[ 
https://issues.apache.org/jira/browse/HADOOP-13974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16293366#comment-16293366
 ] 

Aaron Fabbri commented on HADOOP-13974:
---------------------------------------

Ok, I think this is ready. I did more testing and this really does look like 
occasional consistency issues listing the multipart uploads.  The two 
eventually() wrappers I added have worked for %100 of my tests.

Interesting passage from the AWS Multipart Upload docs:
{quote}
Only use the returned listing for verification. You should not use the result 
of this listing when sending a complete multipart upload request. Instead, 
maintain your own list of the part numbers you specified when uploading parts 
and the corresponding ETag values that Amazon S3 returns.
{quote}

Seems to imply MPU upload listing does suffer from eventual consistency. I'm 
guessing their implementation of the MPU complete operation just uses GET 
internally, or has some retries and/or cache invalidations to make it 
"reliable".

> S3a CLI to support list/purge of pending multipart commits
> ----------------------------------------------------------
>
>                 Key: HADOOP-13974
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13974
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs/s3
>    Affects Versions: 3.0.0-beta1
>            Reporter: Steve Loughran
>            Assignee: Aaron Fabbri
>         Attachments: HADOOP-13974.001.patch, HADOOP-13974.002.patch, 
> HADOOP-13974.003.patch, HADOOP-13974.004.patch, HADOOP-13974.005.patch, 
> HADOOP-13974.006.patch
>
>
> The S3A CLI will need to be able to list and delete pending multipart 
> commits. 
> We can do the cleanup already via fs.s3a properties. The CLI will let scripts 
> stat for outstanding data (have a different exit code) and permit batch jobs 
> to explicitly trigger cleanups.
> This will become critical with the multipart committer, as there's a 
> significantly higher likelihood of commits remaining outstanding.
> We may also want to be able to enumerate/cancel all pending commits in the FS 
> tree



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to