Windows Codex app: Nonpaged Pool (NtFC/ntfs.sys) grows until reboot is required

Codex
, ,
1

@OpenAI_Support

I am reporting a reproducible Windows Codex Desktop issue that appears to cause severe Nonpaged Pool growth until a full system reboot is required.

Environment

  • OS: Windows 11 x64
  • Codex Desktop: Microsoft Store build 26.616.5445.0
  • Current observed install path: C:\Program Files\WindowsApps\OpenAI.Codex_26.616.5445.0_x64__2p2nqsd0c76g0\app\Codex.exe
  • App server command observed: …\app\resources\codex.exe app-server --analytics-default-enabled
  • Microsoft Store apps were previously configured to install on D:, but Codex is now running from C:. The issue still reproduces, so this does not look like only a Store install-location problem.

Symptoms

After Codex Desktop runs for a while, Windows Nonpaged Pool keeps growing and eventually consumes many GB of memory. The machine becomes sluggish or unstable, and simply closing Codex has not reliably recovered the memory. So far, the only reliable recovery has been a full Windows reboot.

In one affected session, Nonpaged Pool reached roughly 20 GB. In a later measurement after cleanup, it was still growing at roughly 300+ MB/min.

Local observations

I collected several local diagnostics before posting:

  1. Codex.exe showed extremely high IO Other Operations/sec, around 160k+ operations/sec.
  2. Pool tag sampling previously showed ntfs.sys / NtFC as the dominant growing Nonpaged Pool tag.
  3. After restarting Windows and launching Codex again, the Codex app-server quickly accumulated a large child-process tree:
    • direct child processes under codex.exe app-server: about 185
    • many repeated cmd.exe, node.exe, and node_repl.exe processes
  4. Cleaning up duplicate/stale MCP child processes temporarily reduced the process tree, but it did not stop the Nonpaged Pool growth.
  5. The problem still reproduces while Codex is installed/running from C:, so it does not appear to be caused only by my previous D: Store app installation setting.

My working hypothesis is that Codex Desktop on Windows may be triggering very high filesystem/process activity, which then drives ntfs.sys / NtFC Nonpaged Pool growth. I am not claiming that ntfs.sys is the root cause; it may simply be where the kernel memory pressure becomes visible.

Related public reports

These public Codex issue reports look related or adjacent:

They mention Windows instability, possible memory growth, repeated git/process activity, and high system resource usage around Codex. I am posting here because this specific Nonpaged Pool failure mode is severe: once it happens, rebooting Windows is the only reliable recovery I have found.

Questions for the OpenAI / Codex team

Could someone from OpenAI please advise:

  1. Is this a known Codex Desktop Windows issue?
  2. Is there an official workaround to disable, debounce, or limit Codex workspace/Git scanning on Windows?
  3. Is there a supported way to limit repeated MCP server spawning or safely clean up stale MCP child processes without disabling MCP entirely?
  4. What diagnostics would be most useful for the Codex team?
    • PoolMon output?
    • RAMMap snapshot?
    • ProcMon trace?
    • WPR/ETW trace?
    • Codex app logs?
  5. Is a fix planned for the Windows Desktop / Microsoft Store build?

I can provide sanitized logs, screenshots, and traces if needed. I would prefer not to post raw local project paths or private workspace names publicly.

Thanks.