Choose between Scalability (autoscaling), Load balancing or Failover.
When CPU crosses the limit you set, a new replica starts, passes the health check and only then joins the load balancer. When the peak passes, it scales down.
Choose per project between Scalability (autoscaling), Load balancing or Failover. With failover included in every plan, your project stays online: if the server goes down, the standby copy in another country takes over on its own. Load balancing and autoscaling are optional and billed by the hour.
A clone that starts with the load
Normal load: one copy is enough
When CPU crosses the limit, the platform starts a clone of the app and the load balancer splits the load with it. When usage drops, the clone is turned off and stops being billed.
Running copies sharing the traffic
Traffic split between US and Germany
Two or more copies run at the same time on different machines. The load balancer splits every request between them; if one fails, the others keep serving while it is recreated.
Protection without burning resources
Primary serves alone; standby is off, no CPU
A standby copy stays off in another country. If the primary goes down, the standby starts on its own and takes the traffic. Every deploy updates it to the new version and it stays off.
to scale well, keep sessions in Redis and files in S3, and avoid duplicate jobs. App replicas don’t make your database highly available.
Yes. Every project, on any plan, has a stopped failover replica in another country. If the primary fails it starts on its own; when the primary is back, it stops.
Only if it stores sessions or files on local disk. Use Redis and S3.
Failover (a standby copy kept off in another country) costs nothing extra. Load balancing (an always-on copy) and autoscaling (copies that start at peak) are billed by the hour while the copies run: load balancing pays the full hourly price for every copy; with autoscaling the 1st extra copy pays full price and the next ones get 10% off. Usage is invoiced on the 1st of each month, by Pix or card, and needs a paid plan in good standing.
No. Replicas are for the app; the database has its own resources.
Servers with AMD EPYC processors. Replicas and the failover replica sit on different servers, so losing one doesn’t take the app down.