This page explains how to move workloads between machine types in Google Kubernetes Engine (GKE) clusters.
GKE clusters run workloads on worker nodes which are Compute Engine virtual machines (VMs). You can choose which machine types to run your workloads on depending on your cluster mode:
- Autopilot: Use compute classes to choose a specific compute configuration based on your workload requirements. You can choose a specific machine series. To learn more, see Compute classes in Autopilot.
- Standard: Specify a machine type when you add a node pool to a Standard cluster.
Over time, you might want to change the machines that run your workloads to optimize for evolving workload requirements or to benefit from newly released machine series. Newer generation machine series can have advantages such as improved price-to-performance and improved performance consistency. For more details about machine series and choosing between different options, see the Machine families resource and comparison guide.
Before you decide to migrate to a new machine type, ensure that the machine type supports your specific requirements, including regional availability.
To move workloads between machine types, take the following steps:
- Autopilot: Update your workload configuration.
- Standard: Choose one of the following options:
- Change the node machine attributes of your existing node pool. GKE uses the node pool's configured node upgrade strategy to migrate your workloads from the previous machine type to the new machine type.
- Create a new node pool and then migrate workloads between node pools.
After you've migrated between machine types, ensure that your workloads perform as expected on the newer machine type. You can perform the same steps if you need to move back to the original configuration.