[
https://issues.apache.org/jira/browse/PIG-1270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13156580#comment-13156580
]
Min Zhou commented on PIG-1270:
-------------------------------
@Daniel
It improves here, but with a bug. I did the test in a 25-nodes cluster which
such script
{noformat}
A = load '/tpch/orders' USING PigStorage('\u0001') AS (o_orderkey:int,
o_custkey:int, o_orderstatus:chararray, o_totalprice:double,
o_orderdate:chararray, o_orderpriority:chararray, o_clerk:chararray,
o_shippriority:int, o_comment: chararray);
F = FOREACH A GENERATE o_orderkey;
L = LIMIT F 10;
DUMP L;
{noformat}
||case||job cost time||HDFS bytes read||Average time taken by Map tasks|Worst
performing map task
|w/o optimization| 26 sec|12,976,128|1 sec | 1 sec|
|with optimization| 24 sec|19,347,931,305|3 sec | 5 sec|
Since with your patch, the LimitOptimizer would remove LOLimit from logic plans
after set the limit to LOLoad, this would generate a map-only job. Record
number of the result would be map_num * 10, this is incorrect.
I will submit a patch soon.
> Push limit into loader
> ----------------------
>
> Key: PIG-1270
> URL: https://issues.apache.org/jira/browse/PIG-1270
> Project: Pig
> Issue Type: Bug
> Components: impl
> Affects Versions: 0.7.0
> Reporter: Daniel Dai
> Assignee: Daniel Dai
> Attachments: PIG-1270-1.patch, PIG-1270-2.patch
>
>
> We can optimize limit operation by stopping early in PigRecordReader. In
> general, we need a way to communicate between PigRecordReader and execution
> pipeline. POLimit could instruct PigRecordReader that we have already had
> enough records and stop feeding more data.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
