codex - 💡(How to fix) Fix `codex resume` picker is slow with many local sessions because thread/list scans rollout files [1 comments, 2 participants]

TL;DR

The delay in the codex resume command can be mitigated by optimizing the rollout file listing process, potentially by utilizing the SQLite state DB for sorting and filtering.

Guidance

• Investigate modifying the codex-rs/tui/src/resume_picker.rs to set useStateDbOnly: true when the SQLite state DB exists, to prioritize database queries over filesystem scans.
• Review the codex-rs/rollout/src/list.rs and codex-rs/rollout/src/recorder.rs code paths to optimize the filesystem-first scan-and-repair approach, potentially by introducing a caching mechanism or improving the efficiency of the rollout file listing process.
• Consider introducing a threshold for the number of rollout files to determine when to switch from a filesystem-based approach to a database-based approach for sorting and filtering.
• Verify the performance impact of these changes by testing with a large number of rollout files and measuring the time it takes for the resume picker to become usable.

Example

No code example is provided as the issue requires a deeper understanding of the Codex CLI codebase and its internal workings.

Notes

The provided information suggests that the issue is related to the way the Codex CLI handles rollout file listing and sorting. However, without further investigation and testing, it is difficult to provide a definitive solution. The suggested guidance points are intended to help mitigate the issue, but may require additional modifications and testing to fully resolve the problem.

Recommendation

Apply workaround: Modify the codex-rs/tui/src/resume_picker.rs to set useStateDbOnly: true when the SQLite state DB exists, and investigate optimizing the rollout file listing process to improve performance. This approach prioritizes database queries over filesystem scans, which may help reduce the delay in the codex resume command.