hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From John Sichi <jsi...@fb.com>
Subject HIVE-1538/HIVE-2344
Date Thu, 11 Aug 2011 20:21:41 GMT
I just committed the fix from Amareshwari, so after this gets pushed, it should be possible
to back out the conf changes which were applied to avoid the bug from HIVE-1538.

Read the release notes I added on HIVE-2344 and chime in with a new JIRA issue if you think
there are cases where it's important to do finer discrimination in what kinds of SELECT expressions
to allow for ppd...in general it's a cost-based optimizer problem.

As an example, consider

select * from
(select f(x,y) as z from t) s 
where z > 3;

Before HIVE-1538, there was a bug where we would push down f(x,y)>3 even when f was non-deterministic.

HIVE-1538 made that bug much more obvious.

HIVE-2344 fixes it, but also prevents the pushdown even in cases where f is deterministic.
 This is good in some cases (e.g. when f is expensive to compute and the filter selectivity
is poor), but could be bad in others (e.g. when f is something simple like a CAST and the
filter is highly selective).

JVS


Mime
View raw message