Merge pull request #12 from digitalocean/varsha/pod-state

Pod state check: Checks the cluster for pods in unknown or failed state.

checks/basic/pod_status.go

@@ -0,0 +1,46 @@
+package basic
+
+import (
+	"fmt"
+
+	"github.com/digitalocean/clusterlint/checks"
+	"github.com/digitalocean/clusterlint/kube"
+	corev1 "k8s.io/api/core/v1"
+)
+
+func init() {
+	checks.Register(&podStatusCheck{})
+}
+
+type podStatusCheck struct{}
+
+// Name returns a unique name for this check.
+func (p *podStatusCheck) Name() string {
+	return "pod-state"
+}
+
+// Groups returns a list of group names this check should be part of.
+func (p *podStatusCheck) Groups() []string {
+	return []string{"workload-health"}
+}
+
+// Description returns a detailed human-readable description of what this check
+// does.
+func (p *podStatusCheck) Description() string {
+	return "Check if there are unhealthy pods in the cluster"
+}
+
+// Run runs this check on a set of Kubernetes objects. It can return warnings
+// (low-priority problems) and errors (high-priority problems) as well as an
+// error value indicating that the check failed to run.
+func (p *podStatusCheck) Run(objects *kube.Objects) (warnings []error, errors []error, err error) {
+	var e []error
+
+	for _, pod := range objects.Pods.Items {
+		if corev1.PodFailed == pod.Status.Phase || corev1.PodUnknown == pod.Status.Phase {
+			e = append(e, fmt.Errorf("Pod '%s' in namespace '%s' has state: %s. Pod state should be `Running`, `Pending` or `Succeeded`.", pod.GetName(), pod.GetNamespace(), pod.Status.Phase))
+		}
+	}
+
+	return nil, e, nil
+}

checks/basic/pod_status_test.go

@@ -0,0 +1,79 @@
+package basic
+
+import (
+	"fmt"
+	"testing"
+
+	"github.com/digitalocean/clusterlint/kube"
+	"github.com/stretchr/testify/assert"
+	corev1 "k8s.io/api/core/v1"
+)
+
+func TestMeta(t *testing.T) {
+	podStatusCheck := podStatusCheck{}
+	assert.Equal(t, "pod-state", podStatusCheck.Name())
+	assert.Equal(t, "Check if there are unhealthy pods in the cluster", podStatusCheck.Description())
+	assert.Equal(t, []string{"workload-health"}, podStatusCheck.Groups())
+}
+
+func TestPodStateError(t *testing.T) {
+	scenarios := []struct {
+		name     string
+		arg      *kube.Objects
+		expected []error
+	}{
+		{
+			name:     "no pods",
+			arg:      initPod(),
+			expected: nil,
+		},
+		{
+			name:     "pod with running status",
+			arg:      status(corev1.PodRunning),
+			expected: nil,
+		},
+		{
+			name:     "pod with pending status",
+			arg:      status(corev1.PodPending),
+			expected: nil,
+		},
+		{
+			name:     "pod with succeeded status",
+			arg:      status(corev1.PodSucceeded),
+			expected: nil,
+		},
+		{
+			name: "pod with failed status",
+			arg:  status(corev1.PodFailed),
+			expected: []error{
+				fmt.Errorf("Pod 'pod_foo' in namespace 'k8s' has state: Failed. Pod state should be `Running`, `Pending` or `Succeeded`."),
+			},
+		},
+		{
+			name: "pod with unknown status",
+			arg:  status(corev1.PodUnknown),
+			expected: []error{
+				fmt.Errorf("Pod 'pod_foo' in namespace 'k8s' has state: Unknown. Pod state should be `Running`, `Pending` or `Succeeded`."),
+			},
+		},
+	}
+
+	podStatusCheck := podStatusCheck{}
+
+	for _, scenario := range scenarios {
+		t.Run(scenario.name, func(t *testing.T) {
+			w, e, err := podStatusCheck.Run(scenario.arg)
+			assert.ElementsMatch(t, scenario.expected, e)
+			assert.Empty(t, w)
+			assert.Nil(t, err)
+		})
+	}
+}
+
+func status(status corev1.PodPhase) *kube.Objects {
+	objs := initPod()
+	objs.Pods.Items[0].Status = corev1.PodStatus{
+		Phase: status,
+	}
+	return objs
+}