5.7Sampling Strategies

In high-traffic systems, collecting every trace is expensive. Choose a sampling strategy:

Strategy	Description	Pros	Cons
Head-based (probability)	Decide at trace start whether to sample	Simple, predictable cost	May miss rare errors
Tail-based	Decide after trace completes, based on outcome	Keeps all errors and slow traces	Higher memory usage in collector
Rate-limited	Sample N traces per second	Predictable cost	Misses bursts
Always-on for errors	Sample 100% of error traces	Never misses failures	Does not reduce volume of error traces

Recommendation: Use tail-based sampling with always-on for errors and slow traces. This gives you 100% visibility into problems while keeping costs manageable for successful fast requests.

Distributed tracing is the backbone of observability in microservices. Combined with structured logging and metrics, it provides the complete picture needed for effective production testing.