Fix: Prevent retry of termination signals (Celery SoftTimeLimitExceeded)

[yashwantbezawada]

• I understand that this repository is auto-generated and my pull request may not be merged

Changes being requested

This PR fixes issue #2737 where the OpenAI client's broad exception handling was catching Celery's SoftTimeLimitExceeded exception and retrying the request instead of allowing it to propagate for graceful task shutdown.

Changes:

• Added _should_not_retry() helper function to identify termination signals that should propagate immediately
• Modified both sync and async exception handlers to check for termination signals before attempting retry
• Added test coverage to verify termination signals are not retried

Affected exception types:

• Celery task limits: SoftTimeLimitExceeded, TimeLimitExceeded, Terminated
• Asyncio cancellation: CancelledError

The fix preserves all existing retry logic while ensuring task executor termination signals are properly handled.

Additional context & links

Fixes #2737

Background:
When using the OpenAI client in Celery tasks with soft time limits, if the task exceeds its limit during an API call, Celery raises SoftTimeLimitExceeded. The client's except Exception clause was catching this and treating it as a retryable connection error, preventing cleanup logic from running.

Solution approach:
Instead of narrowing the exception catch (which could miss legitimate connection errors), we check if caught exceptions are termination signals and re-raise them immediately without retry. This approach:

• Maintains backward compatibility
• Doesn't require importing Celery (checks by module name)
• Handles both sync and async code paths
• Is extensible for other task executors

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Ensure asyncio cancellations bypass retry loop

_should_not_retry checks for CancelledError by requiring exc.__class__.__module__ == "asyncio", but asyncio.CancelledError is defined in asyncio.exceptions across supported Python versions. The predicate never matches, so cancelled tasks are still retried and eventually surface as APIConnectionError instead of propagating the cancellation. This defeats the intended behavior for async termination signals.

Useful? React with 👍 / 👎.

[yashwantbezawada]

Thanks for the review! You're absolutely right - asyncio.CancelledError is in asyncio.exceptions, not asyncio directly.

I've updated the check to use startswith("asyncio") (consistent with the Celery check pattern) and added a test case specifically for asyncio.CancelledError to ensure it properly bypasses the retry loop.

Changes in bd96a86:

• Fixed module check from == "asyncio" to startswith("asyncio")
• Added test_asyncio_cancelled_error_not_retried test case

[yashwantbezawada]

Rebased on latest main to bring branch up-to-date (was 10 commits behind).