Cephceph

cosbench使用方法

2018-11-09  本文已影响1人  Joncc

转自 : https://ceph.com/planet/cosbench%E4%BD%BF%E7%94%A8%E6%96%B9%E6%B3%95/
前言
cosbench的功能很强大,但是配置起来可能就有点不是太清楚怎么配置了,本篇将梳理一下这个测试的配置过程,以及一些测试注意项目,以免无法完成自己配置模型的情况

安装
cosbench模式是一个控制端控制几个driver向后端rgw发起请求

下载最新版本

https://github.com/intel-cloud/cosbench/releases/download/v0.4.2.c4/0.4.2.c4.zip

[root@lab102 cosbench]# yum install java-1.7.0-openjdk nmap-ncat
[root@lab102 cosbench]# unzip 0.4.2.zip

[root@localhost /data]# cd 0.4.2.c4
[root@localhost /data/0.4.2.c4]# chmod +x *.sh

启动cosbench , ./start-all.sh

[root@localhost /data/0.4.2.c4]# ./start-all.sh 
Launching osgi framwork ... 
Successfully launched osgi framework!
Booting cosbench driver ... 
Starting    cosbench-log_0.4.2    [OK]
.
Starting    cosbench-tomcat_0.4.2    [OK]
Starting    cosbench-config_0.4.2    [OK]
Starting    cosbench-http_0.4.2    [OK]
Starting    cosbench-cdmi-util_0.4.2    [OK]
Starting    cosbench-core_0.4.2    [OK]
Starting    cosbench-core-web_0.4.2    [OK]
Starting    cosbench-api_0.4.2    [OK]
Starting    cosbench-mock_0.4.2    [OK]
Starting    cosbench-ampli_0.4.2    [OK]
Starting    cosbench-swift_0.4.2    [OK]
Starting    cosbench-keystone_0.4.2    [OK]
Starting    cosbench-httpauth_0.4.2    [OK]
Starting    cosbench-s3_0.4.2    [OK]
Starting    cosbench-librados_0.4.2    [OK]
Starting    cosbench-scality_0.4.2    [OK]
Starting    cosbench-cdmi-swift_0.4.2    [OK]
Starting    cosbench-cdmi-base_0.4.2    [OK]
Starting    cosbench-driver_0.4.2    [OK]
Starting    cosbench-driver-web_0.4.2    [OK]
Successfully started cosbench driver!
Listening on port 0.0.0.0/0.0.0.0:18089 ... 
Persistence bundle starting...
Persistence bundle started.
----------------------------------------------
!!! Service will listen on web port: 18088 !!!
----------------------------------------------

======================================================

Launching osgi framwork ... 
Successfully launched osgi framework!
Booting cosbench controller ... 
Starting    cosbench-log_0.4.2    [OK]
.
Starting    cosbench-tomcat_0.4.2    [OK]
Starting    cosbench-config_0.4.2    [OK]
Starting    cosbench-core_0.4.2    [OK]
Starting    cosbench-core-web_0.4.2    [OK]
Starting    cosbench-controller_0.4.2    [OK]
Starting    cosbench-controller-web_0.4.2    [OK]
Successfully started cosbench controller!
Listening on port 0.0.0.0/0.0.0.0:19089 ... 
Persistence bundle starting...
Persistence bundle started.
----------------------------------------------
!!! Service will listen on web port: 19088 !!!
----------------------------------------------

访问 http://ip:port/controller/, 端口是上一步成功后的端口!!! Service will listen on web port: 19088 !!!这里是19088

image.png

同时可以执行的workloads的个数通过下面的control参数控制

concurrency=1

默认是一个,这个为了保证单机的硬件资源足够,保持单机启用一个workload

创建一个s3用户

[root@lab101 ~]# radosgw-admin user create --uid=test1 --display-name="test1" --access-key=test1 --secret-key=test1
{
"user_id": "test1",
"display_name": "test1",
"email": "",
"suspended": 0,
"max_buckets": 1000,
"auid": 0,
"subusers": [],
"keys": [
{
"user": "test1",
"access_key": "test1",
"secret_key": "test1"
}
],
"swift_keys": [],
"caps": [],
"op_mask": "read, write, delete",
"default_placement": "",
"placement_tags": [],
"bucket_quota": {
"enabled": false,
"max_size_kb": -1,
"max_objects": -1
},
"user_quota": {
"enabled": false,
"max_size_kb": -1,
"max_objects": -1
},
"temp_url_keys": []
}
配置相关
cosbench的配置文件结构
image.png-47.7kB

一个workload 可以定义一个或者多个work stages
执行多个work stages是顺序的,执行同一个work stage里面的work是可以并行执行的
每个work里面,worker是来调整负载的
认证可以多个级别的定义,低级别的认证会覆盖高级别的配置
可以通过配置多个work的方式来实现并发,而在work内通过增加worker的方式增加并发,从而实现多对多的访问,worker的分摊是分到了driver上面,注意多work的时候的containers不要重名,划分好bucker的空间

image.png-144.5kB

work相关的说明

可以通过写入时间,写入容量,写入iops来控制什么时候结束
interval默认是5s是用来对性能快照的间隔,可以理解为采样点
division 控制workers之间的分配工作的方式是bucket还是对象还是none
默认全部的driver参与工作,也可以通过参数控制部分driver参与
时间会控制执行,如果时间没到,但是指定的对象已经写完了的话就会去进行复写的操作,这里要注意是进行对象的控制还是时间的控制进行的测试
如果读取测试的时候,如果没有那个对象,会中断的提示,所以测试读之前需要把测试的对象都填充完毕(最好检查下先)

单项的配置文件
通过单网关创建bucket

<?xml version="1.0" encoding="UTF-8"?>
<workload name="create-bucket" description="create s3 bucket" config="">
    <auth type="none" config=""/>
    <workflow config="">
        <workstage name="create bucket" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1" type="init" workers="2" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(1,32)">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="init" ratio="100" division="container"
                    config="containers=r(1,32);containers=r(1,32);objects=r(0,0);sizes=c(0)B;containers=r(1,32)" id="none"/>
            </work>
        </workstage>
    </workflow>
</workload>

如上配置的时候,如果设置的是workers=1,那么就会从当前的driver中挑选一个driver出来,然后选择配置storage进行bucket的创建,如果设置的是workers=2,那么就会挑选两个driver出来进行创建,一个driver负责一半的工作,相当于两个客户端同时向一个网关发起创建的操作

rgw的网关是对等的关系,那么这里肯定就有另外一种配置,我想通过不只一个网关进行创建的操作,那么这个地方是通过增加work的配置来实现的,我们看下配置

通过多网关创建bucket

<?xml version="1.0" encoding="UTF-8"?>
<workload name="create-bucket" description="create s3 bucket" config="">
    <auth type="none" config=""/>
    <workflow config="">
        <workstage name="create bucket" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1" type="init" workers="2" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16)">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="init" ratio="100" division="container"
                    config="containers=r(1,16);containers=r(1,16);objects=r(0,0);sizes=c(0)B;containers=r(1,16)" id="none"/>
            </work>
            <work name="rgw2" type="init" workers="2" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32)">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="init" ratio="100" division="container"
                    config="containers=r(17,32);containers=r(17,32);objects=r(0,0);sizes=c(0)B;containers=r(17,32)" id="none"/>
            </work>
        </workstage>
    </workflow>
</workload>

以上配置就实现了通过两个网关进行创建bucket的配置了,下面是做prepare的相关配置,在cosbench里面有两个部分可以进行写操作,在prepare stage里面和 main stage里面
这个地方这样设置的理由是:
如果有读和写混合测试的时候,那么就需要提前进行读数据的准备,然后再开始进行读写并发的测试,所以会有一个prepare的阶段,这个在配置文件里面只是type设置的不同,其他没区别,我们可以看下这里web界面里面提供的配置项目,下面其他项目默认都是采取双并发的模式

prepare

在写的部分是一样的

通过多网关写数据

<workstage name="putobject" closuredelay="0" config="">
    <auth type="none" config=""/>
    <work name="rgw1-put" type="normal" workers="2" interval="5"
        division="container" runtime="60" rampup="0" rampdown="0"
        afr="200000" totalOps="0" totalBytes="0" config="">
        <auth type="none" config=""/>
        <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
        <operation type="write" ratio="100" division="container"
            config="containers=u(1,16);objects=u(1,5);sizes=u(2,2)MB" id="none"/>
    </work>
    <work name="rgw2-put" type="normal" workers="2" interval="5"
        division="container" runtime="60" rampup="0" rampdown="0"
        afr="200000" totalOps="0" totalBytes="0" config="">
        <auth type="none" config=""/>
        <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
        <operation type="write" ratio="100" division="container"
            config="containers=u(17,32);objects=u(1,5);sizes=u(2,2)MB" id="none"/>
    </work>     
</workstage>

这里有几个参数可以注意一下:

containers=u(1,16);objects=u(1,5);sizes=u(2,2)MB

控制写入的bucket的名称的,是全部散列还是把负载均分可以自己去控制,objects是指定写入bucke里面的对象的名称的,sizes是指定大小的,如果两个值不同,就是设置的范围,相同就是设置的指定大小的

runtime="60" rampup="0" rampdown="0" afr="200000" totalOps="0" totalBytes="0"
这个是控制写入什么时候中止的,可以通过时间,也可以通过总的ops,或者总的大小来控制,这个需求可以自己定,afr是控制允许的失效率的,单位为1百万分之

interval="5"
这个是控制抓取性能数据的周期的

写入的配置就完了

并发读取的配置

<workstage name="getobj" closuredelay="0" config="">
    <auth type="none" config=""/>
    <work name="rgw1-get" type="normal" workers="2" interval="5"
        division="none" runtime="30" rampup="0" rampdown="0"
        afr="200000" totalOps="0" totalBytes="0" config="">
        <auth type="none" config=""/>
        <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
        <operation type="read" ratio="100" division="none"
            config="containers=u(1,16);objects=u(1,5);" id="none"/>
    </work>
    <work name="rgw2-get" type="normal" workers="2" interval="5"
        division="none" runtime="30" rampup="0" rampdown="0"
        afr="200000" totalOps="0" totalBytes="0" config="">
        <auth type="none" config=""/>
        <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
        <operation type="read" ratio="100" division="none"
            config="containers=u(17,32);objects=u(1,5);" id="none"/>
    </work>     
</workstage>

删除对象的配置

<workstage name="cleanup" closuredelay="0" config="">
          <auth type="none" config=""/>
          <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
          <work name="rgw1-cleanup" type="cleanup" workers="1" interval="5"
              division="object" runtime="0" rampup="0" rampdown="0"
              afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16);objects=r(1,5);">
              <auth type="none" config=""/>
              <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
              <operation type="cleanup" ratio="100" division="object"
                  config="containers=r(1,16);objects=r(1,5);;deleteContainer=false;" id="none"/>
          </work>
          <work name="rgw2-cleanup" type="cleanup" workers="1" interval="5"
              division="object" runtime="0" rampup="0" rampdown="0"
              afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32);objects=r(1,5);">
              <auth type="none" config=""/>
              <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
              <operation type="cleanup" ratio="100" division="object"
                  config="containers=r(17,32);objects=r(1,5);;deleteContainer=false;" id="none"/>
          </work>
      </workstage>

删除bucket的配置

<workstage name="dispose" closuredelay="0" config="">
          <auth type="none" config=""/>
          <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
          <work name="rgw1-dispose" type="dispose" workers="1" interval="5"
              division="container" runtime="0" rampup="0" rampdown="0"
              afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16);">
              <auth type="none" config=""/>
              <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
              <operation type="dispose" ratio="100"
                  division="container"
                  config="containers=r(1,16);;objects=r(0,0);sizes=c(0)B;;" id="none"/>
          </work>
          <work name="rgw2-dispose" type="dispose" workers="1" interval="5"
              division="container" runtime="0" rampup="0" rampdown="0"
              afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32);">
              <auth type="none" config=""/>
              <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
              <operation type="dispose" ratio="100"
                  division="container"
                  config="containers=r(17,32);;objects=r(0,0);sizes=c(0)B;;" id="none"/>
          </work>     
      </workstage>

上面的workstage一共包括下面几种

init 创建bucket
normal write 写入对象
normal read 读取对象
cleanup 清理对象
dispose 清理bucket
division是控制多个worker之间的操作怎么去分的控制,最好在operation那层进行控制

测试前自我提问
单机用了几个workload(默认一般一个,保证单个测试资源的独占)
采用了几个driver(决定了客户端的发起是有几个客户端,单机一个就可以)
测试了哪几个项目(init,prepare or normal,remove),单独测试还是混合测试
单个项目的workstage里面启动了几个work(work可以控制请求发向哪里)
单个work里面采用了几个workers(这个是控制几个driver进行并发的)
测试的ceph集群有多少个rgw网关,创建了多少个bucket测试
设置的写入每个bucket的对象为多少?对象大小为多少?测试时间为多久?
测试很多文件的时候,可以用ops控制,并且将ops设置大于想测试的文件数目,保证能写入那么多的数据,或者比较确定性能,也可以通过时间控制

那么我来根据自己的需求来进行一个测试模型说明,然后根据说明进行配置

采用两个客户端测试,那么准备两个driver
准备配置两个rgw的网关,那么在配置workstage的时候配置两个work对应到两个storage
测试创建,写入,读取,删除对象,删除bucket一套完整测试
wokers设置为2的倍数,初始值为2,让每个driver分得一半的负载,在进行一轮测试后,成倍的增加driver的数目,来增大并发,在性能基本不增加,时延在增加的时候,记录性能值和参数值,这个为本环境的最大性能
创建32个bucket,每个bucket写入5个2M的对象,测试写入时间为600s,读取时间为60s
简单框架图
cosbench

配置文件如下:

<?xml version="1.0" encoding="UTF-8"?>
<workload name="create-bucket" description="create s3 bucket" config="">
    <auth type="none" config=""/>
    <workflow config="">
        <workstage name="create bucket" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1-create" type="init" workers="2" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16)">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="init" ratio="100" division="container"
                    config="containers=r(1,16);objects=r(0,0);sizes=c(0)B" id="none"/>
            </work>
            <work name="rgw2-create" type="init" workers="2" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32)">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="init" ratio="100" division="container"
                    config="containers=r(17,32);objects=r(0,0);sizes=c(0)B" id="none"/>
            </work>
        </workstage>
    
        <workstage name="putobject" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1-put" type="normal" workers="2" interval="5"
                division="container" runtime="600" rampup="0" rampdown="0"
                afr="200000" totalOps="0" totalBytes="0" config="">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="write" ratio="100" division="container"
                    config="containers=u(1,16);objects=u(1,5);sizes=u(2,2)MB" id="none"/>
            </work>
            <work name="rgw2-put" type="normal" workers="2" interval="5"
                division="container" runtime="600" rampup="0" rampdown="0"
                afr="200000" totalOps="0" totalBytes="0" config="">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="write" ratio="100" division="container"
                    config="containers=u(17,32);objects=u(1,5);sizes=u(2,2)MB" id="none"/>
            </work>     
        </workstage>

        <workstage name="getobj" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1-get" type="normal" workers="2" interval="5"
                division="none" runtime="60" rampup="0" rampdown="0"
                afr="200000" totalOps="0" totalBytes="0" config="">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="read" ratio="100" division="none"
                    config="containers=u(1,16);objects=u(1,5);" id="none"/>
            </work>
            <work name="rgw2-get" type="normal" workers="2" interval="5"
                division="none" runtime="60" rampup="0" rampdown="0"
                afr="200000" totalOps="0" totalBytes="0" config="">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="read" ratio="100" division="none"
                    config="containers=u(17,32);objects=u(1,5);" id="none"/>
            </work>     
        </workstage>
    
    <workstage name="cleanup" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1-cleanup" type="cleanup" workers="1" interval="5"
                division="object" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16);objects=r(1,100);">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="cleanup" ratio="100" division="object"
                    config="containers=r(1,16);objects=r(1,100);;deleteContainer=false;" id="none"/>
            </work>
            <work name="rgw2-cleanup" type="cleanup" workers="1" interval="5"
                division="object" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32);objects=r(1,100);">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="cleanup" ratio="100" division="object"
                    config="containers=r(17,32);objects=r(1,100);;deleteContainer=false;" id="none"/>
            </work>
        </workstage>

    <workstage name="dispose" closuredelay="0" config="">
            <auth type="none" config=""/>
            <work name="rgw1-dispose" type="dispose" workers="1" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(1,16);">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7481;path_style_access=true"/>
                <operation type="dispose" ratio="100"
                    division="container"
                    config="containers=r(1,16);;objects=r(0,0);sizes=c(0)B;;" id="none"/>
            </work>
            <work name="rgw2-dispose" type="dispose" workers="1" interval="5"
                division="container" runtime="0" rampup="0" rampdown="0"
                afr="0" totalOps="1" totalBytes="0" config="containers=r(17,32);">
                <auth type="none" config=""/>
                <storage type="s3" config="accesskey=test1;secretkey=test1;endpoint=http://192.168.19.101:7482;path_style_access=true"/>
                <operation type="dispose" ratio="100"
                    division="container"
                    config="containers=r(17,32);;objects=r(0,0);sizes=c(0)B;;" id="none"/>
            </work>     
        </workstage>    
    
    </workflow>
</workload>

上面的测试是为了做测试模板,所以采用了比较小的对象数目和比较小的测试时间

可以根据自己的硬件环境或者客户的要求来设计测试模型,环境够大的时候,提供足够的rgw和足够的客户端才能测出比较大的性能值

测试的时候,尽量把写入和读取的测试分开,也就是分两次测试,避免第一次的写入没写足够对象,读取的时候读不到中断了,对于长达数小时的测试的时候,中断是很令人头疼的,分段可以减少这种中断后的继续测试的时间

写入的测试在允许的范围内,尽量写入多点对象,尽量避免复写,也能够在读取的时候尽量能够足够散列

测试时间能够长尽量长

测试结果
result
graph

可以通过线图来看指定测试项目的中间情况,一般是去关注是否出现比较大的抖动 ,相同性能下,抖动越小越好

其他调优
在硬件环境一定的情况下,可以通过增加nginx负载均衡,或者lvs负载均衡来尝试增加性能值,这个不在本篇的讨论范围内

上一篇下一篇

猜你喜欢

热点阅读