Resources and replicas
CPU and memory per replica, the number of replicas and temporary disk.
CPU and memory
Each replica can use up to your plan's ceiling of CPU and memory. Raseef sets it; you do not choose it per service. The ceiling is a limit, not a reservation: you pay for what the service actually uses. See pricing for each plan's ceiling.
Some templates need more than others and never run below their own minimum size.
A replica that goes past its memory ceiling is stopped and restarted. Watch memory in the service's Metrics and size your app's
runtime accordingly (for Node.js, --max-old-space-size; for Java, -Xmx).
Replicas
Under Scaling, choose how many replicas run: 1, 2 or 3, up to what your plan allows. Replicas run on different machines, and requests are spread across them.
- A service with a volume runs one replica.
- A scheduled job runs one replica.
- There is no automatic scaling of a service yet; you set the number.
Two replicas keep your app answering while one is restarted or moved to another machine. Your app must not keep state in memory or on local disk that the other replica needs.
Temporary disk
Each replica has temporary disk for files it writes outside a volume, sized by your plan. It is emptied whenever the replica restarts or a new version deploys. A replica that writes past it is restarted. Keep data that must last on a volume or in a database.