[Linux-cluster] fence_ipmilan not working

[Jan Pokorný]

Just a quick correction...

On 22/11/13 19:29 +0100, Jan Pokorný wrote:
> On 22/11/13 14:24 +0800, Zama Ques wrote:
>> I  configured fence_ipmilan as fencing device for my cluster .  From
>> the command line , it works fine but when I try to fence a node from
>> luci interface it is failing .
>> 
>> fencing config info from cluster.conf  
>> 
>> ====
>> 
>>     <cman expected_votes="1" two_node="1"/>
>>         <fencedevices>
>>                 <fencedevice agent="fence_ipmilan" power_wait="10" auth="password" ipaddr="192.168.2.181" lanplus="1" login="aman" name="DB1" passwd="xxxx"/>
>>                 <fencedevice agent="fence_ipmilan" power_wait="10" auth="password" ipaddr="192.168.2.183" lanplus="1" login="aman" name="DB2" passwd="xxxx"/>
>>         </fencedevices>
>> ====
>> 
>>> From system log file (/var/log/messages..)
>> 
>> 
>> ====
>> Apr 15 23:32:18 db2 fence_node[17909]: agent "fence_ipmilan" reports: Rebooting machine @ IPMI:192.168.2.181...Failed
>> Apr 15 23:32:18 db2 fence_node[17909]: Fence of "db1.example.com" was unsuccessful
>> ====
>> 
>> From the command line , it works fine with the parameters provided above
>> 
>> #  fence_ipmilan -t 10 -A password -l aman -p xxxx -P  -a 192.168.2.183
>> Rebooting machine @ IPMI:192.168.2.183...
>> 
>> 
>> Version and hardware related info
>> 
>> ====
>> # rpm -q cman
>> cman-2.0.115-34.el5
>> 
>>  ~]# cat /etc/redhat-release
>> CentOS release 5.5 (Final)
>> 
>> # dmidecode -t system
>> 
>> System Information
>>         Manufacturer: HP
>>         Product Name: ProLiant DL580 G7
>> ====
> 
> 
> so far, I can see two problematic places:
> 
> 1. "power_wait" parameter for fence_ipmilan seems not be supported yet
>    in el5.5 based distro; to achieve the same as with "-t 10", please
>    use "timeout" parameter instead
> 
>    - not sure if this will explain what you observe or not
> 
> 2. as of el5.5, conga is still using direct /sbin/fence_node
>    invocation (direct library call is used since some later version),
>    so you can mimic (and possibly debug, e.g., with strace) what luci
>    triggers with "fence this node" by running on the
>    respective node:

actually, run the command on the other node, not only this how luci
does it, it also to prevent unexpected "blackout" when expecting
the command to return...

>    /sbin/fence_node -O NODENAME
> 
>    where NODENAME is the respective clusternode's "name" parameter in
>    cluster.conf
> 
> 
> Hope this helps.

-- 
Jan