@AgentWrapperchecks…lint-check-as-completed → next1 files · +20 −0updated 4mo agoPrevent reintroduction of the bug that caused workflow ada5svth to hang for 7+ hours and cost $1.7K.
as_completed() blocks indefinitely if tasks hang on LLM calls - this was the root cause of #925.
Add a CI lint check that flags as_completed( usage in production code (cortex/).
Excludes:
/tests/ - test files_test.py, test_ - test file patterns/eval.py, long_eval - benchmarking/evaluation codeOn failure, suggests:
Use wait(futures, timeout=60, return_when=FIRST_COMPLETED) with manual timeout checking instead.
See cortex/workflows/response_schema_modifier.py for the correct pattern.
# Should pass (all production code now uses wait())
grep -r "as_completed(" cortex/ --include="*.py" \
| grep -v "/tests/" | grep -v "_test.py" | grep -v "test_" \
| grep -v "/eval.py" | grep -v "long_eval"
This is a required CI check - any new as_completed() usage in production code will fail the lint job.
🤖 Generated with Claude Code
Fixed in commit 5616066ca. The grep patterns now use path separators (/) to anchor them to filenames rather than matching the entire grep output line:
/test_ - matches files starting with test_ (the leading slash ensures it's a path component)/long_eval - matches any path containing /long_eval (covers both directory and files)Added a comment explaining this approach in the workflow file.
Fixed in commit dae0d2b1a. Removed the leading / from the pattern so it now correctly matches any file ending in _test.py (e.g., foo_test.py). The trailing colon still ensures it matches the filename portion of grep output, not code content.
Fixed in commit 0a26da0f8. Changed /test_ to /test_[^/]*\.py: which matches test files (like test_foo.py) but not directories containing test_ in their name (like test_and_fix_agent/).