[
https://issues.apache.org/jira/browse/TAJO-275?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13804937#comment-13804937
]
Jihoon Son commented on TAJO-275:
---------------------------------
Keuntae, thanks for the further explanation.
You are right. To implement C++ TaskRunner, QueryMaster and TaskRunner roles in
java should be distinguished first.
But, the configuration still looks little complicated.
I mean that added configurations are not intuitive.
For example, when both tajo.worker.mode.querymaster and
tajo.worker.mode.taskrunner are set to false, it is hard to suppose what will
happen.
> Separating QueryMaster and TaskRunner roles in worker
> -----------------------------------------------------
>
> Key: TAJO-275
> URL: https://issues.apache.org/jira/browse/TAJO-275
> Project: Tajo
> Issue Type: Improvement
> Reporter: Keuntae Park
> Assignee: Keuntae Park
> Attachments: TAJO-275.patch
>
>
> In C++ implementation of Tajo worker,
> I think it is better to maintain Java code of QueryMaster and implement only
> TaskRunner as C++ code, rather than implementing both QueryMaster and
> TaskRunner as C++ code.
> Accordingly, standby mode worker will have following 3 modes:
> 1) TaskRunner + QueryMaster (current implementation)
> 2) TaskRunner only (C++)
> 3) QueryMaster only (java)
> And, most workers work as C++ TaskRunner, and 1 or 2 workers work as
> separate QueryMasters (java process).
> For backward compatibility, default mode should be mode 1 (TaskRunner +
> QueryMaster)
> To achieve the goal, it needs to separate java TaskRunner and java
> QueryMaster Worker first.
> (And then, implement C++ TaskRunner)
--
This message was sent by Atlassian JIRA
(v6.1#6144)
