Tail sampling with the Collector

Oriel ingests every span it receives; it has no in-ingest tail sampling. When you need to keep error and slow traces while dropping the bulk of healthy ones, run the OpenTelemetry Collector’s tail_sampling processor in front of Oriel and export the sampled output over OTLP.

Tail sampling buffers the spans of a trace until the trace is complete, then applies its policies to the whole trace. That requires all spans of a trace to reach the same Collector instance: run a single sampling Collector, or shard with a loadbalancing exporter keyed on trace ID ahead of a sampling tier.

Requirements

• An OpenTelemetry Collector that receives the full trace before export
• An Oriel ingest token that permits traces
• Enough Collector memory to hold traces during decision_wait

Collector configuration

receivers:
  otlp:
    protocols:
      grpc:
      http:

processors:
  tail_sampling:
    decision_wait: 10s
    policies:
      [
        { name: errors, type: status_code, status_code: { status_codes: [ERROR] } },
        { name: slow, type: latency, latency: { threshold_ms: 500 } },
        { name: baseline, type: probabilistic, probabilistic: { sampling_percentage: 5 } },
      ]

exporters:
  otlp/oriel:
    endpoint: oriel.example.com:4317
    headers:
      Authorization: Bearer ${ORIEL_INGEST_TOKEN}

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [tail_sampling]
      exporters: [otlp/oriel]

The decision_wait is how long the Collector holds a trace’s spans before deciding; size it above your longest expected trace duration so late spans are not evaluated against an already-flushed decision. Only sampled traces are forwarded, so Oriel stores the kept subset.

Verify in Oriel

Open Explore and query recent traces:

traces | sort start_time desc | limit 20

Then confirm that expected error traces still arrive:

traces | where error_count > 0 | limit 20