-Re: Hive mapper creation
Bejoy KS 2012-06-28, 19:29
If it is to control the split size and there by the number of map tasks, you just need to play with min and max split size properties.
Sent from handheld, please excuse typos.
From: Mohammad Tariq <[EMAIL PROTECTED]>
Date: Fri, 29 Jun 2012 00:55:54
To: <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Re: Hive mapper creation
Thanks a lot for the valuable response Bejoy. Actually I wanted to
know if it is possible to set the size of filesplits or the criterion
on which filesplits are created (in turn controlling the creation of
mappers) for a Hive query. For example, If I want to take 'n' lines
from a file as one split instead of taking each individual row, I can
use nlineinput format.Is it possible to do something similar at Hive's
level or do I need to look into the source code??
On Fri, Jun 29, 2012 at 12:37 AM, Bejoy KS <[EMAIL PROTECTED]> wrote:
> Hi Mohammed
> Splits are associated with MapReduce framework and not necessarily with hive. It is the data processed by a mapper. Based on your InputFormat, min and max split size properties MR framework considers hdfs blocks that a mapper should process.( It can be just one block or more if CombineFileInputFormat is used.) This choice of which all hdfs blocks forms a split is determined under the consideration of data locality. Number of mappers/map tasks created by a job is equal to the number of splits thus determined. ie one map task per split.
> Hope it is clear. Feel free to revert if you still have any queries.
> Bejoy KS
> Sent from handheld, please excuse typos.
> -----Original Message-----
> From: Mohammad Tariq <[EMAIL PROTECTED]>
> Date: Fri, 29 Jun 2012 00:29:13
> To: <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
> Reply-To: [EMAIL PROTECTED]
> Subject: Re: Hive mapper creation
> Hello Nitin, Bejoy,
> Thanks a lot for the quick response. Could you please tell me
> what is the default criterion of split creation??How the splits for a
> Hive query are created??(Pardon my ignorance).
> Mohammad Tariq
> On Fri, Jun 29, 2012 at 12:22 AM, Bejoy KS <[EMAIL PROTECTED]> wrote:
>> Hi Mohammed
>> Internally In hive the processing is done using MapReduce. So like in mapreduce the splits are calculated on job submission and a mapper is assigned per split. So a mapper ideally process a split and not a row.
>> You can store data in various formats as text, sequence files, RC files etc. No restriction just on text files.
>> Bejoy KS
>> Sent from handheld, please excuse typos.
>> -----Original Message-----
>> From: Mohammad Tariq <[EMAIL PROTECTED]>
>> Date: Fri, 29 Jun 2012 00:17:05
>> To: user<[EMAIL PROTECTED]>
>> Reply-To: [EMAIL PROTECTED]
>> Subject: Hive mapper creation
>> Hello list,
>> Since Hive tables are assumed to be of text input format, is
>> it right to assume that a mapper is created per row of a particular
>> table??Please correct me if my understanding is wrong. Also let me
>> know how mappers are created corresponding to a Hive query. Many
>> Mohammad Tariq