[
https://issues.apache.org/jira/browse/HIVE-74?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zheng Shao updated HIVE-74:
---------------------------
Description:
There are cases when the input to a Hive job are thousands of small files. In
this case, there is a mapper for each file. Most of the overhead for spawning
all these mappers can be avoided if Hive used CombineFileInputFormat introduced
via HADOOP-4565
Options to control this behavior:
{code}
hive.input.format (org.apache.hadoop.hive.ql.io.CombineHiveInputFormat
(default, if empty), or org.apache.hadoop.hive.ql.io.HiveInputFormat)
mapred.min.split.size.per.node (the minimum bytes of data to create a
node-local partition, otherwise the data will combine to rack level. default:0)
mapred.min.split.size.per.rack (the minimum bytes of data to create a
rack-local partition, otherwise the data will combine to global level.
default:0)
mapred.max.split.size (the max size of each split, will be exceeded because we
stop accumulating *after* reaching it, instead of before)
{code}
The 3 numbers above must be in non-descending order.
was:There are cases when the input to a Hive job are thousands of small
files. In this case, there is a mapper for each file. Most of the overhead for
spawning all these mappers can be avoided if Hive used CombineFileInputFormat
introduced via HADOOP-4565
> Hive can use CombineFileInputFormat for when the input are many small files
> ---------------------------------------------------------------------------
>
> Key: HIVE-74
> URL: https://issues.apache.org/jira/browse/HIVE-74
> Project: Hadoop Hive
> Issue Type: Improvement
> Components: Query Processor
> Reporter: dhruba borthakur
> Assignee: Namit Jain
> Fix For: 0.5.0
>
> Attachments: hive.74.1.patch, hive.74.2.patch,
> hiveCombineSplit.patch, hiveCombineSplit.patch, hiveCombineSplit2.patch
>
>
> There are cases when the input to a Hive job are thousands of small files. In
> this case, there is a mapper for each file. Most of the overhead for spawning
> all these mappers can be avoided if Hive used CombineFileInputFormat
> introduced via HADOOP-4565
> Options to control this behavior:
> {code}
> hive.input.format (org.apache.hadoop.hive.ql.io.CombineHiveInputFormat
> (default, if empty), or org.apache.hadoop.hive.ql.io.HiveInputFormat)
> mapred.min.split.size.per.node (the minimum bytes of data to create a
> node-local partition, otherwise the data will combine to rack level.
> default:0)
> mapred.min.split.size.per.rack (the minimum bytes of data to create a
> rack-local partition, otherwise the data will combine to global level.
> default:0)
> mapred.max.split.size (the max size of each split, will be exceeded because
> we stop accumulating *after* reaching it, instead of before)
> {code}
> The 3 numbers above must be in non-descending order.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
