Parallel pytest tracing

[misrasaurabh1]

PR Type

Enhancement

---

Description

• Introduce pytest_split for test-file splitting

• Detect pytest runs and split tests by CPU count

• Spawn parallel tracing subprocesses per split

• Aggregate multiple replay test paths and replay

---

Diagram Walkthrough

flowchart LR
  A["tracer main"] -- "detect pytest run" --> B["pytest_split"]
  B -- "multiple splits" --> C["spawn subprocesses"]
  C -- "wait & load pickles" --> D["collect replay paths"]
  D -- "invoke codeflash replay" --> E["optimizer.run"]
  A -- "non-pytest or single" --> F["single-process trace"]

Loading

File Walkthrough

Relevant files

Enhancement

tracer.py Add parallel pytest tracing logic                                                codeflash/tracer.py Import pytest_split helper function Refactor args_dict setup before subprocess calls Detect pytest invocation and split test paths Launch multiple or single subprocesses based on splits Collect and combine multiple replay_test_paths Update sys.argv to include all replay test files +91/-39  pytest_parallelization.py Implement pytest test splitting logic                                        codeflash/tracing/pytest_parallelization.py Add pytest_split to parse pytest args and paths Recursively gather test_*.py files Calculate splits based on CPU count and file count Distribute test files into balanced groups +81/-0

---

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Missing Flag Update The flag added_paths is never set to True after extending updated_sys_argv, causing repeated insertion of the same test splits for each matching element. added_paths = False updated_sys_argv = [] for elem in sys.argv: if elem in test_paths_set: if not added_paths: updated_sys_argv.extend(test_split) Nil Split Handling There is no check for pytest_split returning (None, None) before calling len(pytest_splits), which could raise a TypeError and crash the process. if parsed_args.module and unknown_args[0] == "pytest": pytest_splits, test_paths = pytest_split(unknown_args[1:]) print(pytest_splits) if len(pytest_splits) > 1: Invalid Parser Usage Using pytest.Parser() assumes a public API that doesn’t exist in pytest, leading to an AttributeError. Consider using pytest’s recommended CLI argument parsing approach. import pytest parser = pytest.Parser() pytest_args = parser.parse_known_args(arguments)

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Prevent duplicate test arguments After extending updated_sys_argv with a split, set added_paths = True so you don’t repeatedly inject the same tests on subsequent matches. This prevents duplicated test arguments in each subprocess call. codeflash/tracer.py [119-126] added_paths = False updated_sys_argv = [] for elem in sys.argv: - if elem in test_paths_set: - if not added_paths: - updated_sys_argv.extend(test_split) + if elem in test_paths_set and not added_paths: + updated_sys_argv.extend(test_split) + added_paths = True else: updated_sys_argv.append(elem) Suggestion importance[1-10]: 8 __ Why: Setting added_paths = True prevents repeatedly injecting test_split into updated_sys_argv, fixing duplicate test arguments in subprocess calls.	Medium
General	Rebuild pytest args from unknown_args Instead of iterating over the full sys.argv, rebuild the pytest invocation from unknown_args to ensure only the intended test paths and flags are included. This simplifies argument handling and avoids mixing tracer flags with pytest flags. codeflash/tracer.py [121-126] -for elem in sys.argv: - if elem in test_paths_set: - if not added_paths: - updated_sys_argv.extend(test_split) - else: - updated_sys_argv.append(elem) +# Rebuild pytest command with split tests +base_pytest = unknown_args[0] +other_args = [arg for arg in unknown_args[1:] if arg not in test_paths_set] +updated_sys_argv = [base_pytest, *other_args, *test_split] Suggestion importance[1-10]: 4 __ Why: Rebuilding from unknown_args could simplify argument handling, but it risks dropping tracer-specific flags and may not cover all edge cases.	Low