Instrument Node.js Apps with Azure Monitor

This skill auto-instruments Node.js applications with distributed tracing, metrics, and logs sent to Azure Monitor (Application Insights) using @azure/monitor-opentelemetry, the recommended "distro" package for zero-code auto-instrumentation, with @azure/monitor-opentelemetry-exporter available for lower-level custom OpenTelemetry setups and @azure/monitor-ingestion for custom logs ingestion. The critical setup rule is calling useAzureMonitor({ azureMonitorExporterOptions: { connectionString } }) before importing any other application modules (including the framework itself, e.g. Express), so instrumentation hooks are installed before those modules load; for ESM projects on Node.js 18.19+, this is done instead via a --import @azure/monitor-opentelemetry/loader flag. Full configuration covers offline storage (a local directory buffering telemetry when disconnected, toggleable), a samplingRatio (0-1) for high-traffic trace volume control, feature toggles (enableLiveMetrics, enableStandardMetrics, enablePerformanceCounters), per-library instrumentation toggles (Azure SDK, HTTP, MongoDB, MySQL, PostgreSQL, and Redis enabled by default; Bunyan and Winston logging disabled by default), and a custom resource via resourceFromAttributes (e.g. setting service.name). Custom traces use the OpenTelemetry API directly (trace.getTracer → startSpan, setting attributes, adding events, recording exceptions with a status code, then ending the span), and custom metrics cover a Counter (e.g. request totals with route/method labels), a Histogram (e.g. request duration), and an ObservableGauge with a polling callback (e.g. active connections). For fully manual setups bypassing the distro, AzureMonitorTraceExporter/AzureMonitorMetricExporter/AzureMonitorLogExporter wire directly into OpenTelemetry SDK providers (NodeTracerProvider with a BatchSpanProcessor, MeterProvider with a PeriodicExportingMetricReader, LoggerProvider with a BatchLogRecordProcessor). Separately, LogsIngestionClient uploads arbitrary custom log records to a Data Collection Endpoint/Rule and custom stream name, with isAggregateLogsUploadError distinguishing partial upload failures and surfacing exactly which log entries failed. A custom SpanProcessor example shows enriching spans with custom dimensions on onEnd and filtering out INTERNAL-kind spans by clearing their trace flags. Sampling can also be controlled directly via ApplicationInsightsSampler (e.g. sampling 75% of traces) passed to a NodeTracerProvider. Graceful shutdown flushes buffered telemetry via shutdownAzureMonitor(), typically wired to a SIGTERM handler.

Use this skill when adding distributed tracing, metrics, and log collection to a Node.js/TypeScript application reporting to Azure Monitor/Application Insights - whether via zero-code auto-instrumentation or fully custom OpenTelemetry pipeline wiring. It is not for applications reporting to a non-Azure observability backend (though the underlying OpenTelemetry API code is portable, the exporters here are Azure Monitor-specific), and the auto-instrumentation approach requires useAzureMonitor() to run before any instrumented library is imported - getting that ordering wrong silently loses instrumentation coverage.

Input is an Application Insights connection string, sampling/feature configuration, and application code emitting spans/metrics/logs (auto-instrumented or manual). Output is traces, metrics, and logs exported to Azure Monitor, plus live metrics and standard performance counters when enabled.

Built on @azure/monitor-opentelemetry (auto-instrumentation distro) and @azure/monitor-opentelemetry-exporter (manual exporters), layered on the OpenTelemetry JS SDK and API, with @azure/monitor-ingestion's LogsIngestionClient for custom log data reporting to a Data Collection Rule.

npm install @azure/monitor-opentelemetry

Node.js/TypeScript developers instrumenting an application for observability in Azure Monitor/Application Insights, from zero-code auto-instrumentation to fully custom trace/metric/log pipelines, including custom log streams (e.g. a Custom-MyTable_CL stream name) uploaded to a Data Collection Rule.