Grafana Dashboard

A ready-made Grafana dashboard is included with 9 rows covering all metric categories.

Auto-Provisioning via Helm

The dashboard auto-provisions via the Grafana sidecar when deployed with Helm:

grafana:
  dashboard:
    enabled: true   # default
    label: "1"      # matches Grafana sidecar default

The Helm chart creates a ConfigMap with the dashboard JSON, labeled for automatic pickup by the Grafana sidecar.

Manual Import

For non-Helm setups, import grafana/kpod-metrics-dashboard.json directly via the Grafana UI:

1. Open Grafana
2. Go to Dashboards > Import
3. Upload grafana/kpod-metrics-dashboard.json
4. Select your Prometheus data source

Dashboard Rows

The dashboard includes 9 rows:

1. Overview — Pod count, collection cycle duration, collector health
2. CPU — Run queue latency (p50/p90/p99), context switches
3. Network — TCP bytes, connections, retransmits, RTT
4. Memory — OOM kills, page faults, cgroup usage
5. Syscalls — Per-syscall count, errors, latency
6. Disk I/O — Block I/O latency, read/write throughput
7. Filesystem — Capacity, usage, available space
8. Interrupts — Hardware/software IRQ latency
9. Topology — Service dependency Node Graph (requires Infinity plugin)

Recording Rules

When Prometheus Operator is enabled, 17 recording rules are provisioned for precomputed p50/p90/p99 aggregations. The dashboard uses these recording rules for efficient rendering.