Disk limits and volumes
Each server's data lives in a sparse ext4 image at DATA_ROOT/.volumes/<id>.img, sized to its disk allowance and loop-mounted at DATA_ROOT/<id>. Because the filesystem itself is that size, the kernel returns no space left to the game, SFTP and the panel alike.
Requirements
Root, /dev/loop-control, and the tools mkfs.ext4, losetup, mount, resize2fs, e2fsck, plus fsfreeze and cp for snapshot backups. The enforcement mode (auto, require, off) comes from the panel (Settings → Game nodes → Disk limits):
| Mode | Behaviour |
|---|---|
| auto | Use volumes when the node can; otherwise fall back to soft limits |
| require | Refuse to run servers on nodes that cannot enforce limits |
| off | Never use volumes |
Growing and shrinking
- Growing is applied live where the kernel allows it.
- Shrinking, or growing on hosts that refuse online resizing, briefly stops the server: it is stopped, the volume is checked and resized with
e2fsckandresize2fs, and the server starts again. - The panel refuses to shrink below the space in use.
- Usable space is about 2 % less than the allowance because of filesystem metadata.
- The unmount before a resize is strict: it retries while something (a stopping container) still holds the mount, waits until the loop device is released, and never falls back to a lazy unmount that would report success while the filesystem is still mounted.
Existing servers
Servers created before volumes existed move into a volume the next time they are restarted.
Snapshot backups
On volume nodes a running server is backed up from a frozen point-in-time copy of its volume while it keeps running (Minecraft first flushes and pauses world saving). The result is crash-consistent. On other nodes the game is stopped cleanly.
What is not limited
The container's writable layer (anything a game writes outside the data mount) is not limited. Volumes need a Linux node with root.
Troubleshooting
- "e2fsck: ... is mounted" in an old agent's logs: fixed in 0.6.1.1 by the strict unmount.
- A full host disk can hang the Docker engine; see Troubleshooting.