Nodes
Primary, standby and read-only nodes.
A database runs on one or more nodes, each on its own server. You can run a single node, just a primary, or grow it into a highly available cluster by adding standbys and read-only nodes at any time.
Primary
The primary serves reads and writes. Every database has exactly one.
Standby nodes
A standby keeps a full copy of the data and takes over if the primary fails.
- A database has at most 2 standbys.
- A standby's server must share the primary's private network and have about 1 GB of memory.
- With a primary and 2 standbys, failover is automatic. With a primary and 1 standby, use Recover when the primary goes down. Valkey and Redis also count read-only nodes toward the 3 voting nodes Sentinel needs.
Clients can connect to standbys directly to spread read traffic. See Connecting.
Read-only nodes
A read-only node has its own name and host, serves reads from its own copy on the same port, and is never promoted to primary.
- Read-only nodes do not count toward the standby limit. Each engine caps them at its own replica limit.
- When a healthy standby exists, a new read-only node copies its data from it to spare the primary (Valkey and Redis always replicate from the primary).
- Its name is used for its hostname when the database has a domain.
Add a node
On the database's Overview tab, choose to add a standby or a read-only node, pick a free server on the same private network, and confirm. Read-only nodes also need a name.
Remove a node
Open the node and choose Remove on its settings. The node leaves the database and its copy of the data is erased. Turn on Force to remove it even if its server cannot be reached.
Reboot servers
Actions → Reboot servers reboots the database's servers one at a time: read-only nodes and standbys first, each must serve again before the next. Then the primary moves to a standby and its server reboots last, so clients may reconnect once.