性能测试

目标

对ShardingSphere-JDBC,ShardingSphere-Proxy及MySQL进行性能对比。从业务角度考虑,在基本应用场景(单路由,主从+加密+分库分表,全路由)下,INSERT+UPDATE+DELETE通常用作一个完整的关联操作,用于性能评估,而SELECT关注分片优化可用作性能评估的另一个操作;而主从模式下,可将INSERT+SELECT+DELETE作为一组评估性能的关联操作。 为了更好的观察效果,设计在一定数据量的基础上,使用jmeter 20并发线程持续压测半小时,进行增删改查性能测试,且每台机器部署一个MySQL实例,而对比MySQL场景为单机单实例部署。

测试场景

单路由

在1000数据量的基础上分库分表,根据id分为4个库,部署在同一台机器上,根据k分为1024个表,查询操作路由到单库单表; 作为对比,MySQL运行在1000数据量的基础上,使用INSERT+UPDATE+DELETE和单路由查询语句。

主从

基本主从场景,设置一主库一从库,部署在两台不同的机器上,在10000数据量的基础上,观察读写性能; 作为对比,MySQL运行在10000数据量的基础上,使用INSERT+SELECT+DELETE语句。

主从+加密+分库分表

在1000数据量的基础上,根据id分为4个库,部署在四台不同的机器上,根据k分为1024个表,c使用aes加密,pad使用md5加密,查询操作路由到单库单表; 作为对比,MySQL运行在1000数据量的基础上,使用INSERT+UPDATE+DELETE和单路由查询语句。

全路由

在1000数据量的基础上,分库分表,根据id分为4个库,部署在四台不同的机器上,根据k分为1个表,查询操作使用全路由。 作为对比,MySQL运行在1000数据量的基础上,使用INSERT+UPDATE+DELETE和全路由查询语句。

测试环境搭建

数据库表结构

此处表结构参考sysbench的sbtest表

CREATE TABLE `tbl` (
  `id` bigint(20) NOT NULL AUTO_INCREMENT,
  `k` int(11) NOT NULL DEFAULT 0,
  `c` char(120) NOT NULL DEFAULT '',
  `pad` char(60) NOT NULL DEFAULT '',
  PRIMARY KEY (`id`)
);

测试场景配置

ShardingSphere-JDBC使用与ShardingSphere-Proxy一致的配置,MySQL直连一个库用作性能对比,下面为四个场景的具体配置:

单路由配置

schemaName: sharding_db

dataSources:
  ds_0:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password: 
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_1:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password: 
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_2:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test 
    password: 
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_3:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
rules:
- !SHARDING
  tables:
    tbl:
      actualDataNodes: ds_${0..3}.tbl${0..1023}
      tableStrategy:
        standard:
          shardingColumn: k
          shardingAlgorithmName: tbl_table_inline
      keyGenerateStrategy:
          column: id
          keyGeneratorName: snowflake
  defaultDatabaseStrategy:
    standard:
      shardingColumn: id
      shardingAlgorithmName: default_db_inline
  defaultTableStrategy:
    none:
  shardingAlgorithms:
    tbl_table_inline:
      type: INLINE
      props:
        algorithm-expression: tbl${k % 1024}
    default_db_inline:
      type: INLINE
      props:
        algorithm-expression: ds_${id % 4}
  keyGenerators:
    snowflake:
      type: SNOWFLAKE
      props:
        worker-id: 123

主从配置

schemaName: sharding_db

dataSources:
  primary_ds:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  replica_ds_0:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
rules:
- !READWRITE_SPLITTING
  dataSources:
    pr_ds:
      writeDataSourceName: primary_ds
      readDataSourceNames:
        - replica_ds_0

主从+加密+分库分表配置

schemaName: sharding_db

dataSources:
  primary_ds_0:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  replica_ds_0:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  primary_ds_1:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  replica_ds_1:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  primary_ds_2:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  replica_ds_2:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  primary_ds_3:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  replica_ds_3:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
rules:
- !SHARDING
  tables:
    tbl:
      actualDataNodes: pr_ds_${0..3}.tbl${0..1023}
      databaseStrategy:
        standard:
          shardingColumn: id
          shardingAlgorithmName: tbl_database_inline
      tableStrategy:
        standard:
          shardingColumn: k
          shardingAlgorithmName: tbl_table_inline
      keyGenerateStrategy:
        column: id
        keyGeneratorName: snowflake
  bindingTables:
    - tbl
  defaultDataSourceName: primary_ds_1
  defaultTableStrategy:
    none:
  shardingAlgorithms:
    tbl_database_inline:
      type: INLINE
      props:
        algorithm-expression: pr_ds_${id % 4}
    tbl_table_inline:
      type: INLINE
      props:
        algorithm-expression: tbl${k % 1024}
  keyGenerators:
    snowflake:
      type: SNOWFLAKE
      props:
          worker-id: 123
- !READWRITE_SPLITTING
  dataSources:
    pr_ds_0:
      writeDataSourceName: primary_ds_0
      readDataSourceNames:
        - replica_ds_0
      loadBalancerName: round_robin
    pr_ds_1:
      writeDataSourceName: primary_ds_1
      readDataSourceNames:
        - replica_ds_1
      loadBalancerName: round_robin
    pr_ds_2:
      writeDataSourceName: primary_ds_2
      readDataSourceNames:
        - replica_ds_2
      loadBalancerName: round_robin
    pr_ds_3:
      writeDataSourceName: primary_ds_3
      readDataSourceNames:
        - replica_ds_3
      loadBalancerName: round_robin
  loadBalancers:
    round_robin:
      type: ROUND_ROBIN
- !ENCRYPT:
  encryptors:
    aes_encryptor:
      type: AES
      props:
        aes-key-value: 123456abc
    md5_encryptor:
      type: MD5
  tables:
    sbtest:
      columns:
        c:
          cipherColumn: c_cipher
          encryptorName: aes_encryptor
        pad:
          cipherColumn: pad_cipher
          encryptorName: md5_encryptor

全路由

schemaName: sharding_db

dataSources:
  ds_0:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_1:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_2:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
  ds_3:
    url: jdbc:mysql://***.***.***.***:****/ds?serverTimezone=UTC&useSSL=false
    username: test
    password:
    connectionTimeoutMilliseconds: 30000
    idleTimeoutMilliseconds: 60000
    maxLifetimeMilliseconds: 1800000
    maxPoolSize: 200
rules:
- !SHARDING
  tables:
    tbl:
      actualDataNodes: ds_${0..3}.tbl1
      tableStrategy:
        standard:
          shardingColumn: k
          shardingAlgorithmName: tbl_table_inline
      keyGenerateStrategy:
        column: id
        keyGeneratorName: snowflake
  defaultDatabaseStrategy:
    standard:
      shardingColumn: id
      shardingAlgorithmName: default_database_inline
  defaultTableStrategy:
    none:  
  shardingAlgorithms:
    default_database_inline:
      type: INLINE
      props:
        algorithm-expression: ds_${id % 4}
    tbl_table_inline:
      type: INLINE
      props:
        algorithm-expression: tbl1    
  keyGenerators:
    snowflake:
      type: SNOWFLAKE
      props:
        worker-id: 123

测试结果验证

压测语句

INSERT+UPDATE+DELETE语句:
INSERT INTO tbl(k, c, pad) VALUES(1, '###-###-###', '###-###');
UPDATE tbl SET c='####-####-####', pad='####-####' WHERE id=?;
DELETE FROM tbl WHERE id=?

全路由查询语句:
SELECT max(id) FROM tbl WHERE id%4=1

单路由查询语句:
SELECT id, k FROM tbl ignore index(`PRIMARY`) WHERE id=1 AND k=1

INSERT+SELECT+DELETE语句:
INSERT INTO tbl1(k, c, pad) VALUES(1, '###-###-###', '###-###');
SELECT count(id) FROM tbl1;
SELECT max(id) FROM tbl1 ignore index(`PRIMARY`);
DELETE FROM tbl1 WHERE id=?

压测类

参考shardingsphere-benchmark实现,注意阅读其中的注释

编译

git clone https://github.com/apache/shardingsphere-benchmark.git
cd shardingsphere-benchmark/shardingsphere-benchmark
mvn clean install

压测执行

cp target/shardingsphere-benchmark-1.0-SNAPSHOT-jar-with-dependencies.jar apache-jmeter-4.0/lib/ext
jmeter –n –t test_plan/test.jmx
test.jmx参考https://github.com/apache/shardingsphere-benchmark/tree/master/report/script/test_plan/test.jmx

压测结果处理

注意修改为上一步生成的result.jtl的位置。

sh shardingsphere-benchmark/report/script/gen_report.sh

历史压测数据展示

正在进行中,请等待。