Reduce (Map-Reduce) support for entry processor

[enesakar]

No description provided.

[pveentjer]

The current approach is dangerous because the resulting map is fully kept in memory.

Christopher also has been working on a map/reduce solution. I think it would be a good idea to look at that. We should prevent that we get a lot of similar but a little bit different solutions in the system. I would rather have a single solution that is configurable.

[pveentjer]

The thing I'm also worried about using the entry processor, is that it will run on top of the partition (operation) thread. If there for example are a million entries in a given partition, no other operations can be executed during the processing of these million map entries and this could cause not only starvation problems (to other partitions.. and not only of this particular map.. but to all distributed objects within a given member), it can also lead to uncontrollable members; partitions operations can't be executed.