[Serve] Avoid self-cause on non-gRPC replica exceptions

[Ziy1-Tan]

Why

Fixes #62358.

For non-gRPC Serve requests, Replica.handle_request* currently does:

raise self._maybe_wrap_grpc_exception(e, request_metadata) from e

When _maybe_wrap_grpc_exception() returns the original exception unchanged, this degenerates into raise e from e, creating a self-referential __cause__ chain (e.__cause__ is e).

That breaks error-reporting tools that walk __cause__ recursively and can also confuse debugging and logging paths for normal HTTP and handle-based requests.

Root cause

This regression came from #60482 (221a19395a), which unified exception handling across the handle_request* paths so gRPC requests could preserve user-set status codes.

That behavior is correct when a new gRPC wrapper exception is created. It is wrong for non-gRPC requests, because _maybe_wrap_grpc_exception() just returns e, so the code effectively executes raise e from e.

Fix

Introduce a small _raise_user_exception() helper in replica.py:

• if _maybe_wrap_grpc_exception() returns the original exception, raise it directly
• if it returns a wrapped gRPC exception, raise the wrapped exception from e

This preserves the intended gRPC chaining behavior while eliminating the circular __cause__ chain on non-gRPC paths.

Testing

source /home/simple/github/ray/.venv/bin/activate
PYTHONPATH=/home/simple/github/ray/.worktrees/fix-issue-62358-replica-exception-chaining/python \
python -m pytest python/ray/serve/tests/unit/test_replica_exception_chaining.py -v

[gemini-code-assist]

Code Review

This pull request refactors exception handling within the Replica class to prevent self-referential __cause__ chains, particularly for non-gRPC requests. It introduces a centralized _raise_user_exception helper method and adds unit tests to verify the fix. The feedback points out that the unit tests currently redefine the production logic within a fake class, which risks testing the test code itself rather than the actual implementation; it is recommended to reference the production method directly in the test setup.

[gemini-code-assist]

The FakeReplica class redefines _raise_user_exception with the exact same logic as the production code in replica.py. This means the test is verifying the logic defined within the test file itself rather than the actual implementation in the codebase. To ensure the production code is correctly verified, you should remove this redundant definition and instead assign the production method to the fake class.

    _raise_user_exception = Replica._raise_user_exception

[abrarsheikh]

i suggest dropping this unit test. instead use the repro from here #62358 and add a integration test in test_handle.py

[Ziy1-Tan]

Ok. Let me take a look.

[abrarsheikh]

CI is failing.

[abrarsheikh]

how can we make the test deterministic, relying on sleep can cause the test to flake in CI.

[abrarsheikh]

this test is a little hard to follow, is there a more simpler version we can write?

[Ziy1-Tan]

I can reuse the issue repro as the basis, but I don’t think the script should be copied directly into CI because it relies on sleep and health-check timing.

I’ll either reduce it to a focused unit test around _raise_user_exception, or keep a thin integration test with an explicit synchronization signal instead of timing-based waiting.

[abrarsheikh]

explicit synchronization signal sounds good to me.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit 09866b6. Configure here.}

[cursor]

Return type annotation hides unconditional raise from callers

Low Severity

_raise_user_exception always raises and never returns, but is annotated -> None instead of -> NoReturn. This means type checkers and readers at the three call sites (e.g., handle_request which expects Tuple[bytes, Any]) cannot tell the function is unconditionally raising. If someone later adds a conditional return path to this helper, callers would silently return None instead of propagating the exception, introducing a subtle bug.

 

^{Reviewed by Cursor Bugbot for commit 09866b6. Configure here.}