Skip to content

Ingress and Networking Metrics

Ingress and networking metrics help you decide whether traffic failures come from the frontend path, DNS latency, or outbound exhaustion.

Topic Groups

Traffic-path and name-resolution metrics

Metric Source What it means Common use Denominator / cardinality notes
Load balancer active connections Azure Monitor platform metric Concurrent frontend or backend connection load on the Azure Load Balancer path used by AKS services. Spot sudden traffic spikes, connection leaks, or unexpected load concentration on ingress paths. Trending is more useful than absolute thresholds. Split by frontend/backend rule only when a specific service is under investigation.
SNAT port usage Container Insights and Azure-side diagnostics Outbound port consumption for egress flows. High utilization can cause intermittent outbound failures even when pods look healthy. Validate CNI IP or egress exhaustion hypotheses for image pulls, external API calls, or DNS recursion. The real denominator is available ports per outbound path, not node count. Review pool growth and egress design together.
dns_request_duration_seconds Managed Prometheus CoreDNS request latency. Increased resolution time can slow service discovery and external name lookups. Distinguish DNS slowness from application slowness or ingress-controller defects. Filter by server or zone only when necessary. Label-heavy DNS charts become noisy quickly.

Interpretation priorities

  • If ingress appears broken but pod health is normal, check traffic-path and DNS metrics before changing application code.
  • SNAT pressure often presents as random outbound timeouts, not as a clean cluster-wide outage.
  • DNS latency is most useful as a comparative signal: look for sudden deviation from the cluster baseline.

Usage Notes

  • Azure Monitor platform metrics are usually the fastest way to confirm load balancer-side pressure.
  • Use Container Insights and diagnostics when you need node-level evidence for outbound exhaustion or related event history.
  • Keep Prometheus DNS charts focused on a narrow label set; CoreDNS metrics can become hard to read when over-split.

See Also

Sources