Fix various issues with validating deployment

Signed-off-by: Kyle Squizzato <[email protected]>

.github/workflows/test.yml

@@ -71,4 +71,4 @@ jobs:
         uses: azure/setup-kubectl@v4
       - name: Run E2E tests
         run: |
-          make e2e-test
+          make test-e2e

Makefile

@@ -106,7 +106,7 @@ test: generate-all fmt vet envtest tidy external-crd ## Run tests.
 # Utilize Kind or modify the e2e tests to load the image locally, enabling compatibility with other vendors.
 .PHONY: test-e2e # Run the e2e tests against a Kind k8s instance that is spun up.
 test-e2e: cli-install
-	KIND_CLUSTER_NAME=$(KIND_CLUSTER_NAME) KIND_VERSION=$(KIND_VERSION) go test ./test/e2e/ -v -ginkgo.v
+	KIND_CLUSTER_NAME="hmc-test" KIND_VERSION=$(KIND_VERSION) go test ./test/e2e/ -v -ginkgo.v -timeout=2h
 
 .PHONY: lint
 lint: golangci-lint ## Run golangci-lint linter & yamllint
@@ -270,10 +270,10 @@ dev-aws: yq
 	@$(YQ) e ".data.credentials = \"${AWS_CREDENTIALS}\"" config/dev/awscredentials.yaml | $(KUBECTL) -n $(NAMESPACE) apply -f -
 
 .PHONY: dev-apply
-dev-apply: kind-deploy registry-deploy dev-push dev-deploy dev-templates dev-aws
+dev-apply: kind-deploy registry-deploy dev-push dev-deploy dev-templates dev-aws ## Deploy a kind cluster, local registry, push the image, deploy the controller, and apply the templates.
 
 .PHONY: dev-destroy
-dev-destroy: kind-undeploy registry-undeploy
+dev-destroy: kind-undeploy registry-undeploy ## Destroy the development environment by deleting the kind cluster and local registry.
 
 .PHONY: dev-aws-apply
 dev-aws-apply: ## Apply the AWS deployment
@@ -286,11 +286,17 @@ dev-aws-destroy: ## Delete the AWS deployment
 .PHONY: dev-aws-nuke
 dev-aws-nuke: ## Warning: Destructive! Nuke all AWS resources deployed by 'dev-aws-apply', prefix with CLUSTER_NAME to nuke a specific cluster.
 	@CLUSTER_NAME=$(CLUSTER_NAME) envsubst < config/dev/cloud_nuke.yaml.tpl > config/dev/cloud_nuke.yaml
-	$(CLOUDNUKE) aws --region us-west-2 --force --config config/dev/cloud_nuke.yaml --resource-type vpc,eip,nat-gateway,ec2-subnet,internet-gateway,network-interface,security-group
+	DISABLE_TELEMETRY=true $(CLOUDNUKE) aws --region $$AWS_REGION --config config/dev/cloud_nuke.yaml --resource-type vpc,eip,nat-gateway,ec2-subnet,elb,elbv2,internet-gateway,network-interface,security-group
 	@rm config/dev/cloud_nuke.yaml
 
+.PHONY: test-apply
+test-apply: kind-deploy registry-deploy dev-push dev-deploy dev-templates
+
+.PHONY: test-destroy
+test-destroy: kind-undeploy registry-undeploy
+
 .PHONY: cli-install
-cli-install: clusterawsadm clusterctl cloud-nuke
+cli-install: clusterawsadm clusterctl cloud-nuke yq ## Install the necessary CLI tools for deployment, development and testing.
 
 ##@ Dependencies
 

config/dev/deployment.yaml

@@ -1,17 +1,17 @@
 apiVersion: hmc.mirantis.com/v1alpha1
 kind: Deployment
 metadata:
-  name: aws-dev
+    name: bba1743d-e2e-test
 spec:
-  template: aws-standalone-cp
-  config:
-    region: us-east-2
-    publicIP: true
-    controlPlaneNumber: 1
-    workersNumber: 1
-    controlPlane:
-      amiID: ami-02f3416038bdb17fb
-      instanceType: t3.small
-    worker:
-      amiID: ami-02f3416038bdb17fb
-      instanceType: t3.small
+    config:
+        controlPlane:
+            amiID: ami-0989c067ff3da4b27
+            instanceType: t3.small
+        controlPlaneNumber: 1
+        publicIP: true
+        region: us-west-2
+        worker:
+            amiID: ami-0989c067ff3da4b27
+            instanceType: t3.small
+        workersNumber: 1
+    template: aws-standalone-cp

test/e2e/e2e_test.go

@@ -32,19 +32,19 @@ import (
 const namespace = "hmc-system"
 
 var _ = Describe("controller", Ordered, func() {
-	BeforeAll(func() {
-		By("building and deploying the controller-manager")
-		cmd := exec.Command("make", "dev-apply")
-		_, err := utils.Run(cmd)
-		Expect(err).NotTo(HaveOccurred())
-	})
-
-	AfterAll(func() {
-		By("removing the controller-manager")
-		cmd := exec.Command("make", "dev-destroy")
-		_, err := utils.Run(cmd)
-		Expect(err).NotTo(HaveOccurred())
-	})
+	// BeforeAll(func() {
+	// 	By("building and deploying the controller-manager")
+	// 	cmd := exec.Command("make", "test-apply")
+	// 	_, err := utils.Run(cmd)
+	// 	Expect(err).NotTo(HaveOccurred())
+	// })
+
+	// AfterAll(func() {
+	// 	By("removing the controller-manager")
+	// 	cmd := exec.Command("make", "test-destroy")
+	// 	_, err := utils.Run(cmd)
+	// 	Expect(err).NotTo(HaveOccurred())
+	// })
 
 	Context("Operator", func() {
 		It("should run successfully", func() {
@@ -54,7 +54,7 @@ var _ = Describe("controller", Ordered, func() {
 			By("validating that the controller-manager pod is running as expected")
 			verifyControllerUp := func() error {
 				// Ensure only one controller pod is running.
-				podList, err := kc.Client.CoreV1().Pods(namespace).List(context.Background(), metav1.ListOptions{
+				podList, err := kc.Client.CoreV1().Pods(kc.Namespace).List(context.Background(), metav1.ListOptions{
 					LabelSelector: "control-plane=controller-manager,app.kubernetes.io/name=cluster-api",
 				})
 				if err != nil {
@@ -81,8 +81,16 @@ var _ = Describe("controller", Ordered, func() {
 				}
 
 				return nil
-			}()
-			EventuallyWithOffset(1, verifyControllerUp, time.Minute, time.Second).Should(Succeed())
+			}
+			EventuallyWithOffset(1, func() error {
+				err := verifyControllerUp()
+				if err != nil {
+					_, _ = fmt.Fprintf(GinkgoWriter, "Controller pod validation failed: %v\n", err)
+					return err
+				}
+
+				return nil
+			}(), 5*time.Minute, time.Second).Should(Succeed())
 		})
 	})
 
@@ -100,29 +108,30 @@ var _ = Describe("controller", Ordered, func() {
 		})
 
 		AfterAll(func() {
-			// Purge the AWS resources, the AfterAll for the controller will
-			// clean up the management cluster.
-			cmd := exec.Command("make", "dev-aws-nuke")
-			_, err := utils.Run(cmd)
-			ExpectWithOffset(2, err).NotTo(HaveOccurred())
+			// // Purge the AWS resources, the AfterAll for the controller will
+			// // clean up the management cluster.
+			// cmd := exec.Command("make", "dev-aws-nuke")
+			// _, err := utils.Run(cmd)
+			// ExpectWithOffset(2, err).NotTo(HaveOccurred())
 		})
 
 		It("should work with an AWS provider", func() {
 			By("using the aws-standalone-cp template")
-			clusterName, err := utils.ConfigureDeploymentConfig(utils.ProviderAWS, utils.AWSStandaloneCPTemplate)
-			ExpectWithOffset(2, err).NotTo(HaveOccurred())
+			//clusterName, err := utils.ConfigureDeploymentConfig(utils.ProviderAWS, utils.TemplateAWSStandaloneCP)
+			//ExpectWithOffset(1, err).NotTo(HaveOccurred())
+
+			clusterName := "bba1743d-e2e-test"
 
 			cmd := exec.Command("make", "dev-aws-apply")
 			_, err = utils.Run(cmd)
 			ExpectWithOffset(2, err).NotTo(HaveOccurred())
-			EventuallyWithOffset(2, func() error {
+			_, _ = fmt.Fprintf(GinkgoWriter, "Waiting for resource validation to succeed\n")
+			Eventually(func() error {
 				return verifyProviderDeployed(context.Background(), kc, clusterName)
-			}(), 30*time.Minute, 10*time.Second).Should(Succeed())
-
+			}).WithTimeout(30 * time.Minute).WithPolling(5 * time.Second).Should(Succeed())
 			By("using the aws-hosted-cp template")
 			// TODO: Use the standalone control plane resources to craft a hosted
 			// control plane and test it.
-
 		})
 	})
 })