OpenAI Codex: Logging Bug Causes Massive SSD Storage Consumption

OpenAI Codex has released a fix for a critical bug where the application writes excessive amounts of data to local SSDs. This issue, tracked in GitHub issue #28224, caused some users to experience storage exhaustion and potential SSD wear due to unbounded logging.

The Logging Bug and Its Impact

OpenAI Codex was found to be shipping with trace-level debug logging enabled by default for all users. Because the application uses SQLite for logging and lacks log rotation or capacity caps, the log files grew indefinitely.

This "firehose" effect resulted in some users reporting that their local storage was consumed by massive SQLite files. One user reported that running a VACUUM FULL command on their log file shrunk it from 27GB down to 73MB, indicating a significant amount of wasted space.

Temporary Workarounds for Affected Users

Before the official patch, community members identified several ways to mitigate the storage impact:

  • SQLite Trigger: Users can prevent further log inserts by creating a SQLite trigger to ignore new entries. The suggested command is: sqlite3 ~/.codex/logs_2.sqlite "CREATE TRIGGER IF NOT EXISTS block_log_inserts BEFORE INSERT ON logs BEGIN SELECT RAISE(IGNORE); END;"
  • Database Vacuuming: To reclaim space, users can run VACUUM FULL on the ~/.codex/logs_2.sqlite file.
  • Filesystem Symlinking: Some users suggested symlinking the log directory to a tmpfs (RAM disk) to prevent SSD wear.

Community Feedback and Performance Issues

Discussion among developers on Hacker News highlighted broader concerns regarding the quality and stability of Codex. Beyond the logging bug, users reported other performance regressions:

"Just having the window unhidden on my mac will cause it to use 100% of the GPU displaying the spinner message."

Users also compared Codex to competitors like Claude Code, noting that Codex currently suffers from typing latency and general instability. Some users reported that Claude Code has similar issues with massive debug logs, suggesting a pattern of "vibe coding"—prioritizing speed of release over rigorous QA and testing.

Resolution

An OpenAI engineer confirmed that a fix for this issue has been published in an update to the CLI and the Codex App. Users are encouraged to update their installations to the latest version to resolve the storage consumption bug.

Sources