Observability

The Observability page gives you live metrics and streamed logs for a chosen cluster, so you can see what your workload is doing without leaving the console.

Metrics

Pick a cluster and a time window. DC Suite charts the signals that matter for GPU work:

• GPU utilization — per-GPU utilization across the cluster.
• SLURM nodes by state — how many nodes are idle, allocated, down.
• SLURM jobs by state — pending, running, completed.

Charts poll periodically and build a rolling window, so they fill in over the first few polls after you open the page.

Logs

The Logs panel streams control-plane logs for the selected cluster. You can:

• Choose a time window (last 15 minutes, hour, or 6 hours).
• Follow the tail live, or pause to scroll back.
• Refresh on demand.

Logs are scoped to what you're allowed to see — you only get logs for clusters within your access scope.

When the backend isn't available

Observability depends on a metrics backend and a logs backend being wired up by your operators. If they aren't:

• You'll see an informational note ("observability backend not configured") rather than charts — this is expected on installations that haven't enabled the stack.
• If logs show a backend error, it's an operator-side issue (backend unreachable or misconfigured); metrics may still work independently.

API

For automation, the same data is available read-only:

• GET /v1/obs/metrics — instant metric queries.
• GET /v1/obs/logs — recent log lines for a cluster.
• GET /v1/obs/usage — usage counters (node-hours, GPU-hours) that also feed Cost & Billing.

---

Next: Cost & Billing.