Skip to content

aiohttp timeout and wrapped google-auth transport exceptions are not retried despite HttpRetryOptions being configured #3329

Description

@rladeira

Describe the bug

When HttpRetryOptions is configured and aiohttp is used as the async HTTP transport (google-genai[aiohttp]), request timeouts and wrapped google.auth.aio transport exceptions bypass retry_args() and fail on the first attempt without retrying.

Specifically:

  1. asyncio.TimeoutError (and aiohttp.ServerTimeoutError / aiohttp.ConnectionTimeoutError):
    While _HTTPX_TRANSIENT_EXC includes httpx.TimeoutException and httpx2.TimeoutException (added in fix(retry): retry on httpx.TimeoutException with HttpRetryOptions #2345 for httpx.TimeoutException is not retried despite HttpRetryOptions being configured #2337), _AIOHTTP_TRANSIENT_EXC only lists (aiohttp.ClientConnectorError, aiohttp.ClientOSError, aiohttp.ServerDisconnectedError). In aiohttp, ClientTimeout(total=...) (TimerContext.__exit__) raises asyncio.TimeoutError directly, and aiohttp.ServerTimeoutError / ConnectionTimeoutError inherit from (aiohttp.ClientConnectionError, asyncio.TimeoutError), not ClientConnectorError, ClientOSError, or ServerDisconnectedError.
  2. google.auth.exceptions.TimeoutError (auth_exceptions.TimeoutError):
    When Vertex AI mTLS is used (_use_google_auth_async() == True), _get_aiohttp_session() returns google.auth.aio.transport.sessions.AsyncAuthorizedSession. Both google.auth.aio.transport.aiohttp.Request.__call__ and google.auth.aio.transport.sessions.timeout_guard catch asyncio.TimeoutError and raise google.auth.exceptions.TimeoutError (which inherits from GoogleAuthError, not builtins.TimeoutError or aiohttp.ClientError).
  3. google.auth.exceptions.TransportError wrapping _AIOHTTP_TRANSIENT_EXC:
    Under AsyncAuthorizedSession, google.auth.aio.transport.aiohttp.Request.__call__ catches aiohttp.ClientError and raises google.auth.exceptions.TransportError(...) from caught_exc. In _async_request_once(), the except (..., auth_exceptions.TransportError) block re-raises when retry_options is configured (raise # HttpRetryOptions owns retries; tenacity decides.), but retry_args() does not match auth_exceptions.TransportError wrapping _AIOHTTP_TRANSIENT_EXC.

Steps to reproduce

import asyncio
from google.auth import exceptions as auth_exceptions
from google.genai import _api_client as api_client
from google.genai import types
import aiohttp

retry = api_client.retry_args(types.HttpRetryOptions())['retry']
print("asyncio.TimeoutError retried:", retry.predicate(asyncio.TimeoutError()))
print("aiohttp.ServerTimeoutError retried:", retry.predicate(aiohttp.ServerTimeoutError()))
print("google.auth.exceptions.TimeoutError retried:", retry.predicate(auth_exceptions.TimeoutError("Request timed out")))

Output on main:

asyncio.TimeoutError retried: False
aiohttp.ServerTimeoutError retried: False
google.auth.exceptions.TimeoutError retried: False

Expected behavior

When HttpRetryOptions is configured, retry_args() should retry asyncio.TimeoutError (including aiohttp.ServerTimeoutError and aiohttp.ConnectionTimeoutError), google.auth.exceptions.TimeoutError, and google.auth.exceptions.TransportError whose __cause__ is in _AIOHTTP_TRANSIENT_EXC, matching the httpx.TimeoutException behavior from #2337 / #2345 while keeping non-transient errors (such as aiohttp.ClientPayloadError and google.auth.exceptions.RefreshError) and the no-retry_options path unchanged.

Activity

  1. added a commit that references this issue on Nov 10, 2026
    6e7ea76
  2. added
    type: bugError or flaw in code with unintended results or allowing sub-optimal usage patterns.
    priority: p2Moderately-important priority. Fix may not be included in next release.
    on Nov 11, 2026
  3. Venkaiahbabuneelam commented on Nov 11, 2026

    @Venkaiahbabuneelam

    Hi @rladeira, Thanks for reaching out us!
    I have successfully reproduced the issue when HttpRetryOptions is enabled in google-genai, async requests using aiohttp and google-auth token refreshes do not retry when a timeout or temporary aiohttp disconnect happens because _api_client.py is missing those timeout/error types in its retry list, causing the request to fail immediately on the first try, we will file a bug internally.
    Thanks

  4. added a commit that references this issue on Nov 13, 2026
    f40084c
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

priority: p2Moderately-important priority. Fix may not be included in next release.type: bugError or flaw in code with unintended results or allowing sub-optimal usage patterns.

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions