hc-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Scott Stanton (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HTTPCORE-334) https request to a non-responsive but alive port results in a busy loop in one thread
Date Wed, 20 Feb 2013 00:01:14 GMT

    [ https://issues.apache.org/jira/browse/HTTPCORE-334?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13581779#comment-13581779

Scott Stanton commented on HTTPCORE-334:

It looks like the culprit is in SSLIOSession.java.  When the server doesn't send any data
back to the client, the session ends up in the NEED_UNWRAP state with the appEventMask indicating
a desire to write.  updateEventMask ends up setting both read and write in the event mask
despite there being no data in the handshake write buffer.  

I'm not sure why updateEventMask is mixing in the appEventMask during handshaking.  That seems
wrong.  I have added a patch with a unittest and a possible fix for this problem.  I'm not
sure if it is the right approach, but it passes all of the tests I have tried.

> https request to a non-responsive but alive port results in a busy loop in one thread
> -------------------------------------------------------------------------------------
>                 Key: HTTPCORE-334
>                 URL: https://issues.apache.org/jira/browse/HTTPCORE-334
>             Project: HttpComponents HttpCore
>          Issue Type: Bug
>          Components: HttpCore NIO
>    Affects Versions: 4.2.4
>         Environment: Windows 8 64-bit
> JDK 1.7.0_11
>            Reporter: Scott Stanton
>         Attachments: HTTPCORE-334.patch
> Steps to reproduce:
> * Start a process that listens on a socket, accepts the connection, but does not read
any data
> * In a second processs, establish an https connection 
> At this point, one reactor thread will go into a busy loop getting writable events but
not writing any data.  As far as I can tell, this continues indefinitely until either the
client or the server closes the port.  Ideally this scenario would result in no cpu overhead
as the selector should only be waiting for read events if it doesn't have any data to write.
 Eventually it should hit an idle timeout so the client can recover from a non responsive
> Here is a stack snapshot for the busy thread:
> Agent-003 [RUNNABLE, IN_NATIVE] CPU time: 39s
> com.yourkit.probes.Table.createRow()
> com.yourkit.probes.Table.createRow(int)
> com.yourkit.probes.builtin.Sockets$SocketChannel_write_Probe.onEnter(SocketChannel)
> sun.nio.ch.SocketChannelImpl.write(ByteBuffer)
> org.apache.http.nio.reactor.ssl.SSLIOSession.sendEncryptedData()
> org.apache.http.nio.reactor.ssl.SSLIOSession.outboundTransport()
> org.apache.http.impl.nio.reactor.AbstractIODispatch.outputReady(IOSession)
> org.apache.http.impl.nio.reactor.BaseIOReactor.writable(SelectionKey)
> org.apache.http.impl.nio.reactor.AbstractIOReactor.processEvent(SelectionKey)
> org.apache.http.impl.nio.reactor.AbstractIOReactor.processEvents(Set)
> org.apache.http.impl.nio.reactor.AbstractIOReactor.execute()
> org.apache.http.impl.nio.reactor.BaseIOReactor.execute(IOEventDispatch)
> org.apache.http.impl.nio.reactor.AbstractMultiworkerIOReactor$Worker.run()
> java.lang.Thread.run()
> com.electriccloud.util.EcThread.run()

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

To unsubscribe, e-mail: dev-unsubscribe@hc.apache.org
For additional commands, e-mail: dev-help@hc.apache.org

View raw message