[infinispan-issues] [JBoss JIRA] Commented: (ISPN-1205) Redesign marshallers to be per named cache instance, per thread

[Galder Zamarreño (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-1205?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12619682#comment-12619682 ] 

Galder Zamarreño commented on ISPN-1205:
----------------------------------------

Noting a future improvement that is not likely to go in this version because it'd require a fairly big restructuring:

Enable a multiplexing marshaller architecture where the cache name is marshalled before any CacheRpcCommand which then can be read by the unmarshalling side and select the unmarshaller to unmarshall the rest of the command. Currently the cache name is embedded within the command's parameters, which means that it is marshalled by ReplicableCommandExternalizer, if the cache name is amongst the parameters. At the moment the cache name is embedded far enough in the marshalling chain that is not easy to carry on reading the stream with a different marshaller.

If I resolve this JIRA and this is not in, I'll create a JIRA for future improvements in this area. To clarify, the aim of this JIRA is to solve the use case where lazy deserialization can be done using the cache loader provided by the user. This, and the fact that a cache can now be associated with a classloader means that we need a marshaller per cache.

> Redesign marshallers to be per named cache instance, per thread
> ---------------------------------------------------------------
>
>                 Key: ISPN-1205
>                 URL: https://issues.jboss.org/browse/ISPN-1205
>             Project: Infinispan
>          Issue Type: Feature Request
>    Affects Versions: 5.0.0.CR6
>            Reporter: Pete Muir
>            Assignee: Galder Zamarreño
>             Fix For: 5.1.0.BETA1
>
>
> Since we have switched to explicit classloaders in Infinispan 5.0, we can now pass the correct classloader to use into JBoss Marshalling, rather than rely on the TCCL. However at the moment marshallers are per thread, not per named cache instance. We will need to change the way marshallers are created to make a marshaller per-thread-per-named-cache-instance. This will also require investigation of the best way of propagating this through the DI system. Dan has previously suggested using the InvocationContext

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira