[jira] [Commented] (REEF-290) Clean up the remoting part in Wake

["Markus Weimer (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/REEF-290?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14519511#comment-14519511 ] 

Markus Weimer commented on REEF-290:
------------------------------------

One key challenge in Wake is that some parts of it don't agree with the singleton assumption of Tang. For instance, users might want to instantiate many {{Transport}} or {{RemoteManager}} instances.

In [REEF-283], we implemented an injectable factory class for this case. In REEF, we face a similar issue: Every evaluator is represented by one instance of {{EvaluatorManager}} in the driver. Again, we have an injectable factory class that creates these. However, the {{EvaluatorManagerFactory}} uses a forked injector to make those instances. This is a nicer design than what we have in {{RemoteManagerFactory}}, as the factory doesn't have to know about all the parameters of the classes it instantiates and can "pass-through" configuration.

> Clean up the remoting part in Wake
> ----------------------------------
>
>                 Key: REEF-290
>                 URL: https://issues.apache.org/jira/browse/REEF-290
>             Project: REEF
>          Issue Type: Improvement
>          Components: Wake
>            Reporter: Byung-Gon Chun
>
> The remoting part of Wake and Tang were simultaneously developed. So, the current code base does not take advantage of Tang. We should clean up this part.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)