›_ Codex Issues
Codex Issues / #16828

Linux: long-lived Codex session appears to accumulate memory and can hard-freeze host

Open 💬 0 comments Opened Apr 5, 2026 by vladk-ml

What version of Codex CLI is running?

0.118.0

What subscription do you have?

Plus

Which model were you using?

gpt-5.4

What platform is your computer?

cachyos

What terminal emulator and version are you using (if applicable)?

kitty

What issue are you seeing?

Environment

  • OS: CachyOS Linux
  • Session: Wayland / niri
  • Terminal: kitty
  • Codex version: codex-cli 0.118.0
  • RAM: 64 GiB
  • GPU: NVIDIA

Summary

  • A long-lived Codex session was left open in a single terminal for most of the day.
  • Near the end of the session, the machine hard-froze.
  • Recovery required a full power cut.
  • The first reboot attempt produced no display output.
  • The second reboot completed successfully.

Observed evidence

  • The prior boot journal shows the terminal scope reaching a very large memory peak:

Apr 04 20:39:11 ... kitty-20694-0.scope: Consumed 2h 59min 55.297s CPU time over 10h 43min 8.879s wall clock time, 49.4G memory peak, 3.1G memory swap peak.

  • The parent launcher scope remained small:

Apr 04 20:39:11 ... app-niri-inir-20694.scope: ... 136.9M memory peak

  • The long-lived Codex thread in that same terminal had very large token totals by approximately 2026-04-04 20:05 PDT:
  • input_tokens: 64,403,106
  • cached_input_tokens: 62,698,496
  • output_tokens: 255,556
  • reasoning_output_tokens: 88,037
  • total_tokens: 64,658,662
  • That thread shut down around 20:37:41 PDT, roughly 90 seconds before the host froze.

What was not observed

  • No kernel panic
  • No OOM killer event
  • No systemd-oomd event
  • No thermal shutdown
  • No NVMe/storage fault
  • No machine check / hardware error
  • No NVIDIA Xid before the freeze

Interpretation

  • This does not look like a normal kernel-level crash.
  • The strongest signal is memory growth inside a long-lived terminal scope hosting a long-lived Codex session.
  • kitty may contribute via scrollback or terminal state, but the issue appears correlated with Codex session lifetime/size.

Expected behavior

  • Long-running Codex sessions should degrade more safely.
  • A CLI session should not be able to silently grow until it destabilizes a 64 GiB workstation.

Actual behavior

  • Memory usage appears to grow large enough to hang the machine.
  • The host became unresponsive and required hard power cycling.

Notes

  • This was a heavy all-day session, not a short normal run.
  • Project-specific details are intentionally omitted.
  • More sanitized logs can be provided if needed.

What steps can reproduce the bug?

  • Start a Codex session on Linux in kitty.
  • Keep the same session/thread open for a long period, with repeated interactive use across the day.
  • In this case, the session lived for about 10h 43m.
  • The affected long-lived thread had very large token totals by around 2026-04-04 20:05 PDT:
  • input_tokens: 64,403,106
  • cached_input_tokens: 62,698,496
  • output_tokens: 255,556
  • reasoning_output_tokens: 88,037
  • total_tokens: 64,658,662
  • Near the end of the session, the host hard-froze.
  • Recovery required a full power cut.
  • The first reboot attempt produced no display output.
  • The second reboot succeeded.

Empirical evidence from the previous boot journal:

Apr 04 20:39:11 ... kitty-20694-0.scope: Consumed 2h 59min 55.297s CPU time over 10h 43min 8.879s wall clock time, 49.4G memory peak, 3.1G memory swap peak.
Apr 04 20:39:11 ... app-niri-inir-20694.scope: ... 136.9M memory peak

Codex version:

codex-cli 0.118.0

What is the expected behavior?

  • A long-running Codex session should not silently grow until it destabilizes the host.
  • If session state becomes unusually large, Codex should fail more gracefully, warn, or limit resource growth.
  • A CLI session should not be able to hard-freeze a 64 GiB workstation during normal interactive use.

Additional information

  • Environment:
  • CachyOS Linux
  • Wayland / niri
  • kitty
  • NVIDIA GPU
  • 64 GiB RAM
  • No matching kernel panic, OOM killer, systemd-oomd, thermal shutdown, NVMe/storage fault, machine check, or NVIDIA Xid was found before the freeze.
  • This appears more consistent with memory growth inside a long-lived terminal scope hosting a long-lived Codex session than with a normal kernel-level crash.
  • Project-specific details intentionally omitted, but more sanitized logs can be provided if useful.

Additional information

_No response_

View original on GitHub ↗