Merge pull request #554 from replicatedhq/laverya/pods-with-notready-containers-are-unhealthy

pods with containers that are not 'ready' are unhealthy

Author: laverya

pkg/analyze/cluster_pod_statuses.go

@@ -22,7 +22,7 @@ func clusterPodStatuses(analyzer *troubleshootv1beta2.ClusterPodStatuses, getChi
 
 	var pods []corev1.Pod
 	for fileName, fileContent := range collected {
-		podsNs := strings.TrimSuffix(fileName, ".json")
+		podsNs := strings.TrimSuffix(filepath.Base(fileName), ".json")
 		include := len(analyzer.Namespaces) == 0
 		for _, ns := range analyzer.Namespaces {
 			if ns == podsNs {
@@ -33,9 +33,14 @@ func clusterPodStatuses(analyzer *troubleshootv1beta2.ClusterPodStatuses, getChi
 		if include {
 			var nsPods corev1.PodList
 			if err := json.Unmarshal(fileContent, &nsPods); err != nil {
-				return nil, errors.Wrapf(err, "failed to unmarshal pods list for namespace %s", podsNs)
+				var nsPodsArr []corev1.Pod
+				if err := json.Unmarshal(fileContent, &nsPodsArr); err != nil {
+					return nil, errors.Wrapf(err, "failed to unmarshal pods list for namespace %s", podsNs)
+				}
+				pods = append(pods, nsPodsArr...)
+			} else {
+				pods = append(pods, nsPods.Items...)
 			}
-			pods = append(pods, nsPods.Items...)
 		}
 	}
 

pkg/k8sutil/pod.go

@@ -114,13 +114,20 @@ func IsPodUnhealthy(pod *corev1.Pod) bool {
 	}
 
 	reason := GetPodStatusReason(pod)
+	if PodStatusReason(reason) == PodStatusReasonCompleted {
+		return false // completed pods are healthy pods
+	}
+
+	if PodStatusReason(reason) != PodStatusReasonRunning {
+		return true // pods that are not completed or running are unhealthy
+	}
 
-	switch PodStatusReason(reason) {
-	case PodStatusReasonRunning:
-		fallthrough
-	case PodStatusReasonCompleted:
-		return false
+	// running pods with unready containers are not healthy
+	for _, containerStatus := range pod.Status.ContainerStatuses {
+		if !containerStatus.Ready {
+			return true
+		}
 	}
 
-	return true
+	return false
 }

pkg/k8sutil/pod_test.go

@@ -0,0 +1,101 @@
+package k8sutil
+
+import (
+	"github.com/stretchr/testify/require"
+	corev1 "k8s.io/api/core/v1"
+	"testing"
+)
+
+func TestIsPodUnhealthy(t *testing.T) {
+	andTrue := true
+	tests := []struct {
+		name string
+		pod  *corev1.Pod
+		want bool
+	}{
+		{
+			name: "healthy pod with init containers",
+			want: false,
+			pod: &corev1.Pod{
+				Status: corev1.PodStatus{
+					Phase:      "Running",
+					Conditions: []corev1.PodCondition{
+						// ignored here
+					},
+					InitContainerStatuses: []corev1.ContainerStatus{
+						{
+							Name: "init",
+							State: corev1.ContainerState{
+								Terminated: &corev1.ContainerStateTerminated{
+									ExitCode: 0,
+									Reason:   "Completed",
+								},
+							},
+							Ready:        true,
+							RestartCount: 2,
+							Image:        "projectcontour/contour:v1.11.0",
+							ImageID:      "docker://sha256:12878e02b6f969de1456b51b0093f289fd195db956837ccd75aa679e9dac24d9",
+							ContainerID:  "docker://50fc0794fa1402b1a48521e2fd380a92521a69db70ad25c9699c91fccd839d70",
+						},
+					},
+					ContainerStatuses: []corev1.ContainerStatus{
+						{
+							Name: "contour",
+							State: corev1.ContainerState{
+								Running: &corev1.ContainerStateRunning{},
+							},
+							Ready:        true,
+							RestartCount: 5,
+							Image:        "projectcontour/contour:v1.11.0",
+							ImageID:      "docker://sha256:12878e02b6f969de1456b51b0093f289fd195db956837ccd75aa679e9dac24d9",
+							ContainerID:  "docker://50fc0794fa1402b1a48521e2fd380a92521a69db70ad25c9699c91fccd839d70",
+							Started:      &andTrue,
+						},
+					},
+					QOSClass: corev1.PodQOSBestEffort,
+				},
+			},
+		},
+		{
+			name: "running pod with one unhealthy container",
+			want: true,
+			pod: &corev1.Pod{
+				Status: corev1.PodStatus{
+					Phase:      "Running",
+					Conditions: []corev1.PodCondition{
+						// ignored here
+					},
+					ContainerStatuses: []corev1.ContainerStatus{
+						{
+							Name: "envoy",
+							State: corev1.ContainerState{
+								Running: &corev1.ContainerStateRunning{},
+							},
+							Ready:        false,
+							RestartCount: 5,
+							Started:      &andTrue,
+						},
+						{
+							Name: "shutdown-manager",
+							State: corev1.ContainerState{
+								Running: &corev1.ContainerStateRunning{},
+							},
+							Ready:        true,
+							RestartCount: 5,
+							Started:      &andTrue,
+						},
+					},
+					QOSClass: corev1.PodQOSBestEffort,
+				},
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := require.New(t)
+
+			got := IsPodUnhealthy(tt.pod)
+			req.Equal(tt.want, got)
+		})
+	}
+}