Server cleanup
Docker cleanup, reclaiming disk on a server. One schedule for the instance, individual hosts opt out.
Reclaiming disk. This is the setting people wish they had found earlier.
Settings -> Servers -> the server -> Cleanup. One schedule for the whole instance, and individual hosts opt out of it with the Scheduled cleanup switch in that card's header, so a server you add tomorrow can never silently go unswept. The checkbox below it turns the whole thing off instance-wide.
It ships on, with all seven scopes enabled, a nightly 0 4 * * * schedule, a 24-hour minimum
age and one image kept per app. A new scope added by a later release enables itself on a policy
you have already saved.
What gets cleaned
Prop
Type
Every scope judges against what this Deplo still knows about, so a live app keeps its images, files and network even while it is moving between servers, and a stopped app keeps everything.
Unused app images and the build cache are also swept right after every deploy, so a day of builds cannot fill the disk before the schedule fires.
The build cache also has a ceiling: a tenth of the server's disk. Past that, the oldest cache is evicted even when it is younger than the minimum age. Without it, a host that builds all day never has a cache old enough to qualify, and the ceiling is the only thing that still bites.
Settings
Prop
Type
What it deliberately never does: system prune, container prune, volume prune, network prune. A
container and anything a container still uses are never touched, because on a Deplo host a
stopped app is a live app. A named volume goes only when it is one of your apps' and is
empty: no bytes, nothing to lose. Anything with data in it stays, whatever its name.
What cleanup does not touch
- Cleanup and rollbacks fight over the same images. If rollbacks stop working, Images kept per app is too low for the app's Keep.
- This is not what frees the disk after you delete something. Cleanup reclaims images, caches and the directories of apps that are gone. It never touches a named volume, and it never touches a server folder an app reached through a Bind - those are yours to remove.
- A pulled image comes back with the next deploy. Removing one costs a pull, never data. Untick Unused pulled images on a server without registry access.
- Backup files are not on this list either. Deleting a backup schedule leaves the copies it already made in the destination.
- A migration source is never swept, scheduled or on demand. That machine is somebody else's platform until you are done with it.
Troubleshooting
| Symptom | What to check |
|---|---|
| Cleanup frees nothing | The build cache is already under its ceiling and the minimum age is protecting the rest, or the only large thing on disk is images your apps still use. |
| Reclaim disk now ran on a host you opted out | It does: the manual run ignores the opt-out list. The schedule honours it. |
| Rollbacks stopped working after a sweep | Images kept per app is too low for an app that cannot roll back. For apps that can, their own Keep governs. |
More in Servers and agents.
Next steps
Did this page help you?