thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-15 20:24:07 +00:00
Code Issues Packages Projects Releases Wiki Activity
ff191b0b08
ClickHouse/docs/en/operations/optimizing-performance/profile-guided-optimization.md
Alexander Zaitsev d001e25787
doc: update profile-guided-optimization.md 
- update PGO recommendations according to the https://github.com/llvm/llvm-project/issues/45668
2023-11-30 06:41:59 +03:00

2.0 KiB
Raw Blame History

sidebar_position sidebar_label
54 Profile Guided Optimization (PGO)

import SelfManaged from '@site/docs/en/_snippets/_self_managed_only_no_roadmap.md';

Profile Guided Optimization

Profile-Guided Optimization (PGO) is a compiler optimization technique where a program is optimized based on the runtime profile.

According to the tests, PGO helps with achieving better performance for ClickHouse. According to the tests, we see improvements up to 15% in QPS on the ClickBench test suite. The more detailed results are available here. The performance benefits depend on your typical workload - you can get better or worse results.

More information about PGO in ClickHouse you can read in the corresponding GitHub issue.

How to build ClickHouse with PGO?

There are two major kinds of PGO: Instrumentation and Sampling (also known as AutoFDO). In this guide is described the Instrumentation PGO with ClickHouse.

  1. Build ClickHouse in Instrumented mode. In Clang it can be done via passing -fprofile-generate option to CXXFLAGS.
  2. Run instrumented ClickHouse on a sample workload. Here you need to use your usual workload. One of the approaches could be using ClickBench as a sample workload. ClickHouse in the instrumentation mode could work slowly so be ready for that and do not run instrumented ClickHouse in performance-critical environments.
  3. Recompile ClickHouse once again with -fprofile-use compiler flags and profiles that are collected from the previous step.

A more detailed guide on how to apply PGO is in the Clang documentation.

If you are going to collect a sample workload directly from a production environment, we recommend trying to use Sampling PGO.