Robert Kanter created HIVE-8059:
-----------------------------------

             Summary: MiniHS2 improvements
                 Key: HIVE-8059
                 URL: https://issues.apache.org/jira/browse/HIVE-8059
             Project: Hive
          Issue Type: Improvement
          Components: HiveServer2, Testing Infrastructure
    Affects Versions: 0.13.1
            Reporter: Robert Kanter


In Oozie, we recently added a Hive 2 action that talks to Hive Server 2.  With 
actions, we typically like to have unit tests that actually run some simple job 
with that action type.  For the Hive 2 action, this required using the MiniHS2, 
but we ran into a few difficulties:
# The artifact containing MiniHS2 ("hive-it-unit") isn't published to the maven 
repo.  Luckily, there were only 2 classes that we needed so copying them into 
Oozie for now wasn't a big deal.
# Figuring out which system properties, config properties, etc needed to be set 
for MiniHS2 required looking at some code, the MiniHS2 pom file, and asking 
people.  Better documentation on MiniHS2 would be good.
# In Oozie, we use our own MiniHDFS and MiniMR clusters for tests.  As-is, 
MiniHS2 can either start and use it's own Mini clusters, or it can use an 
existing cluster.  However, to use an existing cluster, it will fork out to a 
shell script that we don't have.  There was no way to tell MiniHS2 to use an 
existing cluster without forking.  In the end, I had to hack the MiniHS2 
constructor to allow this (see 
[here|https://github.com/apache/oozie/commit/e332299dc86298e5f6e457f12cc089c354d8db3f#diff-cfd2359ee992ff940b4b06871eb5da24R64])
It would be great if these could be fixed so that we can get rid of our hacks 
and just use MiniHS2 directly.

Special thanks to [~brocknoland], [~szehon], and [~mohitsabharwal] for their 
help through this.



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

Reply via email to