Clusters and credits
Compute services run on clusters that you provision. A cluster is defined by three things:
- its name
- its node size — how big each node in the cluster is
- its replica range — how many nodes it may run
Every Compute service — EDR, Tiles, WMS, DAP2, openEO — runs on a cluster, and all of them draw on the same pool of credits.
Zax SQL is the exception. A Zax SQL service runs on a single node rather than a cluster you size, but it draws on the same credits.
Node sizes
Sizes step up by a factor of two at each rung.
| Node size | Credits per hour |
|---|---|
| XS | 1 |
| S | 2 |
| M | 4 |
| L | 8 |
| XL | 16 |
| 2XL | 32 |
| 3XL | 64 |
A cluster consumes credits for each running node. A cluster of two M nodes costs 8 credits per hour, not 4.
Scaling
A cluster scales out as traffic rises and back in as it falls, within the replica range you set. You are only charged for the nodes that are running.
A cluster runs up to 24 nodes. Talk to us if your workload needs more.
On Community, the minimum is always zero, so a cluster scales all the way down after 15 minutes with no traffic and comes back on the next request — usually in a few seconds, and up to about two minutes if we have no spare capacity waiting.
On Professional, setting a minimum above zero keeps that many nodes running at all times. The cluster never goes cold and never pays a start-up delay, at the cost of burning credits while idle.
Choosing a size
Start smaller than you think you need and watch it.
- Tile and map workloads are bursty and mostly I/O. They usually want more nodes rather than bigger ones.
- Batch analytics are the opposite. A single larger node beats several small ones when the work does not parallelize cleanly.
When you run out
Your data stays where it is and stays readable. Your Compute services stop serving until your credits reset or you upgrade. Nothing is deleted.