fix(cluster): 等待 Worker 完全退出后执行跨节点验收

修正缩容期间 Deployment 已显示零就绪但旧 Worker Pod 尚在退出时仍可能领取验收任务的竞态。真实任务提交前现在会确认宁波 Worker Pod 数量为零，并在失败路径尝试取消未完成的测试任务。\n\n验证：node --check、bash -n、ShellCheck、git diff --check。

scripts/cluster/cross-node-file-e2e.mjs

@@ -129,6 +129,8 @@ assert(typeof apiKey === 'string' && apiKey.startsWith('sk-gw-'), 'API key creat
 assert(typeof apiKeyId === 'string' && apiKeyId.length > 0, 'API key ID is unavailable')
 
 let result
+let taskId
+let taskSucceeded = false
 try {
   const assignableModels = await request('/api/v1/api-keys/assignable-models', { token: jwt })
   const targetModels = (assignableModels.payload.items || []).filter(
@@ -185,7 +187,7 @@ try {
     form,
   })
   assert(accepted.response.status === 202, `image edit returned ${accepted.response.status}, expected 202`)
-  const taskId = accepted.payload.taskId || accepted.payload.task_id || accepted.payload.task?.id
+  taskId = accepted.payload.taskId || accepted.payload.task_id || accepted.payload.task?.id
   assert(/^[0-9a-f-]{36}$/.test(taskId || ''), 'image edit response has no task ID')
 
   let task
@@ -202,8 +204,12 @@ try {
   assert(task?.status === 'succeeded', `task ${taskId} did not succeed before timeout`)
   const resultURLs = [...new Set(collectHTTPSURLs(task.result))]
   assert(resultURLs.length > 0, 'task result contains no shared HTTPS URL')
+  taskSucceeded = true
   result = { taskId, sourceSHA256, resultURL: resultURLs[0], model }
 } finally {
+  if (taskId && !taskSucceeded) {
+    await request(`/api/v1/tasks/${taskId}/cancel`, { method: 'POST', token: apiKey }).catch(() => {})
+  }
   if (apiKeyId) {
     await request(`/api/v1/api-keys/${apiKeyId}`, { method: 'DELETE', token: jwt }).catch(() => {})
   }

scripts/cluster/run-cross-node-file-e2e.sh

@@ -26,24 +26,29 @@ set -euo pipefail
 k3s kubectl scale deployment/easyai-worker-hongkong -n easyai --replicas=1 >/dev/null
 k3s kubectl rollout status deployment/easyai-worker-hongkong -n easyai --timeout=300s
 k3s kubectl scale deployment/easyai-worker-ningbo -n easyai --replicas=0 >/dev/null
-for _ in $(seq 1 60); do
+for _ in $(seq 1 180); do
   pods=$(k3s kubectl get pods -n easyai \
     -l app.kubernetes.io/name=easyai-worker,easyai.io/site=hongkong \
     -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')
   pod_count=$(wc -w <<<"$pods" | tr -d ' ')
+  ningbo_pods=$(k3s kubectl get pods -n easyai \
+    -l app.kubernetes.io/name=easyai-worker,easyai.io/site=ningbo \
+    -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}')
+  ningbo_pod_count=$(wc -w <<<"$ningbo_pods" | tr -d ' ')
   if [[ $pod_count == 1 ]]; then
     pod=$pods
     process_role=$(k3s kubectl get pod "$pod" -n easyai \
       -o jsonpath='{.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}')
     pod_ready=$(k3s kubectl get pod "$pod" -n easyai \
       -o jsonpath='{.status.containerStatuses[0].ready}')
-    if [[ $process_role == worker && $pod_ready == true ]]; then
+    if [[ $process_role == worker && $pod_ready == true && $ningbo_pod_count == 0 ]]; then
       break
     fi
   fi
   sleep 1
 done
 [[ ${pod_count:-0} == 1 && ${process_role:-} == worker && ${pod_ready:-} == true ]]
+[[ ${ningbo_pod_count:-1} -eq 0 ]]
 [[ $(k3s kubectl get deployment easyai-api-ningbo -n easyai \
   -o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="AI_GATEWAY_PROCESS_ROLE")].value}') == api ]]
 ningbo_worker_ready=$(k3s kubectl get deployment easyai-worker-ningbo -n easyai \