hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "John Zhuge (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (MAPREDUCE-6572) Improved incremental data copy of distcp for truncated file
Date Tue, 12 Jan 2016 22:30:39 GMT

     [ https://issues.apache.org/jira/browse/MAPREDUCE-6572?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

John Zhuge updated MAPREDUCE-6572:
----------------------------------
    Description: 
MAPREDUCE-5899 improves distcp by supporting incremental data copy. That is, if a file is
only appended since it was copied last time, only new data need to be copied. 

This improvement was done before HDFS truncate feature (HDFS-3107) was implemented. Since
we support truncate, if a large file is truncated a little bit, the whole file will still
need to be copied, even with the solution of MAPREDUCE-5899.

Creating this jira to improve the situation, by possibly remembering the smallest truncated
size, so there is chance to only append from that size on.

HDFS tasks
* Add field *minTruncateLength* to *FileWithSnapshotFeature*, default to file size.
* Whenever a file is truncated, update the field.
* Pass the field to HDFS client in MODIFY entry of *SnapshotDiffReport*.

CopyMapper tasks
* If *minTruncateLength* < *target_file_length*, CopyMapper should perform a *truncate(target_file_length
- minTruncateLength)* operation.
* If *minTruncateLength* < *source_file_length*, CopyMapper should perform an *append(source_file_length
- minTruncateLength)* operation.

In some cases, CopyMapper may perform a *truncate* followed by an *append*.

  was:
MAPREDUCE-5899 improves distcp by supporting incremental data copy. That is, if a file is
only appended since it was copied last time, only new data need to be copied. 

This improvement was done before HDFS truncate feature (HDFS-3107) was implemented. Since
we support truncate, if a large file is truncated a little bit, the whole file will still
need to be copied, even with the solution of MAPREDUCE-5899.

Creating this jira to improve the situation, by possibly remembering the smallest truncated
size, so there is chance to only append from that size on.

Thanks.




> Improved incremental data copy of distcp for truncated file
> -----------------------------------------------------------
>
>                 Key: MAPREDUCE-6572
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-6572
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>            Reporter: Yongjun Zhang
>            Assignee: John Zhuge
>
> MAPREDUCE-5899 improves distcp by supporting incremental data copy. That is, if a file
is only appended since it was copied last time, only new data need to be copied. 
> This improvement was done before HDFS truncate feature (HDFS-3107) was implemented. Since
we support truncate, if a large file is truncated a little bit, the whole file will still
need to be copied, even with the solution of MAPREDUCE-5899.
> Creating this jira to improve the situation, by possibly remembering the smallest truncated
size, so there is chance to only append from that size on.
> HDFS tasks
> * Add field *minTruncateLength* to *FileWithSnapshotFeature*, default to file size.
> * Whenever a file is truncated, update the field.
> * Pass the field to HDFS client in MODIFY entry of *SnapshotDiffReport*.
> CopyMapper tasks
> * If *minTruncateLength* < *target_file_length*, CopyMapper should perform a *truncate(target_file_length
- minTruncateLength)* operation.
> * If *minTruncateLength* < *source_file_length*, CopyMapper should perform an *append(source_file_length
- minTruncateLength)* operation.
> In some cases, CopyMapper may perform a *truncate* followed by an *append*.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message