Managed PostgreSQL
PostgreSQL 15 to 18, as one machine or a three-machine cluster that fails over.
At a glance
- Engine
- PostgreSQL 15, 16, 17 and 18; 17 by default
- Topology
- Single node, or a 3-node cluster with up to 5 readers
- Replication
- Synchronous by default, or asynchronous
- Backups
- Daily plus continuous WAL, kept up to 35 days
- Extensions
- 72, including pgvector, PostGIS and TimescaleDB
What it does
- High availability across zones
- A cluster runs Patroni and etcd on three machines spread over Falkenstein and Nuremberg. If the leader fails, a replica takes over and the endpoint follows it. Replication is synchronous by default and can be switched live. Add up to five readers behind a separate reader endpoint.
- Backups and point-in-time restore
- Daily backups plus archived WAL, encrypted before upload and kept for up to 35 days. Restore to any point in that window, as a single node or as a cluster.
- Minor and major upgrades
- Minor releases are applied in the weekly window you choose, a cluster one node at a time with a switchover. Major upgrades run when you start them, after a full backup, and can skip versions.
- Branches
- Create a copy-on-write branch of a database in seconds, from now or from a point in time, to test a migration on real data. Branches expire after seven days unless you choose otherwise.
- Encrypted, under your key
- Every new database has LUKS2-encrypted disks, optionally under your own KMS key, and its backups are encrypted with a key of its own. A database created before 12 November 2026 moves onto encrypted disks by restoring it into a new one.
- Parameters, sized to the machine
- Memory settings follow the machine size and move with it when you resize. Override 24 tunable parameters, and the console tells you which need a restart.
- TLS, personal logins, pooling
- TLS is required by default, with certificates from our own CA so
sslmode=verify-fullworks. Give each person their own login with an expiry date. PgBouncer pools connections. - Logical replication
- Turn on logical replication and declare
pgoutputslots that survive a failover, to stream changes into another system. - Logs, metrics and query statistics
- Read the server log, see connections, disk and replication lag, and find slow queries in the query statistics, all without opening a shell.
Command line
A three-node cluster with 35 days of backups, then a reader and a branch.
$ dcloud rds instances create --identifier orders --instance-class performance-4 \
--topology cluster --backup-retention-days 35
$ dcloud rds instances add-reader <db>
$ dcloud rds branches create <db> --name try-migration
$ dcloud rds logins create <db> --name alice Price
As of 19 November 2026
A database is billed as the machines it runs on: a three-node cluster on performance-4 costs three performance-4 machines. Storage and backups are priced per GB-month.
| Type | vCPU | Memory | Per hour | Month, at most |
|---|---|---|---|---|
standard-2 | 2 | 4 GiB | €0.0096 | €5.99 |
performance-1x1 | 1 | 1 GiB | €0.0104 | €6.49 |
dedicated-2 | 2 | 8 GiB | €0.0697 | €43.49 |
dedicated-4 | 4 | 16 GiB | €0.1386 | €86.49 |
| What | Price | Unit |
|---|---|---|
| Managed database storage † | €0.0572 | per GB-month |
| Database backups † | €0.0143 | per GB-month |
† Charged after notice: metered today and shown in the console with what it would cost. It appears on invoices from a date we announce in advance, never retroactively.
Measured
Measured 12 November 2026, not guaranteed
Single node, a 21 GB database, transactions per second as pgbench reports them. Reads grow with memory; writes are bounded by the data volume (about 5 ms per commit).
| Machine | tpcb | select | shop-read | shop-write |
|---|---|---|---|---|
performance-4 | 1,141 | 2,810 | 820 | 875 |
performance-8 | 1,206 | 5,365 | 1,665 | 838 |
dedicated-4 | 1,574 | 5,003 | 2,674 | 1,086 |
dedicated-8 | 1,880 | 17,036 | 24,425 | 1,338 |