Discussion on fence devices in RHCS

In the linux-ha project, more and more attention has been paid to io fence devices. And in rhes4 ha also supports more io fence devices.

1. Why use io fence equipment: Because purely software does not provide full HA. So additional hardware is needed to support it. 2. Those devices can be used as io fence:eg: SCSI/FC disk array; Power swith; LAN/FC swith; Network adapter; 3.io fence device's most fundamental role is to prevent split-brainRHAS4+RHCS4, unplug the power will not switch hqonline

Nothing else to say, fence device is using IBM BMC, using IPMI LAN The problem now is that the network cable can be switched normally. The backup machine will restart the host and take over the service and resources. After unplugging the power supply, it shows: fence node db2: then always shows fence node db2 failed, Then the service does not switch any problems, thank you for your answer, expert help

hqonline

fence device should be no problem, because when I do the test of pulling the network cable, unplug the network cable of the A machine. The B opportunity sends a fence signal, then the A chance is restarted, and then the B opportunity takes over the service normally.

hqonline

I used the main integrated IPMI device, I can choose the fence device as IPMI LAN, and I can drive it directly with RHAS4. Work, the IPMI network port of 2 machines is directly connected, and the other network port is connected to the switch. When testing, unplug the network cable of the A machine to the switch. The B machine can restart the A machine with the fence signal, return the success information, and serve And the resources take over; after the A machine is up, the network cable of the B machine is also unplugged. The A machine can restart the B machine with the fence signal, return the success information, and take over the service and resources; however, if the A is unplugged If there is a problem with the power cord of the machine, the B will also send a fence signal. However, because the power of the A machine is broken, it cannot be restarted, and there is no successful signal return. As a result, the standby will always display fence failed. Do not take over the service, is this a bug in RHCS? Or IPMI device does not support the power supply, the landlord please advise.

hqonline

Attach the cluster.conf file to help you see if there is any problem, thank you Bamboo

<?xml version="1.0"?>< ;cluster alias="dbcluster" config_version="2" name="alpha_cluster"><fence_daemon post_fail_delay="0" post_join_delay="3"/><clusternodes><clusternode name=" ;netview1" votes="1"><fence><method name="1"><device name="bmc"/></method></fence>< /clusternode><clusternode name="netview2" votes="1"><fence><method name="1"><device name="bmc"/></Method></fence></clusternode></clusternodes><cman expected_votes="1" two_node="1"/><fencedevices><fencedevice agent="fence_ipmilan" auth= "none" ipaddr="10.10.10.100" login="USERID" name="bmc" passwd="PASSW 0RD"/></fencedevices><rm><failoverdomains><failoverdomain name="sybase" ordered="0" restricted="1"><failoverdomainnode name="netview2" Priority="1"/><failoverdomainnode name="netview1" priority="1"/></failoverdomain></failoverdomains><resources><fs device="/dev /sdb1" force_fsck="0" force_unmount="1" fsid="57438" fstype="ext3" mountpoint="/sybasedata" name="shdisk" options="" self_fence=" 0"/><ip address="192.168.10.91" monitor_link="1"/><ip address="192.168.16.92" monitor_link="1"/><ip Address="192.168.16.93" monitor_link="1"/><script file="/cluster/dbstart.sh" name="dbscript"/></resources><service autostart ="1" domain="sybase" name ="sybase"><ip ref="192.168.16.93"><fs ref="shdisk"/><script ref="dbscript"/></ip> </service></rm></cluster>

hqonline

There was a result

In the afternoon, I found REDHAT 800 and finally asked a result. There are two types of REDHAT's fence devices, internal fence devices (such as IBM RSAII cards, HP's iLO cards, and IPMI devices) and external fence devices (such as UPS, SAN SWITCH, NETWORK SWITCH, etc.). For the external fence device, you can test the power supply. Because the standby device can receive the signal returned by the fence device, the standby device can take over the service normally. For the internal fence device, the power supply cannot be tested. The machine can not accept the motherboard chip as a signal for the fence device to return, you can not take over the service, clustat will see the owner of the resource is unknow, view the log will see the information reported continuously failed. So in the future, when you are doing the project, you should clearly distinguish it. When writing the test report, don't give yourself the next set. When writing the power supply, you can switch the power supply.

RHAS4+RHCS4, ask a question

aquaofchina

Do a hot standby, but now there is no additional fence device, is it only With manual mode? Also equipped with other modes, what equipment is generally needed? Can you introduce the following, thank you!

hqonline

Recently read a lot of fence device, just talk about what you understand, if there is no fence device, you can only configure it as a manual fence. When you switch the side room, you must Manually enter fence_man_alk on the standby machine (it seems to be this, you can see it in the log, confirm it yourself) that command, the standby machine can take over the resources and start the service.