[
https://issues.apache.org/jira/browse/PHOENIX-876?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gabriel Reid updated PHOENIX-876:
---------------------------------
Attachment: PHOENIX-876.patch
Patch to remove the mapreduce assembly. I've tested this on a real cluster (3.0
branch, HBase 1) and it works.
The most important change that this makes is that it's no longer possible to
run the client jar as a standalone jarfile, as there's no longer a mainClass
value set in the manifest of that jar. However, this wasn't necessary anyhow
when running it via psql.py, so I assume that's ok.
With this change, the mapreduce import can be started up as
{code}./bin/hadoop jar phoenix-3.0.0-incubating-SNAPSHOT-client.jar
org.apache.phoenix.mapreduce.CsvBulkLoadTool -t mytab -i /tmp/data.csv{code}
In other words, starting the MR job also requires supplying a class name on the
commandline (which is also different than how it was with a specific MR
assembly).
BTW, the Hadoop client jars are not necessary in order to run the MR job --
that's what running it with "hadoop jar" provides you with.
> Remove phoenix-*-mapreduce.jar if not necessary
> -----------------------------------------------
>
> Key: PHOENIX-876
> URL: https://issues.apache.org/jira/browse/PHOENIX-876
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 3.0.0, 4.0.0
> Reporter: James Taylor
> Assignee: Gabriel Reid
> Attachments: PHOENIX-876.patch
>
>
> Do we need the
> phoenix-assembly/target/phoenix-3.0.0-incubating-SNAPSHOT-mapreduce.jar to be
> produced, as it pulls in a ton of stuff? Can our standard client jar be used
> plus the hadoop.jar instead? If we don't need it, we should update the
> pom/build stuff to not produce it.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
