Fix missing pause reason in status

* Add more `ExtendedDaemonSetStatusReason`
* Check also Init and Ephemeral container state

api/v1alpha1/extendeddaemonset_types.go

@@ -131,6 +131,18 @@ const (
 	ExtendedDaemonSetStatusRestartsTimeoutExceeded ExtendedDaemonSetStatusReason = "RestartsTimeoutExceeded"
 	// ExtendedDaemonSetStatusSlowStartTimeoutExceeded represents timeout on slow starts as the reason for the ExtendedDaemonSet status
 	ExtendedDaemonSetStatusSlowStartTimeoutExceeded ExtendedDaemonSetStatusReason = "SlowStartTimeoutExceeded"
+	// ExtendedDaemonSetStatusReasonErrImagePull represent ErrImagePull as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonErrImagePull ExtendedDaemonSetStatusReason = "ErrImagePull"
+	// ExtendedDaemonSetStatusReasonImagePullBackOff represent ImagePullBackOff as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonImagePullBackOff ExtendedDaemonSetStatusReason = "ImagePullBackOff"
+	// ExtendedDaemonSetStatusReasonCreateContainerConfigError represent CreateContainerConfigError as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonCreateContainerConfigError ExtendedDaemonSetStatusReason = "CreateContainerConfigError"
+	// ExtendedDaemonSetStatusReasonCreateContainerError represent CreateContainerError as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonCreateContainerError ExtendedDaemonSetStatusReason = "CreateContainerError"
+	// ExtendedDaemonSetStatusReasonPreStartHookError represent PreStartHookError as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonPreStartHookError ExtendedDaemonSetStatusReason = "PreStartHookError"
+	// ExtendedDaemonSetStatusReasonPostStartHookError represent PostStartHookError as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonPostStartHookError ExtendedDaemonSetStatusReason = "PostStartHookError"
 	// ExtendedDaemonSetStatusReasonUnknown represents an Unknown reason for the status state
 	ExtendedDaemonSetStatusReasonUnknown ExtendedDaemonSetStatusReason = "Unknown"
 )

controllers/extendeddaemonsetreplicaset/strategy/canary.go

@@ -147,7 +147,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 		restartingPodStatus string
 
 		cannotStart          bool
-		cannotStartPodReason string
+		cannotStartPodReason v1alpha1.ExtendedDaemonSetStatusReason
 		cannotStartPodStatus string
 	)
 
@@ -167,7 +167,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 		cannotStart, cannotStartReason = podUtils.CannotStart(pod)
 		if cannotStart {
 			cannotStartPodStatus = fmt.Sprintf("Pod %s cannot start with reason: %s", pod.ObjectMeta.Name, string(cannotStartReason))
-			cannotStartPodReason = string(cannotStartReason)
+			cannotStartPodReason = cannotStartReason
 		} else if autoPauseEnabled && podUtils.PendingCreate(pod) && params.Strategy.Canary.AutoPause.MaxSlowStartDuration != nil {
 			if time.Now().After(pod.Status.StartTime.Time.Add(params.Strategy.Canary.AutoPause.MaxSlowStartDuration.Duration)) {
 				params.Logger.Info(
@@ -178,7 +178,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 				cannotStart = true
 				cannotStartReason = v1alpha1.ExtendedDaemonSetStatusSlowStartTimeoutExceeded
 				cannotStartPodStatus = fmt.Sprintf("Pod %s cannot start with reason: %s", pod.ObjectMeta.Name, cannotStartReason)
-				cannotStartPodReason = string(cannotStartReason)
+				cannotStartPodReason = cannotStartReason
 			}
 		}
 
@@ -240,7 +240,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 			metav1.NewTime(newRestartTime),
 			v1alpha1.ConditionTypePodRestarting,
 			v1.ConditionTrue,
-			cannotStartPodReason,
+			string(cannotStartPodReason),
 			restartingPodStatus,
 			false,
 			true,
@@ -263,7 +263,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 		metav1.NewTime(now),
 		v1alpha1.ConditionTypePodCannotStart,
 		conditionStatus,
-		cannotStartPodReason,
+		string(cannotStartPodReason),
 		cannotStartPodStatus,
 		false,
 		true,

pkg/controller/utils/pod/pod.go

@@ -99,6 +99,11 @@ func GetPodConditionFromList(conditions []v1.PodCondition, conditionType v1.PodC
 	return -1, nil
 }
 
+func containerStatusList(pod *v1.Pod) []v1.ContainerStatus {
+	containersStatus := append(pod.Status.ContainerStatuses, pod.Status.InitContainerStatuses...)
+	return append(containersStatus, pod.Status.EphemeralContainerStatuses...)
+}
+
 // HighestRestartCount checks if a pod in the Canary deployment is restarting
 // This returns the count and the "reason" for the pod with the most restarts
 func HighestRestartCount(pod *v1.Pod) (int, datadoghqv1alpha1.ExtendedDaemonSetStatusReason) {
@@ -108,7 +113,7 @@ func HighestRestartCount(pod *v1.Pod) (int, datadoghqv1alpha1.ExtendedDaemonSetS
 		reason       datadoghqv1alpha1.ExtendedDaemonSetStatusReason
 	)
 
-	for _, s := range pod.Status.ContainerStatuses {
+	for _, s := range containerStatusList(pod) {
 		if s.RestartCount > restartCount {
 			restartCount = s.RestartCount
 			reason = datadoghqv1alpha1.ExtendedDaemonSetStatusReasonUnknown
@@ -126,7 +131,7 @@ func MostRecentRestart(pod *v1.Pod) (time.Time, datadoghqv1alpha1.ExtendedDaemon
 		restartTime time.Time
 		reason      datadoghqv1alpha1.ExtendedDaemonSetStatusReason
 	)
-	for _, s := range pod.Status.ContainerStatuses {
+	for _, s := range containerStatusList(pod) {
 		if s.RestartCount != 0 && s.LastTerminationState != (v1.ContainerState{}) && s.LastTerminationState.Terminated != (&v1.ContainerStateTerminated{}) {
 			if s.LastTerminationState.Terminated.FinishedAt.After(restartTime) {
 				restartTime = s.LastTerminationState.Terminated.FinishedAt.Time
@@ -158,17 +163,36 @@ func IsCannotStartReason(reason string) bool {
 
 // CannotStart returns true if the Pod is currently experiencing abnormal start condition
 func CannotStart(pod *v1.Pod) (bool, datadoghqv1alpha1.ExtendedDaemonSetStatusReason) {
-	for _, s := range pod.Status.ContainerStatuses {
+	for _, s := range containerStatusList(pod) {
 		if s.State.Waiting != nil && IsCannotStartReason(s.State.Waiting.Reason) {
-			return true, datadoghqv1alpha1.ExtendedDaemonSetStatusReason(s.State.Waiting.Reason)
+			return true, convertReasonToEDSStatusReason(s.State.Waiting.Reason)
 		}
 	}
 	return false, datadoghqv1alpha1.ExtendedDaemonSetStatusReasonUnknown
 }
 
+func convertReasonToEDSStatusReason(reason string) datadoghqv1alpha1.ExtendedDaemonSetStatusReason {
+	t := datadoghqv1alpha1.ExtendedDaemonSetStatusReason(reason)
+	switch t {
+	case datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCLB,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonOOM,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusRestartsTimeoutExceeded,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusSlowStartTimeoutExceeded,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonErrImagePull,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonImagePullBackOff,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerConfigError,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerError,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPreStartHookError,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPostStartHookError:
+		return t
+	default:
+		return datadoghqv1alpha1.ExtendedDaemonSetStatusReasonUnknown
+	}
+}
+
 // PendingCreate returns true if the Pod is pending create (may be an eventually resolving state)
 func PendingCreate(pod *v1.Pod) bool {
-	for _, s := range pod.Status.ContainerStatuses {
+	for _, s := range containerStatusList(pod) {
 		if s.State.Waiting != nil && s.State.Waiting.Reason == "ContainerCreating" {
 			return true
 		}

pkg/controller/utils/pod/pod_test.go

@@ -6,6 +6,7 @@
 package pod
 
 import (
+	"reflect"
 	"testing"
 	"time"
 
@@ -278,3 +279,62 @@ func newPod(now metav1.Time, ready bool, beforeSec int) *v1.Pod {
 		},
 	}
 }
+
+func Test_convertReasonToEDSStatusReason(t *testing.T) {
+	tests := []struct {
+		reason string
+		want   datadoghqv1alpha1.ExtendedDaemonSetStatusReason
+	}{
+		{
+			reason: "CrashLoopBackOff",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCLB,
+		},
+		{
+			reason: "OOMKilled",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonOOM,
+		},
+		{
+			reason: "RestartsTimeoutExceeded",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusRestartsTimeoutExceeded,
+		},
+		{
+			reason: "SlowStartTimeoutExceeded",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusSlowStartTimeoutExceeded,
+		},
+		{
+			reason: "ErrImagePull",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonErrImagePull,
+		},
+		{
+			reason: "ImagePullBackOff",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonImagePullBackOff,
+		},
+		{
+			reason: "CreateContainerConfigError",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerConfigError,
+		},
+		{
+			reason: "CreateContainerError",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerError,
+		},
+		{
+			reason: "PreStartHookError",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPreStartHookError,
+		},
+		{
+			reason: "PostStartHookError",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPostStartHookError,
+		},
+		{
+			reason: "does not exist",
+			want:   datadoghqv1alpha1.ExtendedDaemonSetStatusReasonUnknown,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.reason, func(t *testing.T) {
+			if got := convertReasonToEDSStatusReason(tt.reason); !reflect.DeepEqual(got, tt.want) {
+				t.Errorf("convertReasonToEDSStatusReason() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}