Github user yhuai commented on the pull request:
https://github.com/apache/spark/pull/13236#issuecomment-221050299
@rdblue Thank you for looking at this!
The reason that I added the flag to disable sharing Hadoop classes is that
hadoops used by Spark and metastore client may not be binary compatible (e.g.
hadoop 2 used by spark and hadoop 1 used by the metastore client).
For HiveClientImpl, it is a wrapper of a Hive's metastore client. As long
as we can propagate user set configurations to its internal HiveConf (to make
it talk the metastore correctly), it is fine.
Regarding the classloader, we actually do not use the classloader
originally associated with the Hadoop Configuration and we always explicitly
set the classloader associated with the HiveConf created inside HiveClientImpl.
Regarding Configuration.addDefaultResource, before 2.0, we did not pass
Hadoop Configuration to HiveClientImpl (it was called ClientWrapper in Spark
1.6). Since we were not relying on Configuration.addDefaultResource, using a
Map should not change anything.
btw, I am fine to change the default value of the flag to true (sharing
hadoop classes) if you think it represents common use cases.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
