Troubleshooting response errors

This page describes how to troubleshoot errors that you receive in a response from a request to your API.

BAD_GATEWAY

If you receive error code 13 and the message BAD_GATEWAY, this indicates that the Extensible Service Proxy (ESP) can't reach the service's backend. Check the following:

  • Make sure the backend service is running. How you do that depends on the backend.

  • The correct IP address port of the backend service is specified:
    • For GKE, check the ESP --backend flag value (the short option is -a) in your deployment manifest file (often called deployment.yaml).
    • For Compute Engine check the ESP --backend flag value (the short option is -a) in the docker run command.

reset reason: connection failure

If you receive HTTP code 503 or gRPC code 14 and the message upstream connect error or disconnect/reset before headers. reset reason: connection failure, this indicates that ESPv2 can't reach the service's backend.

To troubleshoot, double check the items below.

Backend Address

ESPv2 should be configured with the correct backend address. Common issues include:

  • The scheme of the backend address should match the backend application type. OpenAPI backends should be http:// and gRPC backends should be grpc://.
  • For ESPv2 deployed on Cloud Run, the scheme of the backend address should be either https:// or grpcs://. The s tells ESPv2 to set up TLS with the backend.

DNS Lookup

By default, ESPv2 attempts to resolve domain names to IPv6 addresses. If the IPv6 resolution fails, ESPv2 falls back to IPv4 addresses.

For some networks, the fallback mechanism may not work as intended. Instead, you can force ESPv2 to use IPv4 addresses via the --backend_dns_lookup_family flag.

This error is common if you configure a Serverless VPC Connector for ESPv2 deployed on Cloud Run. VPCs do not support IPv6 traffic.

API is not enabled for the project

If you sent an API key in the request, an error message like "API my-api.endpoints.example-project-12345.cloud.goog is not enabled for the project" indicates that the API key was created in a different Google Cloud project than the API. To fix this, you can either create the API key in the same Google Cloud project that the API is associated with, or you can enable the API in the Google Cloud project that the API key was created in.

Service control request failed with HTTP response code 403

If you receive error code 14 and the message Service control request failed with HTTP response code 403, this indicates that the Service Control API (servicecontrol.googleapis.com) isn't enabled on the project.

  1. See Checking required services to make sure that all the services that Endpoints and ESP require are enabled on your project.

  2. See Checking required permissions to make sure that all the required permissions to the service account associated with the instance running ESP.

Method doesn't allow unregistered callers

ESP responds with the error, Method doesn't allow unregistered callers, when you have specified an API key in the security section in your OpenAPI document, but the request to your API doesn't have an API key assigned to a query parameter named key.

If you need to generate an API key to make calls to your API, see Creating an API key.

Method does not exist

The response, Method does not exist, means that the HTTP method (GET, POST, or other) on the specified URL path wasn't found. To troubleshoot, compare the service configuration that you have deployed to make sure that the method name and URL path that you are sending in the request match:

  1. In the Google Cloud console, go to the Endpoints Services page for your project.

    Go to the Endpoints Services page

  2. If you have more than one API, select the API that you sent the request to.

  3. Click the Deployment history tab.

  4. Select the latest deployment to see the service configuration.

If you don't see the method you are calling specified in the paths section of your OpenAPI document, either add the method, or add the x-google-allow flag at the top level of the file:

x-google-allow: all

This flag means that you can avoid listing all methods supported in your backend in your OpenAPI document. When all is used, all calls—with or without an API key or user authentication—pass through ESP to your API. See x-google-allow for more information.

Errors specific to the App Engine flexible environment

This section describes error responses from APIs deployed on the App Engine flexible environment.

Error code 502 or 503

App Engine may take a few minutes to respond successfully to requests. If you send a request and get back an HTTP 502, 503, or some other server error, wait a minute and try the request again.

Error message BAD_GATEWAY

An error code 502 with BAD_GATEWAY in the message usually indicates that App Engine terminated the application because it ran out of memory. The default App Engine flexible VM only has 1GB of memory, with only 600MB available for the application container.

To troubleshoot error code 502:

  1. In the Google Cloud console, go to the Logging page:

    Go to the Logs Explorer page

  2. Select the applicable Google Cloud project at the top of the page.

  3. Select Google App Engine Application and open vm.syslog.

  4. Look for a log entry similar to the following:

    kernel: [  133.706951] Out of memory: Kill process 4490 (java) score 878 or sacrifice child
    kernel: [  133.714468] Killed process 4306 (java) total-vm:5332376kB, anon-rss:2712108kB, file-rss:0kB
    

    If you see an Out of memory entry in the log:

    1. Add the following to your app.yaml file to increase the size of the default VM:

      resources:
        memory_gb: 4
      
    2. Redeploy your API:

      gcloud app deploy
      

If you have the rollout_strategy: managed option specified in the endpoints_api_service section of the app.yaml file, use the following command to redeploy your API:

  gcloud app deploy

See Deploying your API and ESP for more information.

Checking the Cloud Logging logs

To use the Cloud Logging logs to help troubleshoot response errors:

  1. In the Google Cloud console, go to the Logging page.

    Go to the Logs Explorer page

  2. At the top of the page, select the Google Cloud project.

  3. Using the drop-down menu on the left, select Produced API > [YOUR_SERVICE_NAME].

  4. Adjust the time range until you see a row that shows your response error.

  5. Expand the JSON payload and look for error_cause.

    • If the error_cause is set to application, this indicates an issue in your code.

    • If the error cause is anything else and you are unable to fix the issue, export the log and include it in any communication that you have with Google.

See the following for more information:

Issues with the example Invoke-WebRequest

In some versions of Windows PowerShell, the example Invoke-WebRequest in the tutorials fails. We have also received a report that the response contained a list of unsigned bytes that had to be converted to characters. If the example Invoke-WebRequest didn't return the expected result, try sending the request using another application. Following are a few suggestions:

  • Start Cloud Shell and follow the Linux steps in the tutorial that you were using to send the request.
  • Install a third-party application, such as the Chrome browser extension Postman (offered by www.getpostman.com). When creating the request in Postman:

    • Select POST as the HTTP verb.
    • For the header, select the key content-type and the value application/json.
    • For the body, enter: {"message":"hello world"}
    • In the URL, use the actual API key rather than the environment variable. For example:

      • On the App Engine flexible environment: https://example-project-12345.appspot.com/echo?key=AIza...
      • On other backends: http://192.0.2.0:80/echo?key=AIza...
  • Download and install curl, which you run in the command prompt. Because Windows doesn't handle double quotation marks nested inside single quotation marks, you have to change the --data option in the example to:

    --data "{\"message\":\"hello world\"}"