjectMeta.Name != blueGreenActiveService {
+		if len(blueGreenActiveService) > 0 && service.ObjectMeta.Name != blueGreenActiveService && service.ObjectMeta.Name != blueGreenPreviewService {
 			log.Infof("Skipping service=%s for rollout=%s in namespace=%s and cluster=%s", service.Name, rollout.Name, rollout.Namespace, rc.ClusterID)
 			continue
 		}
@@ -865,8 +868,11 @@ func getServiceForRollout(rc *RemoteController, rollout *argo.Rollout) map[strin
 		if match {
 			ports := GetMeshPortsForRollout(rc.ClusterID, service, rollout)
 			if len(ports) > 0 {
-				//if the strategy is bluegreen or using canary with NO istio traffic management, pick the first service that matches
-				if len(istioCanaryWeights) == 0 {
+				// if the strategy is bluegreen return matched services
+				// else if using canary with NO istio traffic management, pick the first service that matches
+				if rolloutStrategy.BlueGreen != nil {
+					matchedServices[service.Name] = &WeightedService{Weight: 1, Service: service}
+				} else if len(istioCanaryWeights) == 0 {
 					matchedServices[service.Name] = &WeightedService{Weight: 1, Service: service}
 					break
 				}

diff --git a/admiral/pkg/clusters/serviceentry.go b/admiral/pkg/clusters/serviceentry.go
@@ -62,6 +62,7 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 	var serviceEntries = make(map[string]*networking.ServiceEntry)
 
 	var cname string
+	cnames := make(map[string]string)
 	var serviceInstance *k8sV1.Service
 	var weightedServices map[string]*WeightedService
 	var rollout *admiral.RolloutClusterEntry
@@ -104,6 +105,7 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 			localMeshPorts := GetMeshPortsForRollout(rc.ClusterID, serviceInstance, rolloutInstance)
 
 			cname = common.GetCnameForRollout(rolloutInstance, common.GetWorkloadIdentifier(), common.GetHostnameSuffix())
+			cnames[cname] = "1"
 			sourceRollouts[rc.ClusterID] = rolloutInstance
 			createServiceEntryForRollout(event, rc, remoteRegistry.AdmiralCache, localMeshPorts, rolloutInstance, serviceEntries)
 		} else {
@@ -132,10 +134,14 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 		localFqdn := serviceInstance.Name + common.Sep + serviceInstance.Namespace + common.DotLocalDomainSuffix
 		rc := remoteRegistry.RemoteControllers[sourceCluster]
 		var meshPorts map[string]uint32
+		isBlueGreenStrategy := false
+
+		if len(sourceRollouts) > 0 {
+			isBlueGreenStrategy = sourceRollouts[sourceCluster].Spec.Strategy.BlueGreen != nil
+		}
+
 		if len(sourceDeployments) > 0 {
 			meshPorts = GetMeshPorts(sourceCluster, serviceInstance, sourceDeployments[sourceCluster])
-		} else {
-			meshPorts = GetMeshPortsForRollout(sourceCluster, serviceInstance, sourceRollouts[sourceCluster])
 		}
 
 		for key, serviceEntry := range serviceEntries {
@@ -147,9 +153,20 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 			for _, ep := range serviceEntry.Endpoints {
 				//replace istio ingress-gateway address with local fqdn, note that ingress-gateway can be empty (not provisoned, or is not up)
 				if ep.Address == clusterIngress || ep.Address == "" {
-					// see if we have weighted services (rollouts with canary strategy)
-					if len(weightedServices) > 1 {
+					// Update endpoints with locafqdn for active and preview se of bluegreen rollout
+					if isBlueGreenStrategy {
+						oldPorts := ep.Ports
+						updateEndpointsForBlueGreen(sourceRollouts[sourceCluster], weightedServices, cnames, ep, sourceCluster, key)
+
+						AddServiceEntriesWithDr(remoteRegistry.AdmiralCache, map[string]string{sourceCluster: sourceCluster}, remoteRegistry.RemoteControllers,
+							map[string]*networking.ServiceEntry{key: serviceEntry})
+						//swap it back to use for next iteration
+						ep.Address = clusterIngress
+						ep.Ports = oldPorts
+						// see if we have weighted services (rollouts with canary strategy)
+					} else if len(weightedServices) > 1 {
 						//add one endpoint per each service, may be modify
+						meshPorts = GetMeshPortsForRollout(sourceCluster, serviceInstance, sourceRollouts[sourceCluster])
 						var se = copyServiceEntry(serviceEntry)
 						updateEndpointsForWeightedServices(se, weightedServices, clusterIngress, meshPorts)
 						AddServiceEntriesWithDr(remoteRegistry.AdmiralCache, map[string]string{sourceCluster: sourceCluster}, remoteRegistry.RemoteControllers,
@@ -169,7 +186,7 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 		}
 
 		for _, val := range dependents {
-			remoteRegistry.AdmiralCache.DependencyNamespaceCache.Put(val, serviceInstance.Namespace, localFqdn, map[string]string{cname: "1"})
+			remoteRegistry.AdmiralCache.DependencyNamespaceCache.Put(val, serviceInstance.Namespace, localFqdn, cnames)
 		}
 
 		if common.GetWorkloadSidecarUpdate() == "enabled" {
@@ -179,6 +196,26 @@ func modifyServiceEntryForNewServiceOrPod(event admiral.EventType, env string, s
 	return serviceEntries
 }
 
+func updateEndpointsForBlueGreen(rollout *argo.Rollout, weightedServices map[string]*WeightedService, cnames map[string]string,
+	ep *networking.ServiceEntry_Endpoint, sourceCluster string, meshHost string) {
+	activeServiceName := rollout.Spec.Strategy.BlueGreen.ActiveService
+	previewServiceName := rollout.Spec.Strategy.BlueGreen.PreviewService
+
+	if previewService, ok := weightedServices[previewServiceName]; strings.HasPrefix(meshHost, common.BlueGreenRolloutPreviewPrefix+common.Sep) && ok {
+		previewServiceInstance := previewService.Service
+		localFqdn := previewServiceInstance.Name + common.Sep + previewServiceInstance.Namespace + common.DotLocalDomainSuffix
+		cnames[localFqdn] = "1"
+		ep.Address = localFqdn
+		ep.Ports = GetMeshPortsForRollout(sourceCluster, previewServiceInstance, rollout)
+	} else if activeService, ok := weightedServices[activeServiceName]; ok {
+		activeServiceInstance := activeService.Service
+		localFqdn := activeServiceInstance.Name + common.Sep + activeServiceInstance.Namespace + common.DotLocalDomainSuffix
+		cnames[localFqdn] = "1"
+		ep.Address = localFqdn
+		ep.Ports = GetMeshPortsForRollout(sourceCluster, activeServiceInstance, rollout)
+	}
+}
+
 //update endpoints for Argo rollouts specific Service Entries to account for traffic splitting (Canary strategy)
 func updateEndpointsForWeightedServices(serviceEntry *networking.ServiceEntry, weightedServices map[string]*WeightedService, clusterIngress string, meshPorts map[string]uint32) {
 	var endpoints = make([]*networking.ServiceEntry_Endpoint, 0)
@@ -570,6 +607,17 @@ func createServiceEntryForRollout(event admiral.EventType, rc *RemoteController,
 
 	san := getSanForRollout(destRollout, workloadIdentityKey)
 
+	if destRollout.Spec.Strategy.BlueGreen != nil && destRollout.Spec.Strategy.BlueGreen.PreviewService != "" {
+		rolloutServices := getServiceForRollout(rc, destRollout)
+		if _, ok := rolloutServices[destRollout.Spec.Strategy.BlueGreen.PreviewService]; ok {
+			previewGlobalFqdn := common.BlueGreenRolloutPreviewPrefix + common.Sep + common.GetCnameForRollout(destRollout, workloadIdentityKey, common.GetHostnameSuffix())
+			previewAddress := getUniqueAddress(admiralCache, previewGlobalFqdn)
+			if len(previewGlobalFqdn) != 0 && len(previewAddress) != 0 {
+				generateServiceEntry(event, admiralCache, meshPorts, previewGlobalFqdn, rc, serviceEntries, previewAddress, san)
+			}
+		}
+	}
+
 	tmpSe := generateServiceEntry(event, admiralCache, meshPorts, globalFqdn, rc, serviceEntries, address, san)
 	return tmpSe
 }