fix: clean stuck argo hook image pull jobs

scripts/src/hwlab-node/control-actions.ts

@@ -503,8 +503,12 @@ export function nodeRuntimeSync(scoped: ReturnType<typeof parseNodeScopedDelegat
     "  tracking=$(kubectl -n \"$runtime_namespace\" get \"$job\" -o go-template='{{ index .metadata.annotations \"argocd.argoproj.io/tracking-id\" }}' 2>/dev/null || true)",
     "  failed=$(kubectl -n \"$runtime_namespace\" get \"$job\" -o go-template='{{ range .status.conditions }}{{ if or (eq .type \"Failed\") (eq .type \"FailureTarget\") }}{{ .status }} {{ end }}{{ end }}' 2>/dev/null || true)",
     "  case \"$tracking\" in \"$app:\"*) ;; *) continue ;; esac",
-    "  case \"$failed\" in *True*) ;; *) continue ;; esac",
     "  hook_name=${job#job.batch/}",
+    "  waiting=$(kubectl -n \"$runtime_namespace\" get pod -l \"job-name=$hook_name\" -o 'jsonpath={range .items[*].status.containerStatuses[*]}{.state.waiting.reason}{\" \"}{end}' 2>/dev/null || true)",
+    "  case \"$failed\" in",
+    "    *True*) ;;",
+    "    *) case \"$waiting\" in *ImagePullBackOff*|*ErrImagePull*|*CrashLoopBackOff*) ;; *) continue ;; esac ;;",
+    "  esac",
     "  failed_hook_count=$((failed_hook_count + 1))",
     "  if [ -z \"$failed_hooks\" ]; then failed_hooks=$hook_name; else failed_hooks=$failed_hooks,$hook_name; fi",
     "  if [ \"$dry_run\" = false ]; then",