One node per machine
A cluster instance has three or five nodes. Each runs unilayer with its own PostgreSQL and Valkey; PostgreSQL only listens inside the node. The nodes are spread so that no two of them share a machine.
What runs behind your instance.
No shared database, no noisy neighbours in the same process: every instance is a unilayer cluster of its own, on Kubernetes, kept running by an operator that watches it around the clock.
Its own namespace
Every instance gets a Kubernetes namespace of its own, with a resource quota and network policies: only its own nodes and the ingress may talk to it.
TLS, inside and out
Your instance answers on its own address with a certificate. Inside the cluster, the nodes talk to each other over TLS with certificates of the instance's own CA.
Volumes that survive a machine
The database of each node lives on a replicated volume. Raft keeps a copy on every node anyway; the replica lets a node restart elsewhere without rebuilding.
Backups and files in object storage
Backups and uploaded files go to S3-compatible object storage in the same region, in a folder per instance. Backups can be downloaded from the portal.
Secrets stay secret
Root tokens are shown only to who may see them, and can be rotated in the portal. Our own keys are stored encrypted (AES-256-GCM), and our systems talk to the clusters over mutual TLS.
When something fails
A machine goes down. Your app doesn't.
- 1
The nodes of a cluster never share a machine. When one machine fails, the others still form the majority Raft needs: reads and writes go on.
- 2
If the failed node was the leader, the remaining nodes elect a new one by themselves.
- 3
The missing node starts again on another machine with a copy of its volume and catches up with the cluster.
Data protection
Your data, in the region you choose.
An instance, its backups and its files stay in the region it was booked in. What your instance stores is processed on your behalf, under a data processing agreement (Art. 28 GDPR) that is part of the contract.
When you delete an instance, a last backup is taken and kept with its files for 30 days – so a mistake can still be undone on request. After that it is deleted for good.