You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@tez.apache.org by "Sangjin Lee (JIRA)" <ji...@apache.org> on 2016/07/15 16:04:20 UTC

[jira] [Commented] (TEZ-3338) Support classloader isolation

    [ https://issues.apache.org/jira/browse/TEZ-3338?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15379611#comment-15379611 ] 

Sangjin Lee commented on TEZ-3338:
----------------------------------

One high level decision we need to make is this. Do you desire isolation of tez + user code from the hadoop libraries or user code from the tez + hadoop libraries?

If it is the former, it would mean that the biggest concern is tez (and its dependencies) having collisions with hadoop's dependencies. Users of tez would still need to ensure they harmonize their dependencies with tez.

On the other hand, if it is the latter, the biggest concern would be user code (and its dependencies) having collisions with tez and hadoop's dependencies. Tez would still need to ensure it harmonizes its dependencies with hadoop.

Doing both would be a much more complex behavior/feature, and we would probably need to choose one over the other. IMO, the latter (separating the user code) might make more sense, but I'd like to hear what tez folks think.

> Support classloader isolation
> -----------------------------
>
>                 Key: TEZ-3338
>                 URL: https://issues.apache.org/jira/browse/TEZ-3338
>             Project: Apache Tez
>          Issue Type: Improvement
>            Reporter: Ming Ma
>
> HADOOP-10893 and MAPREDUCE-1700 provide classloader isolation at both client side and container side for MR. We should add the same support for Tez. Given we use hadoop command to launch Tez, it appears the client side has been taken care of. Only the container side support is needed.



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