sysbench安装及性能测试

现在的压力测试工具各种各样，只要上手好几款功能强大点的而且比较大众化的压力测试工具即可，以下跟大家交流下sysbench的安装和压力测试

sysbench支持以下几种测试模式：

1、CPU运算性能

2、磁盘IO性能

3、内存分配及传输速度

4、内存分配及传输速度

5、POSIX线程性能

6、数据库性能（OLTP基准测试）

目前sysbench主要支持mysql,drizzle,pgsql,oracle等几种数据库

下面安装sysbench

一、 下载源码包，安装epel包后以便安装bzr客户端：

[root ~]$ rpm -Uvh http://dl.fedoraproject.org/pub/epel/5/i386/epel-release-5-4.noarch.rpm
Retrieving http://dl.fedoraproject.org/pub/epel/5/i386/epel-release-5-4.noarch.rpm
warning: /var/tmp/rpm-tmp.wgHSDP: Header V3 DSA/SHA1 Signature, key ID 217521f6: NOKEY
Preparing...                ########################################### [%]
   :epel-release           ########################################### [%]

然后就可以开始安装bzr客户端了：

以下安装报错了：

[root yum.repos.d]$ yum install bzr
Loaded plugins: fastestmirror
Loading mirror speeds from cached hostfile
 * base: centos.ustc.edu.cn
 * epel: mirrors.ustc.edu.cn
 * extras: mirror.bit.edu.cn
 * updates: mirror.bit.edu.cn
epel                                                                                                                                                                                                               | 3.7 kB     :
Setting up Install Process
Resolving Dependencies
--> Running transaction check
---> Package bzr.x86_64 :2.1.-.el5 will be installed
--> Processing Dependency: python(abi) = 2.4 for package: bzr-2.1.-.el5.x86_64
--> Processing Dependency: python-paramiko for package: bzr-2.1.-.el5.x86_64
--> Processing Dependency: python-elementtree for package: bzr-2.1.-.el5.x86_64
--> Running transaction check
---> Package bzr.x86_64 :2.1.-.el5 will be installed
--> Processing Dependency: python(abi) = 2.4 for package: bzr-2.1.-.el5.x86_64
--> Processing Dependency: python-elementtree for package: bzr-2.1.-.el5.x86_64
---> Package python-paramiko.noarch :1.7.-.el5 will be installed
--> Processing Dependency: python(abi) = 2.4 for package: python-paramiko-1.7.-.el5.noarch
--> Finished Dependency Resolution
Error: Package: bzr-2.1.-.el5.x86_64 (epel)
           Requires: python(abi) = 2.4
           Installed: python-2.6.-.el6.x86_64 (@updates)
               python(abi) = 2.6
           Available: python26-2.6.-.el5.x86_64 (epel)
               python(abi) = 2.6
Error: Package: python-paramiko-1.7.-.el5.noarch (epel)
           Requires: python(abi) = 2.4
           Installed: python-2.6.-.el6.x86_64 (@updates)
               python(abi) = 2.6
           Available: python26-2.6.-.el5.x86_64 (epel)
               python(abi) = 2.6
Error: Package: bzr-2.1.-.el5.x86_64 (epel)
           Requires: python-elementtree
 You could try using --skip-broken to work around the problem
 You could try running: rpm -Va --nofiles --nodigest

解决方法把旧版本的epel源删除，使用新的：（这个坑说白了，下载的epel源要跟自己使用系统符合，例如你用的是32位系统和64位的系统，下载epel源是不同的）

[root ~]$ cd /etc/yum.repos.d/
[root yum.repos.d]$ ls
CentOS-Base.repo  CentOS-Debuginfo.repo  CentOS-Media.repo  epel.repo  epel-testing.repo
[root yum.repos.d]$ rm -rf epel.repo
[root yum.repos.d]$ rpm -ivh http://dl.fedoraproject.org/pub/epel/6/x86_64/epel-release-6-8.noarch.rpm
Retrieving http://dl.fedoraproject.org/pub/epel/6/x86_64/epel-release-6-8.noarch.rpm
warning: /var/tmp/rpm-tmp.eyCDYb: Header V3 RSA/SHA256 Signature, key ID 0608b895: NOKEY
Preparing...                ########################################### [%]
        file /etc/yum.repos.d/epel-testing.repo from install of epel-release--.noarch conflicts with file from package epel-release--.noarch
        file /etc/yum.repos.d/epel.repo from install of epel-release--.noarch conflicts with file from package epel-release--.noarch

二、再安装安装bzr客户端：

[root yum.repos.d]$ yum install bzr -y
.....
.......
Running Transaction
  Installing : python-paramiko-1.7.-2.1.el6.noarch                                                                                                                                                                                   /
  Installing : bzr-2.1.-.el6.x86_64                                                                                                                                                                                                 /
Installed:
  bzr.x86_64 :2.1.-.el6
Dependency Installed:
  python-paramiko.noarch :1.7.-2.1.el6
Complete!

之后就可以用bzr客户端下载sysbench源码了。

[root ~]$ bzr branch lp:sysbench
You have not informed bzr of your Launchpad ID, and you must do this to
write to Launchpad or access private data.  See "bzr help launchpad-login".
Branched  revision(s). 
[root ~]$ ll -d /root/sysbench/
drwxr-xr-x 7 root root 4096 11月 25 14:45 /root/sysbench/

可以看到，已经下载好sysbench目录了

三、 以下准备安装sysbench,涉及到sysbench源码的配置和编译，首先确认系统安装了gcc gcc-c++编译器:

[root sysbench]$ yum install autoconf automake libtool make cmake gcc gcc-c++ gcc4 -y 
[root sysbench]$ pwd
/root/sysbench
[root sysbench]$ ./autogen.sh

查看./counfiure支持的选项再进行编译安装：

[root sysbench]$ ./configure --help | grep with | grep mysql
  --with-mysql            compile with MySQL support (default is enabled)
  --with-mysql-includes   path to MySQL header files
  --with-mysql-libs       path to MySQL libraries

[root sysbench]$ ./configure --with-mysql=/usr/local/mysql-5.5./ --with-mysql-includes=/usr/local/mysql-5.5./include/ \
> --with-mysql-libs=/usr/local/mysql-5.5./lib/
[root sysbench]$ make && make insall

第二种安装方法：

在 http://sourceforge.net/projects/sysbench 下载sysbench的源码包。

也可以在大牛的MySQL中文网上下载：http://imysql.com/wp-content/uploads/2014/09/sysbench-0.4.12-1.1.tgz

[root src]$wget http://imysql.com/wp-content/uploads/2014/09/sysbench-0.4.12-1.1.tgz
[root src]$ tar zxf sysbench-0.4.-1.1.tgz
[root sysbench-0.4.-1.1]$ ls
aclocal.m4  autom4te.cache  config      config.status  configure.ac  doc         libtool  Makefile     Makefile.in  mkinstalldirs  README-WIN.txt  TODO
autogen.sh  ChangeLog       config.log  configure      COPYING       install-sh  m4       Makefile.am  missing      README         sysbench
[root sysbench-0.4.-1.1]$ ./autogen.sh
[root sysbench-0.4.12-1.1]$./configure --with-mysql=/usr/local/mysql-5.5.40/ --with-mysql-includes=/usr/local/mysql-5.5.40/include/ \
> --with-mysql-libs=/usr/local/mysql-5.5.40/lib/
[root sysbench]$ make && make insall

以上的安装只是下载方式不同，安装都是一样的^.^

四、下面进行使用说明：

测试磁盘IOPS可以分为：prepare、run、cleanup三个阶段：（准备数据、运行压测、清理数据）

创建测试数据库：

mysql> create database sbtest;
Query OK, 1 row affected (0.08 sec)
mysql> 

[root sysbench]$ pwd
/root/sysbench-0.4.-1.1/sysbench
 [root sysbench]$  ./sysbench --mysql-host=localhost --mysql-port= --mysql-user=root --mysql-password= --mysql-socket=/data/mysql-5.5/mysql.sock \
> --test=tests/db/oltp.lua --oltp_tables_count= --oltp-table-size= --rand-init=on prepare

若运行过程报以下错：

./sysbench: error while loading shared libraries: libmysqlclient.so.: cannot open shared object file: No such file or directory
解决方法：
ln -s /usr/local/mysql-5.5./lib/libmysqlclient.so. /usr/lib64/

关于这几个参数的解释：

--test=tests/db/oltp.lua 表示调用 tests/db/oltp.lua 脚本进行 oltp 模式测试
--oltp_tables_count=10 表示会生成 10 个测试表
--oltp-table-size=50000 表示每个测试表填充数据量为 50000
--rand-init=on 表示每个测试表都是用随机数据来填充的

五、进行OLTP测试

./sysbench --mysql-host=localhost --mysql-port=3308 --mysql-user=root --mysql-password=123456 --mysql-socket=/data/mysql-5.5/mysql.sock \
> --test=tests/db/oltp.lua --oltp_tables_count=10 --oltp-table-size=50000 --num-threads=8 \
> --oltp-read-only=off --report-interval=10 --rand-type=uniform --max-time=5  --max-requests=0 --percentile=99 run

参数解释：

--num-threads=8      表示发起 8个并发连接
--oltp-read-only=off 表示不要进行只读测试，也就是会采用读写混合模式测试
--report-interval=10 表示每10秒输出一次测试进度报告
--rand-type=uniform  表示随机类型为固定模式，其他几个可选随机模式：uniform(固定),gaussian(高斯),special(特定的),pareto(帕累托)
--max-time=120       表示最大执行时长为 120秒
--max-requests=0     表示总请求数为 0，因为上面已经定义了总执行时长，所以总请求数可以设定为 0；也可以只设定总请求数，不设定最大执行时长
--percentile=99      表示设定采样比例，默认是 95%，即丢弃1%的长请求，在剩余的99%里取最大值

六、结果解读：

sysbench 0.5:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads:
Report intermediate results every  second(s)
Random number generator seed is  and will be ignored

Threads started!
-- 每10秒钟报告一次测试结果，tps、每秒读、每秒写、%以上的响应时长统计
[  10s] threads: , tps: 119.03, reads/s: 1677.65, writes/s: 478.93, response time: .40ms (%)
[  20s] threads: , tps: 76.80, reads/s: 1074.47, writes/s: 307.19, response time: .75ms (%)
[  30s] threads: , tps: 88.20, reads/s: 1232.94, writes/s: 351.61, response time: .98ms (%)
[  40s] threads: , tps: 149.79, reads/s: 2099.73, writes/s: 600.78, response time: .18ms (%)
[  50s] threads: , tps: 78.09, reads/s: 1093.32, writes/s: 312.38, response time: .12ms (%)
[  60s] threads: , tps: 79.30, reads/s: 1109.76, writes/s: 316.82, response time: .75ms (%)
[  70s] threads: , tps: 35.50, reads/s: 497.52, writes/s: 142.41, response time: .36ms (%)
[  80s] threads: , tps: 45.40, reads/s: 635.57, writes/s: 180.39, response time: .67ms (%)
[  90s] threads: , tps: 73.00, reads/s: 1019.58, writes/s: 290.49, response time: .84ms (%)
[ 100s] threads: , tps: 148.01, reads/s: 2069.14, writes/s: 591.64, response time: .85ms (%)
[ 110s] threads: , tps: 60.60, reads/s: 850.90, writes/s: 242.90, response time: .81ms (%)
[ 120s] threads: , tps: 88.20, reads/s: 1237.71, writes/s: 354.20, response time: .94ms (%)
OLTP test statistics:
    queries performed:
        read:                              -- 读总数
        write:                              -- 写总数
        other:                              -- 其他操作总数(SELECT、INSERT、UPDATE、DELETE之外的操作，例如COMMIT等)
        total:                             -- 全部总数
    transactions:                          (86.74 per sec.)  -- 总事务数(每秒事务数)
    deadlocks:                                 (0.00 per sec.)   -- 发生死锁总数
    read/write requests:                  (1561.37 per sec.)  -- 读写总数(每秒读写次数)
    other operations:                      (173.49 per sec.)   -- 其他操作总数(每秒其他操作次数)

General statistics:  -- 一些统计结果
    total time:                          .1714s   -- 总耗时
    total number of events:                     -- 共发生多少事务数
    total time taken by event execution: .6432s   -- 所有事务耗时相加(不考虑并行因素)
    response time:
         min:                                  .35ms  -- 最小耗时
         avg:                                 .16ms  -- 平均耗时
         max:                               .05ms  -- 最长耗时
         approx.   percentile:             .23ms  -- 超过99%平均耗时

Threads fairness:
    events (avg/stddev):           1303.0000/24.81
    execution time (avg/stddev):   120.0804/0.00

测试建议：
1、真实测试场景中，数据表建议不低于10个，单表数据量不低于500万行，当然了，要视服务器硬件配置而定。如果是配备了SSD或者PCIE SSD这种高IOPS设备的话，则建议单表数据量最少不低于1亿行。

2、真实测试场景中，建议持续压测时长不小于30分钟，否则测试数据可能不具参考意义。

想了解更多关于sysbench的相关说明，可以参考大牛的博客：http://imysql.com/2014/10/17/sysbench-full-user-manual.shtml

作者：陆炫志

出处：xuanzhi的博客 http://www.cnblogs.com/xuanzhi201111

您的支持是对博主最大的鼓励，感谢您的认真阅读。本文版权归作者所有，欢迎转载，但请保留该声明。