Skip to main content
Environment classes define computing resources (instance type, disk size) for development environments.

ā€‹
Default classes

ā€‹
Create custom classes

  1. Go to Organization settings ā†’ Environment Classes (under AWS runners)
  2. Click Add Environment Class
  3. Configure:
    • Name and Description
    • Disk size: Storage capacity
    • Instance type: Any x64 EBS-enabled type (e.g., t3.medium, m5.large, c5.xlarge)

ā€‹
Modify classes

Only name and description can be edited after creation. To change disk size or instance type, create a new class and optionally disable the old one.

ā€‹
Disable classes

When disabled:
  • Existing environments continue running
  • No new environments can use this class

ā€‹
Use classes

Select environment classes when creating environments manually or via project configuration.

ā€‹
Supported instance types

Requirements:
  • EBS volume support
  • X64 architecture (Intel or AMD)
ARM-based instance types (e.g., Graviton) are not currently supported.
Examples by category:
  • General purpose: t3.medium, t3.large, m5.large, m5.xlarge
  • Compute optimized: c5.large, c5.xlarge
  • Memory optimized: r5.large, r5.xlarge
  • Storage optimized: i3.large, d2.xlarge

ā€‹
Considerations

  • Regional availability: Not all types available in every region
  • Quotas: Ensure sufficient AWS account quota
  • Costs: Pricing varies by instance family
  • Instance generations: Newer generations offer better price-performance
  • Storage: Choose instances with more storage than RAM
See AWS EC2 Instance Types for details.

ā€‹
Spot instance reclamation

When using AWS Spot instances, Ona automatically handles reclamation:
  1. The environment on the reclaimed instance is terminated
  2. A new instance is provisioned automatically
  3. The EBS volume is reattached to the new instance
  4. All data on the EBS volume is preserved

ā€‹
GPU / CUDA workloads

To run NVIDIA CUDA workloads, select the Data Science environment class (g5.4xlarge) when creating an environment or configuring a project. See the gpu-demo repository for a working devcontainer.json and Dockerfile with CUDA support via containers.dev features.

ā€‹
Best practices

  • Standardize classes across your organization
  • Limit to classes teams actually need
  • Create specialized classes for specific workloads (data processing, frontend, etc.)
  • Review usage patterns and adjust regularly