apache segfault on Enable/Disable contexts

[stourwalk]

Hi,

We just starting to make use of mod_cluster as an experiment to see if we can simplify our production environment.

Currently we are using mod_cluster 1.0.1GA, a custom built Apache 2.2.11 on CentOS5 (virtualised under XenServer) to talk to a number (between 1 and 6) of Jboss 4 nodes running our web application.

Long term we will move to Jboss5, but current Cache and AOP dependencies make that a project in its own right.

Once we get to over 2 nodes, we invariably find that clicking on the Enable/Disable contexts button on mod_cluster-manager causes an apache segfault.

[notice] child pid 15229 exit signal Segmentation fault (11)

Is this a known issue, is there anything we can do to help debug the cause?

Thanks
Chris

[jfclere]

Weird that looks like MODCLUSTER-80

[stourwalk]

It does look rather similar doesn't it.

We downloaded the binaries from

http://jboss.org/mod_cluster/downloads/latest/

yesterday - so hopefully they should be the correct ones - is there any other way of telling which version they are?

Is there anything we can do to help debug the issue?

Also we've noticed that if a node starts up but doesn't yet have a virtual host then the html output is missing a - so it makes a mess of the layout for the following node. e.g.

</pre><h1> Node l3205 (ajp://192.168.1.15:8009): <a href="/mod_cluster-manager?nonce=c3c4190e-8720-11de-897b-1956cd71a2c6&Cmd=ENABLE-APP&Range=NODE&JVMRoute=l3205">Enable Contexts</a> <a href="/mod_cluster-manager?nonce=c3c4190e-8720-11de-897b-1956cd71a2c6&Cmd=DISABLE-APP&Range=NODE&JVMRoute=l3205">Disable Contexts</a></h1>
Balancer: mycluster,Domain: l32,Flushpackets: Off,Flushwait: 10000,Ping: 10000000,Smax: 0,Ttl: 31,Elected: 0,Read: 0,Transferred: 0,Connected: 0,Load: 0
</pre><h1> Node l3206 (ajp://192.168.1.16:8009): <a href="/mod_cluster-manager?nonce=c3c4190e-8720-11de-897b-1956cd71a2c6&Cmd=ENABLE-APP&Range=NODE&JVMRoute=l3206">Enable Contexts</a> <a

I guess you output the on the new node regardless of whether it outputted one previously or not?

Thanks
Chris

[stourwalk]

"stourwalk" wrote:
It does look rather similar doesn't it.

Also we've noticed that if a node starts up but doesn't yet have a virtual host then the html output is missing a <pre> - so it makes a mess of the layout for the following node. e.g.

I think (despite not having programmed in C for a long long time) that the fault might lie in mod_manager.c -> static void manager_info_hosts(), in that it always outputs a at the end of the method regardless of whether size is greater than 0 or not. I think it should only output it if size>0 (ie there are some)

Chris

[jfclere]

The mess in the layout is MODCLUSTER-95

[jfclere]

I can't reproduce the core... Does it happen in the mod_cluster-manager page or on the other requests?

[stourwalk]

I can't reproduce the core... Does it happen in the mod_cluster-manager page or on the other requests?

Specifically when we click the enable/disable context links on the mod_cluster-manager page.

Chris

[stourwalk]

The case in question was that two nodes were already started and we were starting a third, whilst it was starting we tried to disable the first two.

So it might be related to that (ie trying to disable one whilst one is starting)

Chris

[jfclere]

I can't reproduce it... Use the bundle corresponding to you platform. If it cores open a JIRA and attach the core to it.