[Linux-cluster] Re: apache resource problem in RHCS 5.1

[Lon Hohberger]

On Tue, 2008-06-17 at 09:52 +0200, denis wrote:
> Lorenz Pfiffner wrote:
> > I want to add something I recently found out about the relocation time 
> > of IP resources, which I complained about.
> > 
> > The reason why it takes 10 seconds per IP can be found here: 
> > http://sources.redhat.com/cluster/faq.html#rgm_failovertime
> 
> Can anyone add to this, seems very interesting to cut the failover time 
> for my scenario too.
> 
> Is lowering the sleep time or removing it safe for clusters without NFS?

Generally, whenever you leave a process running to service requests, you
want to leave the 'sleep' in.

EX: nfsd generally stays running because it can be used by multiple
services running on the same node (and there is in no way to make
multiple separate instances of nfsd bind to separate IP addresses
currently).  So, sleep for a bit.

Leaving the app running has some limitations, particularly if the
service comes back to the original host (this was recently discussed on
linux-nfs).


Whenever you stop the application, you can remove the sleep.

EX: Apache is generally shut down for a service relocation.  You may
remove the 'sleep' because the sockets will be closed and the
application will be terminated.

-- Lon