Skip to main content

Observability

The runtime emits both traces and metrics via System.Diagnostics. Hook those signals into your monitoring stack to see orchestration lifecycles, activity timings, and management calls.

Distributed tracing​

OpenTelemetry support is baked in through the Asynkron.DurableFunctions activity source.

Program.cs
using OpenTelemetry.Trace;

builder.Services.AddOpenTelemetry()
    .WithTracing(tracing => tracing
        .AddSource("Asynkron.DurableFunctions")
        .AddAspNetCoreInstrumentation()
        .AddHttpClientInstrumentation()
        .AddConsoleExporter()); // swap for Jaeger, OTLP, etc.

var app = builder.Build();
app.UseDurableTraceContext(); // capture incoming traceparent headers

Key spans:

  • orchestration.start / orchestration.complete – emitted around each orchestrator execution.
  • orchestration.call – wraps every activity or sub-orchestrator call.
  • orchestration.event.receive – recorded when an external event is delivered.
  • state.save, state.load, state.remove – track persistence operations.
  • Management.* – emitted by the HTTP management controllers.

Add a sampler if you need to reduce trace volume:

tracing.SetSampler(new TraceIdRatioBasedSampler(0.1)); // capture 10% of traces

Metrics​

The runtime uses a Meter named Asynkron.DurableFunctions. Attach your exporter of choice.

MetricsExample.cs
services.AddOpenTelemetry()
    .WithMetrics(metrics => metrics
        .AddMeter("Asynkron.DurableFunctions")
        .AddPrometheusExporter());

Notable instruments:

MetricTypeWhat it tells you
orchestrations.started / completed / failedCounterVolume and health of orchestrations per function name and host.
orchestrations.durationHistogramExecution time distribution, great for SLOs.
functions.calls / functions.failuresCounterActivity throughput and failure rate.
state.save.durationHistogramPersistence latency. Spikes usually indicate storage contention.
lease.active_countUpDownCounterNumber of orchestration leases currently held; helps spot runaway hosts.

Expose the metrics endpoint (for example, /metrics via the Prometheus exporter) so your monitoring system can scrape it.