sampling.md

Sampling

A cloud-scale web application generates a large number of traces. You can set up sampling strategies in your application to reduce the volume of trace data that it sends to Wavefront.

You set up a sampling strategy by configuring the WavefrontTracer builder with an implementation of the Sampler interface.

For example, suppose you want to approximately report 1 out of every 5 traces to Wavefront. The following snippet shows how to configure a WavefrontTracer with a RateSampler:

// Create a WavefrontTracer builder
WavefrontTracer.Builder wfTracerBuilder = ...  

// Create a RateSampler with a rate of 20% and add it to the WavefrontTracer builder
wfTracerBuilder.withSampler(new RateSampler(0.2));

... // (Optional) Further configure the WavefrontTracer builder

// Build the WavefrontTracer
Tracer tracer = wfTracerBuilder.build();

Supported Sampling Strategies

The following table lists the supported sampling strategies. You create and configure a sampling strategy by configuring a Sampler implementation:

Sampler	Description
ConstantSampler	Allows either all traces or no traces. Specify true to sample all traces, or false to sample no traces.
DurationSampler	Allows a span if its duration exceeds a specified threshold. Specify the duration threshold as a number of milliseconds.
RateSampler	Allows a specified probabilistic rate of traces to be reported. Specify the rate of allowed traces as a number between 0.0 and 1.0.
CompositeSampler	Delegates the sampling decision to multiple other samplers and allows a span if any delegate decides to allows it. Specify a list of samplers to delegate to.

Note: Regardless of the sampling strategy, the WavefrontTracer:

• Allows all error spans (error=true span tag).
• Allows all spans that have a sampling priority greater than 0 (sampling.priority span tag).
• Includes all spans in the RED metrics that are automatically collected and reported.

Using Multiple Sampling Strategies

You can configure a WavefrontTracer with multiple sampling strategies. To do so, you use a CompositeSampler, which delegates the sampling decision to multiple other samplers and decides to allow a span if any of the delegate samplers decide to allow it.

For instance, suppose you want to report approximately 10% of traces but you also don't want to lose any spans that are over 60 seconds long. The following code snippet shows how to configure a WavefrontTracer with a RateSampler and a DurationSampler:

// Create a WavefrontTracer builder
WavefrontTracer.Builder wfTracerBuilder = ...  

// Create and configure the RateSampler and DurationSampler
Sampler rateSampler = new RateSampler(0.1);
Sampler durationSampler = new DurationSampler(60_000);

// Create and configure the CompositeSampler with a list of samplers
Sampler compositeSampler = new CompositeSampler(Arrays.asList(rateSampler, durationSampler));

// Add the CompositeSampler to the WavefrontTracer builder
wfTracerBuilder.withSampler(compositeSampler);

... // (Optional) Further configure the WavefrontTracer builder

// Build the WavefrontTracer
Tracer tracer = wfTracerBuilder.build();