[infinispan-dev] Switch to AffinityPartitioner by default, or even enforce it

[Dan Berindei]

First, a question: doesn't query already configure its internal caches
to use the AffinityPartitioner? Does it need the application caches to
have AffinityPartitioner enabled, too?

I'm ok with making AffinityPartitioner the default key partitioner:
the less configuration the user has to change to make things work, the
better. I'm also ok with changing it to use delegation, to make it
easy for user to compose it with another key partitioner. We should
just run some benchmarks first to check it doesn't affect performance
for repl mode reads.

But I feel users should be able to disable it if they want to, so I
wouldn't always wrap user partitioners. Also, usually less magic ==
better. Query can validate the configuration and warn the user if the
configuration doesn't have AffinityPartitioner in place.

Cheers
Dan


On Thu, Aug 18, 2016 at 3:43 PM, Sanne Grinovero <sanne at infinispan.org> wrote:
> Hi all,
>
> I would like for Infinispan to use the AffinityPartitioner by default,
> replacing the HashFunctionPartitioner.
>
> This should have zero impact as AffinityPartitioner extends
> HashFunctionPartitioner and only changes semantics when a key is
> implementing AffinityTaggedKey.
>
> So the difference would be that for those using AffinityTaggedKey,
> these would work out of the box without having to change the
> configuration as well.
>
> WDYT?
>
> As a further but separate improvement, I'd change the
> AffinityPartitioner to use delegation to the HashFunctionPartitioner
> instead of extending it, and always wrap any user configured
> partitioner with the AffinityPartitioner.
> This would ensure that AffinityTaggedKey work as expectedf even when
> people experiment with other partitioners, and avoids some complexity
> of configuring Infinispan:
>
> "oh, I didn't know that changing hashing function would break feature [x]..."
>
> For the record, we're using AffinityTaggedKey in our evil plans to
> improve query performance, but it's also sparked interest as a very
> useful function by the HACEP team.
>
> I have patches ready...
>
> Thanks,
> Sanne
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev