PocketBase log retention deletes old rows, but SQLite does not shrink the file. Over time your instance logs.db can balloon even when the admin UI shows a normal row count. We wrote up the full case study in Your logs.db Might Be Gigabytes of Empty Space. Today we are shipping the fix for hosted instances.
Auto Vacuum runs SQLite VACUUM on your instance data.db and logs.db during PocketHost’s nightly maintenance sweep. It is on by default for every instance. You can turn it off per instance if you prefer to manage compaction yourself.
Full details are in Auto Vacuum.
When it runs
The edge runs pockethost health compact once a day via PM2 (health-compact). For each instance with Auto Vacuum enabled, we compact both SQLite files only while the instance is idle (hibernated). Running containers are skipped until traffic stops and the instance sleeps.
If a request wakes your instance while vacuum is in progress, you may see up to about 5 seconds of downtime while the database finishes. We schedule this for idle windows on purpose.
Dashboard toggle
Open any instance and scroll to Danger Zone → Auto Vacuum. Flip the switch off if you want PocketHost to skip your instance during the nightly job. Existing instances were migrated with Auto Vacuum enabled.
Self-hosters
Same command, same behavior:
pockethost health compact --dry-run # preview what would run
pockethost health compact # run the sweepWhen Mothership data lives on the same host, the job briefly stops Mothership to vacuum its data.db and logs.db too. Idle instance mounts are skipped if Docker still has the volume open.
Why bother
Smaller databases mean faster backups, less disk pressure on the edge, and fewer surprises when you ls -lh pb_data/*.db. You should not have to think about SQLite freelists. Auto Vacuum handles the boring part while you sleep.
Our first production sweep across the fleet reclaimed about 52 GB from compacted data.db and logs.db files alone, roughly 10% of used space on the node that ran it.
Questions or before/after numbers? Discord.