Seems to be fixed in 0.6. Here's what I got:

test.q:
alter table tmp_pyang_t ADD PARTITION (ds='2') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='3') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='4') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='5') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='6') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='7') LOCATION '/tmp/blah2';
alter table tmp_pyang_t ADD PARTITION (ds='8') LOCATION '/tmp/blah2';


Hive history file=/tmp/pyang/hive_job_log_pyang_201004081410_378771152.txt
OK
Time taken: 4.101 seconds
OK
Time taken: 0.558 seconds
OK
Time taken: 0.453 seconds
OK
Time taken: 0.416 seconds
OK
Time taken: 0.378 seconds
OK
Time taken: 0.457 seconds
OK
Time taken: 0.454 seconds


Can you the stack trace in /tmp/<username>/hive.log?

-----Original Message-----
From: Prasad Chakka [mailto:[email protected]] 
Sent: Thursday, April 08, 2010 1:03 PM
To: [email protected]
Subject: Re: enough alter tables in the same .q file eventually fail

There was a bug that got fixed where each request was creating a separate 
metastore client. That could be it or something similar that hasn't gotten 
fixed.

On Apr 8, 2010, at 11:47 AM, Edward Capriolo wrote:

> Hive 5.0 mysql as metastore backend. Using external tables with location for 
> partitions...
> 
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000843') LOCATION 'hit_date=20100329/mid=000843';
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000844') LOCATION 'hit_date=20100329/mid=000844';
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000849') LOCATION 'hit_date=20100329/mid=000849';
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000850') LOCATION 'hit_date=20100329/mid=000850';
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000851') LOCATION 'hit_date=20100329/mid=000851';
> alter table XXXX_action ADD PARTITION (hit_date = '20100329' , mid = 
> '000852') LOCATION 'hit_date=20100329/mid=000852';
> 
> Eventually this fails after a number of entries.
> 
> Time taken: 0.159 seconds
> OK
> Time taken: 0.17 seconds
> OK
> Time taken: 0.241 seconds
> FAILED: Error in metadata: Unable to fetch table XXXXX_action
> FAILED: Execution Error, return code 1 from 
> org.apache.hadoop.hive.ql.exec.DDLTask
> 
> Restarting the process after removing the already added tables works until it 
> breaks again. Anyone ever dealt with this?
> 
> Doing one hive -e per table always works but takes a lot longer ...3 seconds 
> a partition rather then ~.5 seconds.
> 
> 

Reply via email to