[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] RHCS separate datacenter



On Fri, 2010-08-13 at 17:09 +0300, Dan Frincu wrote:
> 
> > > Absolutely, RHCS != IPVS ... I do got that, sorry for my mis-stating it as such.
> > > 
> > > My response and focus should not have been squarely on IP takeover and/or load-balancing -- as I was not even thinking (strangely) that CMAN was even part of Ana's question (how the {bleep} did I come to that?)  It could very well be in play, and that it might be an issue considering its LAN requirements.  My bad comes from our tendancies to only implement RHCS for IPVS only; and all that it offers in cman, fencing, clvmd, rgmanager, et al is only configured & started when we have GFS / GFS2 filesystems in play.
> > > 
> > > Slightly OT, I have "heard" that multicasting can be routed -- is that true, and if so, couldn't cman then work on different subnets?  Or is there some other constriction or no-no besides "best practices" that I am missing?  And I know you cannot have a node playing in two clusters, despite configuring it to meet network requirements, which could be construed as a shame.
> > > 
> > >     
> > CMAN relies on multicast that can be routed indeed, but the services IP
> > address (in the HA world) are in general unicast IP addresses. How would
> > you manage these IP failover if on different subnets ?
> >   
> When a company owns an AS number and a Provider Independent subnet, it
> is possible to announce via BGP that subnet via 2 separate
> geographical sites, while running a separate subnet for each site; the
> active site would announce it normally and the backup/failover site
> would announce it by prepending it's own AS number to the AS_PATH
> attribute. Therefore, if the main site fails, IP traffic would go to
> the only remaining option, the backup/failover site. But this is more
> related to IP routing then anything else.

> My 2c.
As you're saying it's IP routing related.
On the failover site (the one with the longest AS_PATH) you have means
to reach that subnet (the one you advertise to the outside) either by a
directly connected interface or by internal routes to that subnet.
 
> > > Ok, have a good night!  Myself, I am off to the first tee ... :)
> > > 
> > > 
> > > -----Original Message-----
> > > From: linux-cluster-bounces redhat com [mailto:linux-cluster-bounces redhat com] On Behalf Of Laszlo Beres
> > > Sent: Thursday, August 12, 2010 3:06 PM
> > > To: linux clustering
> > > Subject: Re: [Linux-cluster] RHCS separate datacenter
> > > 
> > > On Thu, Aug 12, 2010 at 8:35 PM,  <rhurst bidmc harvard edu> wrote:
> > > 
> > >     
> > > > stretches a 100 inches or a 100 miles.  I think Ana should reveal more 
> > > > about her implementation rather than hearing about yours.  ;)
> > > >       
> > > Cannot agree more :)
> > > 
> > >     
> > > > And what part of what I said is "false"?  I didn't say anything that fail-over AND load-balancing were required.  Fail-over can be achieved in numerous ways and without RH supplied tools; the load-balancing is native to Linux using IPVS.  But back to the original question: will Red Hat support ... ?  If you use your OWN fail-over strategy, you OWN it.
> > > >       
> > > It was your statement "RHCS is IPVS" that I felt false.
> > > 
> > >     
> > > > Yes, OpenAIS (and likewise the former pulse on RHEL4, sorry for dating myself) is for fail-over which (either) can operate on different LANs.  And to my knowledge and not practical use, the load-balancing (IPVS) can work on different LANs -- if the tunneling option is used.  But my point was that I have not seen any implementation that also maintains IPVS client-session tracking on DIFFERENT LANs (which is NOT a problem if it is on the same physical LAN, like your setup).  It is that last point that has obvious implications on the scope and objectives for those seeking a "supportable" Linux-based solution.
> > > >       
> > > I'm afraid there's still a misunderstanding there - either on my or your side.
> > > 
> > > Pulse was and is a mechanism to ensure a heartbeat channel between two ipvs primary and backup routers. Pulse never had anything to do in the failover cluster core (which is cman in RHEL4, or OpenAIS starting with RHEL5). cman is not supported to be operated on different subnets.
> > > 
> > >     
> > > > Have a great day!
> > > >       
> > > Rather night here in Europe ;)
> > > 
> > > --
> > > László Béres            Unix system engineer http://www.google.com/profiles/beres.laszlo
> > > 
> > > --
> > > Linux-cluster mailing list
> > > Linux-cluster redhat com
> > > https://www.redhat.com/mailman/listinfo/linux-cluster
> > > 
> > > --
> > > Linux-cluster mailing list
> > > Linux-cluster redhat com
> > > https://www.redhat.com/mailman/listinfo/linux-cluster
> > >     
> > 
> > 
> > --
> > Linux-cluster mailing list
> > Linux-cluster redhat com
> > https://www.redhat.com/mailman/listinfo/linux-cluster
> 
> -- 
> Dan FRINCU
> Systems Engineer
> CCNA, RHCE
> Streamwide Romania
> --
> Linux-cluster mailing list
> Linux-cluster redhat com
> https://www.redhat.com/mailman/listinfo/linux-cluster



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]