What you can do
Add or remove nodes
Ask for more nodes, or take one out. The plan shows the machines, the services that move and the estimated cost after the change. Removing a node can move or stop the workloads on it, so check the list. How: Ask Monk; you approve the plan in the local dashboard. See Scale and resize.Add GPU nodes
Ask for a node with a specific accelerator. Monk can list valid GPU types for a provider, region and instance type first. How: Ask Monk; you approve the plan. See Scale and resize.Change machine size
Nodes aren’t resized in place. Add nodes of the new size, move work onto them, then remove the old ones, each as a reviewed change. How: Ask Monk; you approve each plan. See Scale and resize.Scale a service by CPU, memory or disk
A scaling rule in the service’s template lets the orchestrator add or remove copies by CPU, memory or disk use, between a minimum and a maximum you choose. Copies spread across the cluster’s nodes, so a high maximum needs enough nodes to hold them. How: In your Monk config, where your coding agent writes the rule and deploys after a plan review. See Scale and resize.Give a service more copies or resources
Run three copies of a worker, or give the API more memory or CPU, as settings in the template. How: In your Monk config. See Scale and resize.Grow a volume
Change the volume size in the template and deploy. Growing a cloud volume works on AWS and GCP. How: In your Monk config. See Scale and resize.Upgrade the Monk runtime on a cluster
Nodes are upgraded one at a time. How: Ask Monk; you approve in the local dashboard. See Scale and resize.Try it
Good to know
- No automatic node scaling: nothing adds or removes machines based on load.
- A service’s maximum copies is limited by the nodes you have.
- You can estimate cost before you grow. See Costs.
Scale and resize
The common requests, step by step
Costs
Estimates and usage
Clusters and clouds
Nodes, tags and GPUs
Watcher and recovery
What recovers by itself and what doesn’t