Avoid shell in async command runner

[NguyenCong2k]

Summary

• replace asyncio.create_subprocess_shell() with asyncio.create_subprocess_exec() in async_run_command()
• keep command arguments as separate argv entries instead of joining them into a shell string
• avoid sending SIGINT after the subprocess has already exited
• add regression coverage for shell metacharacters in command arguments

Test

$env:PYTHONPATH='src'
python -m pytest tests\test_common.py -q

[Copilot]

Pull request overview

This PR hardens openllm.common.async_run_command() by avoiding shell invocation when running async subprocesses, reducing the risk of shell metacharacters being interpreted and improving cleanup behavior.

Changes:

• Switch from asyncio.create_subprocess_shell() to asyncio.create_subprocess_exec() and pass argv entries directly.
• Avoid sending SIGINT to the subprocess if it has already exited (via returncode check).
• Add a regression test to ensure shell metacharacters in arguments don’t trigger command chaining.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/openllm/common.py	Uses create_subprocess_exec(*cmd) and adds a returncode guard before sending SIGINT during teardown.
tests/test_common.py	Adds regression coverage to ensure async_run_command() doesn’t interpret shell metacharacters as command separators.

Comments suppressed due to low confidence (1)

src/openllm/common.py:440

• Even with the returncode check, there is still a race where the process can exit but returncode hasn't been updated yet; send_signal() can raise ProcessLookupError in that window. To avoid intermittent failures during context manager teardown, catch/ignore ProcessLookupError (or similar) around proc.send_signal(signal.SIGINT).

def md5(*strings: str) -> int:
  m = hashlib.md5()

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.