Resize

The supervisor reports sustained cpu and memory pressure (kernel PSI), and pier resize fixes it without losing anything:

$ pier ls
NAME           REPO  STATE               AGE  COST
checkout-flow  shop  working (strained)  2h   ~$0.04/h

$ pier resize checkout-flow t4g.xlarge

Or press m in the TUI. It lists same-arch machines with their vCPU, memory and hourly cost, so nobody memorizes instance type names.

One park/resume cycle, about a minute on AWS and about two on GCP, disk and state intact.

Deliberately not automatic

The VM holds no cloud credentials, so it cannot resize itself. Auto-scaling is also a surprise-cost footgun. You see strain, you decide.

Limits

Resize stays within the CPU arch (t4g to t4g). Providers only allow type changes on stopped instances, which is where the park/resume cycle comes from. On AWS it needs one extra permission, ec2:ModifyInstanceAttribute.

On GCP, ARM64 sessions can resize among the T2A and C4A choices shown by the picker. Those machine families are not available in every zone; if a target is unavailable, pier restarts the session on its original machine type.