Tickless Backups

First public release of Tickless Backups: reliable world backups for Fabric servers that never block the server thread.

Features

• Asynchronous backups: copying, compression and verification run on background threads, so ticks are not blocked.
• Atomic snapshots: autosave is paused only while the snapshot is taken, then resumed immediately. Compression works on the copy. Autosave is never left off, even if a backup fails.
• Copy-on-write snapshots: uses reflink on Linux and macOS where the filesystem supports it (Btrfs, XFS with reflink, APFS), and falls back to a safe copy otherwise.
• Verified archives: every backup is a zip with a SHA-256 manifest. It is re-read and verified before it becomes final.
• Retention: keep backups by count, age and/or total size. The newest backup is never deleted.
• Restore: /backup restore <name> verifies the backup and applies it when the server stops. The previous world is kept as <world>.pre-restore-<date> and is never deleted.
• Optional save before backup, so the backup holds the current state.

Commands (owner permission level)

/backup start [name], /backup status, /backup list, /backup reload, /backup restore <name>, /backup restore cancel

Configuration

config/tickless-backups.json (created on first start, reload it with /backup reload).

Requirements

Fabric Loader 0.17+, Fabric API, Java 25.

Known limitations

• A restore is applied at a normal server stop, not after a crash.
• .pre-restore-* folders are not cleaned automatically.
• Only the zip format for now. Off-site upload and zstd are planned.