reef-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Chung (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (REEF-1339) Adding IInputPartition.Cache() for data download and cache
Date Mon, 25 Apr 2016 20:15:13 GMT

    [ https://issues.apache.org/jira/browse/REEF-1339?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15256962#comment-15256962
] 

Andrew Chung commented on REEF-1339:
------------------------------------

I'm opposed to this for the following reasons:
Caching and data fetching should be able to be completely separated, and "hints" implies that
1) every {{IInputPartition}} implementation has to potentially provide a non-generic implementation,
2) documentation becomes a lot trickier, and 3) user won't know how the implementation behaves
until runtime. I've provided an alternative suggestion in [REEF-1357](https://issues.apache.org/jira/browse/REEF-1357),
[~markus.weimer] what do you think?

> Adding IInputPartition.Cache() for data download and cache
> ----------------------------------------------------------
>
>                 Key: REEF-1339
>                 URL: https://issues.apache.org/jira/browse/REEF-1339
>             Project: REEF
>          Issue Type: Task
>            Reporter: Julia
>            Assignee: Andrew Chung
>              Labels: FT
>
> Currently, in FileSystemInputPartition, data downloading is implemented in Initilaize()
and called from GetPartitionHandle. It doesn't give client a flexibility to decide when to
download data. Besides, if client wants to cache data in advance, they need to call GetPartitionHandle()
and iterate the data. 
> We would like to expose a new API Cache() in IInputPartition which performs data download
to RAM, SSD, HDD, etc based on client's configuration. 
> The method should be called in ContextStartHandler  in IMRU scenarios. 



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

Mime
View raw message