OpenAI Codex SQLite logging bug writes 640 TB annually, risks SSD failure within one year

According to GitHub issue #28224, OpenAI Codex writes approximately 640 terabytes of diagnostic logs annually to a local SQLite database at ~/.codex/logs_2.sqlite, potentially exhausting a typical 1 TB consumer SSD's 600 TBW lifetime endurance in under a year. The logging runs at TRACE level by default, logging WebSocket payloads and filesystem events with no standard way to reduce verbosity. OpenAI merged three pull requests in June 2026 claiming to reduce logging by 85 percent, and a community-maintained workaround repository (codexfixes.com) provides a SQLite trigger to block further log growth.

Update (24 June 2026): The Register reports developer Rui Fan first identified the pattern after 21 days of uptime produced 37 TB of local writes on a single machine, extrapolating to the same ~640 TB annual figure. The Register frames this as a systemic infrastructure problem, with affected organizations facing millions of dollars in hardware replacement costs from premature SSD failure and warranty exhaustion across enterprise deployments.