Borealis Docs
Monitor sandboxes and capacity
Combine the resource's state with execution results and fresh host telemetry. Any single signal can be incomplete.
Inspect the workload
Start with the sandbox detail view: status, assigned resources, placement, and pending operations. Then look at the specific execution or interactive session that is doing the work.
borealis sandbox get SANDBOX_ID --json
borealis sandbox usage SANDBOX_ID --json
borealis sandbox logs SANDBOX_ID
Sandbox logs describe the container workload; a tracked execution has its own output, status, and exit code. Check the appropriate stream when a command appears to be missing.
Check capacity and freshness
Host management views expose health, heartbeat freshness, and resource information to authorised users. Operator analytics can include CPU, memory, disk, slots, and historical trends.
Check when a measurement was recorded. Old telemetry from a disconnected host is not evidence that the host is currently healthy or has available capacity.
Dedicated or client-managed host views depend on your permissions. Do not expect platform-wide operator information in an ordinary sandbox account.
Follow changes in integrations
The public API provides a sandbox event stream at /api/v1/sandbox-events. Use its cursor-based reconnect contract and reconcile with current resource state when a client reconnects or needs to rebuild its view.
An event is a signal to update your view, not proof that an unrelated execution completed. Keep operation-specific polling and execution results where they are needed.
Collect useful diagnostics
Record the sandbox and operation IDs, time and timezone, selected organisation, visible status, and the last successful action. Include a small redacted log excerpt when reporting a problem.
Never include access tokens, registry credentials, enrollment codes, or full workspace archives in an initial report. Follow the troubleshooting guide to narrow the failure first.
Runtime v2 state and private connectivity
For logical VMs, inspect availability, execution residency and the latest committed durable generation separately. A wake request must complete before the machine is ready; a suspended machine does not necessarily have a current recovery checkpoint.
For private-network failures, inspect the attachment’s membership, connectivity and reason, then its mutation operation. A service DNS record may resolve while the target is inactive. Check the target VM and service readiness before changing DNS or reattaching the network.