[Linux-cluster] fenced segfault

[Sebastian Reitenbach]

Hi James,

James Parsons <jparsons at redhat.com> wrote: 
> Sebastian Reitenbach wrote:
> There is no attribute in the conf file named 'ipaddr' for fence_ilo. 
> Change it to 'hostname'...you can still use a dotted quad as the 
> arg...it's just that the name is not correct.
> 
thanks for your suggestion, but the fenced still segfaults when I start 
cman. I found the man page of fence_ilo, I assume the section about the 
parameters listed in the STDIN PARAMETERS I can add in the cluster.conf file 
as the parameter to the fence_agent? 

I also removed the <fencedevices> section in the cluster.conf file 
completely, but the fenced still segfaults, so in my eyes, my configuration 
file doesn't seem to be the problem.

I recognized the short time delay between these two log statements in the 
messages file:
Jul 17 08:01:20 srv4 fenced[32601]: srv5 not a cluster member after 6 sec 
post_join_delay
Jul 17 08:01:21 srv4 kernel: fenced[32601]: segfault at 0000000000000000 rip 
0000000000405e97 rsp 00007fff818454e0 error 4

I only started the srv4, but not srv5, so the first message was right, it 
detected the not running srv5 node. Therefore I tried to start both at the 
same time, and the fenced did not crash anymore. But now the init script of 
the cman hangs forever at "Starting fencing..." 

I have these processes running, since 20 minutes:
srv4:/usr/src/cluster-2.00.00 # ps ax | grep fence
32688 ?        Ss     0:00 /sbin/fenced
32700 pts/1    S+     0:00 /sbin/fence_tool -w -t 300 join
32733 pts/3    S+     0:00 grep fence

when I make a tcpdump on node srv4, I see a lot of UDP packets going out to 
srv5:5404 and to 239.192.25.250.5405.
when I make the same on srv5, I see a lot of UDP packets going to srv4.
But I do not see packets from srv5 arriving at srv4, and vice versa, despite 
both hosts can ping each other and the firewall is deactivated.

Anybody has an idea what still is wrong with my setup?

kind regards
Sebastian