[
https://issues.apache.org/jira/browse/HADOOP-13826?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15687111#comment-15687111
]
Steve Loughran commented on HADOOP-13826:
-----------------------------------------
This is an interesting problem.
Now, one thing I don't understand. "rename with multipart uploads". copy
doesn't, AFAIK, use upload bandwidth; it's just a PUT with an x-copy-from
header. A call to s3a.rename() shouldn't be doing multipart uploads, just PUT
calls whose duration is {{filesize/in-s3-bandwidth}}. Can you post/attach a
stack trace of the blocked process to help clarify what's happening her?
In HADOOP-13600, I'm setting up for parallel copy in rename...I was going to
have a separate thread pool there just to allow for blocking COPY commands to
not block, but sharing the same xfer manager. I think I'd need to hook this in
somehow. I am cautious there of creating too many thread pools up front or
support unbounded threadpools, as that has led to memory problems already; I'd
been planning to have a v. low minimum thread count (1?), maxing out something
else < the threadpool size
>From a quick look at the code, codewise, tests are pretty raw, production code
>less so. You do know that slf4J supports logs of the form {{LOG.info("logging
>{}", path)}} and the strings are only built up if the log is needed? It's a
>key benefit from adopting that API —lets us avoid the isDebugEnabled clauses.
> [s3a] Deadlock possible using Amazon S3 SDK
> -------------------------------------------
>
> Key: HADOOP-13826
> URL: https://issues.apache.org/jira/browse/HADOOP-13826
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Sean Mackrory
> Attachments: HADOOP-13826.001.patch, HADOOP-13826.002.patch
>
>
> In testing HIVE-15093 we have encountered deadlocks in the s3a connector. The
> TransferManager javadocs
> (http://docs.aws.amazon.com/AWSJavaSDK/latest/javadoc/com/amazonaws/services/s3/transfer/TransferManager.html)
> explain how this is possible:
> {quote}It is not recommended to use a single threaded executor or a thread
> pool with a bounded work queue as control tasks may submit subtasks that
> can't complete until all sub tasks complete. Using an incorrectly configured
> thread pool may cause a deadlock (I.E. the work queue is filled with control
> tasks that can't finish until subtasks complete but subtasks can't execute
> because the queue is filled).{quote}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
