How it works
An instance group starts from a template: flavor, image, SSH keys, boot data, security groups and firewall. You set a minimum and maximum number of VMs (up to 20), and simple rules, for example "CPU above 70% for 3 minutes: add a VM", or "below 20% for 10 minutes: remove one".
Rules based on CPU and network
Rules apply to the average CPU utilization of the VMs in the group, or to inbound and outbound network traffic. The platform evaluates them every 15 seconds, then observes a cooldown period between two changes (3 minutes by default) to avoid oscillations. If no measurement is available, nothing changes: this is the safest behavior.
With your load balancer
When linked to a load balancer, the group registers its new VMs with it as soon as they boot; the load balancer only sends them traffic once their health check has passed. A VM being removed first leaves the load balancer, then is deleted after 30 seconds, giving in-flight requests time to complete.
Replacement and distribution
A VM that is out of service for longer than the replacement delay is replaced automatically. New VMs are placed preferably on the server hosting the fewest members of the group: losing a server only means losing part of the group, which rebuilds itself elsewhere.
VMs without replicas
Group members are identical VMs with no data to protect: they run without replicas, which saves space and network bandwidth. Keep your data in a replicated database or in S3 object storage.
Management
Groups are managed from the portal, the API or the AI assistant. You can also set the number of VMs manually, for example ahead of an expected peak; the rules then take over again.