Node Operations

DGX Cloud Lepton provides node-level operations to help you manage nodes more effectively.

Actions

Pause Schedule

Pause scheduling on a specific node. Once paused, no new workloads are scheduled to the node.

Resume Schedule

Resume scheduling on a specific node. Once resumed, the node is available for new workloads.

Force Reboot

Force a reboot of a specific node.

Drain

Choose from two drain policies:

• Graceful Drain (recommended): Migrates current workloads to other nodes, then terminates all workloads on the node.
• Drain: Evicts all pods from the node and terminates any active workloads.

Release Protection

Protect a node from being released. When enabled, the node cannot be released through the dashboard or API.

Release

Choose from two release policies:

• Graceful Release (recommended): Migrates workloads to an idle node, then terminates all remaining workloads and releases the node.
• Release: Immediately evicts all pods, terminates any active workloads, and releases the node.

Health Check

DGX Cloud Lepton provides health checks to monitor every node across the following dimensions:

• GPU: GPU temperature, utilization, power usage, and more.
• General hardware: CPU, memory, disk, network, PCI, and more.
• System: FUSE, OS, library status, and more.
• Others: Tailscale, kubelet, etc.

Trigger Health Check

Health checks run periodically, but you can also manually trigger a health check for a specific node.

Each health check card includes a trigger button with a play icon. Hover over the icon to see a tooltip showing when the check last ran.

Press the button to trigger the health check.