[Spread-users] network partitions

[Jonathan Stanton]

They each construct a new view showing what daemons they are currently in contact with. So 
in your example, the cluster by itself will deliver a new NETWORK view showing that only 
it is in the membership, while the other two clusters will all deliver a new NETWORK view 
showing that only they are in the membership.

Since spread supports partitioned operation, it does not distinguish between crashed 
daemons and daemons separated by a network partition. In both cases the membership simply 
doesn't include them. However the daemons themselves do know the difference (i.e. they do 
know if they crashed or not) And correspondingly, your applications using spread know the 
difference as if they crash, then they know they crashed (because they restart) but if 
they were running and a network partition occurred, when the network is healed and a merge 
occurs, they can learn the status of the other copies of the applications that were 
running on the partitioned daemons.

Some of the academic papers linked from the spread.org website documentation page have a 
detailed explanation of the model in which Spread supports network partitions.

Cheers,

Jonathan

On Fri, Nov 09, 2007 at 10:42:38AM -0800, Alex Jacobson wrote:
> I've read some of the documentation but am not clear on how spread 
> handles actual partitions.  If you have 3 clusters and one of them can't 
> see the other two, does the separated one believe it is down or that the 
> other two are?
> 
> -Alex-
> 
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users

-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------