By default, Atla Insights instruments and logs all traces. In high-throughput scenarios, you may want to sample traces rather than logging every execution to manage trace limits while maintaining observability.

​
Built-in Sampling

Use pre-built sampling methods for common scenarios: 
from atla_insights import configure
from atla_insights.sampling import TraceRatioSamplingOptions

# Log 20% of traces
sampling_options = TraceRatioSamplingOptions(ratio=0.20)

configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    sampling=sampling_options,
)

​
Custom Sampling

Implement your own sampling logic using OpenTelemetry samplers: 
from atla_insights import configure
from opentelemetry.sdk.trace.sampling import Sampler, SamplingResult, Decision
from opentelemetry.trace import Link
from opentelemetry.util.types import Attributes
from typing import Optional, Sequence

class CustomSampler(Sampler):
    def __init__(self, base_ratio: float = 0.2):
        self.base_ratio = base_ratio
    
    def should_sample(
        self,
        parent_context,
        trace_id: int,
        name: str,
        kind,
        attributes: Optional[Attributes] = None,
        links: Optional[Sequence[Link]] = None,
        trace_state=None,
    ) -> SamplingResult:
        # Sample all error traces
        if attributes and attributes.get("error", False):
            return SamplingResult(Decision.RECORD_AND_SAMPLE)
        
        # Sample all user-facing interactions
        if attributes and attributes.get("user_facing", False):
            return SamplingResult(Decision.RECORD_AND_SAMPLE)
        
        # For other traces, use base sampling ratio
        if (trace_id % 100) < (self.base_ratio * 100):
            return SamplingResult(Decision.RECORD_AND_SAMPLE)
        
        return SamplingResult(Decision.NOT_RECORD)

my_sampler = CustomSampler(base_ratio=0.05)

configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    sampling=my_sampler,
)
The Atla Insights platform is not intended to work well with partial traces. We highly recommend using either ParentBased or StaticSampler samplers to ensure consistent trace handling.

​
ParentBased Sampling

Ensures all spans in the same trace are treated consistently: 
from atla_insights import configure
from opentelemetry.sdk.trace.sampling import ParentBased, TraceIdRatioBasedSampler

# All spans in a trace will have the same sampling decision
sampler = ParentBased(root=TraceIdRatioBasedSampler(0.1))

configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    sampling=sampler,
)