Preview of pull request #446 at 39b4e62. Not the published site.
Skip to content

Restore from a backup

Silo is in active development before 1.0. These guides are updated frequently.

There are two ways to restore one of Silo’s backups:

  • Restore parts of a backup from Admin > Server > Backups while Silo keeps running. You choose settings pages, plugins, libraries, collections, home sections, and accounts. Silo adds and merges them, and never deletes what’s already on this server.
  • Restore the whole server from the command line, with Silo stopped. It replaces the database, so everything goes back to how it was when the backup was taken.

Both leave a way back. Before a partial restore, Silo backs up the server as it is now. That copy, Taken before a restore, is kept for 7 days, and you can restore it in full. A full restore keeps the previous database and the files it replaced until you remove them.

  1. Open Admin > Server > Backups, find the backup under Saved backups, and select Restore….
  2. Choose what to restore from Settings, Plugins, Libraries, Collections and artwork, Home and library sections, and Users. Each item is marked New, Differs, Same, or Can’t restore, with the reason. Items that already match this server are hidden until you select Show them.
  3. Select Review restore. The review lists what Silo will do, and under Left out, what it won’t restore and why.
  4. Select Restore. Silo backs up this server first, and the restore usually takes about a minute.

When it finishes, the report lists each item as Restored, Created, Already the same, Left out, or Failed, and Download report saves it. Stop restore stops after the item Silo is working on. Items already restored stay restored, and running the same restore again is safe.

Under Users, choose accounts and their profiles. Each account either merges into the matching account here or comes back as a New account with the password it had in the backup. Then choose which data to restore for the chosen profiles: watch history and progress, ratings, favorites and watchlist, personal collections, dismissed titles, playback and display preferences, and notification preferences.

Merging keeps whichever progress is newer and never deletes what’s already here. A title that isn’t in this server’s catalog is left out, and the report names it.

  • Accounts and profiles that are already here. They keep their password, role, access, and PIN; only missing profiles and the chosen data are added. Restoring an administrator account that isn’t here creates an administrator, and the review says so.
  • Access settings that are stricter here. Where the backup’s settings would widen access, the server keeps its current values.
  • Settings that belong to this server, such as its address, storage, and backup settings.
  • Libraries whose folders this server can’t open. Make the folders available first.
  • Plugins that aren’t installed here. A restore never installs a plugin from a backup. Install it from a plugin catalog, then restore it again to get its settings.
  • Settings encrypted with another SECRET_KEY, including plugin settings.

A backup that was uploaded, or taken with a different SECRET_KEY, opens with This server can’t verify this backup. Anyone who could change the file could have changed what it restores. Open it only if you trust where it came from: select Open it anyway, and before you restore, check I trust this backup; restore from it anyway.

From such a backup, Silo leaves off whatever would contact outside services or run for hours:

  • settings that connect to outside services, and plugin settings, aren’t restored
  • new libraries come back with AI metadata translation, seek previews, intro detection, chapter thumbnails, and box sets turned off
  • restored collections have no sync schedule
  • home sections come back turned off

The review lists each one. Turn them on afterwards once you’ve checked them; home sections are turned on in the page editor.

A backup’s restore-set.tar file holds the parts the web app can restore. To bring parts of another Silo server into this one, move that file across.

  1. On the other server, open Admin > Server > Backups, select Download next to a backup, and save restore-set.tar. If that backup is in a bucket, select Download the restore set (decrypted) instead.
  2. On this server, select Upload a backup at the top of Admin > Server > Backups, then Choose a file.
  3. When the file has uploaded and Silo has checked it, select Choose what to restore and continue as in Restore parts of a backup.

An upload can be up to 2 GB. Silo keeps uploads under Uploaded backups for a week after their last use. An unfinished upload is deleted after 24 hours.

An encrypted file (restore-set.tar.age) uploads as it is if it was made with this server’s SECRET_KEY, for example on a new install that kept the old key. Silo can’t tell which key a file uses until it has uploaded, so a file made with another key is refused afterwards with This file is encrypted. Decrypt it first, on a computer with that server’s SECRET_KEY set:

Terminal window
silo backup decrypt restore-set.tar.age -o restore-set.tar

If checking an upload stops, for example because the server restarted, select Check again under Uploaded backups. You don’t have to send the file again. If you’re signed out during an upload, sign in again as an administrator in the same browser and continue: the parts already sent are kept.

The Replace the whole server section of Admin > Server > Backups shows the command for the backup you select in the list, or for the newest one, with a button to copy it.

  1. Stop Silo on every server, and turn off automatic restarts until the restore finishes. With the standard Compose setup, stop Silo inside its container. PostgreSQL keeps running:

    Terminal window
    docker compose exec silo /command/s6-svc -d /run/service/svc-silo
    docker compose exec silo /command/s6-svwait -D /run/service/svc-silo
  2. Run the command from the page inside the container, for example:

    Terminal window
    docker compose exec -w /config/silo silo /command/s6-envdir /run/s6/container_environment \
    /command/s6-setuidgid silo silo backup restore --from /config/silo/artwork/backups/<server-id>/<backup>/

    For a backup in a bucket, --from is the bucket address the page shows, and the command also needs the bucket’s details in SILO_BACKUP_S3_ENDPOINT, SILO_BACKUP_S3_REGION, SILO_BACKUP_S3_ACCESS_KEY, SILO_BACKUP_S3_SECRET_KEY, and SILO_BACKUP_S3_PATH_STYLE. With Compose, pass each one with -e after exec.

  3. Start Silo again with docker compose restart silo.

The command checks the whole backup and loads it into a new database before it touches the live one. If a check fails or the load stops, the live database is left as it was. It refuses a backup from a newer version of Silo. Afterwards it keeps the previous database and every file it replaced, and prints how to remove them once the restored server works, or how to go back. To remove the previous database with the standard Compose setup, run the DROP DATABASE line it printed like this:

Terminal window
docker compose exec -u silo silo psql -h /run/postgresql -d postgres -c 'DROP DATABASE "silo_before_restore_<time>"'

The command also clears Silo’s Redis data, so nothing cached from the old database is served. --keep-redis skips that.

If a restore stops partway, for example because the host restarted, run silo backup restore --recover <time> with the time the command printed. Run it the same way as the restore: as the same user, from the same folder, against the same database. --recover takes no other options, so with Compose, replace --from and the backup’s path with it:

Terminal window
docker compose exec -w /config/silo silo /command/s6-envdir /run/s6/container_environment \
/command/s6-setuidgid silo silo backup restore --recover <time>

It finishes the restore or puts everything back. It doesn’t clear Redis: if it finished the restore, run FLUSHDB on the Redis database Silo uses before you start Silo, as the command says.

  • Every app and browser has to sign in again. Sign-ins aren’t part of a backup, so a sign-in you ended after the backup stays ended.
  • Connections to outside services, such as the email server, metadata and subtitle providers, sign-in providers, and notification channels, keep this server’s current addresses and credentials. A service this server has no settings for comes back turned off until you enter its credentials.
  • API keys and pending invitations come from this server, not the backup, so a key you revoked since the backup stays revoked. Keys and invitations of an account the backup doesn’t have, or that had more authority in the backup, are dropped, and the command lists them.
  • Accounts, sign-in providers, plugins, and libraries that are turned off on this server stay off.
  • Things deleted since the backup come back turned off: libraries, plugins, autoscan and history import sources, request integrations, subtitle providers, marker sharing, notifications, and access group permissions. Turn them on again once you’ve checked them. Accounts are the exception: they come back as the backup has them.
  • Scans, imports, storage moves, and other jobs that were running when the backup was taken don’t start again. They show as interrupted by the restore. Start them again if you still want them.

Resume plugin provisioning after a restore

Section titled “Resume plugin provisioning after a restore”

If the restore prints a plugin provisioning hold ID, Silo can start and the local Owner can still sign in. The hold pauses deployment changes that could reinstall a withheld plugin, turn external sign-in back on, or let a provider create accounts again.

  1. Review the plugin declaration on the destination. Check the plugin versions, the sign-in bindings, and who can get an account or become an administrator. Edit any setting you want to keep restricted. Stop the provisioning instance. With Compose, stop the Silo service inside the container so you can run the recovery command there.
  2. From a local shell with access to the restored DATABASE_URL, run silo plugins restore-hold to read the current ID. With the standard Compose setup, use the same container and command prefix as for the restore, with silo plugins restore-hold at the end.
  3. Run silo plugins acknowledge-restore <hold-id> with that ID. If another restore has replaced the hold, Silo rejects the old ID. Review the new restore before acknowledging it.
  4. Start the designated provisioning instance with the reviewed declaration. Check its status in Admin > Integrations > Plugins.

If the backup came from another server or key

Section titled “If the backup came from another server or key”

An encrypted backup, which is every backup from a bucket, needs the SECRET_KEY it was taken with, even after you download its files.

A backup from another server, including the old server when you restore onto a new install, needs --accept-lost-credentials, even with the same SECRET_KEY. So does an unencrypted backup taken with another SECRET_KEY. Settings encrypted with a key this server doesn’t have, such as storage, plugin, or email server credentials, are cleared, and the command lists them so you can enter them again. The restored server has no API keys, invitations, or links to external sign-in accounts. Set those up again.

These steps are for a server that keeps artwork on its own disk, the standard setup. If the old server uses S3 storage, the new install must use the same buckets, or the restore stops with the backup records another storage location than this deployment uses.

  1. On the new machine, install Silo with the same SECRET_KEY and the same media paths as the old server, and start it. Don’t finish setup in the browser.
  2. Copy a backup’s folder from the old server, or the files from its Download, into config/silo on the new server, owned by the PUID account.
  3. Follow Restore the whole server, with --from set to that folder inside the container, such as /config/silo/<backup>/, and add --accept-lost-credentials. The new install counts as another server.