Concepts

Navarch manages GPU fleets through a few core abstractions. This section explains how they work together.

• Components

  ---

  Control plane and node agent architecture.

  Components

• Pools & Providers

  ---

  Organizing nodes by workload and cloud provider.

  Pools

• Health Monitoring

  ---

  Health checks, status types, and failure detection.

  Health

• Node Lifecycle

  ---

  Instance provisioning, node states, and transitions.

  Lifecycle

• Autoscaling

  ---

  Scaling strategies, limits, and cooldown behavior.

  Autoscaling