obs-sglang-request-trace-doc · inference-research

Fields

id

sources/obs-sglang-request-trace-doc

URL

https://docs.sglang.ai/references/production_request_trace.html

Publisher

sgl-project/sglang

Kind

repo

Accessed

2026-10-04

notes

The tracing guide: --enable-trace + --otlp-traces-endpoint, SGLANG_TRACE_LEVEL 0-3 adjustable at runtime via GET /set_trace_level, and SGLANG_TRACE_ASYNC=1 which moves span creation to a ZMQ exporter process because 'synchronous OTel span creation can degrade inference throughput due to thread-safe locking'.

Notes

The tracing guide: --enable-trace + --otlp-traces-endpoint, SGLANG_TRACE_LEVEL 0-3 adjustable at runtime via GET /set_trace_level, and SGLANG_TRACE_ASYNC=1 which moves span creation to a ZMQ exporter process because 'synchronous OTel span creation can degrade inference throughput due to thread-safe locking'.

Cross-references

This record references no other records.