Codex logging bug may write TBs to local SSDs
Points and comments are a snapshot, not live.
Codex's SQLite feedback logs with global TRACE default can write ~640 TB/year to SSDs.
A bug report shows Codex writes ~37 TB to an SSD in 21 days, extrapolating to ~640 TB/year. Over 70% of retained log bytes come from TRACE-level logs, including inotify events, tokio-tungstenite internals, and raw websocket/SSE payloads. The high write volume results from continuous insert-and-prune amplification in SQLite, with ~36,211 rows inserted every 15 seconds while retained row count stays flat. The report recommends filtering low-value dependency noise and avoiding global TRACE for the feedback log sink.
What commenters are saying
Commenters are critical of Codex's quality, calling it "slopware" and citing other severe issues like 100% GPU usage from a spinner animation. Some defend Codex relative to Claude Code, which they argue is worse. Several commenters argue the logging bug reflects over-reliance on AI-generated code with insufficient human review. One commenter notes the software works as intended, with TRACE logs saved for debugging, but others counter that such excessive logging harms users and shows negligent design.