[ 
https://issues.apache.org/jira/browse/HADOOP-16644?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16947209#comment-16947209
 ] 

Siddharth Seth commented on HADOOP-16644:
-----------------------------------------

Looks like a PUTRequest gives back the modification time, a multipart upload 
does not. Given a multipart upload is likely a long operation anyway - a HEAD 
request following a MultiPartComplete call likely doesn't add a large 
percentage to the operation time (only is S3Guard enabled). For a direct PUT - 
we have the data anyway. Will definitely make me happy to avoid writing to DDB 
during a getSTatus operation.

Using S3 for resource localization - that's got at least one issue which I'm 
aware of. Need to test this, and then file a YARN jira. Essentially - I suspect 
the localizer does not use the JobClient config - so any credentials there will 
not be available to YARN for localization (e.g. client sets up access_key and 
secret_key in config).

> Intermittent failure of ITestS3ATerasortOnS3A: timestamp differences
> --------------------------------------------------------------------
>
>                 Key: HADOOP-16644
>                 URL: https://issues.apache.org/jira/browse/HADOOP-16644
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs/s3, test
>    Affects Versions: 3.3.0
>         Environment: -Dparallel-tests -DtestsThreadCount=8 
> -Dfailsafe.runOrder=balanced -Ds3guard -Ddynamo -Dscale
> h2. Hypothesis:
> the timestamp of the source file is being picked up from S3Guard, but when 
> the NM does a getFileStatus call, a HEAD check is made -and this (due to the 
> overloaded test system) is out of sync with the listing. S3Guard is updated, 
> the corrected date returned and the localisation fails.
>            Reporter: Steve Loughran
>            Priority: Major
>
> Terasort of directory committer failing in resource localisaton -the 
> partitions.lst file has a different TS from that expected
> Happens under loaded integration tests (threads = 8; not standalone); 
> non-auth s3guard
> {code}
> 2019-10-08 11:50:29,774 [IPC Server handler 4 on 55983] WARN  
> localizer.ResourceLocalizationService 
> (ResourceLocalizationService.java:processHeartbeat(1150)) - { 
> s3a://hwdev-steve-ireland-new/terasort-directory/sortout/_partition.lst, 
> 1570531828143, FILE, null } failed: Resource 
> s3a://hwdev-steve-ireland-new/terasort-directory/sortout/_partition.lst 
> changed on src filesystem (expected 1570531828143, was 1570531828000
> java.io.IOException: Resource 
> s3a://hwdev-steve-ireland-new/terasort-directory/sortout/_partition.lst 
> changed on src filesystem (expected 1570531828143, was 1570531828000
> {code}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
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