Built in exponential backoff for python api client?

Question

I noticed that in http.py a lot of the methods support num_retries which I believe is an implementation of exponential backoff (in case an API returns an error). Would like to use this rather than coding my own backoff algorithm each time I call an .execute() method. However, it doesn't seem to work with this code. Anyone have a handy idea on handling this backoff in a reusable way?

from apiclient.discovery import build

import google.auth
from google.auth.transport import requests
from google.auth import iam
from google.oauth2 import service_account

TOKEN_URI = 'https://accounts.google.com/o/oauth2/token'
ADMIN_DIRECTORY_SCOPES = [
    'https://www.googleapis.com/auth/admin.directory.user.readonly'
]
ADMIN_USER = 'my_gsuite_admin@domain.com'

def delegated_credentials(credentials, subject, scopes):
    try:
        admin_creds = credentials.with_subject(subject).with_scopes(scopes)
    except AttributeError:
        request = requests.Request()
        credentials.refresh(request)

        signer = iam.Signer(request, credentials, credentials.service_account_email)
        admin_creds = service_account.Credentials(
            signer, credentials.service_account_email, TOKEN_URI,
            scopes=scopes, subject=subject
        )
    except Exception:
        raise
    return admin_creds
print('Setting up auth')
default_credentials, _ = google.auth.default()
admin_creds = delegated_credentials(
    default_credentials, ADMIN_USER, ADMIN_DIRECTORY_SCOPES
)

directory_service = build('admin', 'directory_v1', credentials=admin_creds)

def get_ou(email):
    response = directory_service.users().get(
        userKey=email,
        fields='primaryEmail,orgUnitPath',
        num_retries=5
    ).execute()
    print('Directory.users.get: {}'.format(response))
    return response['orgUnitPath']

Traceback (most recent call last): File "", line 1, in File "/Users/mryerse001/Documents/GitHub/updateEmail_to_OU_Mappings/main.py", line 122, in get_pubsub_messages success = get_ou(email) File "/Users/mryerse001/Documents/GitHub/updateEmail_to_OU_Mappings/main.py", line 87, in get_ou num_retries=5 File "/Users/mryerse001/Documents/GitHub/updateEmail_to_OU_Mappings/venv/lib/python3.7/site-packages/googleapiclient/discovery.py", line 717, in method raise TypeError('Got an unexpected keyword argument "%s"' % name) TypeError: Got an unexpected keyword argument "num_retries"

The client library already implements exponential backoff why do you want to do it twice? — Linda Lawton - DaImTo, Sep 06 '19 at 11:08
Is there documentation describing how it was implemented? Is there any control over it? — Michael, Sep 07 '19 at 13:24
Probably not i would dig around in the source code see if you can find out where they implement it — Linda Lawton - DaImTo, Sep 09 '19 at 06:16

score 1 · Answer 1 · answered Feb 11 '21 at 11:40

1

You need to pass the num_retries parameter to execute(), not to get().

From google-api-python-client code:

@util.positional(1)
def execute(self, http=None, num_retries=0):
    """Execute the request.

Args:
  http: httplib2.Http, an http object to be used in place of the
        one the HttpRequest request object was constructed with.
  num_retries: Integer, number of times to retry with randomized
        exponential backoff. If all retries fail, the raised HttpError
        represents the last request. If zero (default), we attempt the
        request only once.

answered Feb 11 '21 at 11:40

sanjabhauz

11
2

You're right. Although I still don't like the retry functionality built into this client library because it will retry regardless of the error. For example let's say you're trying to edit a file that you don't have access to - one error is enough for that. I've started to use requests with auth'd sessions instead. – Michael Feb 11 '21 at 17:30
That shouldn't be the case, or at least it hasn't been for several years. I wonder if you're using a very old version of the library, or if Google is throwing unexpected errors? The library should retry 429s, >=500, and 403s with a "userRateLimitExceeded" or "rateLimitExceeded" reason. I'm curious specifically which errors are being retried that should not be? If you enable [logging](https://docs.python.org/3/howto/logging.html), you'll be able to see those retry attempts as warnings. – mary Mar 04 '21 at 16:09

Built in exponential backoff for python api client?

1 Answers1