【mysql】使用tpcc-mysql进行压力测试

Tpcc-mysql是percona基于tpcc衍生出来专用于mysql基准测试的产品 ，可以参见 《高性能MySQL第三版》

一、安装

rpm -Uvh http://dl.fedoraproject.org/pub/epel/5/i386/epel-release-5-4.noarch.rpm
yum install bzr
bzr branch lp:~percona-dev/perconatools/tpcc-mysql

查看 README

[root@localhost tpcc-mysql]# cat README
1. Build binaries
   * cd scr ; make
   ( you should have mysql_config available in $PATH)

2. Load data
   * create database
     mysqladmin create tpcc1000
   * create tables
     mysql tpcc1000 < create_table.sql
   * create indexes and FK ( this step can be done after loading data)
     mysql tpcc1000 < add_fkey_idx.sql
   * populate data
     - simple step
       tpcc_load 127.0.0.1:33000 tpcc1000 root "" 1000
                 |hostname:port| |dbname| |user| |password| |WAREHOUSES|
       ref. tpcc_load --help for all options
     - load data in parallel
       check load.sh script

3. start benchmark
   * ./tpcc_start -h127.0.0.1 -P33000 -dtpcc1000 -uroot -w1000 -c32 -r10 -l10800
                  |hostname| |port| |dbname| |user| |WAREHOUSES| |CONNECTIONS| |WARMUP TIME| |BENCHMARK TIME|
   * ref. tpcc_start --help for all options

二、准备工作

网上的一个脚本：tpcc_load_parallel.sh

这里 WAREHOUSE=10

#!/bin/bash                                                                                                                           

# Configration

MYSQL=mysql
TPCCLOAD=./tpcc_load
TABLESQL=./create_table.sql
CONSTRAINTSQL=./add_fkey_idx.sql
DEGREE=`getconf _NPROCESSORS_ONLN`

SERVER=192.168.1.104
DATABASE=tpcc
USER=root
PASS=123456
WAREHOUSE=10

# Load

set -e
$MYSQL -h$SERVER -u$USER -p$PASS -e "DROP DATABASE IF EXISTS $DATABASE"
$MYSQL -h$SERVER -u$USER -p$PASS -e "CREATE DATABASE $DATABASE"
$MYSQL -h$SERVER -u$USER -p$PASS $DATABASE < $TABLESQL
$MYSQL -h$SERVER -u$USER -p$PASS $DATABASE < $CONSTRAINTSQL
/usr/local/src/tpcc-mysql/tpcc_load_parallel.sh [FORMAT=unix] [TYPE=SH] [POS=1,1][1%] 09/01/16 - 23:26                                

        done

        if [ $STATUS -ne 0 ]; then
            exit $STATUS
        fi

        PIDLIST=()
    fi
done

for PID in ${PIDLIST[@]}; do
    wait $PID

    if [ $? -ne 0 ]; then
        STATUS=1
    fi
done

if [ $STATUS -eq 0 ]; then
    echo 'Completed.'
fi

exit $STATUS

三、开始测试

./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c16 -r10 -l1200 > /tmp/512m-tpcc-data.log
./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c16 -r10 -l1200 > /tmp/1g-tpcc-data.log
./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c16 -r10 -l1200 > /tmp/2g-tpcc-data.log

./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c8 -r10 -l1200 > /tmp/8c-tpcc-data.log
./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c16 -r10 -l1200 > /tmp/16c-tpcc-data.log
./tpcc_start -h192.168.1.104 -dtpcc -uroot -p123456 -w10 -c64 -r10 -l1200 > /tmp/64c-tpcc-data.log

-d 测试库
-u 用户名
-p 密码
-w 使用多少个数据仓库
-c 测试的并发数
-r 预热几秒
-l 运行测试多长时间

测试结果解读

***************************************
*** ###easy### TPC-C Load Generator ***
***************************************
option h with value '192.168.1.104'
option d with value 'tpcc'
option u with value 'root'
option p with value '123456'
option w with value '20'//仓库
option c with value '16'//并发线程数
option r with value '10'//预热时长
option l with value '1200'
<Parameters>
     [server]: 192.168.1.104
     [port]: 3306
     [DBname]: tpcc
       [user]: root
       [pass]: 123456
  [warehouse]: 20
 [connection]: 16
     [rampup]: 10 (sec.)
    [measure]: 1200 (sec.)

RAMP-UP TIME.(10 sec.)

//预热结束，开始进行压测

MEASURING START.

//每10秒钟输出一次压测数据

  10, 0(0):17.297|29.366, 2(1):6.077|6.390, 0(0):0.000|0.000, 0(0):0.000|3.968, 0(0):0.000|0.000
  20, 2(2):19.999|40.895, 3(2):5.218|10.209, 0(0):0.000|0.000, 1(0):0.000|53.658, 1(1):0.000|140.858
  30, 0(0):0.000|0.000, 1(1):0.000|5.888, 0(0):0.000|0.000, 0(0):0.000|0.000, 0(0):0.000|0.000
  40, 4(4):19.999|25.264, 4(3):9.853|13.864, 0(0):0.000|0.000, 1(0):0.000|3.219, 0(0):0.000|0.000
  50, 5(5):19.999|40.184, 4(2):6.046|7.648, 0(0):0.000|0.000, 0(0):0.000|0.000, 0(0):0.000|0.000
  60, 0(0):0.000|0.000, 1(0):0.000|3.568, 1(1):0.000|24.226, 1(0):0.000|1.637, 0(0):0.000|0.000
……
1170, 4(4):19.437|25.995, 4(3):7.417|9.241, 0(0):0.000|0.000, 0(0):0.000|0.000, 0(0):0.000|0.000
1180, 6(6):19.999|26.409, 9(2):5.568|8.834, 1(0):0.000|1.762, 1(0):0.000|36.276, 1(1):0.000|86.459
1190, 2(2):11.617|12.754, 2(2):5.420|8.147, 0(0):0.000|0.000, 0(0):0.000|0.000, 0(0):0.000|0.000
1200, 2(2):16.664|22.551, 1(0):0.000|1.486, 0(0):0.000|0.000, 1(0):0.000|30.834, 1(1):0.000|256.143

//以逗号分隔，共6列
//第一列，第N次10秒
//第二列，新订单成功执行压测的次数(推迟执行压测的次数):90%事务的响应时间|本轮测试最大响应时间，新订单事务数也被认为是总有效事务数的指标
//第三列，支付业务成功执行次数(推迟执行次数):90%事务的响应时间|本轮测试最大响应时间
//第四列，订单状态业务的结果，后面几个的意义同上
//第五列，物流发货业务的结果，后面几个的意义同上
//第六列，库存仓储业务的结果，后面几个的意义同上

-- 压测结束

STOPPING THREADS................

<Raw Results>
  [0] sc:4  lt:243  rt:546999  fl:275 //New-Order，新订单业务成功(success,简写sc)次数，延迟(late,简写lt)次数，重试(retry,简写rt)次数，失败(failure,简写fl)次数
  [1] sc:144  lt:92  rt:565412  fl:281 //Payment，支付业务统计，其他同上
  [2] sc:24  lt:3  rt:50802  fl:26 //Order-Status，订单状态业务统计，其他同上
  [3] sc:52  lt:0  rt:0  fl:0 //Delivery，发货业务统计，其他同上
  [4] sc:0  lt:27  rt:50788  fl:26 //Stock-Level，库存业务统计，其他同上
 in 1200 sec.

<Raw Results2(sum ver.)>
  [0] sc:4  lt:243  rt:547120  fl:275
  [1] sc:144  lt:92  rt:565566  fl:281
  [2] sc:24  lt:3  rt:50802  fl:26
  [3] sc:52  lt:0  rt:0  fl:0
  [4] sc:0  lt:27  rt:50788  fl:26 

<Constraint Check> (all must be [OK]) //下面所有业务逻辑结果都必须为 OK 才行
 [transaction percentage]
        Payment: 40.07% (>=43.0%) [NG] * //支付成功次数(上述统计结果中 sc + lt)必须大于43.0%，否则结果为NG，而不是OK
   Order-Status: 4.58% (>= 4.0%) [OK] //订单
       Delivery: 8.83% (>= 4.0%) [OK] //发货
    Stock-Level: 4.58% (>= 4.0%) [OK] //库存
 [response time (at least 90% passed)] //响应耗时指标必须超过90%通过才行
      New-Order: 1.62%  [NG] *
        Payment: 61.02%  [NG] *
   Order-Status: 88.89%  [NG] *
       Delivery: 100.00%  [OK]
    Stock-Level: 0.00%  [NG] *

<TpmC>
                 12.350 TpmC  -- TpmC结果值（每分钟事务数）

四、使用gnuplot 绘图

绘图脚本

./tpcc_analyze.sh 512m-tpcc-data.log > 512m-tpcc-data.data
./tpcc_analyze.sh 1g-tpcc-data.log > 1g-tpcc-data.data
./tpcc_analyze.sh 2g-tpcc-data.log > 2g-tpcc-data.data

paste 512m-tpcc-data.data  1g-tpcc-data.data 2g-tpcc-data.data > tpcc-data.data
./tpcc-graph.sh tpcc-data.data  tpcc.jpg

tpcc_analyze.sh 

#!/bin/bash
TIMESLOT=1

if [ -n "$2" ]
then
    TIMESLOT=$2
    echo "Defined $2"
fi   

cat $1 | grep -v HY000 | grep -v payment | grep -v neword | \
awk -v timeslot=$TIMESLOT ' BEGIN { FS="[,():]"; s=0; cntr=0; aggr=0 } \
/MEASURING START/ { s=1} /STOPPING THREADS/ {s=0} /0/ { if (s==1) { cntr++; aggr+=$2; } \
if ( cntr==timeslot ) { printf ("%d %3f\n",$1,$5) ; cntr=0; aggr=0  }  } '

tpcc_graph.sh

#!/bin/bash
gnuplot << EOP
set style line 1 lt 1 lw 3
set style line 2 lt 5 lw 3
set style line 3 lt 7 lw 3
set terminal png size 960,480
set grid x y
set xlabel "Time(sec)"
set ylabel "Transactions"
set output "$2"
plot "$1" using 1:2 title "PS 5.1.56 buffer pool 512MM" ls 1 with lines,\
     "$1" using 3:4 title "PS 5.1.56 buffer pool 1g" ls 2 with lines,\
     "$1" using 3:6 title "PS 5.1.56 buffer pool 2g" ls 3 with lines axes x1y1
EOP

可能出现的错误

Could not find/open font when opening font "arial", using internal non-scalable font

export GDFONTPATH=/usr/share/fonts/liberation
export GNUPLOT_DEFAULT_GDFONT=LiberationSans-Regular
source ~/.bashrc

　　

参考文章

《高性能MySQL第三版》

http://blog.itpub.net/26250550/viewspace-1102011/

http://imysql.com/2014/10/10/tpcc-mysql-full-user-manual.shtml

https://securfox.wordpress.com/2009/09/02/plotting-cpu-temperature-with-gnuplot/

https://niftybits.wordpress.com/2008/03/28/gnuplot-dstat-easy-graphing-on-linux/

http://blog.chinaunix.net/uid-12087380-id-4152297.html

https://github.com/yejr/MyDBAtools/tree/master/bench-tools