[Spread-users] Spread for HA rpc services
yairamir at cs.jhu.edu
Wed Aug 8 16:47:16 EDT 2007
Spread guarantees that the order of members in the membership list is
consistent across all members. You can just pick the first member and
be assured all other members will pick the same member.
In fact, if you only use AGREED messages, then the order of the message
can tell you who should serve it. For example, suppose you have 5 members
in the membership, the first member (based on the order in the member
list in the membership message) should serve all the "1 modulo 5" messages,
the second member should serve all the "2 modulo 5" messages and so
on. In such a way, there is no need for a leader - just count the messages.
Colin Meyer wrote:
> On Wed, Aug 08, 2007 at 02:14:41PM -0400, Daniel F. Savarese wrote:
>> There are a number of different ways to do what you describe with
>> Spread, but you have to select the most appropriate algorithm for
>> your application and implement it. For example, one approach is
>> for only the "leader" process to process the request, giving up
>> leadership and passing it to the next available process. Another
>> approach is to implement a work queue and have each worker process
>> ask the work queue/manager for work.
> Ok, this makes sense. Could you point me to any examples or recipes for
> this sort of implementation?
> It seems easy enough for the non-lead processes to ignore requests. What
> happens when the current lead dies unexpectedly? The pool will receive
> the disconnect message, but how do all of the processes come to a
> consensus on which process is going to be the next lead?
> Spread-users mailing list
> Spread-users at lists.spread.org
More information about the Spread-users