In an issue filed on the openai/codex GitHub repository (issue #28224), a user reported that the Codex CLI continuously writes large SQLite feedback logs, located at ~/.codex/logs_2.sqlite, ~/.codex/logs_2.sqlite-wal, and ~/.codex/logs_2.sqlite-shm. The reporter measured about 37 TB written to their main SSD after roughly 21 days of uptime, which they extrapolated to roughly 640 TB/year. The issue note connects that extrapolation to endurance figures by observing that 1 TB consumer SSDs would see about 640 full-drive writes per year, and many consumer drives list ratings near 600 TBW, implying the observed logging rate could consume an SSD's warranted write endurance in under a year. The report includes retained-row counts and process-level checks attributing the continuous writes to Codex's SQLite logs.
What happened
In an issue filed on the openai/codex GitHub repository (issue #28224), a user reported that the Codex CLI is continuously writing large amounts of data to the local SQLite feedback log database, specifically to ~/.codex/logs_2.sqlite, ~/.codex/logs_2.sqlite-wal, and ~/.codex/logs_2.sqlite-shm. The reporter measured approximately 37 TB of writes on their main SSD after about 21 days of uptime, and the issue text extrapolates that to roughly 640 TB/year, noting that on a 1 TB SSD this equates to about 640 full-drive writes per year, compared to common consumer endurance ratings near 600 TBW.
Technical details
The issue includes process- and file-level checks that identify the Codex SQLite files as the primary continuous writers. The reporter provided retained-row counts inside logs_2.sqlite and observed continuous WAL activity tied to the Codex process. Those artifacts are logged within the GitHub issue as evidence.
Editorial analysis
Persistent, high-volume local logging into a small SQLite store can generate sustained write amplification via the WAL (-wal) file and frequent checkpoints; industry experience shows this pattern can rapidly increase SSD write cycles on long-running developer tooling and client-side agents.
Context and significance
For practitioners, this is a reminder that local feedback collection and verbose client logging can have hardware costs, especially on consumer-grade SSDs with finite TBW ratings. Similar issues have prompted changes to logging frequency, batching, or configurable retention in other developer tools.
What to watch
Observers should watch for responses or fixes on the openai/codex repository, such as opt-in logging, reduced write frequency, configurable retention, or migration to less write-intensive telemetry mechanisms. Also monitor related issue threads for reproducible metrics on different platforms and SSD models.
Scoring Rationale #
The issue documents a concrete measurement that could affect developer workflows and device reliability, making it notable for practitioners running long-lived client tooling. It is not a broad-platform vulnerability, so its industry impact is meaningful but limited to deployments and tooling configurations.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.