kserve / kserve/website

docs: add standalone distributed tracing guide for LLMInferenceService

Open Beginner friendly
#739 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
MDX
Stars
113
Forks
193
Avg merge
5d 20h
Merged PRs (30d)
3

Description

## Summary

KServe v0.20 added a first-class distributed tracing API to LLMInferenceService. While `spec.tracing` is partially documented within the canary rollout observability page, there is no standalone tracing guide.

## What needs to be documented

- Standalone guide for enabling distributed tracing on LLMInferenceService
- Per-component tracing: `spec.tracing` (inference server) vs `spec.router.scheduler.tracing` (scheduler)
- "Omit vs empty" pattern: omit = no tracing, `tracing: {}` = defaults, override individual fields
- Available fields: `serviceName`, `otelExporterEndpoint`, `tracesSamplerArg`
- Automatic Kubernetes resource attributes injection (`OTEL_RESOURCE_ATTRIBUTES`)
- Disaggregated (P/D) behavior: automatic `-decode`/`-prefill` suffix on service names
- Example YAML for minimal, custom, and disaggregated tracing configurations

## Related PR

- https://github.com/kserve/kserve/pull/5481

## Since

KServe v0.20

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with the existing canary rollout observability page and related PR #5481 in the KServe website. Add a standalone guide covering per-component tracing, omit-versus-empty behavior, available fields, resource attributes, disaggregated service-name suffixes, and minimal, custom, and disaggregated YAML examples; done means each listed configuration is documented.

Written by the indexing model from the issue text.

Assessment

Tech stack
kubernetes
Domain
documentation, observability
Issue type
Documentation
Difficulty
2/5
Estimated time
1-2 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
72/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.