Skip to content

Storage & Retention

Everything fetchly persists lives under one directory: DATA_DIR (/app/data in the container). Mount it, back it up, and the instance is fully reproducible.

Layout

$DATA_DIR/
├── jobs.db               # SQLite (WAL): jobs, settings, share links, analysis cache
├── jobs.db-wal
├── jobs.db-shm
├── <job-uuid>/           # One directory per job
│   ├── <title>.source.<ext>
│   ├── <title>.mp3
│   ├── thumbnail.jpg
│   ├── trim_<start>_<end>.wav
│   └── trim_<...>_vocals.mp3
├── cookies/              # Per-platform Netscape jars, mode 0600
├── logo/                 # Your uploaded watermark logo, if any
├── thumb-cache/          # Cached remote thumbnails
├── watermark-cache/      # Rendered video watermark badges
├── update_check.json     # Cached upstream release check
└── .cache/torch/         # beat_this model checkpoint (~81 MB)

This directory is a credentials store

cookies/ holds live, signed-in sessions for your own platform accounts, and jobs.db holds the admin password hash and your Lalal.ai activation key. Restrict its permissions and encrypt your backups.

Retention

Settings → General → Retention (0–365 days, default 0)

0 means unlimited: job files are kept until you remove them explicitly. Any other value is the number of days a finished job's artifacts survive.

The housekeeping sweep

A background task runs every hour and, in one pass:

  1. Reads the current retention setting
  2. Deletes the artifacts of jobs past their retention age
  3. Deletes the share links for those jobs — their targets are gone, so the links could only 404 from here on, and the table would otherwise grow without bound
  4. Removes thumbnail cache entries older than 7 days
  5. Removes orphaned directories — directories on disk with no matching job row

Retained rows protect their directories

The orphan sweep only removes directories with no corresponding job record, and it skips anything younger than a 15-minute grace period so a directory being written by a running job is never mistaken for an orphan.

Manual cleanup

Action Where Effect
Remove all jobs Settings → System Removes every row and every directory
Delete a trim Job page Removes trim outputs, keeps the source

There is no per-job delete action in the UI or API — only the bulk Remove all jobs action and the automatic retention sweep remove job rows and their directories.

Removing jobs breaks their share links

Links to a removed job return the standard "link unavailable" page immediately.

Disk usage

Settings → System → Host resources shows storage alongside CPU, RAM, and uptime.

du -sh data/
du -sh data/*/ | sort -rh | head

Video jobs at max quality dominate. If space is tight:

  • Set a retention period instead of leaving it unlimited
  • Prefer 720p or audio-only where the source resolution does not matter
  • Cap the input in Settings → General → Runtime limits

The database

SQLite in WAL mode. On a clean shutdown the WAL is checkpointed and the database closed, which is why the container's stop_grace_period must stay above GRACEFUL_TIMEOUT — a kill mid-checkpoint leaves the WAL to be replayed on next start.

The schema migrates forward automatically at startup; there is no manual migration step.

Backup and restore

docker compose stop
tar czf fetchly-backup-$(date +%F).tar.gz data/
docker compose start

docker compose down
rm -rf data/
tar xzf fetchly-backup-2026-09-01.tar.gz
docker compose up -d

Stopping first is what makes the copy consistent. A hot copy can catch the database mid-write.

A smaller backup

To back up only the configuration and job history, take jobs.db* and cookies/ and skip the job directories. You lose the downloaded media but keep every setting, credential, and record.

Moving to another host

The data directory is portable across hosts of the same architecture family — it is SQLite plus plain files. Copy it, point the new instance's DATA_DIR at it, and set the same FETCHLY_SECRET_KEY if you want existing sessions to survive.