tomcat-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Konstantin Preißer <kpreis...@apache.org>
Subject RE: svn commit: r1530057 - /tomcat/trunk/java/org/apache/tomcat/util/net/NioEndpoint.java
Date Mon, 07 Oct 2013 21:29:17 GMT
Mark,

> -----Original Message-----
> From: markt@apache.org [mailto:markt@apache.org]
> Sent: Monday, October 7, 2013 10:44 PM
> To: dev@tomcat.apache.org
> Subject: svn commit: r1530057 -
> /tomcat/trunk/java/org/apache/tomcat/util/net/NioEndpoint.java
> 
> Author: markt
> Date: Mon Oct  7 20:44:04 2013
> New Revision: 1530057
> 
> URL: http://svn.apache.org/r1530057
> Log:
> Fix https://issues.apache.org/bugzilla/show_bug.cgi?id=55633 for NIO.
> The Comet code that ensured that multiple threads didn't process the same
> socket when the selector indicated that a socket was ready for read and
> write pre-dated r1001698 where syncs where added to the SocketProcessor
> to achieve the same aim for Servlet 3.0 asyncs processing.
> The Comet code was re-used to handle upgraded connections.
> The upgrade code did not handle the case where a socket was registered for
> read and write but only a write event occurred. In this case the read
> registration was lost. This is the root cause of the lack of responsiveness
> observed in bug 55633.
> With the changes in r1001698, a simpler solution can be used for both
> Comet and HTTP upgrade. The new approach unregisters the socket
> operations the selector has reported ready for and then triggers a read
> and/or write as appropriate. For Comet the syncs will ensure that read and
> write aren't processed in parallel. For HTTP upgrade such parallel processing
> is permitted.

Thank you very much. I can confirm that for NIO the behavior described in the report is fixed
- Tomcat will correctly call the OnMessage and OnClose methods.

I noticed that there still seems to be an issue when the connection is aborted before the
data is read:
1) Follow steps 1-4 from bugzilla 55633
2) Instead of resuming the firefox.exe, kill the process.
3) On IE/other browser, the snakes will stand still for ~ 15 seconds, then continue to move
- but the snake from the killed firefox is still visible.


Regards,
Konstantin Preißer


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


Mime
View raw message