Summary

Fixes pod restart issues caused by blocking subprocess operations during large PR reviews by converting git operations to non-blocking async subprocess calls.

Problem

• Large PRs (like Tr/readme links The-PR-Agent/pr-agent#760 with 476 files) caused pod restarts due to liveness probe failures
• Root cause: blocking subprocess.run() calls freeze event loop for 5-6+ seconds
• Critical git operations: clone (~5-6s), fetch (~2-3s), show (~1-2s)

Solution

Converted blocking subprocess calls to asyncio.create_subprocess_exec():

• _clone_inner() - git clone operation
• _get_diff_files_with_git_clone() - git fetch operation
• get_file_from_git() - git show operation

Changes

Core Implementation

• Added async helper functions in git_provider.py:
  • run_subprocess_async() - non-blocking subprocess execution
  • run_subprocess_with_output() - non-blocking with output capture
• Made clone() and get_diff_files() async in base class
• Updated all 3 blocking git operations to use async helpers

Propagated Changes

• Made methods async across tool files (pr_reviewer, pr_description, pr_code_suggestions, etc.)
• Added await to all get_diff_files() calls
• Made helper functions async (dedent_code, validate functions)

Test Results

✅ All unit tests pass: 204 tests passed in 4.03s

✅ Async functionality tests pass:

• Basic async subprocess: 0.006s
• Output capture: 0.008s
• Timeout handling: Correct
• Concurrent execution verified: 3 commands in 1.026s (would be 3s if blocking)
• Real git operations: 0.019s

Merge Conflict Risk

• bitbucket_server_git_provider.py: 0% risk (Rakuten-specific file)
• git_provider.py: 30-40% risk (base class)
• Tool files: 50-70% risk (actively developed)
• Overall: Medium-High (50-60%) but EASY to resolve (systematic async/await additions)

Deployment Plan

1. Deploy to staging environment
2. Test with large PR (e.g., Tr/readme links The-PR-Agent/pr-agent#760)
3. Monitor pod health (should not restart during processing)
4. Verify PR review completes successfully

Documentation

• docs/rakuten/ASYNC_BLOCKING_ANALYSIS.md - Problem analysis
• docs/rakuten/SUBPROCESS_FIX_PLAN.md - Implementation plan

Testing

Local async tests verified:

• Non-blocking subprocess execution works
• Concurrent operations run in parallel
• Event loop remains responsive
• Git operations are properly async

Ready for deployment testing on k8s.