[Linux-cluster] cluster with mixed versions

[Daniel McNeil]

On Fri, 2004-10-29 at 00:19, Patrick Caulfield wrote:
> On Thu, Oct 28, 2004 at 05:12:35PM -0700, Daniel McNeil wrote:
> > When upgrading my test cluster from 2.6.8.1 to 2.6.9,
> > I rebooted my cluster 1 node at a time.
> > 
> > I noticed that the 2.6.9 node did not join the cluster
> > with the 2.6.8.1 nodes.  When I reboot the 2nd node to
> > 2.6.9, the 2 nodes running 2.6.9 formed a cluster, but
> > the 3rd node running 2.6.8.1 was left in his own cluster
> > without quorum.
> > 
> > Does this mean one cannot have a cluster of mixed versions?
> > 
> > This seems like it could be a problem for rolling upgrades
> > (not a big deal right now).
> 
> You can run mixed kernel version, but you should (at least at this stage) use the
> same cluster patches on all nodes. The code is still in a state of change
> (though stabilising now) and I made a change to the wire protocol fairly
> recently to accommodate a needed feature. This explains why the newer nodes are
> talking to each other but not the older one.
> 
> If you put the new code into a 2.6.8.1 (it should work) kernel then it should
> connect OK.

Patrick,

Thanks for the info.  Once things stabilize, I assume supporting
rolling upgrades is requirement, right? 

What needed feature required changing the protocol?
Is the protocol documented anywhere?
When changes like this are checked in, it would be good to
announce it on the mailing list.

Thanks,

Daniel