Observability

Mediant emits OpenTelemetry-compatible traces and metrics using the System.Diagnostics primitives built into .NET. There is no dependency on the OpenTelemetry SDK — you wire it up only if you want to collect them.

Wire it up

builder.Services.AddOpenTelemetry()
    .WithTracing(t => t.AddSource(MediatorDiagnostics.ActivitySourceName))   // "Mediant"
    .WithMetrics(m => m.AddMeter(MediatorDiagnostics.MeterName));            // "Mediant"

Both constants resolve to the string "Mediant". Use the constants rather than the literal — they are the contract.

Spans

Span	When
mediator.send <RequestType>	Every Send and CreateStream
mediator.publish <NotificationType>	Every Publish

Each span carries the request or notification type as a tag, and an Ok or Error status. Failures record error.type.

Because behaviors run inside the span, a slow request shows you which behavior was slow — the retry backoff, the cache miss, the transaction commit.

Metrics

Instrument	Kind
mediant.send.count	Counter
mediant.send.duration	Histogram
mediant.publish.count	Counter
mediant.publish.duration	Histogram

Cost when nothing listens

This is the part that matters. With no listener registered — the default — the hot path performs a cheap ActivitySource.HasListeners() / Instrument.Enabled check and skips all activity creation and measurement. The benchmark numbers are measured with instrumentation compiled in and nobody listening; that is the cost you pay for having it available.

Structured logging

Separately from OpenTelemetry, the logging behavior (order 200, always on when you use Mediant.Behaviors) writes a structured log line per request through ILogger. It masks properties marked [SensitiveData], plus anything you name explicitly, and it is safe against circular references in the request graph.

builder.Services.AddMediantAllBehaviors(opts =>
{
    opts.ConfigureLogging = log =>
    {
        log.MaskProperties.Add("CardNumber");
        log.MaxSerializedLength = 4096;   // truncate large payloads
    };
});

For a durable, queryable trail rather than log lines, use [Auditable] with an IAuditStore — see Pipeline behaviors and EF Core.

Performance thresholds

[PerformanceThreshold] logs a warning past WarningMs and an error past CriticalMs, with the elapsed time. It’s the cheapest way to notice a handler that has quietly gotten slower, without standing up a tracing backend first.

[PerformanceThreshold(WarningMs = 500, CriticalMs = 5000)]
public record GenerateMonthlyReport(int Year, int Month) : ICommand<Result<Uri>>;