Google Batch任务通知未送达Pub/Sub主题问题求助

我正在测试Google Batch任务，目标是任务成功完成后自动向指定Pub/Sub主题发送通知，但配置完成后，任务成功却没有通知送达主题，且无任何错误提示。

预期结果 #

• Batch任务成功运行，状态变为SUCCEEDED时，自动向指定Pub/Sub主题发送通知
• 通知能正常送达关联订阅，可通过控制台查看

执行的实验步骤 #

1. 编写config.json定义Batch任务，包含3个打印字符串的子任务，同时配置通知规则。使用的自定义服务账号已分配以下角色：Batch agent reporter、Batch job editor、Compute Storage Admin、Google Batch service agent、Logs Writer、Pub/Sub Publisher。
   config.json内容：

{"taskGroups": [{"taskSpec": {"runnables": [{"script": {"text": "echo Hello ${VARIABLE_NAME}! This is task ${BATCH_TASK_INDEX}. This job has a total of ${BATCH_TASK_COUNT} tasks."},"environment": {"variables": {"VARIABLE_NAME": "WORLD"}}},"computeResource": {"cpuMilli": 2000,"memoryMib": 16},"maxRetryCount": 0,"maxRunDuration": "7200s"},"taskCount": 3,"parallelism": 1}],"allocationPolicy": {"serviceAccount": {"email": "my-custom-service-account@my-gcp-project.iam.gserviceaccount.com"},"instances": [{"policy": {"machineType": "e2-standard-4"}}]},"logsPolicy": {"destination": "CLOUD_LOGGING"},"notifications": [{"pubsubTopic": "projects/my-gcp-project/topics/my-topic","message": {"type": "JOB_STATE_CHANGED","newJobState": "SUCCEEDED","newTaskState": "STATE_UNSPECIFIED"}}]}

通知规则配置：任务状态变为SUCCEEDED时，向projects/my-gcp-project/topics/my-topic发送JOB_STATE_CHANGED类型通知。

2. 在Cloud Pub/Sub控制台创建主题my-topic，关联AVRO schemanoti-avro-schema，消息编码类型设为JSON。AVRO schema内容：

{"type": "record","name": "message","fields": [{"name": "type","type": {"type": "enum","name": "Type","symbols": ["TYPE_UNSPECIFIED","JOB_STATE_CHANGED","TASK_STATE_CHANGED"]}},{"name": "newJobState","type": {"type": "enum","name": "State","symbols": ["STATE_UNSPECIFIED","QUEUED","SCHEDULED","RUNNING","SUCCEEDED","FAILED","DELETION_IN_PROGRESS"]}},{"name": "newTaskState","type": "State"}]}

3. 创建主题订阅projects/my-gcp-project/subscriptions/my-topic-sub

4. 提交Batch任务：

gcloud batch jobs submit job-test --location us-central1 --config config.json

5. 任务成功完成后，在Pub/Sub控制台拉取订阅消息检查通知是否送达

实际结果 #

订阅中未收到任何来自Batch任务的通知，且无错误提示

排查建议 #

• 验证服务账号主题级权限：确认自定义服务账号的Pub/Sub Publisher角色是否绑定到目标主题（而非仅项目级绑定），可执行以下命令校验：

  gcloud pubsub topics get-iam-policy projects/my-gcp-project/topics/my-topic --filter="bindings.members:serviceAccount:my-custom-service-account@my-gcp-project.iam.gserviceaccount.com"
  

• 检查通知配置合法性：确认notifications字段结构正确，JOB级通知无需指定newTaskState，可尝试移除该字段后重新提交任务测试。
• 排查Schema兼容性问题：Batch发送的通知消息会包含jobName、timestamp等额外字段，当前AVRO schema未定义这些字段，可能导致消息被主题静默拒收。建议先移除主题的Schema校验，测试通知是否能正常送达，再调整Schema匹配Batch的标准通知格式。
• 查看Batch任务日志：在Cloud Logging中过滤Batch任务的日志，搜索notification或pubsub关键词，排查是否有隐藏的权限或格式错误信息。
• 确认任务状态触发条件：检查任务是否确实进入SUCCEEDED状态，可执行以下命令查看：

  gcloud batch jobs describe job-test --location us-central1 --format="value(status.state)"
  

内容的提问来源于stack exchange，提问作者agus.dinamarca