What PARTHA keeps on disk, what it discards on its own, and what is left to you.
| Where | What | Grows with |
|---|---|---|
STORAGE_PATH/repositories/<id>/ |
the extracted source of each imported repository | each imported repository or revision |
the database (ri_* tables) |
the analysis of each repository revision: files, symbols, relationships, evidence | each analysed revision (tens of MB for a repository the size of PARTHA itself) |
- A failed or cancelled analysis keeps its snapshot row (so the failure stays explainable) but its partial facts are deleted as soon as it fails. A cancelled run on a large repository used to leave tens of megabytes behind.
- Facts left by earlier failures (from before this behaviour, or by a rejected seal) are discarded a few at a time by the worker's periodic sweep.
- Deleting a repository in the app removes its source directory and its analysis.
- Which old revisions to keep. Each re-import of a source is a new repository with its own analysis, and the older one remains so its history can be viewed. Delete the ones you no longer need in the app; PARTHA does not expire completed analyses on its own, because that would silently remove history.
- Returning space to the operating system. SQLite reuses freed pages for new analyses but never shrinks the file until it is vacuumed.
cd apps/backend
python scripts/cleanup_storage.py # report only
python scripts/cleanup_storage.py --apply # sweep failed-analysis facts, remove orphaned directories
python scripts/cleanup_storage.py --apply --vacuum # also shrink the SQLite fileIt reports by default and deletes only with --apply. It never touches a completed analysis or a repository that still exists. Stop PARTHA before --vacuum (it rewrites the whole file and needs free disk of about the database's size). On PostgreSQL, autovacuum reclaims space and --vacuum does nothing.