how to get error if starting compute engine instance fails

Question

I am starting an instance using PHP using this code:

function startInstance($g_project,$g_instance, $g_zone){

    $client = new Google_Client();
    $client->setApplicationName('Google-ComputeSample/0.1');
    $client->useApplicationDefaultCredentials();
    $client->addScope('https://www.googleapis.com/auth/cloud-platform');

    $service = new Google_Service_Compute($client);
    $response = $service->instances->start($g_project, $g_zone, $g_instance);
    echo json_encode($response);

}

Today I was lucky enough to realize that for unknown reason the instance I wanted to start failed to do so. I tried starting it using GUI and got an error via GUI: Zone "some-zone" does not have enough resources available to fulfill the request. Try a different zone, or try again later.

I echoed out the PHP response and compared it to the one I get when an instance start successfully. My findings are shocking. The responses were exactly the same (not counting timestamps and ids). How on earth can I differentiate between failed instance starts and successful, if the response is the same?

https://cloud.google.com/compute/docs/reference/rest/v1/instances/start suggests that there will be an error object present in case of error. I can confirm that there is none.

Response of both failed an successful start:

{
    "clientOperationId": null,
    "creationTimestamp": null,
    "description": null,
    "endTime": null,
    "httpErrorMessage": null,
    "httpErrorStatusCode": null,
    "id": "id",
    "insertTime": "2019-01-28T14:22:36.664-08:00",
    "kind": "compute#operation",
    "name": "operation-name",
    "operationType": "start",
    "progress": 0,
    "region": null,
    "selfLink": "link/operation-name",
    "startTime": null,
    "status": "PENDING",
    "statusMessage": null,
    "targetId": "targetIdHere",
    "targetLink": "linkhere",
    "user": "user",
    "zone": "zone-in-question"
}

What do you suggest that I do? Switching to different zone is probably the best solution. But there is one problem, I don't even that the instance didn't start successfully so I can't react to it. Is this the expected behavior? What did you do mitigate this problem?

Answer 1

The response you get from the start call is just a (zone) operation resource corresponding to your asynchronous API call.

In order to determine the final status of your API call you'll need to poll that operation with a get call until its status is DONE. Then, if error field is not empty it will contain the details of went wrong, for example:

{
  ...
  "error": {
    "errors": [
      {
        "code": "ZONE_RESOURCE_POOL_EXHAUSTED",
        "message": "The zone 'projects/you-project/zones/us-central1-c' does not have enough resources available to fulfill the request."
      }
    ]
  }
  ...
}

otherwise the operation has been completed successfully.