MCO-2371: Add back "should match os version" test

This change reverts MCO-2370 and restores the test that checks if the OS
of the job node's is the expected one based on the job name.
This change also introduced a few changes compared to the original
implementation:
- Assumes that the default OS is now RHEL 10
- Assumes OSImageStreams is now always available
- Doesn't use the debug pods to check mixed clusters

Signed-off-by: Pablo Rodriguez Nava <git@amail.pablintino.eu>

Author: pablintino

test/extended/ci/job_names.go

@@ -10,8 +10,8 @@ import (
 	g "github.com/onsi/ginkgo/v2"
 	o "github.com/onsi/gomega"
 	v1 "github.com/openshift/api/config/v1"
+	mcfgv1 "github.com/openshift/api/machineconfiguration/v1"
 	mcv1client "github.com/openshift/client-go/machineconfiguration/clientset/versioned"
-	kapierrs "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	e2e "k8s.io/kubernetes/test/e2e/framework"
 	e2eskipper "k8s.io/kubernetes/test/e2e/framework/skipper"
@@ -173,97 +173,129 @@ var _ = g.Describe("[sig-ci] [Early] prow job name", func() {
 	})
 
 	g.It("should match os version", func() {
-		// TODO: @pablintino https://redhat.atlassian.net/browse/MCO-2371
-		e2eskipper.Skipf("Temporarily disabled until RHEL-10 switchover, see MCO-2371")
 		if jobName == "" {
 			e2eskipper.Skipf("JOB_NAME env var not set, skipping")
 		}
-
-		jobIsRHCOS10 := strings.Contains(jobName, "rhcos10")
-		jobIsMixedRHCOSVer := strings.Contains(jobName, "rhcos9-10")
+		nodeOSJobType := getNodeOSJobType(jobName)
 
 		isMicroShift, err := exutil.IsMicroShiftCluster(oc.AdminKubeClient())
 		o.Expect(err).NotTo(o.HaveOccurred())
-		if isMicroShift {
-			if jobIsRHCOS10 {
-				// TODO(muller): Assume we do not have RHCOS10 microshift jobs now. If someone adds a RHCOS10 job, this failure
-				// should force them to figure out how to detect RHCOS10 in microshift and update this test.
-				e2e.Failf("TODO: job name %q indicates RHCOS10 which cannot be checked for MicroShift clusters now", jobName)
-				return
-			}
-
-			e2eskipper.Skip("Cannot check RHCOS for MicroShift clusters")
-		}
-
 		isHyperShift, err := exutil.IsHypershift(context.TODO(), oc.AdminConfigClient())
 		o.Expect(err).NotTo(o.HaveOccurred())
-		if isHyperShift {
-			if jobIsRHCOS10 {
-				// TODO(muller): Assume we do not have RHCOS10 hypershift jobs now. If someone adds a RHCOS10 job, this failure
-				// should force them to figure out how to detect RHCOS10 in hypershift and update this test.
-				e2e.Failf("TODO: job name %q indicates RHCOS10 which cannot be checked for HyperShift clusters now", jobName)
-				return
-			}
 
-			e2eskipper.Skip("Cannot check RHCOS for HyperShift clusters")
+		if isMicroShift {
+			validateMicroshiftNodeOS(nodeOSJobType, jobName)
+		} else if isHyperShift {
+			validateHypershiftNodeOS(nodeOSJobType, jobName)
+		} else {
+			validateStandaloneNodeOS(oc, nodeOSJobType)
 		}
+	})
+})
 
-		clusterIsRHCOS10 := isRHCOS10(oc.MachineConfigurationClient())
+func fetchMCPStreams(machineConfigClient mcv1client.Interface) map[string]string {
+	mcps, err := machineConfigClient.MachineconfigurationV1().MachineConfigPools().List(context.TODO(), metav1.ListOptions{})
+	o.Expect(err).NotTo(o.HaveOccurred(), "Error fetching MCPs")
 
-		// Mixed RHCOS version clusters (e.g. rhcos9-10) have nodes running both
-		// RHCOS 9 and RHCOS 10. Validate that we see both versions across nodes.
-		if jobIsMixedRHCOSVer {
-			hasRHCOS9, hasRHCOS10 := hasMixedRHCOSNodes(oc)
-			if !hasRHCOS9 || !hasRHCOS10 {
-				e2e.Failf("job name %q is a mixed RHCOS version cluster but nodes do not have both versions (hasRHCOS9=%v, hasRHCOS10=%v)", jobName, hasRHCOS9, hasRHCOS10)
-			}
-			e2e.Logf("job name %q is a mixed RHCOS version cluster with both RHCOS 9 and RHCOS 10 nodes", jobName)
-			return
-		}
+	// Fetch the OSImageStream CR to get the cluster default
+	osImageStream, err := machineConfigClient.MachineconfigurationV1().OSImageStreams().Get(context.TODO(), "cluster", metav1.GetOptions{})
+	o.Expect(err).NotTo(o.HaveOccurred(), "Error getting OSImageStream singleton")
 
-		if clusterIsRHCOS10 && !jobIsRHCOS10 {
-			e2e.Failf("cluster runs RHCOS10 so job name %q must contain 'rhcos10'", jobName)
+	var workerMCP *mcfgv1.MachineConfigPool
+	for _, mcp := range mcps.Items {
+		if mcp.Name == "worker" {
+			workerMCP = &mcp
+			break
 		}
-		if !clusterIsRHCOS10 && jobIsRHCOS10 {
-			e2e.Failf("cluster does not run RHCOS10 so job name %q must not contain 'rhcos10')", jobName)
+	}
+	o.Expect(workerMCP).NotTo(o.BeNil(), "Cluster worker MCP does not exist")
+
+	mcpsStreams := make(map[string]string)
+	for _, mcp := range mcps.Items {
+		stream := mcp.Spec.OSImageStream.Name
+		if stream == "" {
+			// It can be a custom MCP using the stream of the parent
+			if mcp.Name == "master" || mcp.Name == "worker" || mcp.Name == "arbiter" {
+				// Not a custom pool: It's using the cluster-wide stream
+				stream = osImageStream.Status.DefaultStream
+			} else if workerMCP.Spec.OSImageStream.Name != "" {
+				stream = workerMCP.Spec.OSImageStream.Name
+			} else {
+				stream = osImageStream.Status.DefaultStream
+			}
 		}
-	})
-})
+		mcpsStreams[mcp.Name] = stream
+	}
 
-// isRHCOS10 checks whether the cluster is running RHEL 10 by examining the worker
-// MCP's OSImageStream setting, falling back to the cluster-wide default stream
-// from the OSImageStream singleton if the MCP does not specify one.
-func isRHCOS10(machineConfigClient mcv1client.Interface) bool {
-	mcp, err := machineConfigClient.MachineconfigurationV1().MachineConfigPools().Get(context.TODO(), "worker", metav1.GetOptions{})
-	o.Expect(err).NotTo(o.HaveOccurred(), "Error getting worker MCP")
+	return mcpsStreams
+}
 
-	if mcp.Spec.OSImageStream.Name != "" {
-		return mcp.Spec.OSImageStream.Name == "rhel-10"
+func validateMicroshiftNodeOS(nodeOSJobType NodeOSJobType, jobName string) {
+	if nodeOSJobType != RHEL9 {
+		// TODO(muller): Assume we do not have RHCOS10 microshift jobs now. If someone adds a RHCOS10 job, this failure
+		// should force them to figure out how to detect RHCOS10 in microshift and update this test.
+		e2e.Failf("TODO: job name %q indicates RHCOS10 which cannot be checked for MicroShift clusters now", jobName)
+		return
 	}
 
-	osImageStream, err := machineConfigClient.MachineconfigurationV1().OSImageStreams().Get(context.TODO(), "cluster", metav1.GetOptions{})
-	if kapierrs.IsNotFound(err) {
-		return false
+	e2eskipper.Skip("Cannot check RHCOS for MicroShift clusters")
+
+}
+
+func validateHypershiftNodeOS(nodeOSJobType NodeOSJobType, jobName string) {
+	if nodeOSJobType != RHEL9 {
+		// TODO(muller): Assume we do not have RHCOS10 hypershift jobs now. If someone adds a RHCOS10 job, this failure
+		// should force them to figure out how to detect RHCOS10 in hypershift and update this test.
+		e2e.Failf("TODO: job name %q indicates RHCOS10 which cannot be checked for HyperShift clusters now", jobName)
+		return
 	}
-	o.Expect(err).NotTo(o.HaveOccurred(), "Error getting OSImageStream singleton")
 
-	return osImageStream.Status.DefaultStream == "rhel-10"
+	e2eskipper.Skip("Cannot check RHCOS for HyperShift clusters")
 }
 
-// hasMixedRHCOSNodes scans all nodes and returns whether both RHCOS 9 and
-// RHCOS 10 nodes are present, based on node OSImage.
-func hasMixedRHCOSNodes(oc *exutil.CLI) (hasRHCOS9, hasRHCOS10 bool) {
-	nodes, err := oc.AdminKubeClient().CoreV1().Nodes().List(context.TODO(), metav1.ListOptions{})
-	o.Expect(err).NotTo(o.HaveOccurred(), "Error listing nodes")
-
-	for _, node := range nodes.Items {
-		osImage := node.Status.NodeInfo.OSImage
-		e2e.Logf("node %s has OSImage %q", node.Name, osImage)
-		if strings.Contains(osImage, "CoreOS 10.") {
-			hasRHCOS10 = true
-		} else if strings.Contains(osImage, "CoreOS 9.") {
-			hasRHCOS9 = true
+func validateStandaloneNodeOS(oc *exutil.CLI, nodeOSJobType NodeOSJobType) {
+	mcpsStreams := fetchMCPStreams(oc.MachineConfigurationClient())
+
+	if nodeOSJobType == RHEL9 {
+		for mcp, stream := range mcpsStreams {
+			o.Expect(stream).To(o.Equal("rhel-9"), "MCP %s uses %s as stream but was expecting rhel-9", mcp, stream)
+		}
+	} else if nodeOSJobType == RHEL10 {
+		for mcp, stream := range mcpsStreams {
+			o.Expect(stream).To(o.Equal("rhel-10"), "MCP %s uses %s as stream but was expecting rhel-10", mcp, stream)
 		}
+	} else if nodeOSJobType == RHELMixed910 {
+		var hasRHCOS9 bool
+		var hasRHCOS10 bool
+		for _, stream := range mcpsStreams {
+			if stream == "rhel-9" {
+				hasRHCOS9 = true
+			} else if stream == "rhel-10" {
+				hasRHCOS10 = true
+			}
+		}
+		o.Expect(hasRHCOS9).To(o.BeTrue(), "The cluster is a RHEL 9 and RHEL 10 mixed cluster but it doesn't use RHEL 9")
+		o.Expect(hasRHCOS10).To(o.BeTrue(), "The cluster is a RHEL 9 and RHEL 10 mixed cluster but it doesn't use RHEL 10")
 	}
-	return
+}
+
+type NodeOSJobType int
+
+const (
+	RHEL10 NodeOSJobType = iota
+	RHEL9
+	RHELMixed910
+)
+
+func getNodeOSJobType(jobName string) NodeOSJobType {
+	if strings.Contains(jobName, "rhcos9-10") {
+		return RHELMixed910
+	} else if strings.Contains(jobName, "rhcos10") {
+		return RHEL10
+	} else if strings.Contains(jobName, "rhcos9") {
+		return RHEL9
+	}
+
+	// If not explicit OS in the name assume RHEL 10
+	return RHEL10
 }