Uningnored large data cause massive CPU bound by the worker that are created by Codex for git add and diffs

Open 💬 1 comment Opened Jun 28, 2026 by arkdong

What version of the Codex App are you using (From “About Codex” dialog)?

26.623.42026

What subscription do you have?

Pro

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

After a recent update, Codex Desktop repeatedly spawned multiple high-CPU git workers in one repo.

Observed commands:

  • git -c core.hooksPath=/dev/null -c core.fsmonitor= add -A
  • git -c core.hooksPath=/dev/null -c core.fsmonitor= add -- <hundreds of files>

Observed behavior:

  • 3-4 git processes at once, each around ~95-100% CPU
  • total CPU around 400-500%
  • commands were children of /Applications/Codex.app/Contents/MacOS/Codex
  • env included GIT_INDEX_FILE=/var/.../codex-index-*/index, so Codex was using temp indexes
  • however lsof showed git writing temp packs into the real repo .git/objects/pack/tmp_pack_*
  • logs showed requestKind=branch-diff-stats and turn-diff-capture-start
  • repo has a large unignored data/ directory: ~83GB, 372 files, 225 files >100MB

Codex should not spawn overlapping git add workers for the same repo, should avoid writing huge untracked blobs into the real object database just to compute diff stats, and should degrade/cap/skip large untracked files.

What steps can reproduce the bug?

Large unignored data like csv and txt

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗