本文共 12023 字，大约阅读时间需要 40 分钟。
a.http://www.gyrate.org/misc/gfs.txt
b.http://www.redhat.com/docs/manuals/enterprise/RHEL-3-Manual/cluster-suite/index.html c.ftp://ftp.redhat.com/pub/redhat/linux/updates/enterprise/3ES/en/RHGFS/SRPMS d.http://distro.ibiblio.org/pub/linux/distributions/caoslinux/centos/3.1/contrib/i386/RPMS/ LVS结构图： eth0=10.3.1.101 eth0:1=10.3.1.254 Load Balance Router eth1=192.168.1.71 eth1:1=192.168.1.1 | | | | Real1 Real2 eth0=192.168.1.68 eth0=192.168.1.67 (eth0 gateway=192.168.1.1) eth1=192.168.0.1---eth1=192.168.0.2 (双机互联线) | | GFS ISCSI Share storage eth0=192.168.1.124 1.Setup ISCSI Server Server: PIII 1.4,512M, Dell 1650,Redhat 9,IP=192.168.1.124 从下载ISCSI TARGET的Source code () 我选了iscsitarget-0.3.8.tar.gz,要求kernel 2.4.29 从kernel.org下载kernel 2.4.29,解开编译重启后编译安装iscsitarget-0.3.8： #make KERNELSRC=/usr/src/linux-2.4.29 #make KERNELSRC=/usr/src/linux-2.4.29 install #cp ietd.conf /etc #vi /etc/ietd.conf # Example iscsi target configuration # # Everything until the first target definition belongs # to the global configuration. # Right now this is only the user configuration used # during discovery sessions: # Users, who can access this target # (no users means anyone can access the target) User iscsiuser 1234567890abc Target iqn.2005-04.com.my:storage.disk2.sys1.iraw1 User iscsiuser 1234567890abc Lun 0 /dev/sda5 fileio Alias iraw1 Target iqn.2005-04.com.my:storage.disk2.sys1.iraw2 User iscsiuser 1234567890abc Lun 1 /dev/sda6 fileio Alias iraw2 Target iqn.2005-04.com.my:storage.disk2.sys2.idisk User iscsiuser 1234567890abc Lun 2 /dev/sda3 fileio Alias idisk Target iqn.2005-04.com.my:storage.disk2.sys2.icca User iscsiuser 1234567890abc Lun 3 /dev/sda7 fileio Alias icca 说明：password 长度必须不小于12个字符, Alias是别名, 不知为何这个别名在Client端显示不出来. 分区：我只有一个SCSI盘，所以： /dev/sda3: Share storage,容量越大越好 /dev/sda5: raw1, 建Cluster要的rawdevice, 我给了900M /dev/sda6: raw2, 建Cluster要的rawdevice, 我给了900M /dev/sda7: cca, 建GFS要的,我给了64M (/dev/sda4是Extended分区,在其中建了sda5,6,7) #Reboot，用service iscsi-target start启ISCSI server(我觉得比建议的好，可以用service iscsi-target status看状态) 2.Setup ISCSI Client(on two real server) Server: PIII 1.4,512M, Dell 1650,Redhat AS3U4(用AS3U5更好),2.4.21-27.EL #vi /etc/iscsi.conf DiscoveryAddress=192.168.1.124 OutgoingUsername=iscsiuser OutgoingPassword=1234567890abc Username=iscsiuser Password=1234567890abc LoginTimeout=15 IncomingUsername=iscsiuser IncomingPassword=1234567890abc SendAsyncTest=yes #service iscsi restart #iscsi-ls -l ..., 精简如下: /dev/sdb:iraw2 /dev/sdc:iraw1 /dev/sdd:idisk /dev/sde:icca 注意: 在real server中ISCSI device的顺序很重要，两个real server中一定要一样，如不一样，就改ISCSI Server中的设置，多试几次。 3.Install Redhat Cluster suite 先下载Cluster Suite的ISO, AS3的我是从ChinaUnix.net找到的下载点, 安装clumanager和redhat-config-cluster。没有Cluster Suite的ISO也没关系，从下载clumanager-1.2.xx.src.rpm,redhat-config-cluster-1.0.x.src.rpm,编译后安装，应该更好: #rpm -Uvh clumanager-1.2.26.1-1.src.rpm #rpmbuild -bs /usr/src/redhat/SPECS/clumanager.spec #rpmbuild --rebuild --target i686 /usr/src/redhat/SRPMS/clumanager-1.2.26.1-1.src.rpm 还有redhat-config-cluster-1.0.x.src.rpm，也装好 4.Setup Cluster as HA module 详细步骤我就不写了，网上有很多文章，我也是看了别人的文章学会的，不过人家是用VMWARE，而我是用真的机子+ISCSI，raw device就是/dev/sdb,/dev/sdc, 然后就mount /dev/sdd /u01, mkfs.ext3 /u01 ...... 设好后会发现ISCSI有问题：同时只能有一个Client联接写盘，如果两个Client同时联ISCSI的Share Storge,一个Client写，另一个Client是看不到的，而且此时文件系统已经破坏了，Client重联ISCSI时会发现文件是坏的，用fsck也修复不了。 ISCSI真的是鸡肋吗？ NO！从GOOGLE上我终于查到ISCSI只有用Cluster File System才能真正用于Share Storage! 而Redhat买下的GFS就是一个! 5.Setup GFS on ISCSI GFS只有Fedora Core4才自带了，而GFS又一定要用到Cluster Suite产生的/etc/cluster.xml文件，我没见FC4有Cluster Suite,真不知Redhat给FC4带GFS干嘛，馋人吗？ 好，闲话少说，下载：c处的GFS-6.0.2.20-2.src.rpm， 按a处的gfs.txt编译安装，不过关于cluster.ccs,fence.ccs,nodes.ccs的设置没说，看b的文档，我总算弄出来了，都存在/root/cluster下，存在别的地方也行，不过我不知道有没有错，我没有光纤卡，文档又没讲ISCSI的例子，不过GFS能启动的。 #cat cluster.ccs cluster { name = "Cluster_1" lock_gulm { servers = ["cluster1", "cluster2"] heartbeat_rate = 0.9 allowed_misses = 10 } } 注：name就是Cluster Suite设置的Cluster name, servers就是Cluster member的Hostname，别忘了加进/etc/hosts;allowed_misses我开始设为1，结果跑二天GFS就会死掉，改为10就没死过了。 #cat fence.ccs fence_devices{ admin { agent = "fence_manual" } } #cat nodes.ccs nodes { cluster1 { ip_interfaces { hsi0 = "192.168.0.1" } fence { human { admin { ipaddr = "192.168.0.1" } } } } cluster2 { ip_interfaces { hsi0 = "192.168.0.2" } fence { human { admin { ipaddr = "192.168.0.2" } } } } } 注：ip就是心跳线的ip 这三个文件建在/root/cluster下，先建立Cluster Configuration System： a.#vi /etc/gfs/pool0.cfg poolname pool0 minor 1 subpools 1 subpool 0 8 1 gfs_data pooldevice 0 0 /dev/sde1 b.#pool_assemble -a pool0 c.#ccs_tool create /root/cluster /dev/pool/pool0 d.#vi /etc/sysconfig/gfs CCS_ARCHIVE="/dev/pool/pool0" 再Creating a Pool Volume，就是我们要的共享磁盘啦， a.#vi /etc/gfs/pool1.cfg poolname pool1 minor 2 subpools 1 subpool 0 128 1 gfs_data pooldevice 0 0 /dev/sdd1 b.#pool_assemble -a pool1 c.#gfs_mkfs -p lock_gulm -t Cluster_1:gfs1 -j 8 /dev/pool/pool1 d.#mount -t gfs -o noatime /dev/pool/pool1 /u01 下面是个GFS的启动脚本，注意real1和real2必须同时启动lock_gulmd进程，第一台lock_gulmd会成为Server并等Client的lock_gulmd，几十秒后没有响应会fail，GFS启动失败。Redhat建议GFS盘不要写进/etc/fstab。 #cat gfstart.sh #!/bin/sh depmod -a modprobe pool modprobe lock_gulm modprobe gfs sleep 5 service iscsi start sleep 20 service rawdevices restart pool_assemble -a pool0 pool_assemble -a pool1 service ccsd start service lock_gulmd start mount -t gfs /dev/pool/pool1 /s02 -o noatime service gfs status 6. Setup Linux LVS LVS是章文嵩博士发起和领导的优秀的集群解决方案，许多商业的集群产品，比如RedHat的Piranha，Turbolinux公司的Turbo Cluster等，都是基于LVS的核心代码的。我的系统是Redhat AS3U4，就用Piranha了。从rhel-3-u5-rhcs-i386.iso安装piranha-0.7.10-2.i386.rpm,ipvsadm-1.21-9.ipvs108.i386.rpm () 装完后service httpd start & service piranha-gui start，就可以从管理或设置了，当然了，手工改/etc/sysconfig/ha/lvs.cf也一样。#cat /etc/sysconfig/ha/lvs.cf serial_no = 80 primary = 10.3.1.101 service = lvs rsh_command = ssh backup_active = 0 backup = 0.0.0.0 heartbeat = 1 heartbeat_port = 1050 keepalive = 6 deadtime = 18 network = nat nat_router = 192.168.1.1 eth1:1 nat_nmask = 255.255.255.0 reservation_conflict_action = preempt debug_level = NONE virtual lvs1 { active = 1 address = 10.3.1.254 eth0:1 vip_nmask = 255.255.255.0 fwmark = 100 port = 80 persistent = 60 pmask = 255.255.255.255 send = "GET / HTTP/1.0rnrn" expect = "HTTP" load_monitor = ruptime scheduler = wlc protocol = tcp timeout = 6 reentry = 15 quiesce_server = 1 server Real1 { address = 192.168.1.68 active = 1 weight = 1 } server Real2 { address = 192.168.1.67 active = 1 weight = 1 } } virtual lvs2 { active = 1 address = 10.3.1.254 eth0:1 vip_nmask = 255.255.255.0 port = 21 send = "n" use_regex = 0 load_monitor = ruptime scheduler = wlc protocol = tcp timeout = 6 reentry = 15 quiesce_server = 0 server ftp1 { address = 192.168.1.68 active = 1 weight = 1 } server ftp2 { address = 192.168.1.67 active = 1 weight = 1 } } 设置完后service pulse start, 别忘了把相关的client加进/etc/hosts #iptables -t mangle -A PREROUTING -p tcp -d 10.3.1.254/32 --dport 80 -j MARK --set-mark 100 #iptables -t mangle -A PREROUTING -p tcp -d 10.3.1.254/32 --dport 443 -j MARK --set-mark 100 #iptables -A POSTROUTING -t nat -p tcp -s 10.3.1.0/24 --sport 20 -j MASQUERADE 运行以上三行命令并存入/etc/rc.d/rc.local，用ipvsadm看状态： #ipvsadm IP Virtual Server version 1.0.8 (size=65536) Prot LocalAddress:Port Scheduler Flags -> RemoteAddress:Port Forward Weight ActiveConn InActConn TCP 10.3.1.254:ftp wlc -> cluster2:ftp Masq 1 0 0 -> cluster1:ftp Masq 1 0 0 FWM 100 wlc persistent 60 -> cluster1:0 Masq 1 0 0 -> cluster2:0 Masq 1 0 0 注意：a.Firewall Mark可以不要，我反正是加了，文档说有https的话加上，值我选了100， b.Virtual IP别加进/etc/hosts，我上过当，80端口时有时无的， c.eth0:1,eth1:1是piranha产生的，别自己手工设置，我干过这画蛇添足的事，网上有 些帖子没说清，最后是看Redhat的文档才弄清楚的。 d.The LVS router can monitor the load on the various real servers by using either rup or ruptime. If you select rup from the drop-down menu, each real server must run the rstatd service. If you select ruptime, each real server must run the rwhod service.Redhat的原话，就是如选rup的监控模式real server上 都要运行rstatd进程，如选ruptime就要运行rwhod进程。 e.Real Server同Router相联的网卡的Gateway必须是Router的那块网卡的VIP，举本例： Router的eth1同两个real server的eth0相联，如VIP eth1:1=192.168.1.1，则real server 的eth0的Gateway=192.168.1.1 7.Setup TOMCAT5.59+JDK1.5(用Redhat自带的Apache) a.#tar xzvf jakarta-tomcat-5.5.9.tar.gz #mv jakarta-tomcat-5.5.9 /usr/local #ln -s /usr/local/jakarta-tomcat-5.5.9 /usr/local/tomcat b.#jdk-1_5_0_04-linux-i586.bin #mv jdk1.5.0_4 /usr/java #ln -s /usr/java/jdk1.5.0_4 /usr/java/jdk c.#vi /etc/profile.d/tomcat.sh export CATALINA_HOME=/usr/local/tomcat export TOMCAT_HOME=/usr/local/tomcat d.#vi /etc/profile.d/jdk.sh if ! echo ${PATH} | grep "/usr/java/jdk/bin" ; then JAVA_HOME=/usr/java/jdk export JAVA_HOME export PATH=/usr/java/jdk/bin:${PATH} export CLASSPATH=$JAVA_HOME/lib fi e.#chmod 755 /etc/profile.d/*.sh f.重新用root登录，让tomcat.sh和jdk.sh起作用， #tar xzvf jakarta-tomcat-connectors-jk2-src-current.tar.gz #cd jakarta-tomcat-connectors-jk2-2.0.4-src/jk/native2/ #./configure --with-apxs2=/usr/sbin/apxs --with-jni --with-apr-lib=/usr/lib #make #libtool --finish /usr/lib/httpd/modules #cp ../build/jk2/apache2/mod_jk2.so ../build/jk2/apache2/libjkjni.so /usr/lib/httpd/modules/ g.#vi /usr/local/tomcat/bin/catalina.sh 在# Only set CATALINA_HOME if not already set后加上以下两行： serverRoot=/etc/httpd export serverRoot h.#vi /usr/local/tomcat/conf/jk2.properties serverRoot=/etc/httpd apr.NativeSo=/usr/lib/httpd/modules/libjkjni.so apr.jniModeSo=/usr/lib/httpd/modules/mod_jk2.so i.#vi /usr/local/tomcat/conf/server.xml, 在前加上以下几行，建了两个VirtualPath：myjsp和local，一个指向share storage， 一个指向real server本地 prefix="cluster.log." suffix=".txt" timestamp="true" /> prefix="cluster_access.log." suffix=".txt" pattern="common" resolveHosts="false" /> j.#vi /etc/httpd/conf/workers2.properties #[logger.apache2] #level=DEBUG [shm] file=/var/log/httpd/shm.file size=1048576 [channel.socket:localhost:8009] tomcatId=localhost:8009 keepalive=1 info=Ajp13 forwarding over socket [ajp13:localhost:8009] channel=channel.socket:localhost:8009 [status:status] info=Status worker, displays runtime informations [uri:/*.jsp] worker=ajp13:localhost:8009 context=/ k.#vi /etc/httpd/conf/httpd.conf 改:DocumentRoot "/u01/www" 加： 在LoadModule最后加： LoadModule jk2_module modules/mod_jk2.so JkSet config.file /etc/httpd/conf/workers2.properties 在#之前加： Order allow,deny Deny from all l:#mkdir /u01/ftproot #mkdir /u01/www #mkdir /u01/www/myjsp m:在每个real server上生成index.jsp #vi /var/www/html/index.jsp 在real server2上就是"test page on real server 2" n:下载jdbc Driver 可惜只有for JDK1.4的,在两台real server上分别 #cp -R /usr/local/tomcat/webapps/webdav/WEB-INF /u01/www/myjsp #cp ojdbc14.jar ojdbc14_g.jar ocrs12.zip /u01/www/myjsp/WEB-INF/lib o:假设我有一台OracleServer，ip=10.3.1.211,sid=MYID,username=my,password=1234,并有Oracle 的例子employees的read权限,或干脆把这个table拷过来,我是Oracle9i中的 #vi /u01/www/myjsp/testoracle.jsp p:#vi /u01/www/index.html WEB'>http://10.3.1.254/local/index.jsp">WEB Local Test'>http://10.3.1.254/myjsp/testoracle.jsp">Test Oracle WEB q:在两台real server上分别 #vi /usr/local/tomcat/conf/tomcat-users.xml 加下面一行，允许页面管理： r:在两台real server上分别 #service httpd restart #/usr/local/tomcat/bin/startup.sh s:打开和，选Tomcat Manager，用manager/tomcat登录，虚拟目录/myjsp和/local应该Start了在两台机子上分别打开网页，选WEB Local，可以看到一台显示： "test page on real server 1"，另一台为"test page on real server 2",同时在Router上 ipvsadm可以看到每个real server的联接数 8.设置FTP服务 #vi /etc/vsftpd/vsftp.conf，在两台real server上分别加入以下几行： anon_root=/u01/ftproot local_root=/u01/ftproot setproctitle_enable=YES #service vsftpd start 现在LVM+GFS+ISCSI+TOMCAT就设置好了，我们可以用Apache Jmeter来测试LVM的性能，两台机子上分别运行jmeter，都指向10.3.1.254/myjsp/testoracle.jsp，各200个threads同时运行，在Router上用ipvsadm可以监控，Oracle Server的性能可要好，否则大量的http进程会hang在real server上，ipvsadm也会显示有个real server失去了。测试时real server的CPU idle会降到70%，而Router 的CPU idle几乎不动。
来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/312079/viewspace-245783/，如需转载，请注明出处，否则将追究法律责任。
转载于:http://blog.itpub.net/312079/viewspace-245783/