Persists pause reason during eds status update

DataDog · Jan 25, 2021 · 2b56fe4 · 2b56fe4
1 parent cb836bd
commit 2b56fe4
Show file tree

Hide file tree

Showing 10 changed files with 14 additions and 8 deletions.
diff --git a/api/v1alpha1/extendeddaemonset_types.go b/api/v1alpha1/extendeddaemonset_types.go
@@ -143,6 +143,8 @@ const (
 	ExtendedDaemonSetStatusReasonPreStartHookError ExtendedDaemonSetStatusReason = "PreStartHookError"
 	// ExtendedDaemonSetStatusReasonPostStartHookError represent PostStartHookError as the reason for the ExtendedDaemonSet status state
 	ExtendedDaemonSetStatusReasonPostStartHookError ExtendedDaemonSetStatusReason = "PostStartHookError"
+	// ExtendedDaemonSetStatusReasonStartError represent StartError as the reason for the ExtendedDaemonSet status state
+	ExtendedDaemonSetStatusReasonStartError ExtendedDaemonSetStatusReason = "StartError"
 	// ExtendedDaemonSetStatusReasonUnknown represents an Unknown reason for the status state
 	ExtendedDaemonSetStatusReasonUnknown ExtendedDaemonSetStatusReason = "Unknown"
 )

diff --git a/controllers/extendeddaemonset_e2e_test.go b/controllers/extendeddaemonset_e2e_test.go
@@ -492,6 +492,7 @@ var _ = Describe("ExtendedDaemonSet e2e successful canary deployment update", fu
 				Name:      eds.Status.ActiveReplicaSet,
 			}
 			Eventually(withERS(ersKey, ers, func() bool {
+				fmt.Fprintf(GinkgoWriter, "ERS status:\n%s\n", spew.Sdump(ers.Status))
 				return ers.Status.Status == "active" && int(ers.Status.Available) == len(nodeList.Items)
 			}), timeout, interval).Should(BeTrue())
 		})
@@ -538,7 +539,6 @@ var _ = Describe("ExtendedDaemonSet e2e successful canary deployment update", fu
 				if eds.Annotations == nil {
 					eds.Annotations = make(map[string]string)
 				}
-
 				eds.Annotations[datadoghqv1alpha1.ExtendedDaemonSetCanaryValidAnnotationKey] = canaryReplicaSet
 				eds.Spec.Template.Spec.Containers[0].Image = fmt.Sprintf("k8s.gcr.io/pause:3.1")
 			}
@@ -548,12 +548,9 @@ var _ = Describe("ExtendedDaemonSet e2e successful canary deployment update", fu
 				func() string { return "Unable to update the EDS" },
 			)
 
-			Expect(k8sClient.Get(ctx, key, eds)).Should(Succeed())
-			fmt.Fprintf(GinkgoWriter, "EDS status:\n%s\n", spew.Sdump(eds.Status))
-
 			Eventually(withEDS(key, eds, func() bool {
 				return eds.Status.ActiveReplicaSet == canaryReplicaSet
-			}), timeout, interval).Should(BeTrue())
+			}), timeout, interval).Should(BeTrue(), "eds status should be updated")
 
 			canaryPods := &corev1.PodList{}
 			listOptions := []client.ListOption{

diff --git a/controllers/extendeddaemonsetreplicaset/strategy/canary.go b/controllers/extendeddaemonsetreplicaset/strategy/canary.go
@@ -200,7 +200,7 @@ func manageCanaryPodFailures(pods []*v1.Pod, params *Parameters, result *Result,
 			// Unpausing is a manual action and takes precedence
 			result.IsPaused = false
 			result.PausedReason = ""
-		case !result.IsPaused && autoPauseEnabled:
+		case autoPauseEnabled:
 			// Handle cases related to failure to start states
 			if cannotStart {
 				result.IsPaused = true

diff --git a/examples/failure-cases/foo-eds_config_error.yaml b/examples/failure-cases/foo-eds_config_error.yaml
@@ -17,6 +17,7 @@ spec:
         maxRestartsDuration: 10m
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec:

diff --git a/examples/failure-cases/foo-eds_image_pull_error.yaml b/examples/failure-cases/foo-eds_image_pull_error.yaml
@@ -17,6 +17,7 @@ spec:
         maxRestartsDuration: 10m
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec:

diff --git a/examples/failure-cases/foo-eds_restarts.yaml b/examples/failure-cases/foo-eds_restarts.yaml
@@ -13,10 +13,11 @@ spec:
         maxRestarts: 2
       autoFail:
         enabled: true
-        maxRestarts: 10
+        maxRestarts: 5
         maxRestartsDuration: 10m
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec:

diff --git a/examples/failure-cases/foo-eds_slow_start_missing_volume.yaml b/examples/failure-cases/foo-eds_slow_start_missing_volume.yaml
@@ -18,6 +18,7 @@ spec:
         maxRestartsDuration: 10m
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec:

diff --git a/examples/foo-eds_v1.yaml b/examples/foo-eds_v1.yaml
@@ -11,6 +11,7 @@ spec:
         enabled: false
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec: 

diff --git a/examples/foo-eds_v2.yaml b/examples/foo-eds_v2.yaml
@@ -12,6 +12,7 @@ spec:
         maxRestarts: 5
     rollingUpdate:
       maxParallelPodCreation: 1
+      maxUnavailable: 2
       slowStartIntervalDuration: 1m
   template:
     spec:

diff --git a/pkg/controller/utils/pod/pod.go b/pkg/controller/utils/pod/pod.go
@@ -183,7 +183,8 @@ func convertReasonToEDSStatusReason(reason string) datadoghqv1alpha1.ExtendedDae
 		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerConfigError,
 		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonCreateContainerError,
 		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPreStartHookError,
-		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPostStartHookError:
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonPostStartHookError,
+		datadoghqv1alpha1.ExtendedDaemonSetStatusReasonStartError:
 		return t
 	default:
 		return datadoghqv1alpha1.ExtendedDaemonSetStatusReasonUnknown