[Linux-cluster] Configuring rgmanager

[Lon Hohberger]

On Tue, 2005-03-01 at 14:28 +0100, Ion Alberdi wrote:
> Ion Alberdi wrote:
> 
> >
> > Results from my tests with two nodes(buba and gump)(and latest 
> > cvs(update done today)):
> > I tried to put a basic script in failover on two nodes.
> > Initialization:
> > .....
> 
> > (buba seems not to have any clurgmrgrd running, even if I started the 
> > rgmanager...)
> >
> > I don't know if it's a bug of the rgmanager or if I'm doing something 
> > wrong, but I don't understand why during the first reboot everything 
> > worked and nothing then...
> >
> Aparently this has nothing to do with the reboot, but with the use of 
> clusvcadm.
> When I only reboot nodes without relocating manually services, it works 
> well: the service switching is always done.
> However, after having relocated manually the service and rebooting the 
> server running the service, it seems that magma doesn't say to the 
> remaining node  that the service must be relocated (membership_update is 
> not called, so nothing is done...)
> Is this a bug?

There was a bug where rgmanager was randomly getting stuck in accept(2)
because the listen sockets weren't set O_NONBLOCK (like they should have
been).  Once it got there, it stopped picking up cluster events
(including membership changes).  This should be fixed in current CVS.

Could this have been the problem?

> If it is not, I would find very usefull to relocate manually a service,
> for example if the cluster is composed of a huge server and a little 
> rescue server, after having repaired the huge server and making it join 
> the cluster, it will be welcome to relocate manually the service.
> Also if the administrator wants to do some operations on a server 
> running a criticial service, it will be necessary to have previously 
> relocated the service.

Relocating services is an obvious feature.

-- Lon