[ http://issues.apache.org/jira/browse/HADOOP-259?page=all ]
Owen O'Malley updated HADOOP-259:
---------------------------------
Attachment: http-client-timeout.patch
Ok, this patch adds a timer for the map output copy threads. If they don't make
progress often enough. (Progress is defined as receiving a block of data from
the http server.) I've been testing it for the last two days and it seems to be
working, but if we go to java 1.5 we should probably use the socket timeouts.
> map output http client does not timeout
> ---------------------------------------
>
> Key: HADOOP-259
> URL: http://issues.apache.org/jira/browse/HADOOP-259
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.3
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3
> Attachments: http-client-timeout.patch
>
> The new map output http client uses java.net.URLConnection to fetch the data
> file. However under Java 1.4 there is no way to specify a timeout and it is
> set to infinite (or if not infinite at least 12 hours). This causes reduce
> tasks to get "stuck" in the "reduce > copy" phase even after the "Task failed
> to report status for 600 seconds. Killing." message.
> I will add the code in the ReduceTaskRunner to make sure that copies
> in-flight don't get stuck, but this is another point where a switch to java
> 1.5 would be helpful. Under 1.5 I could set the timeout on the connection and
> the read would timeout after the given interval and my entire change would be
> local to MapOutputLocation.copyFile.
> For now, I'll assume that we need to maintain support for 1.4 and make the
> corresponding fix, but I'm grumbling...
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
