Creating API requests and handling responses

This document describes how to construct API requests and handle API responses from the Compute Engine API. It covers how to:

  • Construct a request body.
  • Determine the resource URIs necessary for a request.
  • Handle API responses.
  • Determine whether an API request succeeded.

This document does not cover how to authorize to the API. To learn how to authorize to the API, read Authorizing requests.

Before you begin

Constructing an API request

The Compute Engine API expects API requests to be in JSON format. To make an API request, you can either make a direct HTTP request, using tools like curl or httplib2, or you can use one of the available client libraries.

When you make an API request that requires a request body, like a POST, UPDATE, or PATCH request, the request body contains resource properties that you want to set in this request. For example, the following curl command makes a POST request to the Instances resource URI. The request creates an instance with the properties defined in the request body:

curl -X POST -H "Authorization: Bearer [OAUTH_TOKEN]" -H "Content-Type: application/json"[PROJECT_ID]/zones/[ZONE]/instances -d

In this case, the request body is indicated by the -d flag, and looks like this (formatted for easier viewing):


Notice that:

  • When referencing another resource, use the fully qualified resource URI. For example, the network property uses a fully-qualified URI to the default network.

  • The image URI has a different project ID (debian-cloud) from your project ID. This is because images belong to different projects, depending on the type of image. For example, all publicly available Debian images offered by Compute Engine are hosted on the debian-cloud project.

Here are other examples of API requests using the Python and Java client libraries.


def create_instance(compute, project, zone, name, bucket):
    # Get the latest Debian Jessie image.
    image_response = compute.images().getFromFamily(
        project='debian-cloud', family='debian-9').execute()
    source_disk_image = image_response['selfLink']

    # Configure the machine
    machine_type = "zones/%s/machineTypes/n1-standard-1" % zone
    startup_script = open(
            os.path.dirname(__file__), ''), 'r').read()
    image_url = ""
    image_caption = "Ready for dessert?"

    config = {
        'name': name,
        'machineType': machine_type,

        # Specify the boot disk and the image to use as a source.
        'disks': [
                'boot': True,
                'autoDelete': True,
                'initializeParams': {
                    'sourceImage': source_disk_image,

        # Specify a network interface with NAT to access the public
        # internet.
        'networkInterfaces': [{
            'network': 'global/networks/default',
            'accessConfigs': [
                {'type': 'ONE_TO_ONE_NAT', 'name': 'External NAT'}

        # Allow the instance to access cloud storage and logging.
        'serviceAccounts': [{
            'email': 'default',
            'scopes': [

        # Metadata is readable from the instance and allows you to
        # pass configuration from deployment scripts to instances.
        'metadata': {
            'items': [{
                # Startup script is automatically executed by the
                # instance upon startup.
                'key': 'startup-script',
                'value': startup_script
            }, {
                'key': 'url',
                'value': image_url
            }, {
                'key': 'text',
                'value': image_caption
            }, {
                'key': 'bucket',
                'value': bucket

    return compute.instances().insert(


public static Operation startInstance(Compute compute, String instanceName) throws IOException {
  System.out.println("================== Starting New Instance ==================");

  // Create VM Instance object with the required properties.
  Instance instance = new Instance();
          PROJECT_ID, ZONE_NAME));
  // Add Network Interface to be used by VM Instance.
  NetworkInterface ifc = new NetworkInterface();
  List<AccessConfig> configs = new ArrayList<>();
  AccessConfig config = new AccessConfig();

  // Add attached Persistent Disk to be used by VM Instance.
  AttachedDisk disk = new AttachedDisk();
  AttachedDiskInitializeParams params = new AttachedDiskInitializeParams();
  // Assign the Persistent Disk the same name as the VM Instance.
  // Specify the source operating system machine image to be used by the VM Instance.
  // Specify the disk type as Standard Persistent Disk
          PROJECT_ID, ZONE_NAME));

  // Initialize the service account to be used by the VM Instance and set the API access scopes.
  ServiceAccount account = new ServiceAccount();
  List<String> scopes = new ArrayList<>();

  // Optional - Add a startup script to be used by the VM Instance.
  Metadata meta = new Metadata();
  Metadata.Items item = new Metadata.Items();
  // If you put a script called "" in this Google Cloud Storage
  // bucket, it will execute on VM startup.  This assumes you've created a
  // bucket named the same as your PROJECT_ID.
  // For info on creating buckets see:
  item.setValue(String.format("gs://%s/", PROJECT_ID));

  Compute.Instances.Insert insert = compute.instances().insert(PROJECT_ID, ZONE_NAME, instance);
  return insert.execute();

Creating resource URIs

In the earlier examples, a reference to another Google Cloud resource is given as a fully qualified URI that looks like this:[PROJECT_ID]/zones/[ZONE]/[RESOURCE_TYPE]/[SPECIFIC_RESOURCE]

Whenever you specify an image, a machine type, a network, or any other resource, you must provide the URI to the resource when using the API. Client tools like the gcloud command-line tool and the Google Cloud Console hide this complexity and handle creating these resource URIs for you, but when interacting with the API directly, you must create these resource URIs yourself.

There are slightly different resource URIs for different types of resources. For example, a zonal resource has the zone specification in the URI, like so:[PROJECT_ID]/zones/[ZONE]/machineTypes/e2-standard-2

Regional resources replace the zone specification with a region specification:[PROJECT_ID]/regions/[REGION]/addresses/[ADDRESS_NAME]

Similarly, global resources have the global specification:[PROJECT_ID]/global/images/[IMAGE_NAME]

The Compute Engine API also accepts partial URIs because the service can infer information like the project ID. So, the following partial versions of the earlier URIs are also acceptable:


In the partial URIs, both the zonal and regional URIs omitted the project ID but the image URI did not. This is because publicly available images offered by Compute Engine are hosted in other projects, like debian-cloud for all Debian images and ubuntu-os-cloud for all Ubuntu images. Before you can use these images, you need to provide the appropriate project ID. If you omitted the project ID for images, Compute Engine attempts to find the image in your project and the request fails because the image does not exist.

However, if you use a custom image that belongs to your project (the same project you are creating this resource in), you can omit the project specification when providing an image URI.

Determining required properties

The Compute Engine API reference documentation, available for both v1 and beta APIs, describe all of the possible properties you can set for a specific resource. The reference documentation makes a distinction between mutable versus immutable properties (marked by an [Output Only] in the property description), but to determine required properties for a resource, you need to review the documentation specific to that task.

For example, if you are creating an instance, read the Creating an instance from an image documentation to see the API properties required for the request. If you want to create a static external IP address in the API, read the Static external IP addresses documentation.

Alternatively, you can validate your API requests in the API explorer, as a quick and easy way to check your code.

Validating API requests

To validate your API requests:

  1. In the Compute Engine API reference, find the method your code is calling. For example, v1/compute.instances.insert.
  2. From the contents menu, click Try it! This opens the Try this API window.

    The Try it! button on the contents menu.

  3. In the Try this API window, paste your request into the Request body text area. You do not need to provide a project or zone since validation doesn't require submitting the request.

    The Try It window, displaying the request box to show where to paste in a request for validation.

Malformed elements of the request are underlined in blue. Click each underlined section for an explanation of the issue to be addressed.

Handling API responses

If you make a request that mutates data, Compute Engine returns an Operations object that you can then poll to get the status of your request. The Operation resource looks like this:

 "kind": "compute#operation",
 "id": "7127550864823803662",
 "name": "operation-1458856416788-52ed27a803e22-1c3bd86a-9e95017b",
 "zone": "[PROJECT_ID]/zones/[ZONE]",
 "operationType": "insert",
 "targetLink": "[PROJECT_ID]/zones/[ZONE]/instances/[EXAMPLE_INSTANCE]",
 "targetId": "4132355614508179214",
 "status": "PENDING",
 "user": "",
 "progress": 0,
 "insertTime": "2016-03-24T14:53:37.788-07:00",
 "selfLink": "[PROJECT_ID]/zones/[ZONE]/operations/operation-1458856416788-52ed27a803e22-1c3bd86a-9e95017b"

If the original request was to mutate a zonal resource, like an instance or a disk, Compute Engine returns a zoneOperations object. Similarly, regional and global resources return a regionOperations and globalOperations object, respectively. You can get the status of an operation by performing a GET or WAIT request to the specific Operation resource, providing the name of the operation.

Your request is not complete until the Operation resource's status returns as DONE. This can take some time depending on the nature of your request. Then, after the operation's status returns as DONE, you can check to see if the operation succeeded and whether there were any errors.

For example, the following response indicates that the same operation that appears above is now complete, specified by the DONE status:

endTime: '2016-03-24T14:54:07.119-07:00'
id: '7127550864823803662'
insertTime: '2016-03-24T14:53:37.788-07:00'
kind: compute#operation
name: operation-1458856416788-52ed27a803e22-1c3bd86a-9e95017b
operationType: insert
progress: 100
startTime: '2016-03-24T14:53:38.397-07:00'
status: DONE
targetId: '4132355614508179214'

To confirm, make a GET request to the resource to check that it exists and is running. For example:

GET /compute/v1/projects/[PROJECT_ID]/zones/[ZONE]/instances/[EXAMPLE_INSTANCE]

  "cpuPlatform": "Intel Haswell",
  "creationTimestamp": "2016-03-24T14:53:37.170-07:00",
  "disks": [
  "selfLink": "[PROJECT_ID]/zones/[ZONE]/instances/[EXAMPLE_INSTANCE]",
  "status": "RUNNING",
  "tags": {
    "fingerprint": "42WmSpB8rSM="
  "zone": "[PROJECT_ID]/zones/[ZONE]"

Polling operations

You can write some code to periodically poll the operation with a GET or WAIT request that returns when the operation status is DONE. With a GET request, the operation is returned immediately, regardless of the status of the operation. You will then need to poll the operation regularly to know when the operation is done. If you make a WAIT request, the request returns when the operation is DONE or if the request is approaching the 2 minute deadline. You can choose to use WAIT or GET to poll your operations, but the WAIT method provides certain benefits over the GET method:

  • You can set up your clients to poll for the operation status less frequently, reducing your QPS usage of the Compute Engine API.
  • The average latency between when the operation is done and when the client is informed that the operation is done is significantly reduced because the server responds as soon as the operation is done. With the GET request, it is up to the client to check whether the operation is done.
  • The method provides bounded waits. The method waits for no more than the default HTTP timeout (2 minutes) and then returns the current state of the operation, which might be DONE or still in progress.

The WAIT method is intended to be a best-effort API. If the server is overloaded, the request might return before it reaches the default deadline, or after waiting just zero seconds. The method is also not guaranteed to return only when the operation is DONE. For example, if the request approaches the 2 minute deadline, the method returns even if the operation is not done. We still recommend that you regularly poll the operation status with the WAIT or GET methods with sleep in-between, to check on your operations.

Here are some examples of polling in Python and Java using the GET method. You can also replace the GET method with the WAIT method:


def wait_for_operation(compute, project, zone, operation):
    print('Waiting for operation to finish...')
    while True:
        result = compute.zoneOperations().get(

        if result['status'] == 'DONE':
            if 'error' in result:
                raise Exception(result['error'])
            return result



 * Wait until {@code operation} is completed.
 * @param compute the {@code Compute} object
 * @param operation the operation returned by the original request
 * @param timeout the timeout, in millis
 * @return the error, if any, else {@code null} if there was no error
 * @throws InterruptedException if we timed out waiting for the operation to complete
 * @throws IOException if we had trouble connecting
public static Operation.Error blockUntilComplete(
    Compute compute, Operation operation, long timeout) throws Exception {
  long start = System.currentTimeMillis();
  final long pollInterval = 5 * 1000;
  String zone = operation.getZone(); // null for global/regional operations
  if (zone != null) {
    String[] bits = zone.split("/");
    zone = bits[bits.length - 1];
  String status = operation.getStatus();
  String opId = operation.getName();
  while (operation != null && !status.equals("DONE")) {
    long elapsed = System.currentTimeMillis() - start;
    if (elapsed >= timeout) {
      throw new InterruptedException("Timed out waiting for operation to complete");
    if (zone != null) {
      Compute.ZoneOperations.Get get = compute.zoneOperations().get(PROJECT_ID, zone, opId);
      operation = get.execute();
    } else {
      Compute.GlobalOperations.Get get = compute.globalOperations().get(PROJECT_ID, opId);
      operation = get.execute();
    if (operation != null) {
      status = operation.getStatus();
  return operation == null ? null : operation.getError();