Move a database to a new server
Move a database to another server at a time you pick, with about a minute of downtime and a way back.
Move a database to a bigger server, another provider or a new region:
- The new server becomes a standby and catches up. Nothing changes for your apps.
- At the time you pick (or Switch now), it takes over. Nothing is lost. Apps are disconnected for about a minute.
- The old server stays stopped for 7 days (or what you chose), as a way back.
A move needs both servers reporting. If the old one is gone, promote the standby instead. This page describes PostgreSQL; By database says what differs.
What you need
The same as for a standby: the agent and an empty cluster of the same major version on the new server, and restarts allowed on both.
Move
Click Standby, Move to a new server. Pick the server, when to switch, and how long to keep the old server. Confirm the new server's key fingerprint.
Standby shows progress. You can change the time or cancel. The switchover waits until the new server is in sync.
At the switchover
- The old server is fenced (stopped and kept stopped). The new one replays everything and takes over. If it can't within 3 minutes, it doesn't switch and says why; start the old one again or switch anyway.
- Backups continue on the new server.
- Rowsafe poolers and apps that list both servers (Connect your apps) follow by themselves. Standby shows the new address for others.
Switch back or remove the old server
Until the days run out:
- Switch back catches the old server up and moves the database back. Nothing is lost.
- Remove old server (type its name) lets it go. It stays stopped. Remove it on the Servers page when nothing else runs there.
After that, the old server stays stopped until you rebuild it as a standby or forget it.
By database
The new server is set up like a standby for each engine.
- MySQL and MariaDB (standby): the old server is made read-only instead of stopped.
- MongoDB (standby): the switch is MongoDB's own election. The old server stays a read-only secondary.
- ClickHouse (standby): the old server is made read-only. Switching back rebuilds it; its databases are set aside (
rowsafe_before_*). - Redis and Valkey (standby): the old server refuses writes. Backups continue from a fresh snapshot.
Only owners and admins can move a database; every step is audited. AI agents can switch, cancel or finish a move as the person who connected them, but not start one: a person must compare the new server's key.