[infinispan-dev] Map Reduce 2.0

[Manik Surtani]

On 31 May 2012, at 15:34, Vladimir Blagojevic wrote:

> On 12-05-31 10:06 AM, Manik Surtani wrote:
>> Some thoughts:
>> 
>> 1. Failover policy.
>> 
>> * Does this rely on being run in a CDI environment?  (RandomNodeTaskFailoverPolicy has a DistributedExecutorService @Injected).  If so, then -1.  While it *should work* in a CDI environment, it shouldn't *require* CDI.
> I agree. It was more of a wishful thinking on my part, it will not be required.

We should update the interface then to indicate how you intend to make the ExecutorService made available to implementations.  :)

> 
>> 
>> * Do you plan to include any canned failover policies?  If so, what are these?
> 
> Yes, at least random and never failover policy.
>> 
>> * Shouldn't  DistributedTaskFailoverPolicy be parameterised to return the same type as DistributedFuture.get() rather than an untyped Object?
> I would love to make it but how can we do it? DefaultExecutorService is not typed and method of DistributedTaskFailoverPolicy is not either.

Can DTFP not be typed?  And even our sub-interface of ExecutorService?

> 
>> 
>> 2.  Task mapping policy
>> 
>> * So the existing policies of "execute everywhere" and "execute on data owner" (and possibly a new one, "execute on PRIMARY data owner") would all implement this policy interface?
>> 
> Yes, for sure.

--
Manik Surtani
manik at jboss.org
twitter.com/maniksurtani

Lead, Infinispan
http://www.infinispan.org