Replicas

SMLL databases support up to 3 replicas for high availability and read scaling. Replicas are streaming replicas that stay in sync with the primary via PostgreSQL's built-in replication.

How replicas work

• Replicas are read-only copies of your primary database
• Changes on the primary are streamed to replicas in near real-time (typically < 1 second lag)
• If the primary fails, a replica is automatically promoted
• Each replica runs on separate infrastructure for fault isolation

Configuring replicas

During creation

When creating a database, set the Replicas count to 2 or 3.

After creation

1. Navigate to your database's Settings tab
2. Under Replicas, use the slider or input to set the replica count (1–3)
3. Click Save — the new replica provisions in the background

Scaling replicas up or down takes effect within a few minutes. Existing connections are not interrupted.

Connection strings

Read-write (primary)

All writes and reads go to the primary:

postgresql://app:<password>@db-<name>-rw.vpc-<id>.svc:5432/app

Read-only (replicas)

Distribute read queries across replicas using the read-only endpoint:

postgresql://app:<password>@db-<name>-ro.vpc-<id>.svc:5432/app

The -ro endpoint load-balances across all healthy replicas. Use this for analytics queries, reporting, and read-heavy workloads.

External access

With external access enabled, both endpoints are available externally:

• Read-write: <shortID>.eu-central-1.database.smllapp.com:5432
• Read-only: <shortID>-ro.eu-central-1.database.smllapp.com:5432

External connections are routed through PgBouncer. The read-only endpoint routes to replica pods via a separate PgBouncer configuration.

Automatic failover

If the primary instance fails:

1. A healthy replica is automatically promoted to primary
2. Remaining replicas reconfigure to follow the new primary
3. Connection endpoints are updated automatically
4. The failed instance is replaced in the background

This process typically completes in under 30 seconds.

Replication lag

Replication lag is shown on the database Overview tab for each replica. A default alert fires if lag exceeds 30 seconds for 2 minutes.

Pricing

Replicas use the same compute and storage pricing as the primary:

• Compute: £5/vCPU + £2.50/GB RAM per month, per replica
• Storage: Each replica maintains its own copy of the data