Observability — Azure
Service mapping
- Azure Monitor — the umbrella: metrics, Metric Alerts, dashboards, and autoscale signals.
- Application Insights — APM: request RED metrics, distributed tracing, dependency maps, and live metrics. Native OpenTelemetry export is supported.
- Log Analytics (Kusto/KQL) — the log + query store behind Azure Monitor; the
high-volume pipeline concern is
distributed-logging. - Azure Managed Grafana + Azure Monitor managed service for Prometheus — run the generic PromQL/Grafana stack managed, for portability.
- Health probes — App Service / AKS liveness+readiness; gating is
load-balancing.
When to pick which
- Application Insights when the app is .NET/Azure-centric — richest auto-instrumentation and dependency tracing with least effort.
- Azure Monitor managed Prometheus + Managed Grafana to keep the generic recipe and PromQL portability.
- Log Analytics/KQL when you need powerful ad-hoc log querying.
Limits / things that bite (verify against current docs)
- Application Insights samples telemetry by default (adaptive sampling) — great for cost, but verify errors/slow requests aren't being dropped.
- Log Analytics ingestion + retention is the main cost driver; data-cap and retention tiers change the bill sharply.
- Metric Alerts evaluation frequency and granularity bound how fast an SLO burn is detected.
- KQL has query limits/timeouts on large windows.
Pitfalls
- Trusting default adaptive sampling for SLO-critical signals without checking it.
- High-cardinality custom dimensions inflating Log Analytics cost.
- Lock-in to App Insights SDKs/KQL dashboards — use the OpenTelemetry exporter and managed Prometheus/Grafana to stay portable.
- Conflating Log Analytics retention with metrics retention (different knobs/costs).