[
https://issues.apache.org/jira/browse/GIRAPH-737?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mohammad Kamrul Islam updated GIRAPH-737:
-----------------------------------------
Attachment: GIRAPH-737-3.patch
Updated with review comments.
Tested with this two commands:
1. mvn clean verify
2. mvn -Phadoop_yarn -Dhadoop.version=2.2.0 clean verify
> Giraph Application Master: move to new and stable YARN API
> ----------------------------------------------------------
>
> Key: GIRAPH-737
> URL: https://issues.apache.org/jira/browse/GIRAPH-737
> Project: Giraph
> Issue Type: New Feature
> Components: mapreduce
> Reporter: Mohammad Kamrul Islam
> Assignee: Mohammad Kamrul Islam
> Attachments: GIRAPH-737-2.patch, GIRAPH-737-3.patch,
> GIRAPH-737.WIP.patch
>
>
> Giraph was the early adopter of Hadoop YARN AM! Eli successfully wrote a
> Giraph AM based on Hadoop 2.0.x_alpha. However, in last few months, Yarn
> significantly *overhauled* its APIs and associated coding patterns. The new
> beta version is 2.1.x and I was told by Yarn-dev that current APIs will not
> change much.
> In the above circumstances, we need to substantially overhaul Giraph AM as
> well to accommodate with the new Yarn API. Moreover, in newer YARN API,
> supporting kerberos security in AM becomes easier and more transparent.
> Potential impact:
> The upcoming Girpah AM will not work with earlier alpha Hadoop versions such
> as 2.0.3. I'm not sure if anyone is using Giraph AM in production. However,
> the more prevalent way of Giraph processing (MR-based) should continue to
> work.
>
--
This message was sent by Atlassian JIRA
(v6.1#6144)
