Skip to main content
Reduce overhead by sampling a subset of traces.

​
Head-Based Sampling

Sample at trace start: 
const lumina = initLumina({
  endpoint: 'http://localhost:9411/v1/traces',
  service_name: 'high-volume-service',
  samplingRate: 0.1, // Sample 10%
});

​
Conditional Sampling

Always sample important traces: 
const lumina = initLumina({
  endpoint: 'http://localhost:9411/v1/traces',
  service_name: 'service',
  samplingRate: 0.1,
  alwaysSampleConditions: {
    error: true,              // Always sample errors
    costUsd: { $gt: 0.5 },   // Always sample expensive calls
  },
});

​
When to Use Sampling

  • > 100K traces/day — Consider 10-20% sampling
  • > 1M traces/day — Use 1-5% sampling
  • Production — Balance cost vs visibility

​
Impact

10% sampling:
  • 90% reduction in storage costs
  • 90% reduction in ingestion load
  • May miss rare errors
Conditional sampling:
  • Sample routine operations at low rate
  • Capture all errors at 100%
  • Best of both worlds