Codex 0.142.0 reduces the SQLite log churn behind high disk writes. Learn how to upgrade, verify the fix, clean old logs, and remove temporary triggers.
EasyGlobe Team
Global Growth Team
EasyGlobe helps teams expand into global markets with practical SEO, localization, LLM optimization, paid advertising, and growth operations. We turn complex international growth work into clear systems, high-quality content, and measurable execution.
Codex 0.142.0 is the version to install if you were affected by heavy SQLite log writes in ~/.codex/logs_2.sqlite. OpenAI published the release on June 22, 2026, and the release notes say it reduced persistent-log churn by removing per-event WebSocket payload logging and filtering duplicated telemetry records.
My practical recommendation is simple: upgrade first, then clean the old database. The fix changes future logging behavior, but it does not automatically shrink a database that already grew on your machine.
For related AI operations notes, browse the EasyGlobe Blog. This guide is focused only on the Codex SQLite log fix and the cleanup steps after upgrading.
The official Codex 0.142.0 release notes list a logging-related chore: reduced persistent-log churn. The same release also includes startup and session latency work, configurable rollout token budgets, indexed web search mode, and multi-agent delegation settings.
The direct logging fix is split across two merged pull requests. PR #29432 stops logging every successful Responses WebSocket event. PR #29457 filters noisy persistent log targets, including bridged dependency logs and duplicated OpenTelemetry mirror targets.
That means the fix is not a promise that Codex will never write local logs. It is a fix for the high-churn behavior that caused many unnecessary SQLite inserts, WAL writes, and prune cycles.
Run this command in a terminal:
codex --version
If the output is older than 0.142.0, update. If you installed through several channels in the past, also check which binary your shell is using.
which codex
The official Codex README documents multiple install paths. Use the one that matches how you installed Codex.
curl -fsSL https://chatgpt.com/codex/install.sh | sh
npm install -g @openai/codex@latest
brew upgrade --cask codex
After upgrading, verify again:
codex --version
First check the files:
ls -lh ~/.codex/logs_2.sqlite ~/.codex/logs_2.sqlite-wal ~/.codex/logs_2.sqlite-shm 2>/dev/null
Then check the log level distribution:
sqlite3 ~/.codex/logs_2.sqlite "SELECT level, COUNT(*) FROM logs GROUP BY level ORDER BY COUNT(*) DESC;"
Finally compare row counts over a short window:
sqlite3 ~/.codex/logs_2.sqlite "SELECT COUNT(*) FROM logs"; sleep 15; sqlite3 ~/.codex/logs_2.sqlite "SELECT COUNT(*) FROM logs";
Some small movement can be normal. The suspicious pattern is thousands of rows being inserted in seconds while the retained row count stays nearly flat.
Upgrading fixes future behavior, but old database pages and WAL files can still occupy disk space. Close Codex first if possible, then run:
sqlite3 ~/.codex/logs_2.sqlite "PRAGMA wal_checkpoint(TRUNCATE); VACUUM; PRAGMA wal_checkpoint(TRUNCATE);"
If you do not need old local debug logs, you can delete the old rows and reclaim more space:
sqlite3 ~/.codex/logs_2.sqlite "DELETE FROM logs; PRAGMA wal_checkpoint(TRUNCATE); VACUUM; PRAGMA wal_checkpoint(TRUNCATE);"
This removes local debug log rows. It does not delete your Codex conversations or project files.
If you previously created a trigger to block log inserts, keep it until you have upgraded and verified that row counts no longer jump rapidly. After that, you can remove it to restore normal local logging.
sqlite3 ~/.codex/logs_2.sqlite "DROP TRIGGER IF EXISTS block_log_inserts;"
Then rerun the 15-second row-count test. If the count stays stable or changes only slightly, the workaround is no longer needed.
No. It reduces the noisy persistent logging that caused abnormal churn. Normal local logging may still happen.
No. Use VACUUM after upgrading if the old SQLite file is still large.
For normal users, yes: these are local debug logs, not the main conversation store. Keep a backup only if you need diagnostic evidence for a bug report.
