UPSTREAM: <carry>: migrate olmv1 QE stress cases

openshift/tests-extension/.openshift-tests-extension/openshift_payload_olmv1.json

@@ -565,6 +565,22 @@
       "exclude": "topology==\"External\""
     }
   },
+  {
+    "name": "[sig-olmv1][Jira:OLM] OLM v1 for stress PolarionID:81509-[OTP][Skipped:Disconnected][OlmStress]olmv1 create mass operator to see if they all are installed successfully [Slow][Timeout:330m]",
+    "labels": {
+      "Extended": {},
+      "NonHyperShiftHOST": {},
+      "StressTest": {}
+    },
+    "resources": {
+      "isolation": {}
+    },
+    "source": "openshift:payload:olmv1",
+    "lifecycle": "blocking",
+    "environmentSelector": {
+      "exclude": "topology==\"External\""
+    }
+  },
   {
     "name": "[sig-olmv1][OCPFeatureGate:NewOLM][Skipped:Disconnected] OLMv1 Catalogs should be installed",
     "labels": {},

openshift/tests-extension/test/qe/README.md

@@ -244,7 +244,20 @@ All migrated test case code needs the following changes to run in the new test f
 
 **Environment Validation for Disconnected-Supporting Migrated Test Cases:**
 
-If your test case supports disconnected environments, you MUST call `ValidateAccessEnvironment` at the beginning of the test:
+**When to use `ValidateAccessEnvironment`:**
+
+1. **Test cases that create ClusterCatalog or ClusterExtension**:
+   - If your test supports disconnected environments (both connected+disconnected, or disconnected-only)
+   - AND your test creates ClusterCatalog or ClusterExtension resources
+   - **MUST** call `ValidateAccessEnvironment(oc)` at the beginning of the test
+   - This applies to both newly created test cases and migrated test cases
+
+2. **Test cases that do NOT create both ClusterCatalog or ClusterExtension**:
+   - Optional to use `ValidateAccessEnvironment(oc)`
+   - Using it won't cause errors, but it's not required
+   - The validation is primarily for ensuring catalog images can be mirrored
+
+**Usage example:**
 
 ```go
 g.It("test case supporting disconnected", func() {

openshift/tests-extension/test/qe/specs/olmv1_stress.go

@@ -0,0 +1,146 @@
+package specs
+
+import (
+	"fmt"
+	"path/filepath"
+	"time"
+
+	g "github.com/onsi/ginkgo/v2"
+	o "github.com/onsi/gomega"
+	e2e "k8s.io/kubernetes/test/e2e/framework"
+
+	exutil "github.com/openshift/operator-framework-operator-controller/openshift/tests-extension/test/qe/util"
+	olmv1util "github.com/openshift/operator-framework-operator-controller/openshift/tests-extension/test/qe/util/olmv1util"
+)
+
+var _ = g.Describe("[sig-olmv1][Jira:OLM] OLM v1 for stress", func() {
+
+	defer g.GinkgoRecover()
+	var (
+		oc = exutil.NewCLIWithoutNamespace("default")
+	)
+
+	g.BeforeEach(func() {
+		exutil.SkipMicroshift(oc)
+		exutil.SkipNoOLMv1Core(oc)
+	})
+
+	// author: [email protected]
+	g.It("PolarionID:81509-[OTP][Skipped:Disconnected][OlmStress]olmv1 create mass operator to see if they all are installed successfully [Slow][Timeout:330m]", g.Label("StressTest"), g.Label("NonHyperShiftHOST"), func() {
+		var (
+			caseID                       = "81509"
+			prefixCatalog                = "catalog-" + caseID
+			prefixSa                     = "sa-" + caseID
+			prefixCe                     = "ce-" + caseID
+			prefixNs                     = "ns-" + caseID
+			prefixPackage                = "stress-olmv1-c"
+			prefixImage                  = "quay.io/olmqe/stress-index:vokv"
+			nsOc                         = "openshift-operator-controller"
+			nsCatalog                    = "openshift-catalogd"
+			catalogLabel                 = "control-plane=catalogd-controller-manager"
+			ocLabel                      = "control-plane=operator-controller-controller-manager"
+			baseDir                      = exutil.FixturePath("testdata", "olm")
+			clustercatalogTemplate       = filepath.Join(baseDir, "clustercatalog.yaml")
+			clusterextensionTemplate     = filepath.Join(baseDir, "clusterextension.yaml")
+			saClusterRoleBindingTemplate = filepath.Join(baseDir, "sa-admin.yaml")
+		)
+
+		if !olmv1util.IsPodReady(oc, nsCatalog, catalogLabel) {
+			_, _ = olmv1util.Get(oc, "pod", "-n", nsCatalog, "-l", catalogLabel, "-o", "yaml")
+			exutil.AssertWaitPollNoErr(fmt.Errorf("the pod with %s is not correct", catalogLabel), "the pod with app=catalog-operator is not correct")
+		}
+		if !olmv1util.IsPodReady(oc, nsOc, ocLabel) {
+			_, _ = olmv1util.Get(oc, "pod", "-n", nsOc, "-l", ocLabel, "-o", "yaml")
+			exutil.AssertWaitPollNoErr(fmt.Errorf("the pod with %s is not correct", ocLabel), "the pod with app=olm-operator is not correct")
+		}
+
+		startTime := time.Now().UTC()
+		e2e.Logf("Start time: %s", startTime.Format(time.RFC3339))
+
+		// for i := 0; i < 500; i++ {
+		for i := 900; i < 969; i++ {
+			// it is not enough with 330m for one case if we run 100 times
+			e2e.Logf("=================it is round %v=================", i)
+			ns := fmt.Sprintf("%s-%d", prefixNs, i)
+			clustercatalog := olmv1util.ClusterCatalogDescription{
+				Name:     fmt.Sprintf("%s-%d", prefixCatalog, i),
+				Imageref: fmt.Sprintf("%s%d", prefixImage, i),
+				Template: clustercatalogTemplate,
+			}
+			saCrb := olmv1util.SaCLusterRolebindingDescription{
+				Name:      fmt.Sprintf("%s-%d", prefixSa, i),
+				Namespace: ns,
+				Template:  saClusterRoleBindingTemplate,
+			}
+			ce := olmv1util.ClusterExtensionDescription{
+				Name:             fmt.Sprintf("%s-%d", prefixCe, i),
+				PackageName:      fmt.Sprintf("%s%d", prefixPackage, i),
+				Channel:          "alpha",
+				Version:          ">=0.0.1",
+				InstallNamespace: ns,
+				SaName:           fmt.Sprintf("%s-%d", prefixSa, i),
+				Template:         clusterextensionTemplate,
+			}
+			g.By(fmt.Sprintf("Create namespace for %d", i))
+			// defer oc.WithoutNamespace().AsAdmin().Run("delete").Args("ns", ns, "--ignore-not-found").Execute()
+			// it take time delete ns which is not necessary. currently 5.5h is not enough to delete them.
+			// so I prefer to keep ns to save case duration
+			err := oc.WithoutNamespace().AsAdmin().Run("create").Args("ns", ns).Execute()
+			o.Expect(err).NotTo(o.HaveOccurred())
+
+			o.Expect(olmv1util.Appearance(oc, exutil.Appear, "ns", ns)).To(o.BeTrue())
+
+			g.By(fmt.Sprintf("Create clustercatalog for %d", i))
+			e2e.Logf("=========Create clustercatalog %v=========", clustercatalog.Name)
+			defer clustercatalog.Delete(oc)
+			err = clustercatalog.CreateWithoutCheck(oc)
+			o.Expect(err).NotTo(o.HaveOccurred())
+			clustercatalog.WaitCatalogStatus(oc, "true", "Serving", 0)
+
+			g.By(fmt.Sprintf("Create SA for clusterextension for %d", i))
+			defer saCrb.Delete(oc)
+			saCrb.Create(oc)
+
+			g.By(fmt.Sprintf("check ce to be installed for %d", i))
+			e2e.Logf("=========Create clusterextension %v=========", ce.Name)
+			defer ce.Delete(oc)
+			err = ce.CreateWithoutCheck(oc)
+			o.Expect(err).NotTo(o.HaveOccurred())
+			ce.CheckClusterExtensionCondition(oc, "Progressing", "reason", "Succeeded", 10, 600, 0)
+			ce.WaitClusterExtensionCondition(oc, "Installed", "True", 0)
+		}
+
+		endTime := time.Now().UTC()
+		e2e.Logf("End time:  %v", endTime.Format(time.RFC3339))
+
+		duration := endTime.Sub(startTime)
+		minutes := int(duration.Minutes())
+		if minutes < 1 {
+			minutes = 1
+		}
+
+		podName, err := oc.AsAdmin().WithoutNamespace().Run("get").Args("pods", "-l", catalogLabel, "-o=jsonpath={.items[0].metadata.name}", "-n", nsCatalog).Output()
+		if err == nil {
+			if !olmv1util.WriteErrToArtifactDir(oc, nsCatalog, podName, "error", "Unhandled|Reconciler error|level=info", caseID, minutes) {
+				e2e.Logf("no error log into artifact for pod %s in %s", podName, nsCatalog)
+			}
+		}
+		podName, err = oc.AsAdmin().WithoutNamespace().Run("get").Args("pods", "-l", ocLabel, "-o=jsonpath={.items[0].metadata.name}", "-n", nsOc).Output()
+		if err == nil {
+			if !olmv1util.WriteErrToArtifactDir(oc, nsOc, podName, "error", "Unhandled|Reconciler error|level=info", caseID, minutes) {
+				e2e.Logf("no error log into artifact for pod %s in %s", podName, nsOc)
+			}
+		}
+
+		if !olmv1util.IsPodReady(oc, nsCatalog, catalogLabel) {
+			_, _ = olmv1util.Get(oc, "pod", "-n", nsCatalog, "-l", catalogLabel, "-o", "yaml")
+			exutil.AssertWaitPollNoErr(fmt.Errorf("the pod with %s is not correct", catalogLabel), "the pod with app=catalog-operator is not correct")
+		}
+		if !olmv1util.IsPodReady(oc, nsOc, ocLabel) {
+			_, _ = olmv1util.Get(oc, "pod", "-n", nsOc, "-l", ocLabel, "-o", "yaml")
+			exutil.AssertWaitPollNoErr(fmt.Errorf("the pod with %s is not correct", ocLabel), "the pod with app=olm-operator is not correct")
+		}
+
+	})
+
+})

openshift/tests-extension/test/qe/util/architecture/architecture.go

@@ -80,10 +80,10 @@ func SkipNonAmd64SingleArch(oc *exutil.CLI) Architecture {
 func getNodeArchitectures(oc *exutil.CLI) []string {
 	output, err := oc.WithoutNamespace().AsAdmin().Run("get").Args("nodes", "-o=jsonpath={.items[*].status.nodeInfo.architecture}").Output()
 	if err != nil {
-		e2e.Failf("unable to get cluster node architectures: %v", err)
+		g.Skip(fmt.Sprintf("unable to get cluster node architectures: %v", err))
 	}
 	if output == "" {
-		e2e.Failf("no nodes found or architecture information missing")
+		g.Skip("no nodes found or architecture information missing")
 	}
 	return strings.Fields(output) // Use Fields instead of Split to handle multiple spaces
 }
@@ -97,7 +97,7 @@ func getNodeArchitectures(oc *exutil.CLI) []string {
 func GetAvailableArchitecturesSet(oc *exutil.CLI) []Architecture {
 	architectureStrings := getNodeArchitectures(oc)
 	if len(architectureStrings) == 0 {
-		e2e.Failf("no node architectures found")
+		g.Skip("no node architectures found")
 	}
 
 	// Use map for deduplication with Architecture as key
@@ -199,7 +199,7 @@ func (a Architecture) String() string {
 func ClusterArchitecture(oc *exutil.CLI) Architecture {
 	architectureStrings := getNodeArchitectures(oc)
 	if len(architectureStrings) == 0 {
-		e2e.Failf("no node architectures found")
+		g.Skip("no node architectures found")
 	}
 
 	// Filter out empty strings and convert to Architecture
@@ -211,7 +211,7 @@ func ClusterArchitecture(oc *exutil.CLI) Architecture {
 	}
 
 	if len(architectures) == 0 {
-		e2e.Failf("no valid node architectures found")
+		g.Skip("no valid node architectures found")
 	}
 
 	// Check if all architectures are the same
@@ -267,7 +267,7 @@ func GetControlPlaneArch(oc *exutil.CLI) Architecture {
 
 	architectureStr = strings.TrimSpace(architectureStr)
 	if architectureStr == "" {
-		e2e.Failf("Control plane node %s has no architecture information", masterNode)
+		g.Skip(fmt.Sprintf("Control plane node %s has no architecture information", masterNode))
 	}
 
 	return FromString(architectureStr)

openshift/tests-extension/test/qe/util/olmv1util/helper.go

@@ -590,7 +590,7 @@ func HasExternalNetworkAccess(oc *exutil.CLI) bool {
 	// Note: In disconnected environments, curl will fail and bash will return non-zero exit code,
 	// causing DebugNodeWithChroot to return an error. We ignore this error and rely on output checking.
 	cmd := `timeout 10 curl -k https://quay.io > /dev/null 2>&1; [ $? -eq 0 ] && echo "connected"`
-	output, _ := exutil.DebugNodeWithChroot(oc, masterNode, "bash", "-c", cmd)
+	output, _ := exutil.DebugNodeWithOptionsAndChroot(oc, masterNode, []string{"--to-namespace=default"}, "bash", "-c", cmd)
 
 	// Check if the output contains "connected"
 	// - Connected environment: curl succeeds -> echo "connected" -> output contains "connected"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/metrics-endpoint.yml

@@ -0,0 +1,9 @@
+- endpoint: {{.PROMETHEUS_URL}}
+  token: {{.PROMETHEUS_TOKEN}}
+  step: 10s
+  skipTLSVerify: true
+  metrics:
+    - metrics-profiles/metrics-aggregated.yml
+  indexer:
+    type: local
+    metricsDirectory: collected-metrics-{{.UUID}}

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/metrics-profiles/metrics-aggregated.yml

@@ -0,0 +1,7 @@
+# Containers & pod metrics
+
+- query: (sum(irate(container_cpu_usage_seconds_total{container="manager",namespace="openshift-catalogd"}[2m]) * 100) by (container, pod)) > 0
+  metricName: containerCPU-Catlogd
+
+- query: (sum(irate(container_cpu_usage_seconds_total{container="manager",namespace="openshift-operator-controller"}[2m]) * 100) by (container, pod)) > 0
+  metricName: containerCPU-OpCon

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/pkg-ins-v1.yml

@@ -0,0 +1,67 @@
+---
+global:
+  gc: {{.GC}}
+  gcMetrics: {{.GC_METRICS}}
+  measurements:
+    - name: podLatency
+
+
+jobs:
+  - name: {{.OPERATION}}
+    jobType: create
+    jobIterations: {{.JOB_ITERATIONS}}
+    namespace: {{.OPERATION}}
+    namespacedIterations: {{.NAMESPACED_ITERATIONS}}
+    iterationsPerNamespace: {{.ITERATIONS_PER_NAMESPACE}}
+    cleanup: true
+    podWait: true
+    waitWhenFinished: true
+    maxWaitTimeout: {{.MAX_WAIT_TIMEOUT}}
+    jobIterationDelay: {{.JOB_ITERATION_DELAY}}
+    jobPause: {{.JOB_PAUSE}}
+    qps: {{.QPS}}
+    burst: {{.BURST}}
+    executionMode: parallel
+    verifyObjects: true
+    errorOnVerify: true
+    skipIndexing: false
+    preLoadImages: true
+    preLoadPeriod: 15s
+    churn: false
+    defaultMissingKeysWithZero: false
+    namespaceLabels:
+      security.openshift.io/scc.podSecurityLabelSync: false
+      pod-security.kubernetes.io/enforce: privileged
+      pod-security.kubernetes.io/audit: privileged
+      pod-security.kubernetes.io/warn: privileged
+    objects:
+
+      - objectTemplate: templates/catalogd.yml
+        replicas: 1
+        inputVars:
+          prefixImageName: "quay.io/olmqe/stress-index:vokv"
+        waitOptions:
+          customStatusPaths:
+            - key: ".conditions[] | select(.type==\"Serving\") | .status"
+              value: "True"
+
+      - objectTemplate: templates/clusterrole.yml
+        replicas: 1
+
+      - objectTemplate: templates/sa.yml
+        replicas: 1
+
+      - objectTemplate: templates/clusterrolebinding.yml
+        replicas: 1
+        inputVars:
+          prefixNamespace: {{.OPERATION}}
+
+      - objectTemplate: templates/ce.yml
+        replicas: 1
+        inputVars:
+          prefixNamespace: {{.OPERATION}}
+          prefixPkgName: {{.PREFIX_PKG_NAME_V1}}
+        waitOptions:
+          customStatusPaths:
+            - key: ".conditions[] | select(.type==\"Installed\") | .status"
+              value: "True"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/templates/catalogd.yml

@@ -0,0 +1,9 @@
+apiVersion: olm.operatorframework.io/v1
+kind: ClusterCatalog
+metadata:
+  name: "clustercatalog-{{.Iteration}}"
+spec:
+  source:
+    type: Image
+    image:
+      ref: "{{.prefixImageName}}{{.Iteration}}"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/templates/ce.yml

@@ -0,0 +1,12 @@
+apiVersion: olm.operatorframework.io/v1
+kind: ClusterExtension
+metadata:
+  name: "ce-{{.Iteration}}"
+spec:
+  namespace: "{{.prefixNamespace}}-{{.Iteration}}"
+  serviceAccount:
+    name: "ins-sa-{{.Iteration}}"
+  source:
+    sourceType: Catalog
+    catalog:
+      packageName: "{{.prefixPkgName}}{{.Iteration}}"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/templates/clusterrole.yml

@@ -0,0 +1,11 @@
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: "ins-admin-clusterrole-{{.Iteration}}"
+rules:
+  - apiGroups:
+    - "*"
+    resources:
+    - "*"
+    verbs:
+    - "*"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/templates/clusterrolebinding.yml

@@ -0,0 +1,12 @@
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: "ins-admin-clusterrole-binding-{{.Iteration}}"
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: "ins-admin-clusterrole-{{.Iteration}}"
+subjects:
+  - kind: ServiceAccount
+    name: "ins-sa-{{.Iteration}}"
+    namespace: "{{.prefixNamespace}}-{{.Iteration}}"

openshift/tests-extension/test/qe/util/stress/manifests/config/pkg-ins-v1/templates/sa.yml

@@ -0,0 +1,4 @@
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: "ins-sa-{{.Iteration}}"

openshift/tests-extension/test/qe/util/stress/util/ma.py

@@ -0,0 +1,6 @@
+#!/usr/bin/env python3
+
+from ma.cli.__main__ import main
+
+if __name__ == "__main__":
+    main()