The Node Details dashboard is your single-pane view for everything happening on a specific compute Node in your cluster, from kernel-level events and GPU thermals to Slurm queue metrics and Kubernetes Pod allocations. Use it to troubleshoot anomalies, validate hardware, and verify that each Node meets workload expectations.
The following example shows what you see when you first open the dashboard, so you know what to expect before exploring the other panel groups. When the dashboard first loads, the Summary panel group expands by default. This example shows the Summary section.
The dashboard has several expandable panel groups. Each focuses on a different aspect of Node telemetry. When you view the dashboard, click these group headings to collapse or expand that section. Use the following reference to decide which groups to expand for the kind of troubleshooting or validation you’re doing.
Section
What you’ll find
Summary
High-level “at-a-glance” cards for host identity, CPU and GPU utilization, network traffic, active alerts, and HPC Verification status.
Logs
A rolling log view that merges Kubernetes events, kernel messages, Node Problem Detector findings, and recent alert annotations. Ideal for pinpointing the root cause of spikes or failures.
Network
End-to-end connectivity metrics, including ICMP loss from the Ping Exporter, conntrack table usage, interface throughput, packet error rates, and more.
Resources
Capacity and utilization for CPU, GPU, memory, NFS mounts, and local disk I/O, plus historical usage charts to identify resource pressure.
GPUs
Everything GPU-related, including ECC errors, power draw, clock speeds, memory usage, thermal headroom, fan RPM, and NVSwitch and NVLink health.
Temperatures
Real-time thermal data for GPU cores, HBM memory, motherboard sensors, and per-Pod thermal impact. Helpful for identifying cooling hot spots.
InfiniBand
Port state, link speed, retransmit counts, and congestion indicators for Nodes equipped with InfiniBand adapters.
Slurm Info
Node state within your Slurm cluster (idle, alloc, drain), running jobs, and allocation timelines. Useful for mixed Kubernetes and Slurm environments.
Kubernetes
Pod allocation breakdowns, taints, Reservation status, CPU-hours per job, memory by namespace, and Calico network policy statistics.
Verification
Results from the latest HPC Verification suite, including GPU compute benchmarks, stress tests, and smoke tests.
Hardware
Low-level chassis data, including IPMI power readings, fan speeds, serial numbers, PCIe link status, and NVLink bandwidth graphs.
Unsorted
Additional metrics that do not yet belong to a dedicated group. Check here for new or experimental panels.
The Node Details dashboard lets you correlate sections quickly. For example, expand multiple groups at a time to observe how a GPU thermal event in Temperatures aligns with throttling in GPUs and related alerts in Logs. Grafana’s time-range controls also enable time-shift analysis, so you can rewind and inspect historical Node behavior during an incident window.
