morunner

FE回归

build

安装pcap库

ubuntu : apt install libpcap-dev centos : yum install libpcap-devel

go build -o main *.go

测试OOM

每秒500个短链接。2ms 1个链接。 ./main -reqcount 500

测试查询一致性

./main --loop --url freetier-01.cn-hangzhou.cluster.aliyun-dev.matrixone.tech --user dump --password

http_proxy="" all_proxy="" curl http 127.0.0.1:8080/status

测试ISSUE15190 query_result

./main -testcase 1

测试在FE拼接结果的sql

./main -testcase 2

测试load

./main -testcase 3

分析pcap文件

./main -testcase 4 -pcap-fname path -pcap-filter tcp -regexpr "" -display-bytes-limit 20

PR checklist

• bvt

• ut

  • FE UT

• sca

• 事务

  • 乐观 bvt
  • 悲观 bvt
  • 泄露

• 回归

  • ci regression
  • morunner
    • 短链
    • FE 处理的语句回归

• proxy和session迁移

  测试tpcc场景

  //启动集群
  
  ./mo-service -debug-http 127.0.0.1:6060 -cfg etc/launch-with-proxy/log.toml >log.log
  ./mo-service -debug-http 127.0.0.1:6061 -cfg etc/launch-with-proxy/tn.toml >tn.log
  ./mo-service -debug-http 127.0.0.1:6062 -cfg etc/launch-with-proxy/cn1.toml >cn1.log
  ./mo-service -debug-http 127.0.0.1:6063 -cfg etc/launch-with-proxy/proxy.toml >proxy.log
  
  //跑tpcc
  
  create database tpcc;
  
  cd mo-tpcc
  ./runSQL.sh props.mo tableCreates
  mysql -h127.0.0.1 -udump -P6009 -p111 tpcc <load-tpcc-w1.sql
  ./runBenchmark.sh props.mo
  
  然后再启动cn2:
  
  ./mo-service -cfg etc/launch-with-proxy/cn2.toml >cn2.log
  
  迁移前后用sql命令观察：
  select node_id, conn_id, user, account from processlist() a where user='dump';
  
  有些session的node_id 会变化
  
  //用这个命令把一个cn设置成draining状态，就会迁移到另外一个cn上。uuid换成具体的cn node-id
  
  select mo_ctl('cn', 'workstate', 'uuid:2');
  
  //这个是设置会working状态，就会再次均衡连接
  
  select mo_ctl('cn', 'workstate', 'uuid:1');
  

  测试begin和autocommit = 0场景

  session 1 create account acc1 admin_name "root" identified by '111';

  session2 mysql --local-infile -h 127.0.0.1 -P 6009 -uacc1:root -p111

  begin;
  select * from mo_catalog.mo_user;
  
  {
    //autocommit = 0 的场景
    set @@autocommit = 0;
    select @@autocommit;
    select * from mo_catalog.mo_user;
  }
  

  session1 //将acc1 标记到node5上 select mo_ctl('cn','label','dd1dccb5-4d3c-41f8-b482-5251dc7a41bf:account:acc1');

  //正确结果是acc1 还在node4上。因为事务未结束 
  select node_id,conn_id,account,user from processlist() t  order by conn_id;
  

  session2 commit / rollback;

  {
    //autocommit = 0 的场景
    commit / rollback;
  }
  

  session1 //观察acc1迁移到node5。因为事务结束了。 select node_id,conn_id,account,user from processlist() t order by conn_id;

• Kill问题

• sysbench 点查

  main: 6b1a10d62ec53a54394120d8c6327c7886c1ce15

  sysbench: 内存 1.3g 19000

• tpcc

  main: 6b1a10d62ec53a54394120d8c6327c7886c1ce15

  10 仓 10 并发

  内存4.6g

  avg tpm total 7100

  current tpm total 640000

  memory 420Mb/1000MB