Steve Loughran commented on HADOOP-12020:

Would welcome this in Hadoop 3.1; keep test costs down.

* copy operations in rename would need to set this too
* fault injection AWS client could generate 405 exceptions to see how they map, 
and how apps would handle a translation to FNFE. What would hive & spark do, as 
well as distcp.
* troubleshooting.md needs to cover the issue

Given the file is lost. FNFE is the right outcome; error text needs to include 
details on reduce redundancy loss

> Support AWS S3 reduced redundancy storage class
> -----------------------------------------------
>                 Key: HADOOP-12020
>                 URL: https://issues.apache.org/jira/browse/HADOOP-12020
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs/s3
>    Affects Versions: 2.7.0
>         Environment: Hadoop on AWS
>            Reporter: Yann Landrin-Schweitzer
>            Priority: Major
> Amazon S3 uses, by default, the NORMAL_STORAGE class for s3 objects.
> This offers, according to Amazon's material, 99.99999999% reliability.
> For many applications, however, the 99.99% reliability offered by the 
> REDUCED_REDUNDANCY storage class is amply sufficient, and comes with a 
> significant cost saving.
> HDFS, when using the legacy s3n protocol, or the new s3a scheme, should 
> support overriding the default storage class of created s3 objects so that 
> users can take advantage of this cost benefit.
> This would require minor changes of the s3n and s3a drivers, using 
> a configuration property fs.s3n.storage.class to override the default storage 
> when desirable. 
> This override could be implemented in Jets3tNativeFileSystemStore with:
>       S3Object object = new S3Object(key);
>       ...
>       if(storageClass!=null)  object.setStorageClass(storageClass);
> It would take a more complex form in s3a, e.g. setting:
>     InitiateMultipartUploadRequest initiateMPURequest =
>         new InitiateMultipartUploadRequest(bucket, key, om);
>     if(storageClass !=null ) {
>         initiateMPURequest = 
> initiateMPURequest.withStorageClass(storageClass);
>     }
> and similar statements in various places.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org

Reply via email to