No space left on device: finding what filled the disk

A full disk on a Minecraft server does not just stop saves, it can damage the world. Here is what usually fills it and what to delete first.

This one is urgent in a way most errors are not, because Minecraft's response to a failing write is to try again rather than stop.

java.io.IOException: No space left on device
Failed to save chunk ...

#Deal with it before you diagnose it

If the disk is full right now, free something immediately and restart. Diagnosis can wait; a server writing chunks to a full disk cannot.

The fastest safe deletions:

• /data/logs/. Old rotated logs, often gigabytes on a long-running server. Keep latest.log and delete the dated archives.
• /data/crash-reports/. Useful individually, worthless in bulk. Keep the most recent one if you are debugging something and delete the rest.
• Backups stored on the server itself. If you have been zipping worlds into the server's own filesystem, those are the first thing to go. Backups belong somewhere that is not the machine they protect.

#Then find what actually filled it

Once the server is running again, work out where the space went.

World data. The usual answer on a long-lived server. Every chunk a player has ever loaded is on disk forever, and a few people flying in straight lines with a high view distance generates a surprising amount of it. Modded servers multiply this by the number of dimensions.

Pre-generation. Pre-generating a world is good for performance and expensive in disk. A pre-generated 10,000-block radius is a lot of region files, and people frequently pre-generate a radius far larger than anyone will visit.

Mod caches. Some mods cache downloaded assets under /data. Modpack installs leave staging directories behind. Neither is cleaned automatically.

Dynmap and similar. Web map mods render tile images for every explored chunk, at several zoom levels. On a large explored world this can exceed the world data itself. If you run one, check its directory first.

#Preventing it

Set a world border. The single most effective measure. A border stops the world growing indefinitely and costs nothing. Most servers will never use more than a few thousand blocks in any direction, and the ones that do can raise it deliberately.

Keep backups off the server. Automatic backups on Hostd are stored off-site, which is the point of them. Manual snapshots you want to keep should be downloaded rather than left where they consume the space they are meant to protect.

Watch the disk metric. The dashboard shows disk use. A server that climbs steadily rather than sitting flat is generating something, and finding out what while there is still room is a great deal easier than after.

#When world data is genuinely the bulk

If logs are trimmed, backups are elsewhere, and the world itself is what fills the disk, there are two honest options.

Trim it: tools such as MCA Selector remove regions no player has visited recently, which reclaims a lot on servers where somebody once flew ten thousand blocks in a straight line and never returned. Take a backup first, because this deletes world data by design.

Or upgrade. A world that genuinely needs the space needs the space, and no amount of tidying changes that.

#The corruption risk

Worth stating plainly. When a chunk write fails part-way through, the region file is left partially written. Minecraft may or may not notice.

So after any disk-full incident, watch for chunk errors in the log over the following days. If the disk filled while players were online and building, the odds of at least some damage are not small, and a backup from before the incident is worth keeping until you are confident.

Written and maintained by the Hostd engineering team. Last updated 2026-08-05. Notice a mistake? Tell us.