Meeting date:  Sept 13, 2010

Location:  Cloudera Palo Alto office

Attendees:  http://www.meetup.com/Hive-Contributors-Group/calendar/14689507/

Carl Steinbach gave a status update on the 0.6 release.  Since plans for 
documentation migration have been deferred to the next release, the only 
remaining issues are completion of the CREATE DATABASE feature (HIVE-675), 
metastore VARCHAR precision widening (HIVE-1364), and metastore upgrade scripts 
(HIVE-1427).  HIVE-675 has already been committed to trunk and the backport for 
0.6 is underway.  Carl is still working on companion feature HIVE-1517, but 
unless it is done by Sept 17, we'll drop it from the 0.6 release.  Once a build 
containing the remaining features passes all unit tests, we'll post a release 
candidate and vote on it (no additional acceptance testing is planned for this 
release).

Next, HIVE-1546 (making semantic analysis pluggable) was discussed.  The Howl 
team gave an overview of their use case for reusing Hive DDL processing within 
Howl's CLI, together with a description of the roadmap for Howl/Hive project 
relationship.  Carl raised concerns about potential dependency creep preventing 
future Hive refactoring, and Namit Jain proposed reworking the approach to 
restrict it to pre/post-analysis hooks (limiting the dependencies exposed) 
rather than full-blown analyzer pluggability+subclassing.  It was agreed that 
the hooks approach was the best course for balancing all of the concerns and 
allowing us to achieve the desired project collaboration benefits.  The Howl 
team also committed to getting their Howl checkins going into a public 
repository as soon as possible, together with setting up continuous integration 
to track the health of the combination of Hive+Howl trunks.

Next, HIVE-1609 (partition filtering metastore API) was briefly discussed, and 
it was agreed that the Howl team would move the predicate parser from JavaCC to 
ANTLR and resubmit the patch.

Finally, HIVE-1476 (metastore creating files as service user) was discussed.  
It was agreed that the approach in the proposed patch (performing HDFS 
operations on the metastore client side) was a stopgap that we don't really 
want to include in Hive.  Instead, the correct long-term solution being 
developed by Todd Lipcon is to upgrade the Thrift version used by Hive to a 
recent one containing his SASL support, and then add impersonation support to 
the metastore server.  Since the Howl team's schedule does not allow them to 
wait for that work to complete and get tested, they will keep the HIVE-1476 
patch privately applied within their own branch of Hive; it will not be 
committed on Hive trunk.  Once they are able to move over to the long-term 
solution, the stopgap can be discarded.  (In the meantime, we need to work 
together to minimize the merge-to-branch impact as the metastore code continues 
to change on trunk.)

The October meetup will be at Facebook HQ in Palo Alto.

Reply via email to