[infinispan-dev] Preliminary implementation for ISPN-244 and improvement suggestions.

[Sanne Grinovero]

Hi Galder,
this seems great, I plan to try out soon implementing the ones for
Lucene for a little extra performance boost.
Just to double-check, no annotations are required on the externalized
classes right? That would be a showstopper for example in the JIRA
case as I don't have control over all marshalled source code.

The lucene-infinispan module will definitely use it, do you think
there's a way to make it as simple as possible to have the ones I'll
write registered for the end user?
Maybe in a future release, could we add an SPI style discovery to
self-register externalizers found on classpath?

And in case we have something like that in place, how will we
guarantee ID unique assignments and also consistency across the
cluster; as a starter should I reserve a range of identifiers
in core, even without introducing dependencies; maybe later we should
have a consistency check at node join to verify the same types are
registered.

Sanne


2010/12/7 Galder Zamarreño <galder at redhat.com>:
> Hi all,
>
> Re: https://jira.jboss.org/browse/ISPN-244
>
> This is trunk now and I've written preliminary documentation for it in http://community.jboss.org/docs/DOC-16198. Have a read through it and let me know what you think.
>
> Now, there's a couple of improvements I'm considering doing even for alpha1:
>
> 1. Internally, Infinispan has been mapping Externalizer classes and the types they marshall via @Marshallable interface. I don't think this annotation is really usable as it is for users cos there can be situations where users might not be able to modify the class that they wanna serialize. However, they for sure should be able to modify the Externalizer implementation classes, so I was wondering of converting this annotation into an annotation that an Externalizer class takes and that specifies the type of class that it externalizes and its id. Example:
>
> class Person {
> ...
> }
>
> @Marshalls(typeClass=Person.class, id=768)
> class PersonExternalizer Implements Externalizer<Person> {
> ...
> }
>
> This could be a mechanism that could be used in equal manner by both internal and user defined externalizers. We don't wanna be doing annotation scanning, so in the same fashion that for internal externalizers we list the type classes that can be extermalized, this could lead to a user configuration simplification, i.e:
>
> <global>
>  <serialization>
>    <externalizers>
>      <externalizer class="PersonExternalizer" />
>    </externalizers>
>  </serialization>
> </global>
>
> On startup we'd load the externalizers defined in configuration and search for the @Marshalls annotation which would give us the rest of info which is: typeClass and id. Thoughts?? I think this would make internal and user defined externalizers more homogeneous when it comes to their definition and would simplify configuration.
>
> Chaging @Marshallable to @Marshalls should be easy to do particularly cos this annotation was only used for internal purpouses.
>
> 2. The second improvement I had in mind is around programmatic configuration. As you can see in the docu, configuring programmatically user defined externalizers requires quite a bit of code. I was thinking of adding a helper method in global configuration that would hide all of that away. If you take in account my recommendations for the previous point, that would be limited to something along the lines of:
>
> globalCfg.addExternalizer(PersonExternalizer.class);
>
> Where the method would have a signature like: public void addExternalizer(Class<? externds Externalizer> externalizerClass);
>
> We could also have an alternative that uses reflection to load the class and add it: public void addExternalizer(String externalizerClassName);
>
> Cheers,
> --
> Galder Zamarreño
> Sr. Software Engineer
> Infinispan, JBoss Cache
>
>
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>