启动备份节点的Heartbeat
启动备份节点的Heartbeat,与主节点方法一样,使用如下命令:
#/etc/init.d/heartbeat start
或者执行:
#service heartbeat start
备用节点的Heartbeat日志输出信息与主节点相对应,通过"tail -f /var/log/messages"可以看到如下输出:
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Link node1:eth1 up.
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Status update
for node node1: status active
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Link node1:eth0 up.
Nov 26 07:57:15 node2 harc[2123]: info: Running /etc/ha.d/rc.d/status status
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Comm_now_up():
updating status to active
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Local
status now set to: 'active'
Nov 26 07:57:15 node2 heartbeat: [2110]: info: Starting
child client "/usr/lib/heartbeat/ipfail" (694,694)
Nov 26 07:57:15 node2 heartbeat: [2110]: WARN: G
_CH_dispatch_int: Dispatch function for read child
took too long to execute: 70 ms (> 50 ms) (GSource: 0x8f62080)
Nov 26 07:57:15 node2 heartbeat: [2134]: info:
Starting "/usr/lib/heartbeat/ipfail" as uid 694 gid 694 (pid 2134)
备份节点检测到node1处于活动状态,没有可以接管的资源,因此仅仅启动了网络监听插件ipfail,监控主节点的心跳。
测试Heartbeat
如何才能得知HA集群是否正常工作,模拟环境测试是个不错的方法。在把Heartbeat高可用性集群放到生产环境中之前,需要做如下5个步骤的测试,从而确定HA是否正常工作。
1、正常关闭和重启主节点的Heartbeat
首先在主节点node1上执行"service heartbeat stop"正常关闭主节点的Heartbeat进程,此时通过ifconfig命令查看主节点网卡信息。正常情况下,应该可以看到主节点已经释放了集群的服务IP地址,同时也释放了挂载的共享磁盘分区。然后查看备份节点,现在备份节点已经接管了集群的服务IP,同时也自动挂载上了共享的磁盘分区。
在这个过程中,使用ping命令对集群服务IP进行测试。可以看到,集群IP一致处于可通状态,并没有任何延时和阻塞现象,也就是说在正常关闭主节点的情况下,主备节点的切换是无缝的,HA对外提供的服务也可以不间断运行。
接着,将主节点Heartbeat正常启动。Heartbeat启动后,备份节点是否自动释放资源将取决于auto_failback 中的设置,本文将顶设置为on,备份节点将自动释放资源,而主节点将再次接管集群资源。其实备份节点释放资源与主节点绑定资源是同步进行的,因而,这个过程也是一个无缝切换。
2、在主节点上拔去网线
拔去主节点连接公共网络的网线后,Heartbeat插件ipfail通过ping测试可以立刻检测到网络连接失败,接着自动释放资源。而就在此时,备用节点的ipfail插件也会检测到主节点出现网络故障,在等待主节点释放资源完毕后,备用节点马上接管了集群资源,从而保证了网络服务不间断持续运行。
同理,当主节点网络恢复正常时,由于设置了"auto_failback on"选项,集群资源将自动从备用节点切会主节点。
在主节点拔去网线后日志信息如下:
Nov 26 09:04:09 node1 heartbeat: [3689]: info: Link node2:eth0 dead.
Nov 26 09:04:09 node1 heartbeat: [3689]: info:
Link 192.168.60.1:192.168.60.1 dead.
Nov 26 09:04:09 node1 ipfail: [3712]: info: Status update:
Node 192.168.60.1 now has status dead
Nov 26 09:04:09 node1 harc[4279]: info: Running /etc/ha.d/rc.d/status status
Nov 26 09:04:10 node1 ipfail: [3712]: info: NS: We are dead. :<
Nov 26 09:04:10 node1 ipfail: [3712]: info: Link Status
update: Link node2/eth0 now has status dead
…… 中间部分省略 ……
Nov 26 09:04:20 node1 heartbeat: [3689]: info: node1 wants to go standby [all]
Nov 26 09:04:20 node1 heartbeat: [3689]: info: standby:
node2 can take our all resources
Nov 26 09:04:20 node1 heartbeat: [4295]: info: give up all
HA resources (standby).
Nov 26 09:04:21 node1 ResourceManager[4305]: info: Releasing
resource group: node1 192.168.60.200/24/eth0 Filesystem::/dev/sdb5::/webdata::ext3
Nov 26 09:04:21 node1 ResourceManager[4305]: info: Running
/etc/ha.d/resource.d/ Filesystem/dev/sdb5 /webdata ext3 stop
Nov 26 09:04:21 node1 Filesystem[4343]: INFO: Running stop for /dev/sdb5 on /webdata
Nov 26 09:04:21 node1 Filesystem[4343]: INFO: Trying to unmount /webdata
Nov 26 09:04:21 node1 Filesystem[4343]: INFO: unmounted /webdata successfully
Nov 26 09:04:21 node1 Filesystem[4340]: INFO: Success
Nov 26 09:04:22 node1 ResourceManager[4305]: info: Running
/etc/ha.d/resource.d/IPaddr 192.168.60.200/24/eth0 stop
Nov 26 09:04:22 node1 IPaddr[4428]: INFO: /sbin/ifconfig eth0:0 192.168.60.200 down
Nov 26 09:04:22 node1 avahi-daemon[1854]: Withdrawing
address record for 192.168.60.200 on eth0.
Nov 26 09:04:22 node1 IPaddr[4407]: INFO: Success
备用节点在接管主节点资源时的日志信息如下:
Nov 26 09:02:58 node2 heartbeat: [2110]: info: Link node1:eth0 dead.
Nov 26 09:02:58 node2 ipfail: [2134]: info: Link Status
update: Link node1/eth0 now has status dead
Nov 26 09:02:59 node2 ipfail: [2134]: info: Asking
other side for ping node count.
Nov 26 09:02:59 node2 ipfail: [2134]: info: Checking remote count of ping nodes.
Nov 26 09:03:02 node2 ipfail: [2134]: info: Telling other
node that we have more visible ping nodes.
Nov 26 09:03:09 node2 heartbeat: [2110]: info: node1
wants to go standby [all]
Nov 26 09:03:10 node2 heartbeat: [2110]: info: standby:
acquire [all] resources from node1
Nov 26 09:03:10 node2 heartbeat: [2281]: info: acquire all HA resources (standby).
Nov 26 09:03:10 node2 ResourceManager[2291]: info: Acquiring
resource group: node1 192.168.60.200/24/eth0 Filesystem::/dev/sdb5::/webdata::ext3
Nov 26 09:03:10 node2 IPaddr[2315]: INFO: Resource is stopped
Nov 26 09:03:11 node2 ResourceManager[2291]: info: Running
/etc/ha.d/resource.d/IPaddr 192.168.60.200/24/eth0 start
Nov 26 09:03:11 node2 IPaddr[2393]: INFO: Using calculated
netmask for 192.168.60.200: 255.255.255.0
Nov 26 09:03:11 node2 IPaddr[2393]: DEBUG: Using calculated
broadcast for 192.168.60.200: 192.168.60.255
Nov 26 09:03:11 node2 IPaddr[2393]: INFO: eval /sbin/ifconfig
eth0:0 192.168.60.200 netmask 255.255.255.0 broadcast 192.168.60.255
Nov 26 09:03:12 node2 avahi-daemon[1844]: Registering new
address record for 192.168.60.200 on eth0.
Nov 26 09:03:12 node2 IPaddr[2393]: DEBUG: Sending Gratuitous
Arp for 192.168.60.200 on eth0:0 [eth0]
Nov 26 09:03:12 node2 IPaddr[2372]: INFO: Success
Nov 26 09:03:12 node2 Filesystem[2482]: INFO: Resource is stopped
Nov 26 09:03:12 node2 ResourceManager[2291]: info: Running
/etc/ha.d/resource.d/ Filesystem/dev/sdb5 /webdata ext3 start
Nov 26 09:03:13 node2 Filesystem[2523]: INFO: Running start for /dev/sdb5 on /webdata
Nov 26 09:03:13 node2 kernel: kjournald starting. Commit interval 5 seconds
Nov 26 09:03:13 node2 kernel: EXT3 FS on sdb5, internal journal
Nov 26 09:03:13 node2 kernel: EXT3-fs: mounted filesystem with ordered data mode.
Nov 26 09:03:13 node2 Filesystem[2520]: INFO: Success
3、在主节点上拔去电源线
在主节点拔去电源后,备用节点的Heartbeat进程会立刻收到主节点已经shutdown的消息。如果在集群上配置了Stonith设备,那么备用节点将会把电源关闭或者复位到主节点。当Stonith设备完成所有操作时,备份节点才能拿到接管主节点资源的所有权,从而接管主节点的资源。
在主节点拔去电源后,备份节点有类似如下的日志输出:
Nov 26 09:24:54 node2 heartbeat: [2110]: info:
Received shutdown notice from 'node1'.
Nov 26 09:24:54 node2 heartbeat: [2110]: info:
Resources being acquired from node1.
Nov 26 09:24:54 node2 heartbeat: [2712]: info:
acquire local HA resources (standby).
Nov 26 09:24:55 node2 ResourceManager[2762]:
info: Running /etc/ha.d/resource.d/IPaddr 192.168.60.200/24/eth0 start
Nov 26 09:24:57 node2 ResourceManager[2762]:
info: Running /etc/ha.d/resource.d/ Filesystem /dev/sdb5 /webdata ext3 start
4、切断主节点的所有网络连接
在主节点上断开心跳线后,主备节点都会在日志中输出"eth1 dead"的信息,但是不会引起节点间的资源切换。此时出现脑裂现象,主备节点都认为对方宕机,如果再次拔掉主节点连接公共网络的网线,网络资源由于连通性问题切换到备机,但存储资源没有顺利切换到备机。此时,就会重现很严重的问题,即备机点对外提供服务,但主节点挂载存储,很有可能导致数据的不一致。连上主节点的心跳线,观察系统日志,可以看到,备用节点的Heartbeat进程将会重新启动,进而再次控制集群资源。最后,连上主节点的对外网线,集群资源再次从备用节点转移到主节点。这就是整个切换过程。
5、在主节点上非正常关闭Heartbeat守护进程
在主节点上可通过"killall -9 heartbeat"命令关闭Heartbeat进程。由于是非法关闭Heartbeat进程,因此Heartbeat所控制的资源并没有释放。备份节点在很短一段时间没有收到主节点的响应后,就会认为主节点出现故障,进而接管主节点资源。在这种情况下,就出现了资源争用情况,两个节点都占用一个资源,造成数据冲突。针对这个情况,可以通过Linux提供的内核监控模块watchdog来解决这个问题,将watchdog集成到Heartbeat中。如果Heartbeat异常终止,或者系统出现故障,watchdog都会自动重启系统,从而释放集群资源,避免了数据冲突的发生。
本章节我们没有配置watchdog到集群中,如果配置了watchdog,在执行"killall -9 heartbeat"时,会在/var/log/messages中看到如下信息:
Softdog: WDT device closed unexpectedly. WDT will not stop!
这个错误告诉我们,系统出现问题,将重新启动。