Batch vs Simple Span Processor

A span processor is a component in the OpenTelemetry SDK tracing pipeline that intercepts spans when they start and end, and controls when and how they are exported.

There are two options when it comes to choosing which span processor to use:

• A Simple Span Processor: exports each span immediately when it ends (good for debugging/development but higher overhead)

• A Batch Span Processor: buffers spans and sends them in batches (better for production/performance)

Choosing the right processor is important because your LLM/agent workflows can generate large volumes of spans: using batch processing helps keep tracing efficient and avoids impacting your application’s performance.

	BatchSpanProcessor	SimpleSpanProcessor
Best For	Production & staging environments	Local debugging, demos, CI tests
Export Behavior	Exports spans asynchronously in batches	Exports each span immediately (sync)
Impact on Latency	Low latency – work done off the request path	Higher latency – export blocks the request
Throughput	High throughput, optimized for volume	Low throughput, can bottleneck under load
Resource Usage	Uses memory for queue & background thread	Minimal memory overhead
Reliability on Exit	Requires force_flush() / shutdown() to avoid span loss	Spans exported immediately, low chance of loss
Visibility Speed	Slight delay (buffering) before spans visible	Immediate – spans appear as soon as they end
Failure Surfacing	Export failures logged in background	Failures raised inline (easy to notice)
Tuning Options	Highly configurable (batch size, delay, timeouts)	Minimal configuration

To learn more about how to configure: OTEL Guide to batch processing.