Skip to content
Turm docs

Persistent disks

Each VM has its own disk. It persists across stop, start and restart, and across host reboots. Only rm deletes it.

  • The disk lives on local NVMe storage in the EU.
  • cp <vm> <new> copies the whole disk into a new VM. Use it to snapshot a working setup before trying something risky.
  • Your plan includes an amount of disk; see Billing and plans.
Terminal window
ssh turm.dev cp demo demo-backup

If the host fails

Section titled “If the host fails”

On a multi-host setup each disk is copied to a second host in the background. If the VM’s host stops responding for about 30 seconds, the VM is restarted on a host holding that copy, with a new private IP. Its name, URL and SSH address stay the same.

The copy is asynchronous, so a restart on another host loses the most recent writes: by default up to about 5 minutes of them, including writes the guest had already flushed with fsync. The disk you get back is consistent, as if the VM had crashed at that earlier moment. Keep anything you cannot afford to lose somewhere else as well (a git remote, object storage, a database with its own backups).

If no other host has a copy yet (a VM created moments before the failure), the VM shows status lost until its host comes back.