Enable rate-bytes-per-sec option by default · tikv/tikv#6484

(0 评论) (0 反应) (0 负责人)Rust (15,922 star) (2,189 fork)batch import

component/rocksdbcomponent/titandifficulty/mediumhelp wantedsig/engine

描述

Enabling rate-bytes-per-sec would keep rocksdb background compaction steadily runs, smoothing both IO and CPU usage. However few of our users are using the option. Since ansible runs fio test by default, we can use the result to set an initial default value for rate-bytes-per-sec, if user didn't set it.

To be more specific, we can run fio write only with threads = default max-background-jobs, and use the throughput as upper bound. Then we can maybe set rate-bytes-per-sec to 90% of the upper bound. Other suggestions are welcome.

Currently only kvdb respect rate-bytes-per-sec. We should also make raftdb respect it too.

Task breakdown:

Make raftdb respect rate-bytes-per-sec
Update ansible to set default value for rate-bytes-per-sec
benchmark to show what would be a good default rate-bytes-per-sec

贡献者指南

技术栈: rust
领域: databaseperformance
议题类型: feature
难度: 4
预计时间: 3-5 days
活动状态: stale
清晰度: mostly clear
前置要求: Rust programmingRocksDB configurationAnsible basicsfio benchmarking
新手友好度: 20
研究方向: Start by locating the existing rate bytes per sec usage in kvdb code (likely in src/server or src/raftstore). Then extend raftdb to respect this option by modifying configuration parsing and applying it. For the Ansible component, find the role that sets TiKV configuration (e.g., roles/tikv/templates/tikv.toml.j2) and add logic to compute a default from fio results. Finally, benchmark with fio to determine an appropriate default (e.g., 90% of write throughput). The issue's task list provides clear steps but no assignee or recent activity indicates it requires a fresh start.