Checkpoints creating excessive disk usage (40GB+) in VSCode global storage

[KJ7LNW]

Issue

Checkpoints are creating full git repositories for each task in the VSCode global storage directory, which is particularly problematic for large directories that don't already have git repositories, because it seems to copy the entire directory every single time!

This is causing excessive disk usage: 40GB after just a few minutes of Roo'ing!

Version

v3.14.2

Details

• Each task creates a checkpoint directory with a full .git repository
• Multiple identical .git/objects/pack directories of 3.2GB each
• Total storage usage has reached 40GB in the tasks directory
• This occurs in .config/Code/User/globalStorage/rooveterinaryinc.roo-cline/tasks

The issue is particularly severe when working with large directories that don't already have git repositories, as the entire directory content gets copied into each checkpoint.

Steps to reproduce

1. Open VSCode with Roo extension
2. Start a new task in a large directory that doesn't have an existing git repository
3. Check disk usage in .config/Code/User/globalStorage/rooveterinaryinc.roo-cline/tasks:

   cd .config/Code/User/globalStorage/rooveterinaryinc.roo-cline/tasks
   du -mx | sort -n
   

4. Observe multiple 3.2GB+ .git/objects/pack directories and total usage around 40GB

Output

$ du -mx | sort -n
...
3214./631b3718-285e-4f4e-8868-90eb00c08a45/checkpoints/.git/objects/pack
3214./8b468531-8ec4-4da0-adf8-d476b0b616f8/checkpoints/.git/objects/pack
3214./9f7eb741-fdbf-4009-b636-30cb5521fcbc/checkpoints/.git/objects/pack
[...]
40920.  # 40GB total usage

Potential Solutions

1. Only create checkpoints for files that are actually modified
2. Set size limits for checkpoint directories or individual files

The current implementation creates full repository copies which is particularly problematic for large directories without existing git repositories.

[dosubot]

The issue you're experiencing with excessive disk usage due to checkpoints in the Roo Code extension is indeed problematic, especially when dealing with large directories. Here are some potential solutions and considerations based on the current implementation:

1. Modified Files Only: Implementing logic to create checkpoints only for modified files could significantly reduce disk usage. However, the current logic for determining which files are included in a checkpoint when a task is created is not clearly documented or found in the repository ^[1].

2. Git Sparse-Checkout: Utilizing git sparse-checkout could help manage large directories more efficiently by only including specific files or directories in the checkpoint. Unfortunately, it's unclear if the Roo Code extension currently supports git sparse-checkout ^[1].

3. Cleanup Policy: Introducing a mechanism to clean up old checkpoints could help manage disk usage over time. However, there is no existing mechanism for cleaning up old checkpoints documented or found in the repository ^[1].

4. Size Limits: Setting size limits for checkpoint directories could prevent excessive disk usage. The logic for size limits in checkpoint directories is not currently documented or found in the repository ^[1].

Given these points, it might be beneficial to explore implementing these features or checking for any updates or open issues related to these suggestions in the repository.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[bramburn]

@dosu Elaborate on the implementation of the time based cleanup

[bramburn]

Draft solutions #3413