[
https://issues.apache.org/jira/browse/TEZ-2924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14988760#comment-14988760
]
Siddharth Seth commented on TEZ-2924:
-------------------------------------
Looks good. Adds a dependency on 2.8.0-SNAPSHOT ?
Will the shimLoaders evolve ? i.e. if a 2.9.0 were to be released with
additional features, the 2.8 ShimLoader would be modified.
Existing deployments on 2.9 - would end up picking the old 2.8 shim. (The 2.9
feature isn't used in this case).
Log statements indicating which Shim is being used and the Hadoop version found
would be useful.
This would also help with the case where there's multiple versions of Hadoop on
the classpath.
Will there ever be a situation when multiple shims match a specific version ?
Not sure how that can be resolved though - a naming convention on the shims
maybe.
On the patch itself, some fields can be made final - initialize collapsed into
the constructor, etc. Are multiple instances of HadoopShimLoader meant to be
created within a JVM ? Otherwise this could be static (unless that gets in the
way of testing).
Annotate various interfaces with the Private annotation ?
> Framework for Hadoop shims
> --------------------------
>
> Key: TEZ-2924
> URL: https://issues.apache.org/jira/browse/TEZ-2924
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Hitesh Shah
> Assignee: Hitesh Shah
> Attachments: TEZ-2924.wip.1.patch
>
>
> As Hadoop creates a matrix of which versions support what features, Tez will
> need a shim to be cross-compatible and make use of various fixes/features as
> and when possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
