Skip to content

Minecraft server backups that actually restore

← Back to docs index

Most servers have backups. Rather fewer have backups that work, and the difference is only ever discovered on the worst day. Three things separate them: the copy is consistent, it lives somewhere else, and somebody has restored it at least once.

Copy the world while it is not being written

Section titled “Copy the world while it is not being written”

A running server writes region files continuously. Copy them mid-write and you get a file that is half old and half new, which either fails to load or loads with a hole in the ground where somebody’s base used to be. Nothing warns you, because the copy succeeded.

There are two correct ways to take a copy:

  1. Stop the server, copy, start it again. Simple, honest, and unpopular because of the downtime.
  2. Tell the game to stop writing first. Run save-off, then save-all flush, wait for it to finish, copy, then save-on. The world is complete and quiet for the duration of the copy.

Never skip straight to the copy. A snapshot at the filesystem or volume level is fine only if it is genuinely atomic, and a cp -r while players are placing blocks is not.

Restoring the world alone gets you the terrain and loses the server. You also want:

  • server.properties, and whatever else sets the rules.
  • The mod or plugin jars, at the exact versions that were running. A world will not load without the mods that created its blocks.
  • Player data, including whitelist.json, ops.json and the stats and advancements folders.
  • The pack version, if it came from a pack. Writing it down counts.

The quickest sanity check on any backup: could you rebuild this server on a fresh machine from the archive alone? If the answer needs a sentence starting with “and then I would reinstall”, it is not a complete backup.

A backup that lives on the machine it protects covers exactly one failure: you deleted something. It does not cover the disk dying, the provider suspending the account, or a mistake that deletes the whole directory.

Keep it simple rather than perfect:

  • One local copy, for the common case, which is somebody griefing or a bad update.
  • One copy somewhere else. Another machine, an object store, or a home NAS pulling nightly over SSH.
  • Enough history to survive a problem you did not notice immediately. Corruption is often found days later, and if you keep three days of backups you have already overwritten the good one.

Seven daily copies plus four weekly ones covers nearly every real incident and costs very little space.

Pick a backup, restore it into a spare server, and join. That is the entire test, it takes fifteen minutes, and it catches all of the classic failures: the archive missing the mods folder, a world that needs a version you no longer run, permissions that make the files unreadable, a truncated upload.

Do it once when you set the backups up, and once more after any big change to the server.

Worlds grow because players explore. A world where somebody flew ten thousand blocks in a straight line carries every chunk along that line forever, and most of those chunks will never be visited again.

Three levers, in order of how much they give back:

  1. Trim unvisited chunks. Tools that prune by last-touched time typically remove a large fraction of a well-travelled world with no visible effect in play.
  2. Compress properly. Region files compress well; archives of them do not benefit from anything exotic.
  3. Do not keep every hourly copy for a year. Retention beats compression.

Backups take the save-off, flush, copy, save-on path, and a stop or restart requested mid-backup is refused until the copy finishes rather than racing it. The archive covers the whole server directory, so mods, configuration and player data travel with the world.

Backups can run on a schedule alongside restarts and commands, and every new server starts with a daily backup schedule. Retention counts each kind separately, so the ten scheduled copies that pile up on their own cannot push out the one you took by hand before a risky change, and age and size caps apply on top. Admins change all of it on the Settings page, or per server on its Backups tab. Restoring takes a fresh backup first, so a restore to the wrong point in time is itself undoable. A world can be shrunk on its own or as part of a backup, which is the lever that matters most for size.

The whole panel, including every backup, lives under one directory. Copying that directory elsewhere is the off-machine half of the story, and it is deliberately the only thing you have to copy.

Related: backups in the panel, schedules, shrinking a world, moving a server to another host.