Skip to main content
By default, the GCP runner dynamically creates and destroys Compute Engine VMs on demand. This means every new environment starts from scratch, which adds startup latency. Enabling warm pools keeps a set of pre-created VMs in a suspended (stopped) state using GCP Managed Instance Groups. When an environment is requested, a suspended VM resumes instead of being created from scratch, significantly reducing startup time. Even when scaled down to zero active environments, suspended instances remain in a stopped state. You only pay for their persistent disks, not for compute.

​
Enable warm pools

Set mig_warm_pool_enabled in your Terraform configuration: 
# terraform.tfvars
mig_warm_pool_enabled = true
Then apply the change: 
terraform plan -out=tfplan
terraform apply tfplan

​
Configuration

For details on configuring warm pool sizes and prebuild behavior, see Warm pools.