Scaling & storage
Resize compute and storage, change disk type and IOPS — and why storage can grow but never shrink.
You can adjust a cluster's placement, storage, and disk characteristics after provisioning.
Machine tier
Change the compute tier (R5–R200) to scale CPU and memory. A tier resize recreates the database pods, so it is not zero-downtime. Measured on Relay Cloud:
- Single-node cluster: the lone primary restarts, and writes (and reads) are unavailable for about 70 seconds.
- High-availability cluster: the nodes restart one at a time. Expect write gaps while the primary changes: in testing, one of roughly 10 seconds and one of 30–40 seconds during the rolling restart. Reads served by replicas continue between restarts.
Applications should retry on connection errors, and a resize is best scheduled outside peak traffic. (Tier is changed from the provisioning/edit flow, not the in-place storage editor below.)
Storage
Storage can be grown at any time and is applied in place. Storage cannot shrink — the minimum you can set is the current size.
Storage only grows. Plan capacity accordingly; there is no way to reduce a volume's size once provisioned.
Storage ranges from the tier's base size (5 GiB on R5) up to the disk type's maximum (AWS gp3 and io1 up to 16 TiB, io2 up to 64 TiB).
Disk type, IOPS & throughput
The available disk types depend on the cloud:
Relay Cloud runs on AWS, where the disk types are gp3 (the default), io1
and io2.
- IOPS is tunable within the disk type's range and depends on the volume
size (gp3: 3,000–80,000, with 3,000 included at any size and up to 500 per
GiB above that).
io1/io2require an explicit IOPS value. - Throughput is configurable on AWS
gp3(125–2,000 MiB/s).
Growing the volume applies in place. Changes to IOPS, throughput, or disk type are accepted and saved, but are not yet applied to the running volume. Plan around the values the cluster was provisioned with.