[jira] [Commented] (HADOOP-13203) S3a: Consider reducing the number of connection aborts by setting correct length in s3 request

["Chris Nauroth (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-13203?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15313239#comment-15313239 ] 

Chris Nauroth commented on HADOOP-13203:
----------------------------------------

Rajesh, thank you for the patch.  I have to apologize.  I think this might be a regression that traces back to code review feedback I gave Steve on HADOOP-13028:

https://issues.apache.org/jira/browse/HADOOP-13028?focusedCommentId=15267729&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15267729

My thinking during the HADOOP-13028 patch was that we might want to keep on reading through the same stream, regardless of the limit of the "current" read call, so we might as well request the whole content in the HTTP request.  I was attempting to optimize away extraneous additional HTTP calls.  It appears there was an intended side effect.

I want to make sure I understand the problem here fully.  Right now, I don't think I understand why the aborts were happening.  Is it because requesting the full content, in combination with Hive's random seek workloads, left the underlying HTTP connection untouched and idle for a long time?  Then, after a while, the HTTP connection was deemed inactive/not fully consumed, it assumed there was some kind of client error, and then the whole TCP connection was shut down?

It's nice to see a comment on the {{requestedStreamLen}} calculation.  Thank you for adding that.  I might ask for some further details to be added to that comment, after I feel like I have a full understanding of the issue.

> S3a: Consider reducing the number of connection aborts by setting correct length in s3 request
> ----------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-13203
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13203
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs/s3
>            Reporter: Rajesh Balamohan
>            Assignee: Rajesh Balamohan
>            Priority: Minor
>         Attachments: HADOOP-13203-branch-2-001.patch, HADOOP-13203-branch-2-002.patch
>
>
> Currently file's "contentLength" is set as the "requestedStreamLen", when invoking S3AInputStream::reopen().  As a part of lazySeek(), sometimes the stream had to be closed and reopened. But lots of times the stream was closed with abort() causing the internal http connection to be unusable. This incurs lots of connection establishment cost in some jobs.  It would be good to set the correct value for the stream length to avoid connection aborts. 
> I will post the patch once aws tests passes in my machine.



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

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