[Mondrian] Testing a Hive dialect

Julian Hyde jhyde at pentaho.com
Fri Feb 11 12:51:51 EST 2011

If it is difficult to fix HIVE-1922, we could possibly workaround issue by
changing Dialect.generateOrderItem. (Every dialect tends to have different
rules for how to generate an ORDER BY clause -- order by column name, order
by ordinal, order by expression, order by ordinal only when applied to a set
operation such as union, etc. -- so there's no harm having yet another
Someone also mentioned that Hive only supports ANSI join syntax 'FROM t1
JOIN t2 ON t1.x = t2.y', whereas mondrian only generates 'FROM t1, t2 WHERE
t1.x = t2.y'. Is this still an issue? We would need to fix either Hive or
Mondrian's dialect. Mondrian's dialect it probably easier.
Since other people have expressed an interest in a Hive dialect, it would be
useful if you check in what you have right now, even though there are many
failures. Send me the files and I will check them in.


From: mondrian-bounces at pentaho.org [mailto:mondrian-bounces at pentaho.org] On
Behalf Of Fu Hongwei
Sent: Friday, February 11, 2011 3:53 AM
To: 'Mondrian developer mailing list'
Subject: Re: RE: [Mondrian] Testing a Hive dialect

Hi Julian,
Thanks for the prompt reply.
The test is running on Hive 0.7.0, but a patch will be needed. I've started
a jira on Hive and will submit it soon too.
12987548> &focusedCommentId=12987548#action_12987548 
There are too many failures in the test suite.  I'm still working on it, but
I will submit the DialectTest.java and HiveDialect.java, so people
interested can work on it together.
I think the point is first to make it running then speed it up.


Fu Hongwei 

发件人: Julian Hyde 
发送时间: 2011-02-11  00:19:26 
收件人: 'Mondrian developer mailing list' 
主题: RE: [Mondrian] Testing a Hive dialect 

I am well aware of the compromises with Hive. There is certainly an
impedance mismatch between hadoop and the real-time analysis, and that is
reflected in query response time. It is still useful  to have a Hive
dialect, because as you say, Hive is improving all the time. And Pentaho is
thinking about ways to bridge the impedance mismatch.

Can please you attach your code to the jira case as a patch? I will submit
Also please attach the output of the test suite, and describe the version of
Hive you are running against. That will be a reference point for others who
are working on Hive.


From: mondrian-bounces at pentaho.org [mailto:mondrian-bounces at pentaho.org] On
Behalf Of Fu Hongwei
Sent: Thursday, February 10, 2011 1:29 AM
To: mondrian
Subject: [Mondrian] Testing a Hive dialect 

I'm new to the community. I don't know who to ask this for. 
I've written a Hive dialect for mondrian and been testing it.
The DialectTest is passed, but it's still failing other tests.
>From what I can see it's still a long way to make an integration
of Hive and Mondrian practical.
1. It's really slow. It takes 1 or 2 days to run through all the tests, 
a low end estimation. The latency is too big for most applications.
It might take some major architectural change on the Hive side to solve
this problem.
2. Hive ql is still at a rather immature stage, partly because it's not 
really intended to be a full featured relational database. There are some
 bugs too, like the join behavior is incorrect at the current trunk.
I know there is already a jira about it.
How can I submit the code? Thanks


Fu Hongwei 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.pentaho.org/pipermail/mondrian/attachments/20110211/941e1581/attachment.html 

More information about the Mondrian mailing list