PostgreSQL in your own EU cloud account, run for you
Last updated 17 November 2026
EU clouds are cheaper and under EU law, but on most of them nobody runs the database for you the way RDS does. We install a three-node PostgreSQL cluster in your UpCloud account and keep it running: failover, backups, upgrades, alerts. The servers, the data and the code stay yours.
What we measured
We built it, then broke it on purpose. Test cluster on UpCloud Helsinki, 17 November 2026: small servers (1 CPU, 1 GB) and a near-empty database. Bigger databases restore more slowly; we measure again on your sizes.
| What we did | What happened |
|---|---|
| Powered off the primary with no warning while a client wrote a row every 0.2 s | Writes stopped for 36.4 s and 34.6 s (two runs), then continued on a new primary. The client changed nothing. |
| Counted confirmed writes after that failover | 0 of 146 and 0 of 147 lost |
| Powered the old primary on again | It rejoined as a replica by itself, no commands |
| Deleted 990 of 1,000 rows, then restored a copy to 5 seconds before the delete | All 1,000 rows back in 18 s |
| Powered off all three servers at once during writes and replaced them with empty ones | The new servers restored from the backup bucket by themselves. 711 of 717 confirmed rows back (the last 1.6 s lost). Writable again about 5 minutes after the new servers were created. |
What is inside
- PostgreSQL 18 on three servers, each on a different physical host: one primary, two replicas.
- Patroni and etcd notice a dead primary and promote a replica.
- Synchronous replication: a write is confirmed only when a second server has it.
- pgBackRest backups to UpCloud Object Storage, encrypted, with every change shipped within a minute. Restore to any second.
- A restore test every week on a separate server, with an alert if it fails. A backup nobody restored is not a backup.
- Prometheus with twelve alerts, and a written runbook with the steps for each one.
- All of it as Terraform code. One command builds the cluster in about 12 minutes.
The code is open source (Apache-2.0), with the runbook and the drill script: github.com/moveinfra/postgres-ha-upcloud. You can build it and run the drills yourself before you talk to us.
What we do each month
- Watch the alerts and act on them on working days (CET).
- Minor version and security updates, one server at a time, with one short planned switch of the primary.
- Check the weekly restore test and send you the result.
- A failover drill on a schedule you choose, with the measured numbers.
- Changes you ask for: sizes, settings, extensions, a new replica.
Who holds your data
You do. The cluster runs in your cloud account, under your contract with UpCloud. We work through access you grant and can revoke at any time. Nothing is copied to our systems. We sign a data processing agreement and an NDA before we get access. If you stop the contract, everything keeps running and the code and runbook are already yours.
What it is not, yet
- No customer runs it in production yet. The numbers above are from our own test cluster.
- One data centre. It survives a server or a physical host failing, not a whole site. Backups survive that; the running cluster does not.
- If all three servers are lost at once, up to about one minute of the newest writes can be gone.
- Traffic inside the private network is not encrypted yet. TLS between the parts is the next piece of work.
- No 24/7 on-call. Working days, CET. Around-the-clock cover is priced separately when you need it.
- We hold no SOC 2 or ISO 27001 certificate. If your auditors require one from every supplier, we are not a fit today.
Price
From €1,500 a month, monthly contract, cancel any month. The cloud bill is yours and goes straight to UpCloud: three database servers, a monitoring server and object storage cost roughly €30 a month at the smallest sizes and more at production sizes; your offer shows the number. You get a fixed offer before anything starts. All prices are on the pricing page.
Write to hello@moveinfra.euwith your database size and where it runs today. You get an answer from the engineer who built this, within two working days.