Correlates the service with the resource health of the pods/hosts running it, so you can tell an application problem apart from a capacity/infrastructure problem.
Get here: Service Overview → Infrastructure tab. It has two sub‑tabs: Compute and Network.
Compute
Metric tiles — a row of trend charts, by default CPU, Memory, Network I/O, and Filesystem usage. Each tile has a dropdown to switch it to any other available pod metric, and can be expanded to full‑screen. Your tile selections are remembered with the view.
Pod table — one row per pod running the service:
- Status — health status bar.
- Name — pod name. Click a row to open pod detail (its own metric tiles) or the resource.
- Response Time — the service’s latency attributed to that pod.
- Throughput — requests per minute on that pod.
- Error Rate — error % on that pod.
- CPU Usage — pod CPU utilization (
k8s_pod_cpu_utilization). - Memory Usage — pod memory utilization (
k8s_pod_memory_utilization).
Sort by CPU/Memory to find a saturated pod, or by Error Rate/Response Time to find a bad one, then click it for the pod’s detailed metrics.
Network
Network interface graphs for the workloads (throughput, RTT, retransmissions).
Data source: infrastructure/network metrics (from the resources the trace spans map to — CPU/memory come from k8s_pod_* Prometheus series).
How to use it
- When latency/errors rise, open Infrastructure → Compute for the same time window.
- Sort the pod table by CPU/Memory (or scan the tiles) for saturation, and watch for pod restarts/OOM; use Network for RTT/retransmission spikes.
- Click the pod for its detailed metrics.