openai / openai/codex

Windows: OneDrive SyncEngine log storm and 15.7 GiB file-cache growth while using Codex

Open
#44,427 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug performance windows-os
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

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

26.903.8094.0 (OpenAI.Codex Windows package metadata)

What subscription do you have?

ChatGPT account; exact subscription tier not included (local Windows/filesystem issue)

What platform is your computer?

Windows 11 Home 25H2, build 26200.9445, x64; OneDrive 26.158.0816.0002; 32 GB RAM

What issue are you seeing?

System RAM use reached about 27 GiB while OneDrive continuously generated SyncEngine-*.odl logs. The dominant measured growth was resident Windows system file cache; the measured Codex process-group Private Bytes did not grow. The user observes the problem becoming more apparent after opening Codex, but the exact upstream trigger has not been established.

Measurements on 2026-09-10 (UTC+08:00):

  • 21 samples over five minutes: file cache increased 2.211 GiB; 2,128 new OneDrive logs totaled 2,128.43 MiB. The measured Codex group stayed at 28 processes and aggregate Private Bytes decreased from 3.031 to 2.866 GiB.
  • Complete-minute log growth: approximately 414-439 MiB/minute. Before quitting OneDrive, its log directory contained 14,870 files totaling about 14.41 GiB.
  • A completed approximately 1 MiB ODL sample contained HandleRecycledFile 91 times and FindFileByNameAndRepairFSIDCacheIfRequired 92 times. These are raw string occurrences, not a fully decoded error trace.

Controlled comparison, keeping Codex open with the same main process:

Metric Before OneDrive exit 13:17:44 Stopped 13:21:02 Restarted 13:22:11
RAM in use (GiB) 27.02 11.81 12.28
Available RAM (GiB) 4.13 19.51 18.88
Resident system file cache (GiB) 15.68 0.69 0.86
Measured Codex group Private Bytes (GiB) 3.11 3.11 3.02

Within about eight seconds of normal OneDrive exit, RAM fell to 11.70 GiB and file cache to 0.62 GiB. During about three minutes stopped, RAM stayed at 11.70-12.48 GiB. Restarting OneDrive caused the log storm to recur: net growth 285.68 MiB in approximately 63 seconds (272.1 MiB/minute). Restart was temporary relief, not a durable fix.

What steps can reproduce the bug?

This is an observed recurrence on this machine, not yet a minimal clean-install reproducer.

  1. Leave OneDrive running in the affected environment while using Codex with a project under OneDrive-synchronized Documents.
  2. Observe continuous SyncEngine log generation and increasing resident system file cache/RAM.
  3. Normally quit OneDrive with its /shutdown option while keeping Codex open; RAM and file-cache usage drop immediately.
  4. Restart the same OneDrive executable with /background; excessive log generation resumes.

The exact offending file/junction has not been identified. No sync file or log was manually deleted; no reset, reinstall, unlink, runtime relocation or junction removal was performed. OneDrive itself cleaned most old ODL logs at normal exit, so aggregate measurements remain but the original raw log set was not fully preserved.

Token/context-window usage: not relevant to the measured local Windows/filesystem behavior.

What is the expected behavior?

Using Codex in this environment should not lead to sustained OneDrive scanning/logging or system-wide memory pressure. Please investigate whether the Windows runtime/dependency layout or directory junctions interact with OneDrive FileScanner/FSID handling, coordinate with Microsoft as needed, and provide a supported remedy or runtime-location guidance.

Additional information

Possibly related: https://github.com/openai/codex/issues/29597 and https://github.com/openai/codex/issues/42154 . This report adds quantified 15.68 GiB file-cache impact and a controlled stop/restart comparison on the versions listed above; the root cause has not been proven identical. In particular, #42154's handle leak was not established on this machine.

%USERPROFILE%\.cache\codex-runtimes exists on C: (NTFS), including codex-primary-runtime. The active project is in OneDrive Documents. No File ID has yet been mapped to the offending local file. The earliest retained log predates the current Codex main-process start by roughly 27 seconds, so that launch cannot be established as the start of the incident.

Measurements used Windows PerfOS/PerfProc and GetPerformanceInfo/NtQuerySystemInformation(SystemFileCacheInformation). Units are GiB/MiB. No user documents, account identifiers, business paths, access tokens or raw application logs are included in this report.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing with Codex open and the affected project under OneDrive Documents, recording SyncEngine-*.odl growth and file-cache metrics before and after OneDrive /shutdown and /background. Compare the runtime at %USERPROFILE%.cache\codex-runtimes, including codex-primary-runtime, with the reported junction and FSID behavior; done means a supported remedy or runtime-location guidance is identified.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
operating-systems
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.