Consistent sampling

There are two major components included here.

Original proposal implementation

The original specification for consistent probability sampling is defined by https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md and https://github.com/open-telemetry/opentelemetry-specification/pull/2047. It supports sampling probabilities that are power of 2 (1, 1/2, 1/4, ...), and uses 8-bit r-value and 8-bit p-value in tracestate.

The implementation of this proposal is contained by the package io/opentelemetry/contrib/sampler/consistent in this repository and provides various Sampler implementations.

• ConsistentSampler: abstract base class of all consistent sampler implementations below
• ConsistentAlwaysOffSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md#always-off-sampler
• ConsistentAlwaysOnSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md#always-on-consistent-probability-sampler
• ConsistentComposedAndSampler: allows combining two consistent samplers and samples when both samplers would sample
• ConsistentComposedOrSampler: allows combining two consistent sampler and samples when at least one of both samplers would sample, see https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md#requirement-combine-multiple-consistent-probability-samplers-using-the-minimum-p-value
• ConsistentParentBasedSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md#parentconsistentprobabilitybased-sampler
• ConsistentProbabilityBasedSampler: see https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/trace/4673-experimental-probability-sampling.md#consistentprobabilitybased-sampler
• ConsistentRateLimitingSampler: a rate limiting sampler based on exponential smoothing that dynamically adjusts the sampling probability based on the estimated rate of spans occurring to satisfy a given rate of sampled spans

Current proposal implementation

The current version of the specification for consistent probability sampling is described by https://github.com/open-telemetry/oteps/blob/main/text/trace/0235-sampling-threshold-in-trace-state.md. It uses 56 bits for representing rejection threshold, which corresponds to a much wider range of sampling probabilities than the original proposal.

The implementation of the current proposal is contained by the package io/opentelemetry/contrib/sampler/consistent56 in this repository and provides implementation for a number of different Samplers.

• ConsistentSampler abstract base class for all consistent sampler implementations
• ComposableSampler: interface used to build hierarchies of Samplers, see Composite Samplers
• ConsistentAlwaysOffSampler:
• ConsistentAlwaysOnSampler:
• ConsistentAnyOfSampler: allows combining several consistent samplers; it samples when at least one of them would sample,
• ConsistentParentBasedSampler:
• ConsistentFixedThresholdSampler: consistent probability sampler that uses a predefined sampling probability
• ConsistentRateLimitingSampler: a rate limiting sampler based on exponential smoothing that dynamically adjusts the sampling probability based on the estimated rate of spans occurring to satisfy a given rate of sampled spans
• ConsistentRuleBasedSampler a sampler that performs stratified sampling by evaluating qualifying conditions and propagating the sampling decision from one of its delegate samplers

Component owners

• Otmar Ertl, Dynatrace
• Peter Findeisen, Cisco

Learn more about component owners in component_owners.yml.