{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在大量并行读请求、读多写少的业务场景下，本文利用Sysbench 性能测试工具，研究基于【负载均衡+ProxySQL Cluster MGR 的读写分离架构】能够有效地利用横向扩展MySQL实例的读物能力，并最终提高应用系统 QPS。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"前言"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在大量并行读请求、读多写少的业务场景下，本文利用Sysbench 性能测试工具，研究基于【负载均衡+ProxySQL Cluster MGR 的读写分离架构】能够有效地利用横向扩展MySQL实例的读物能力，并最终提高应用系统 QPS。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"MySQL Group Replication（M GR）于 2016 年 12 月被推出，了高可用、高扩展、高可靠的 MySQL 提供服务。但仅解决数据同步问题和其内部的自动故障转移。应用系统可能需要修改数据库连接地址，才能保证服务的可用性。为解决上述问题，可在MRG上层增加代理层，例如ProxySQL。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"ProxySQL 于 2015 年被推出，是一个开源、高可用性、协议探索的 MySQL 代理。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）可通过节点的只读值，自动调整它们是属于读组还是写组；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）可定制的基于用户、基于schema、基于语句的规则对SQL语句进行访问，实现读写分离；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3) 支持搭建ProxySQL Cluster来达到高可用，节点之间的配置可自动同步。"}]},{"type":"numberedlist","attrs":{"start":3,"normalizeStart":3},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"负载均衡是将流量分发至多台节点设备上的服务。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1) 可通过删除单点故障，提升应用系统的可用性；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2) 可降低大量的综合访问，提高应用系统的处理能力。"}]},{"type":"numberedlist","attrs":{"start":4,"normalizeStart":4},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"Sysbench 是一个开源的、线程探索的测试、跨平台的多性能工具。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"一、压测目的"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"基于Sysbench的oltpreadonly压测模式，对比【负载均衡+ProxySQL Cluster+MGR的读写分离】和【应用直连MySQL Master】这两种架构的只读性能："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"建立读写读写性能数据；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"验证读写架构在大量并发读请求场景下的宣讲；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"分析各模块和参数对拆解资产业绩的影响。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"二、压测结论"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2.1. 读写读写读写性能数据"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在Sysbench oltpreadonly压测模式下，【4层负载均衡+ProxySQL Cluster+MGR 读写分离】架构的QPS与并发线程数关系如下表所示。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/62\/62daa7eec675628cb178e9062dac1ae0.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.2. 只读场景下编写的文档"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先简单浏览下实验对比图和结果，"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/9d\/9db6ee77057473dbeda57f87274d46f7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/92\/9204ea3110bead006cec1080cb45df25.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/f3\/f383fcc59b9280aacf6d47d24b9c9b88.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps“\/”表示：由于Sysbench机器CPU完成，完成测试，无实验结果实验结果证明："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"在不引入负载均衡、ProxySQL Cluster中间件的理想情况下，【连MGR 2个只读实例】QPS应用能达到100w，为【应用连直MySQL Master】的只读QPS等37w的2.7 倍。该结果验证了 MGR 架构在大量并发读请求场景下的有效性。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":2,"normalizeStart":2},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"但实际上，如要保证应用系统高可用，则需要增加负载均衡、ProxySQL Cluster 等中间件，而这些中间件或多或少会带来性能损失。 MGR 读写出版】自己的只读 QPS 阅读为 89w，大约【应用直连 MySQL Master】的只读 QPS 报告 37w 的 2.4 倍，该结果验证了【4 层加载均衡+ProxySQL Cluster+MGR 读】写原稿】架构在大量并发读请求场景下的文章。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.3. 各模块和参数对剥离架构绩效的影响"}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"【4层负载均衡+ProxySQL Cluster+MGR读写分离】框架的QPS 滨海周边【直连MGR 2个只读实例】QPS周边100w的89%。其中，ProxySQL Cluster带来约11%的性能但 ProxySQL 的 CPU 占用率最高只有 57%，还需要 57% 的时间才能进一步有效利用 ProxySQL。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":2,"normalizeStart":2},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"根据 https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724，参考 CPU 核数增加 ProxySQL 的 mysql-threads 变量值，即增加 ProxySQL 用于处理 MySQL 流量的后台线程数，能有效提升 QPS（如将）线程数 4 增加至 16，QPS 从提升了 3.3 倍），但目前尚未压测出 ProxySQL 的 CPU 持续提升到 100% 的场景。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":3,"normalizeStart":3},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"横向扩展ProxySQL实例数，能有效提升QPS（实例数增加1至2，QPS提升1倍）。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":4,"normalizeStart":4},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"将 7 层负载均衡实际 4 层，由在应用层进行流量分配改成在传输层，降低网络性能，在实验中提升了 1 倍 QPS。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":5,"normalizeStart":5},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"可知https:\/\/ProxySQL.com\/blog\/benchmarking-ProxySQL-144\/，本身ProxySQL的mysql-max_stmts_per_connection变量值（20增加至100），让单独可以处理更多的准备语句，但实验中影响 QPS。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"三、压测详情"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.1. 压测环境"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/44\/440bceeff0171c0c8b37e6ab2f4aff02.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，还安装了nodeexporter、mysqlexporter、proxysql_exporter来监控OS、MySQL和ProxySQL，方便定位问题。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.2. 压测指标"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 每秒执行请求数 QPS（Queries Per Second），数据库包含执行的 SQL 数 INSERT、SELECT、UPDATE、DELE、COMMIT 等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 95% Latency (ms) 95% 请求的延迟时间完全查询请求中的 95% 在发出到接收结果的平均时间。延迟越小延迟。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.3. 实验设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为减少干净，每轮实验重复 3 次。每次任务执行完之后，等待 300 秒，让系统及时处理未完成任务，进入下管道压测。压测后除了利用 Sysbench 自带的清理才清理数据，还额外把垃圾桶清理干净，清洁磁盘空间影响一次压测。模块设置下可以看到变。"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.1 MySQL设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• MGR：单模式。共3个节点，其中1个只写节点，2个只读节点。max_connection 设为3000。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• Master-Master：主主同步，其中只有 1 个 Master 提供读写服务。max_connection 设为 3000。"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.2 ProxySQL 设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysql_user 的transaction_persistent 表字段：设置为1，表示在某节点内启动的事务将保留在该节点内，而与其他评论规则。用于避免以下问题的操作操作：事务有混合的读和写组成操作，事务未提交前，如果事务中的读操作和写操作访问到不同节点，读取到的结果是脏数据。"}]},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":" insert into mysql_users(username,password,default_hostgroup,transaction_persistent) values('MGR','MGR',1,1);\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysqls 表的 max_connections：允许连接到该实例的最大连接数，不能超过 MySQL 设置的 max_connections，因此设为 3000。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysql_group_replication_hostgroups 表：配置MGR writerhostgroup、readerhostgroup 的hostgroupidProxy。SQL 会通过视图来监控MGR 节点等是否正常，是否开启了读取结果、挤压事务数调整了单个MGR 节点所属的hostgroupid，具体对应可在runtime_mysql_servers中查看。 insert into mysql_group_replication_hostgroups(writer_hostgroup,backup_writer_hostgroup,reader_hostgroup,offline_hostgroup,active,max_writers,writer_is_also_reader,max_transactions_behind,comment) values(1,2,3,4,1,0,'-test,1r ;"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"-- 可以有写组有1个节点，读组有2个节点，均在正常工作"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/4a\/4a9895d0f8e4ca68c7cfbe853b0dfc4c.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"•查询规则：根据SQL的正则表达式匹配，读请求评论至读组，写请求评论至写组。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"——将写请求评论到写组"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"INSERT INTO mysql_query_rules (rule_id,active,username,match_digest,destination_hostgroup,apply) VALUES (200,1,'mgr','^SELECT.*FOR UPDATE ,1,1);"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"——将读请求评论到读组"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"INSERT INTO mysql_query_rules (rule_id,active,username,match_digest,destination_hostgroup,apply) VALUES (201,1,'mgr','^SELECT',3,1);"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 变量变量mysql-threads：ProxySQL 用于处理MySQL 流量的后台线程数。默认值为4，实验中发现，增加至16 值可致命实验提升QPS，因此除了变量该变量的参数调优，其他实验中该变量值16。 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"set mysql-threads=16;显示变量，如'mysql-threads';"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/51\/511c5478693724fd73216af3437d869e.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.3.3 Sysbench 设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 实验基于Sysbench的oltpreadonly只读模式。该模式下，一个事务包含14个读SQL（10条主键点查询、4条范围查询）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• oltpreadonly 模式的压测命令"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"准备数据："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 -- table-size=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on \/usr\/local\/share\/sysbench\/oltp_read_only.lua prepare"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"运行工作量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 --表大小=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on --skip_trx=on \/usr\/local\/share\/sysbench\/ oltp_read_only.lua 运行"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"清理数据："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 -- table-size=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on \/usr\/local\/share\/sysbench\/oltp_read_only.lua 清理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"普通变量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 时间：压测总持续时间（秒），后任务未完也会被终止。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 线程：一起压测的线程数。取值范围[16,32,64,128,256,512,1024,1500,2048,2500,3000,3500,4096]。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"重点变量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• skip_trx[=on|off]：默认为off，即启动显式事务；启动开启时，不显式事务，以AUTOCOMMIT模式执行所有查询。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 压测时设置：sysbench --skiptrx=on；ProxySQL 的mysqluser 表的transaction_persistent=1。原因如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• Proxy 的 mysqluser 表事务的持久字段设为 1 时，在某节点内事务将保留在该节点启动，而与其他转发规则内的。用于避免以下问题：一个事务有混合的读和写操作组成，事务未提交前，如果事务中的读操作和写操作访问到不同节点，读取到的结果是脏数据。因此，如果不开启skip_trx，sysbench 全部请求都被ProxySQL 评论到写组，这样便便测不了分离的性能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• sysbench 默认使用准备好的语句，因为本实验需要测试使用准备好的语句的情况，因此在此不作关闭该功能的参数说明。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 设置变量--mysql-host=[host1,host2,...,hostN]，对多个MySQL同时发起发起请求。可用于同时压测多个MySQL实例时的QPS。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/github.com\/akopytov\/sysbench\/issues\/19"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.4 实验场景设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"共同设计了6个实验场景（架构图查看实验结果分析），实验目的如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/37\/378e3698c44765cf40630bbf378120b7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.4. 实验结果分析"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验一：[MGR] vs [Master-Master]"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d2\/d2578af70f7b28b5caca521db0090548.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"获取应用（sysbench）直连MGR的2个只读实例数引发的QPS服务，确认该服务和应用直连mysql Master-Master中1台的QPS差异。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/c6\/c6237e70db25a5088d3f1cf41b70b61a.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/b7\/b764394403d05fc669d0a4d97a561de0.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• MGR QPS 物业约100w，约Master-Master 的2.67 倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验二：[单ProxySQL+MGR] vs [MGR]"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/04\/047b8f515b665518edd6673b28ebb17e.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在典型应用和ProxySQL网络延迟的情况下，确认增加ProxySQL中间件会带来的性能损失"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/18\/18475c7ea9a99af559288b8b0c76ed74.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/5a\/5aaf89da635257ab94bb673f8aac2dc1.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 单个ProxySQL+MGR QPS 最大48w，大约MGR 的48%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 单个ProxySQL+MGR QPS最大时，ProxySQL实例CPU占用率最高为57%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在该实验中，ProxySQL Cluster 带来了约 48% 的损失，但此时 ProxySQL 的 CPU 占用率还不算高，性能探索可能进一步有效利用 ProxySQL。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验三：【ProxySQL Cluster+MGR】对比【单个ProxySQL+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/b3\/b30f9212a9fceb71e0f4c49d819e07ce.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"确认横向扩展 ProxySQL 实例化可能进一步提升 QPS"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/9a\/9aeeff47f8b72594ab56efa91a4bcba7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/8f\/8fbb1fc629f935101ff25c69a5b923a2.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• ProxySQL+MGR QPS集群约89w，约为单个ProxySQL+MGR的1.85倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• ProxySQL Cluster+MGR QPS 最大时，ProxySQL 实例 CPU 占用率最高为 56%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"横向拓展代理SQL实例集合可以进一步提升QPS至89w，相对接近MGR的小区100w。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验4：【7层负载均衡+ProxySQL Cluster+MGR】对比【ProxySQL Cluster+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/57\/578f64ba0294dfc727f01d5851012149.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"增加读写架构中的需求均衡服务，并确认其带来的性能损失"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/fb\/fbcc34a35b3d0b654aaecddd139e8860.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/90\/90577f709595f186fd985c760e7fa0e3.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 增加负载均衡后，QPS 时间为42w，约MGR 服务的50%，1 主-主.135 倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"增加负载均衡导致性能损失近 50%，可能是因为网络、配置问题，需要进一步排查。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验5：【4层负载均衡+ProxySQL Cluster+MGR】对比【7层负载均衡+ProxySQL Cluster+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"4层负载均衡工作在OSI模型的传输层（基于IP+端口），7层工作在应用层（基于URL）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"晚上，7 层负载均衡会带来更多的网络性能。因此喂饱为 4 层负载均衡，以减少性能损失。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/7b\/7bd41c1f729f54ee091abde1b3dabd12.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/a0\/a0e15198f36e032306806b91aa24ae5a.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"将7层负载均衡立即4层负载均衡后，QPS观看为89w，几乎没有带来性能损失。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验 6: ProxySQL 参数调优：分析 mysql-threads 参数值对性能的影响"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"根据 https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724，mysql-threads 变量是 ProxySQL 用于处理 MySQL 流量的后台线程数，根据机器 CPU 核数来调整该变量，可提升 ProxySQL 性能。因此实验分析该参数对性能的影响。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"结实验果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/f3\/f395dc6080f1d3b0c71bca56d3157aab.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/07\/0740b8399bcf9f9ddfc9ccb631a9dcd4.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可知机器 CPU 核数来增加 ProxySQL 的 mysql-thread 变量值，可更进一步提升 QPS。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"四、总结"}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"【4层负载均衡+代理SQL集群+MGR读写分离】适用于大量读请求场景，只读QPS能达到89w，左右【应用直连MySQL Master】的只读QPS 37w的2.4倍。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"参考机器的 CPU 核数增加 ProxySQL 的 mysql-threads 变量值，即增加了用于处理 MySQL 流量的后台线程数的 ProxySQL，能有效提升 QPS（如将线程数从 4 增加至 16，QPS 提高了 3.3 倍）。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"横向扩展ProxySQL实例数，能有效提升QPS（实例数增加1至2，QPS提升1倍）。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"将7层负载均衡按4层，由在应用层进行流量分配改成在传输层，能降低网络性能并提高QPS。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"次次实验中，ProxySQL 集群带来约 11% 的性能损失，几乎没有带来性能损失。但是 ProxySQL 的 CPU 占用率最高只有 57%，还需要汽车探索才能进一步有效利用 ProxySQL。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"参考文献："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/dev.MySQL.com\/doc\/refman\/5.7\/en\/group-replication.html"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/ProxySQL.com\/documentation\/ProxySQL-Threads\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/ProxySQL.com\/blog\/ProxySQL-vs-maxscale-persistent-connection-response-time\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/www.percona.com\/blog\/2020\/08\/28\/ProxySQL-overhead-explained-and-measured\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":6,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/www.percona.com\/blog\/2017\/04\/10\/ProxySQL-rules-do-i-have-too-many\/"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"作者简介："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"雷宏姧，网易游戏技术部数据库系统工程师。参与海量高级数据库生产环境故障排查和优化，热衷于 MySQL 技术原理、灾难失败和高研究可用方案。"}]}]}
ProxySQL + MySQL MGR 自拆书架的Sysbench 只读压测报告
{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在大量并行读请求、读多写少的业务场景下，本文利用Sysbench 性能测试工具，研究基于【负载均衡+ProxySQL Cluster MGR 的读写分离架构】能够有效地利用横向扩展MySQL实例的读物能力，并最终提高应用系统 QPS。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"前言"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在大量并行读请求、读多写少的业务场景下，本文利用Sysbench 性能测试工具，研究基于【负载均衡+ProxySQL Cluster MGR 的读写分离架构】能够有效地利用横向扩展MySQL实例的读物能力，并最终提高应用系统 QPS。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"MySQL Group Replication（M GR）于 2016 年 12 月被推出，了高可用、高扩展、高可靠的 MySQL 提供服务。但仅解决数据同步问题和其内部的自动故障转移。应用系统可能需要修改数据库连接地址，才能保证服务的可用性。为解决上述问题，可在MRG上层增加代理层，例如ProxySQL。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"ProxySQL 于 2015 年被推出，是一个开源、高可用性、协议探索的 MySQL 代理。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）可通过节点的只读值，自动调整它们是属于读组还是写组；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）可定制的基于用户、基于schema、基于语句的规则对SQL语句进行访问，实现读写分离；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3) 支持搭建ProxySQL Cluster来达到高可用，节点之间的配置可自动同步。"}]},{"type":"numberedlist","attrs":{"start":3,"normalizeStart":3},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"负载均衡是将流量分发至多台节点设备上的服务。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1) 可通过删除单点故障，提升应用系统的可用性；"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2) 可降低大量的综合访问，提高应用系统的处理能力。"}]},{"type":"numberedlist","attrs":{"start":4,"normalizeStart":4},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"Sysbench 是一个开源的、线程探索的测试、跨平台的多性能工具。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"一、压测目的"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"基于Sysbench的oltpreadonly压测模式，对比【负载均衡+ProxySQL Cluster+MGR的读写分离】和【应用直连MySQL Master】这两种架构的只读性能："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"建立读写读写性能数据；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"验证读写架构在大量并发读请求场景下的宣讲；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"分析各模块和参数对拆解资产业绩的影响。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"二、压测结论"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2.1. 读写读写读写性能数据"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在Sysbench oltpreadonly压测模式下，【4层负载均衡+ProxySQL Cluster+MGR 读写分离】架构的QPS与并发线程数关系如下表所示。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/62\/62daa7eec675628cb178e9062dac1ae0.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.2. 只读场景下编写的文档"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先简单浏览下实验对比图和结果，"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/9d\/9db6ee77057473dbeda57f87274d46f7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/92\/9204ea3110bead006cec1080cb45df25.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/f3\/f383fcc59b9280aacf6d47d24b9c9b88.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps“\/”表示：由于Sysbench机器CPU完成，完成测试，无实验结果实验结果证明："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"在不引入负载均衡、ProxySQL Cluster中间件的理想情况下，【连MGR 2个只读实例】QPS应用能达到100w，为【应用连直MySQL Master】的只读QPS等37w的2.7 倍。该结果验证了 MGR 架构在大量并发读请求场景下的有效性。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":2,"normalizeStart":2},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"但实际上，如要保证应用系统高可用，则需要增加负载均衡、ProxySQL Cluster 等中间件，而这些中间件或多或少会带来性能损失。 MGR 读写出版】自己的只读 QPS 阅读为 89w，大约【应用直连 MySQL Master】的只读 QPS 报告 37w 的 2.4 倍，该结果验证了【4 层加载均衡+ProxySQL Cluster+MGR 读】写原稿】架构在大量并发读请求场景下的文章。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.3. 各模块和参数对剥离架构绩效的影响"}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"【4层负载均衡+ProxySQL Cluster+MGR读写分离】框架的QPS 滨海周边【直连MGR 2个只读实例】QPS周边100w的89%。其中，ProxySQL Cluster带来约11%的性能但 ProxySQL 的 CPU 占用率最高只有 57%，还需要 57% 的时间才能进一步有效利用 ProxySQL。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":2,"normalizeStart":2},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"根据 https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724，参考 CPU 核数增加 ProxySQL 的 mysql-threads 变量值，即增加 ProxySQL 用于处理 MySQL 流量的后台线程数，能有效提升 QPS（如将）线程数 4 增加至 16，QPS 从提升了 3.3 倍），但目前尚未压测出 ProxySQL 的 CPU 持续提升到 100% 的场景。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":3,"normalizeStart":3},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"横向扩展ProxySQL实例数，能有效提升QPS（实例数增加1至2，QPS提升1倍）。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":4,"normalizeStart":4},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"将 7 层负载均衡实际 4 层，由在应用层进行流量分配改成在传输层，降低网络性能，在实验中提升了 1 倍 QPS。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":5,"normalizeStart":5},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"可知https:\/\/ProxySQL.com\/blog\/benchmarking-ProxySQL-144\/，本身ProxySQL的mysql-max_stmts_per_connection变量值（20增加至100），让单独可以处理更多的准备语句，但实验中影响 QPS。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"三、压测详情"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.1. 压测环境"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/44\/440bceeff0171c0c8b37e6ab2f4aff02.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，还安装了nodeexporter、mysqlexporter、proxysql_exporter来监控OS、MySQL和ProxySQL，方便定位问题。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.2. 压测指标"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 每秒执行请求数 QPS（Queries Per Second），数据库包含执行的 SQL 数 INSERT、SELECT、UPDATE、DELE、COMMIT 等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 95% Latency (ms) 95% 请求的延迟时间完全查询请求中的 95% 在发出到接收结果的平均时间。延迟越小延迟。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.3. 实验设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为减少干净，每轮实验重复 3 次。每次任务执行完之后，等待 300 秒，让系统及时处理未完成任务，进入下管道压测。压测后除了利用 Sysbench 自带的清理才清理数据，还额外把垃圾桶清理干净，清洁磁盘空间影响一次压测。模块设置下可以看到变。"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.1 MySQL设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• MGR：单模式。共3个节点，其中1个只写节点，2个只读节点。max_connection 设为3000。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• Master-Master：主主同步，其中只有 1 个 Master 提供读写服务。max_connection 设为 3000。"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.2 ProxySQL 设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysql_user 的transaction_persistent 表字段：设置为1，表示在某节点内启动的事务将保留在该节点内，而与其他评论规则。用于避免以下问题的操作操作：事务有混合的读和写组成操作，事务未提交前，如果事务中的读操作和写操作访问到不同节点，读取到的结果是脏数据。"}]},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":" insert into mysql_users(username,password,default_hostgroup,transaction_persistent) values('MGR','MGR',1,1);\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysqls 表的 max_connections：允许连接到该实例的最大连接数，不能超过 MySQL 设置的 max_connections，因此设为 3000。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• mysql_group_replication_hostgroups 表：配置MGR writerhostgroup、readerhostgroup 的hostgroupidProxy。SQL 会通过视图来监控MGR 节点等是否正常，是否开启了读取结果、挤压事务数调整了单个MGR 节点所属的hostgroupid，具体对应可在runtime_mysql_servers中查看。 insert into mysql_group_replication_hostgroups(writer_hostgroup,backup_writer_hostgroup,reader_hostgroup,offline_hostgroup,active,max_writers,writer_is_also_reader,max_transactions_behind,comment) values(1,2,3,4,1,0,'-test,1r ;"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"-- 可以有写组有1个节点，读组有2个节点，均在正常工作"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/4a\/4a9895d0f8e4ca68c7cfbe853b0dfc4c.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"•查询规则：根据SQL的正则表达式匹配，读请求评论至读组，写请求评论至写组。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"——将写请求评论到写组"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"INSERT INTO mysql_query_rules (rule_id,active,username,match_digest,destination_hostgroup,apply) VALUES (200,1,'mgr','^SELECT.*FOR UPDATE ,1,1);"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"——将读请求评论到读组"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"INSERT INTO mysql_query_rules (rule_id,active,username,match_digest,destination_hostgroup,apply) VALUES (201,1,'mgr','^SELECT',3,1);"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 变量变量mysql-threads：ProxySQL 用于处理MySQL 流量的后台线程数。默认值为4，实验中发现，增加至16 值可致命实验提升QPS，因此除了变量该变量的参数调优，其他实验中该变量值16。 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"set mysql-threads=16;显示变量，如'mysql-threads';"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/51\/511c5478693724fd73216af3437d869e.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.3.3 Sysbench 设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 实验基于Sysbench的oltpreadonly只读模式。该模式下，一个事务包含14个读SQL（10条主键点查询、4条范围查询）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• oltpreadonly 模式的压测命令"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"准备数据："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 -- table-size=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on \/usr\/local\/share\/sysbench\/oltp_read_only.lua prepare"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"运行工作量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 --表大小=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on --skip_trx=on \/usr\/local\/share\/sysbench\/ oltp_read_only.lua 运行"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"清理数据："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"sysbench --db-driver=mysql --mysql-host=XXX --mysql-port=XXX --mysql-user=XXX --mysql-password=XXX --mysql-db=dbtest --tables=1 -- table-size=10000000 --report-interval=1 --threads=XXX --rand-type=uniform --time=120 --auto-inc=on \/usr\/local\/share\/sysbench\/oltp_read_only.lua 清理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"普通变量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 时间：压测总持续时间（秒），后任务未完也会被终止。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 线程：一起压测的线程数。取值范围[16,32,64,128,256,512,1024,1500,2048,2500,3000,3500,4096]。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"重点变量："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• skip_trx[=on|off]：默认为off，即启动显式事务；启动开启时，不显式事务，以AUTOCOMMIT模式执行所有查询。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 压测时设置：sysbench --skiptrx=on；ProxySQL 的mysqluser 表的transaction_persistent=1。原因如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• Proxy 的 mysqluser 表事务的持久字段设为 1 时，在某节点内事务将保留在该节点启动，而与其他转发规则内的。用于避免以下问题：一个事务有混合的读和写操作组成，事务未提交前，如果事务中的读操作和写操作访问到不同节点，读取到的结果是脏数据。因此，如果不开启skip_trx，sysbench 全部请求都被ProxySQL 评论到写组，这样便便测不了分离的性能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• sysbench 默认使用准备好的语句，因为本实验需要测试使用准备好的语句的情况，因此在此不作关闭该功能的参数说明。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 设置变量--mysql-host=[host1,host2,...,hostN]，对多个MySQL同时发起发起请求。可用于同时压测多个MySQL实例时的QPS。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/github.com\/akopytov\/sysbench\/issues\/19"}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"3.3.4 实验场景设置"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"共同设计了6个实验场景（架构图查看实验结果分析），实验目的如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/37\/378e3698c44765cf40630bbf378120b7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.4. 实验结果分析"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验一：[MGR] vs [Master-Master]"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d2\/d2578af70f7b28b5caca521db0090548.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"获取应用（sysbench）直连MGR的2个只读实例数引发的QPS服务，确认该服务和应用直连mysql Master-Master中1台的QPS差异。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/c6\/c6237e70db25a5088d3f1cf41b70b61a.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/b7\/b764394403d05fc669d0a4d97a561de0.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• MGR QPS 物业约100w，约Master-Master 的2.67 倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验二：[单ProxySQL+MGR] vs [MGR]"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/04\/047b8f515b665518edd6673b28ebb17e.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在典型应用和ProxySQL网络延迟的情况下，确认增加ProxySQL中间件会带来的性能损失"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/18\/18475c7ea9a99af559288b8b0c76ed74.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/5a\/5aaf89da635257ab94bb673f8aac2dc1.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 单个ProxySQL+MGR QPS 最大48w，大约MGR 的48%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 单个ProxySQL+MGR QPS最大时，ProxySQL实例CPU占用率最高为57%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在该实验中，ProxySQL Cluster 带来了约 48% 的损失，但此时 ProxySQL 的 CPU 占用率还不算高，性能探索可能进一步有效利用 ProxySQL。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验三：【ProxySQL Cluster+MGR】对比【单个ProxySQL+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/b3\/b30f9212a9fceb71e0f4c49d819e07ce.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"确认横向扩展 ProxySQL 实例化可能进一步提升 QPS"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/9a\/9aeeff47f8b72594ab56efa91a4bcba7.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/8f\/8fbb1fc629f935101ff25c69a5b923a2.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• ProxySQL+MGR QPS集群约89w，约为单个ProxySQL+MGR的1.85倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• ProxySQL Cluster+MGR QPS 最大时，ProxySQL 实例 CPU 占用率最高为 56%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"横向拓展代理SQL实例集合可以进一步提升QPS至89w，相对接近MGR的小区100w。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验4：【7层负载均衡+ProxySQL Cluster+MGR】对比【ProxySQL Cluster+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/57\/578f64ba0294dfc727f01d5851012149.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"增加读写架构中的需求均衡服务，并确认其带来的性能损失"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/fb\/fbcc34a35b3d0b654aaecddd139e8860.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/90\/90577f709595f186fd985c760e7fa0e3.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"• 增加负载均衡后，QPS 时间为42w，约MGR 服务的50%，1 主-主.135 倍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"增加负载均衡导致性能损失近 50%，可能是因为网络、配置问题，需要进一步排查。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验5：【4层负载均衡+ProxySQL Cluster+MGR】对比【7层负载均衡+ProxySQL Cluster+MGR】"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"4层负载均衡工作在OSI模型的传输层（基于IP+端口），7层工作在应用层（基于URL）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"晚上，7 层负载均衡会带来更多的网络性能。因此喂饱为 4 层负载均衡，以减少性能损失。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/7b\/7bd41c1f729f54ee091abde1b3dabd12.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/a0\/a0e15198f36e032306806b91aa24ae5a.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"将7层负载均衡立即4层负载均衡后，QPS观看为89w，几乎没有带来性能损失。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"实验 6: ProxySQL 参数调优：分析 mysql-threads 参数值对性能的影响"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验目的："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"根据 https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724，mysql-threads 变量是 ProxySQL 用于处理 MySQL 流量的后台线程数，根据机器 CPU 核数来调整该变量，可提升 ProxySQL 性能。因此实验分析该参数对性能的影响。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"结实验果："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/f3\/f395dc6080f1d3b0c71bca56d3157aab.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/07\/0740b8399bcf9f9ddfc9ccb631a9dcd4.webp","alt":"图片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ps\/”表示由于Sysbench机器CPU完成，完成测试，无实验结果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"实验结论："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可知机器 CPU 核数来增加 ProxySQL 的 mysql-thread 变量值，可更进一步提升 QPS。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"四、总结"}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"【4层负载均衡+代理SQL集群+MGR读写分离】适用于大量读请求场景，只读QPS能达到89w，左右【应用直连MySQL Master】的只读QPS 37w的2.4倍。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"参考机器的 CPU 核数增加 ProxySQL 的 mysql-threads 变量值，即增加了用于处理 MySQL 流量的后台线程数的 ProxySQL，能有效提升 QPS（如将线程数从 4 增加至 16，QPS 提高了 3.3 倍）。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"横向扩展ProxySQL实例数，能有效提升QPS（实例数增加1至2，QPS提升1倍）。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"将7层负载均衡按4层，由在应用层进行流量分配改成在传输层，能降低网络性能并提高QPS。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"次次实验中，ProxySQL 集群带来约 11% 的性能损失，几乎没有带来性能损失。但是 ProxySQL 的 CPU 占用率最高只有 57%，还需要汽车探索才能进一步有效利用 ProxySQL。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"参考文献："}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/dev.MySQL.com\/doc\/refman\/5.7\/en\/group-replication.html"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/ProxySQL.com\/documentation\/ProxySQL-Threads\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/ProxySQL.com\/blog\/ProxySQL-vs-maxscale-persistent-connection-response-time\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/www.percona.com\/blog\/2020\/08\/28\/ProxySQL-overhead-explained-and-measured\/"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/github.com\/sysown\/ProxySQL\/issues\/1724"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":6,"align":null,"origin":null},"content":[{"type":"text","text":"https:\/\/www.percona.com\/blog\/2017\/04\/10\/ProxySQL-rules-do-i-have-too-many\/"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"作者简介："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"雷宏姧，网易游戏技术部数据库系统工程师。参与海量高级数据库生产环境故障排查和优化，热衷于 MySQL 技术原理、灾难失败和高研究可用方案。"}]}]}
發表評論
所有評論
還沒有人評論，想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章