Guide · Backups

Website, Database and Full-Disk Backups: Choosing the Right Workflow

A website, a database and a whole PC disk fail in different ways, so they are not protected in the same way. This guide shows which backup fits which job, where to send the copy, and how to prove it works before you need it.

In this guide: The short answer · A worked scenario · Websites · Databases · Full-disk backups · Where the copy goes · Prove the restore · Questions

The short answer

Match the backup to what you would have to bring back. For a site, back up its files together with its database on a schedule. For a database instance, back up the instance and keep several runs. For a cloud PC, take a backup of the whole disk and keep snapshots for quick undo. Whatever you protect, send one copy off the machine and restore it once on purpose.

Which Nodekeeper backup fits which job
You need to bring back…UsePlan
A site’s files, or a site and its data togetherA scheduled backup job for websites, or websites plus databasesFree and up
A MySQL, PostgreSQL or MongoDB instanceA scheduled backup job for databases, with retentionFree and up
A virtual machineA scheduled backup job for VMsPro and up
An entire cloud PC, system disk includedA full-disk backup of the cloud PCBusiness and up
The state of a cloud PC a few minutes or hours agoA snapshot, then roll backBusiness and up

A worked scenario

Imagine a small web agency that hosts several client sites and a MySQL instance on its servers, and whose office PCs are network-booted with their system disks on central storage. Three things can go wrong on a normal week:

  • A client deletes a folder, or a plugin update breaks a site. You need yesterday’s files and database for that one site.
  • A server dies. You need every site and the database instance back on new hardware, from a copy that did not live on the dead machine.
  • A PC’s Windows install is damaged by an update. You need that PC’s disk back as it was, without reinstalling and reconfiguring.

One backup type does not answer all three. The sections below pair each failure with the workflow that fits.

Websites: schedule it per site, keep several runs

A website backup job picks a target type (websites, databases, VMs, or websites plus databases), a local save directory, a schedule and a retention count. Choosing websites plus databases keeps a site’s files and its data from the same run, so a restore does not mix Monday’s files with Thursday’s database.

  • Schedule. The schedule is a six-field cron expression (seconds, minutes, hours, day, month, weekday). Run it when the site is quiet, for example daily at 03:00.
  • Retention. Set how many runs to keep. When the count is exceeded, the oldest run and its files are removed, so a job cannot fill the disk unnoticed.
  • Boundaries. The job writes inside the site owner’s own file area and runs as the site’s own system user, not as root, so a job only touches its own owner’s files.

Databases: back up the instance, and do not confuse it with replication

Nodekeeper backs up MySQL, PostgreSQL and MongoDB instances that it manages, with the same schedule, retention and destination options as sites. If a site depends on a database, use the combined websites-plus-databases target so both come from one run.

Replication is a different tool. A database replica keeps a second server ready if the first one fails (Business and up), but a bad query or an accidental delete is copied to the replica as well. A backup is what lets you go back to before the mistake.

Cloud PCs: a full-disk backup, plus snapshots for quick undo

Office PCs that boot from the cluster keep their system disk on central storage. That changes what a backup should be: you want the whole disk, not a folder of documents. Nodekeeper supports full-disk backup and recovery of a cloud PC: the unit you protect is the whole system disk, not a folder of documents. It is part of Business.

Snapshots and backups answer different questions:

Snapshot

  • A quick recovery point of one disk, inside the cluster.
  • Best for “undo this update” or “undo what I just changed”.
  • Crash-consistent: it captures the disk as it would look after a sudden power-off.
  • Rolling back asks for two-factor verification, needs the disk out of use, and takes a “before rollback” snapshot first, so the rollback can be undone.
  • Scheduled snapshots are off until you turn them on.

Full-disk backup

  • A copy of the whole disk, for when the machine itself has to be brought back.
  • Best for a damaged install, where the whole system disk has to come back.
  • A snapshot is not a backup: it lives with the disk it protects, so it cannot cover the loss of that storage.

Where the copy goes

Website and database jobs write to a local directory first. After a successful run the job can upload to a remote, with each run in its own folder.

Local disk

Fastest to restore from, but on the same machine. Treat it as the first copy, not the only one.

S3-compatible and OSS buckets

Object storage you own, added as a remote in the panel. This is the usual off-machine copy.

Other remotes

rclone supports many providers. Treat one as supported only after you have run a test restore from it.

Limits and prerequisites
  • A remote copy is not ransomware protection on its own: whatever can write to the bucket can usually delete from it. Use versioning or object lock on the provider side if you need that.
  • The storage, traffic and request fees of the destination are yours. Nodekeeper does not include cloud storage.
  • On a hosted member account, backups use the backup capacity the administrator assigned to that account.

A rule that works for most teams: one copy on the machine for fast restores, one in a bucket you own for when the machine is gone.

Prove the restore, not just the job

A green “completed” on a job means files were written. It does not mean they are the right files. Before you trust a schedule:

  1. Pick one real run of the job in the panel and restore it.
  2. Open the restored site, and query a table you know the contents of.
  3. Do it again after you change the schedule, the destination or the database layout.
  4. Write down who is allowed to restore and how long a restore of your biggest site takes, so nobody has to find out during an incident.

Every run is recorded with its result in the operation log, which gives you a history to check against when something looks wrong.

Questions

Are backups included in the Free plan?

Yes. Scheduled website and database backups, local and remote destinations, retention and restore are part of every plan, Free included. VM backups start with Pro, and full-disk backups of cloud PCs with Business.

Do I still need backups if I use replicas or erasure coding on distributed storage?

Yes. Replicas and erasure coding protect against a failed disk or host. Deleted or corrupted data is replicated too, which is what a backup is for.

How many runs should I keep?

Enough to reach back past the time it takes you to notice a problem. A site that is edited daily and checked weekly needs more than three runs. Size it from the space each run uses.

Can a client restore their own site?

Permissions are granted by user group, so you decide who may create jobs and who may restore. Members work only in their own file area.

Set it up in Nodekeeper

Backup jobs, destinations and restore are in the same panel as your sites and databases, so you do not need a separate tool for each.

Related: backup documentation · full-disk backup and recovery for office PCs · distributed storage management · web hosting

N Nodekeeper

Manage websites, office PCs and servers from one panel on your own hardware.

Support
support@host.mw WhatsApp: +817093066006

© 2026 Nodekeeper. All rights reserved.