Backups and retention¶

Two operational concerns that often blur together but should be kept separate. Retention decides what to throw away; backups decide what to keep safe.

Retention sweeper¶

A built-in sweeper runs every ARGUS_RETENTION_SWEEP_MINUTES (default 60). It is a singleton coordinated by an fcntl lock under ARGUS_LOCK_DIR — only one uvicorn worker runs the sweep, even with ARGUS_WORKERS>1.

What it purges¶

Type	Knob	Default
Resource snapshots	`ARGUS_RETENTION_SNAPSHOT_DAYS`	7
Log lines	`ARGUS_RETENTION_LOG_LINE_DAYS`	14
Job epochs	`ARGUS_RETENTION_JOB_EPOCH_DAYS`	30
Other events	`ARGUS_RETENTION_EVENT_OTHER_DAYS`	90
Demo data	`ARGUS_RETENTION_DEMO_DATA_DAYS`	1

What it does not purge¶

Batch summary rows (one per batch).
Job summary rows (one per job).
User accounts, projects, tokens.
Audit log (separate retention; not user-configurable today).

A batch from 2 years ago whose detail events have been swept clean still appears in the batch list with its final metrics — the headline data remains visible indefinitely.

Disable¶

ARGUS_RETENTION_SWEEP_MINUTES=0 turns off the in-process sweeper. Pair with a cron-driven external script if you want a different cadence.

Editable from the UI¶

Admins can change retention days on Settings → Admin → Retention. The DB row wins over the env var (see Admin settings).

Built-in SQLite backups¶

When the DB URL points at SQLite, Argus runs an in-process backup loop that calls SQLite's online .backup command (consistent under writes) on a schedule:

Variable	Default
`ARGUS_BACKUP_INTERVAL_H`	6 (`0` disables the loop)
`ARGUS_BACKUP_KEEP_LAST_N`	7

Backups land in data/backups/ next to the live DB file. The newest N are kept; older ones are removed.

This is also fcntl-coordinated, so multi-worker installs only run it once per cycle.

For PostgreSQL the loop is a no-op — back up via pg_dump instead.

Manual SQLite backup¶

The simplest one-shot:

docker compose exec argus sqlite3 /app/data/argus.db \
  ".backup /app/data/backups/manual-$(date +%F).db"

Then copy the file off the host (rclone / restic / rsync). cp argus.db argus.db.bak while the server is running can produce a truncated copy — prefer .backup or stop Argus first.

PostgreSQL backups¶

pg_dump -h db.internal -U argus -F c argus > argus-$(date +%F).dump

Restore:

pg_restore -h db.internal -U argus -d argus_new argus-2026-04-26.dump

What else to back up¶

./data/argus.db* (or your Postgres dump) — the database.
./deploy/.env — secrets. Without ARGUS_JWT_SECRET and ARGUS_CONFIG_KEY you cannot decrypt encrypted runtime config rows.

The frontend bundle and Python code are reproducible from the image — no need to back them up.

Retention of backups¶

A common rotation:

Last 7 daily backups (the in-process default).
Last 4 weekly snapshots (Sunday of each week, copied off-host).
Last 6 monthly snapshots.

Tools like restic or borg handle this rotation for you.