E2E test - 50 nodes (#2260)

examples/e2e-tests/kubernetes/node-count/50-nodes/definition.json

@@ -0,0 +1,39 @@
+{
+  "apiVersion": "vlabs",
+  "properties": {
+    "orchestratorProfile": {
+      "orchestratorType": "Kubernetes",
+      "orchestratorRelease": "1.8"
+    },
+    "masterProfile": {
+      "count": 5,
+      "dnsPrefix": "",
+      "vmSize": "Standard_D2_v2",
+      "OSDiskSizeGB": 200
+    },
+    "agentPoolProfiles": [
+      {
+        "name": "agentpool1",
+        "count": 50,
+        "vmSize": "Standard_D2_v2",
+        "osType": "Linux",
+        "storageProfile": "ManagedDisks",
+        "availabilityProfile": "AvailabilitySet"
+      }
+    ],
+    "linuxProfile": {
+      "adminUsername": "azureuser",
+      "ssh": {
+        "publicKeys": [
+          {
+            "keyData": ""
+          }
+        ]
+      }
+    },
+    "servicePrincipalProfile": {
+      "clientId": "",
+      "secret": ""
+    }
+  }
+}

test/e2e/engine/template.go

@@ -169,13 +169,13 @@ func (e *Engine) HasDashboard() bool {
 }
 
 // HasTiller will return true if tiller addon is enabled
-func (e *Engine) HasTiller() bool {
+func (e *Engine) HasTiller() (bool, api.KubernetesAddon) {
 	for _, addon := range e.ExpandedDefinition.Properties.OrchestratorProfile.KubernetesConfig.Addons {
 		if addon.Name == "tiller" {
-			return *addon.Enabled
+			return *addon.Enabled, addon
 		}
 	}
-	return false
+	return false, api.KubernetesAddon{}
 }
 
 // HasACIConnector will return true if aci-connector addon is enabled

test/e2e/kubernetes/kubernetes_test.go

@@ -125,23 +125,20 @@ var _ = Describe("Azure Container Cluster using the Kubernetes Orchestrator", fu
 		})
 
 		It("should have tiller running", func() {
-			if eng.HasTiller() {
+			if hasTiller, tillerAddon := eng.HasTiller(); hasTiller {
 				running, err := pod.WaitOnReady("tiller", "kube-system", 3, 30*time.Second, cfg.Timeout)
 				Expect(err).NotTo(HaveOccurred())
 				Expect(running).To(Equal(true))
-			} else {
-				Skip("tiller disabled for this cluster, will not test")
-			}
-		})
-
-		It("should have a tiller max-history of 5", func() {
-			if eng.HasTiller() {
-				pods, err := pod.GetAllByPrefix("tiller-deploy", "kube-system")
-				Expect(err).NotTo(HaveOccurred())
-				// There is only one tiller pod and one container in that pod.
-				actualTillerMaxHistory, err := pods[0].Spec.Containers[0].GetEnvironmentVariable("TILLER_HISTORY_MAX")
-				Expect(err).NotTo(HaveOccurred())
-				Expect(actualTillerMaxHistory).To(Equal("5"))
+				if tillerAddon.Config != nil {
+					By("Ensuring that the correct max-history has been applied")
+					maxHistory := tillerAddon.Config["max-history"]
+					pods, err := pod.GetAllByPrefix("tiller-deploy", "kube-system")
+					Expect(err).NotTo(HaveOccurred())
+					// There is only one tiller pod and one container in that pod.
+					actualTillerMaxHistory, err := pods[0].Spec.Containers[0].GetEnvironmentVariable("TILLER_HISTORY_MAX")
+					Expect(err).NotTo(HaveOccurred())
+					Expect(actualTillerMaxHistory).To(Equal(maxHistory))
+				}
 			} else {
 				Skip("tiller disabled for this cluster, will not test")
 			}

test/e2e/remote/ssh.go

@@ -30,7 +30,8 @@ type Connection struct {
 func NewConnection(host, port, user, keyPath string) (*Connection, error) {
 	conn, err := net.Dial("unix", os.Getenv("SSH_AUTH_SOCK"))
 	if err != nil {
-		log.Fatal(err)
+		log.Printf("unable to establish net connection $SSH_AUTH_SOCK has value %s\n", os.Getenv("SSH_AUTH_SOCK"))
+		return nil, err
 	}
 	defer conn.Close()
 	ag := agent.NewClient(conn)
@@ -51,7 +52,8 @@ func NewConnection(host, port, user, keyPath string) (*Connection, error) {
 	ag.Add(addKey)
 	signers, err := ag.Signers()
 	if err != nil {
-		log.Fatal(err)
+		log.Println("unable to add key to agent")
+		return nil, err
 	}
 	auths := []ssh.AuthMethod{ssh.PublicKeys(signers...)}
 

test/e2e/runner/cli_provisioner.go

@@ -32,6 +32,8 @@ type CLIProvisioner struct {
 	Point             *metrics.Point
 	ResourceGroups    []string
 	Engine            *engine.Engine
+	Masters           []azure.VM
+	Agents            []azure.VM
 }
 
 // BuildCLIProvisioner will return a ProvisionerConfig object which is used to run a provision
@@ -155,6 +157,22 @@ func (cli *CLIProvisioner) provision() error {
 		return fmt.Errorf("Error while trying to create deployment:%s", err)
 	}
 
+	// Store the hosts for future introspection
+	hosts, err := cli.Account.GetHosts(cli.Config.Name)
+	if err != nil {
+		return err
+	}
+	var masters, agents []azure.VM
+	for _, host := range hosts {
+		if strings.Contains(host.Name, "master") {
+			masters = append(masters, host)
+		} else if strings.Contains(host.Name, "agent") {
+			agents = append(agents, host)
+		}
+	}
+	cli.Masters = masters
+	cli.Agents = agents
+
 	return nil
 }
 
@@ -205,18 +223,6 @@ func (cli *CLIProvisioner) waitForNodes() error {
 
 // FetchProvisioningMetrics gets provisioning files from all hosts in a cluster
 func (cli *CLIProvisioner) FetchProvisioningMetrics(path string, cfg *config.Config, acct *azure.Account) error {
-	var masters, agents []string
-	hosts, err := acct.GetHosts("")
-	if err != nil {
-		return err
-	}
-	for _, host := range hosts {
-		if strings.Contains(host.Name, "master") {
-			masters = append(masters, host.Name)
-		} else if strings.Contains(host.Name, "agent") {
-			agents = append(agents, host.Name)
-		}
-	}
 	agentFiles := []string{"/var/log/azure/cluster-provision.log", "/var/log/cloud-init.log",
 		"/var/log/cloud-init-output.log", "/var/log/syslog", "/var/log/azure/custom-script/handler.log",
 		"/opt/m", "/opt/azure/containers/kubelet.sh", "/opt/azure/containers/provision.sh",
@@ -228,18 +234,18 @@ func (cli *CLIProvisioner) FetchProvisioningMetrics(path string, cfg *config.Con
 	if err != nil {
 		return err
 	}
-	for _, master := range masters {
+	for _, master := range cli.Masters {
 		for _, fp := range masterFiles {
-			err := conn.CopyRemote(master, fp)
+			err := conn.CopyRemote(master.Name, fp)
 			if err != nil {
 				log.Printf("Error reading file from path (%s):%s", path, err)
 			}
 		}
 	}
 
-	for _, agent := range agents {
+	for _, agent := range cli.Agents {
 		for _, fp := range agentFiles {
-			err := conn.CopyRemote(agent, fp)
+			err := conn.CopyRemote(agent.Name, fp)
 			if err != nil {
 				log.Printf("Error reading file from path (%s):%s", path, err)
 			}