feat(run-pod): allow image pull retries in run pod collector (#1811)

* allow image pull retries in run pod collector

* fix formatting

Author: diamonwiggins

config/crds/troubleshoot.sh_collectors.yaml

@@ -8947,6 +8947,8 @@ spec:
                       type: object
                     runPod:
                       properties:
+                        allowImagePullRetries:
+                          type: boolean
                         annotations:
                           additionalProperties:
                             type: string

config/crds/troubleshoot.sh_preflights.yaml

@@ -10767,6 +10767,8 @@ spec:
                       type: object
                     runPod:
                       properties:
+                        allowImagePullRetries:
+                          type: boolean
                         annotations:
                           additionalProperties:
                             type: string

config/crds/troubleshoot.sh_supportbundles.yaml

@@ -10798,6 +10798,8 @@ spec:
                       type: object
                     runPod:
                       properties:
+                        allowImagePullRetries:
+                          type: boolean
                         annotations:
                           additionalProperties:
                             type: string

pkg/apis/troubleshoot/v1beta2/collector_shared.go

@@ -107,13 +107,14 @@ type Run struct {
 }
 
 type RunPod struct {
-	CollectorMeta   `json:",inline" yaml:",inline"`
-	Name            string            `json:"name,omitempty" yaml:"name,omitempty"`
-	Namespace       string            `json:"namespace" yaml:"namespace"`
-	Timeout         string            `json:"timeout,omitempty" yaml:"timeout,omitempty"`
-	ImagePullSecret *ImagePullSecrets `json:"imagePullSecret,omitempty" yaml:"imagePullSecret,omitempty"`
-	PodSpec         corev1.PodSpec    `json:"podSpec,omitempty" yaml:"podSpec,omitempty"`
-	Annotations     map[string]string `json:"annotations,omitempty" yaml:"annotations,omitempty"`
+	CollectorMeta         `json:",inline" yaml:",inline"`
+	Name                  string            `json:"name,omitempty" yaml:"name,omitempty"`
+	Namespace             string            `json:"namespace" yaml:"namespace"`
+	Timeout               string            `json:"timeout,omitempty" yaml:"timeout,omitempty"`
+	ImagePullSecret       *ImagePullSecrets `json:"imagePullSecret,omitempty" yaml:"imagePullSecret,omitempty"`
+	PodSpec               corev1.PodSpec    `json:"podSpec,omitempty" yaml:"podSpec,omitempty"`
+	Annotations           map[string]string `json:"annotations,omitempty" yaml:"annotations,omitempty"`
+	AllowImagePullRetries bool              `json:"allowImagePullRetries,omitempty" yaml:"allowImagePullRetries,omitempty"`
 }
 
 type RunDaemonSet struct {

pkg/collect/run_pod.go

@@ -171,7 +171,7 @@ func runWithoutTimeout(ctx context.Context, bundlePath string, clientConfig *res
 		}
 		if status.Status.Phase == corev1.PodPending {
 			for _, v := range status.Status.ContainerStatuses {
-				if v.State.Waiting != nil && v.State.Waiting.Reason == "ImagePullBackOff" {
+				if v.State.Waiting != nil && v.State.Waiting.Reason == "ImagePullBackOff" && !runPodCollector.AllowImagePullRetries {
 					return nil, errors.Errorf("run pod aborted after getting pod status 'ImagePullBackOff'")
 				}
 
@@ -355,6 +355,11 @@ func RunPodsReadyNodes(ctx context.Context, client v1.CoreV1Interface, opts RunP
 
 // RunPodLogs runs a pod to completion on a node and returns its logs
 func RunPodLogs(ctx context.Context, client v1.CoreV1Interface, podSpec *corev1.Pod) ([]byte, error) {
+	return RunPodLogsWithOptions(ctx, client, podSpec, false)
+}
+
+// RunPodLogsWithOptions runs a pod to completion on a node and returns its logs with configurable options
+func RunPodLogsWithOptions(ctx context.Context, client v1.CoreV1Interface, podSpec *corev1.Pod, allowImagePullRetries bool) ([]byte, error) {
 	// 1. Create
 	pod, err := client.Pods(podSpec.Namespace).Create(ctx, podSpec, metav1.CreateOptions{})
 	if err != nil {
@@ -380,7 +385,7 @@ func RunPodLogs(ctx context.Context, client v1.CoreV1Interface, podSpec *corev1.
 
 		if pod.Status.Phase == corev1.PodPending {
 			for _, v := range pod.Status.ContainerStatuses {
-				if v.State.Waiting != nil && v.State.Waiting.Reason == "ImagePullBackOff" {
+				if v.State.Waiting != nil && v.State.Waiting.Reason == "ImagePullBackOff" && !allowImagePullRetries {
 					return nil, errors.New("wait for pod aborted after getting pod status 'ImagePullBackOff'")
 				}
 			}

schemas/collector-troubleshoot-v1beta2.json

@@ -7856,6 +7856,9 @@
                   "namespace"
                 ],
                 "properties": {
+                  "allowImagePullRetries": {
+                    "type": "boolean"
+                  },
                   "annotations": {
                     "type": "object",
                     "additionalProperties": {

schemas/preflight-troubleshoot-v1beta2.json

@@ -10642,6 +10642,9 @@
                   "namespace"
                 ],
                 "properties": {
+                  "allowImagePullRetries": {
+                    "type": "boolean"
+                  },
                   "annotations": {
                     "type": "object",
                     "additionalProperties": {

schemas/supportbundle-troubleshoot-v1beta2.json

@@ -10688,6 +10688,9 @@
                   "namespace"
                 ],
                 "properties": {
+                  "allowImagePullRetries": {
+                    "type": "boolean"
+                  },
                   "annotations": {
                     "type": "object",
                     "additionalProperties": {