geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From d..@geronimo.apache.org
Subject [jira] Commented: (GERONIMO-288) NIO Network code might send unordered messages
Date Tue, 24 Aug 2004 18:34:21 GMT
The following comment has been added to this issue:

     Author: Alan Cabrera
    Created: Tue, 24 Aug 2004 11:33 AM
       Body:
It's not clear to me that there is a problem.  First off, each event scheduled by the SelectorManager
handles a specific READ or WRITE events.  A READ event will not perform WRITE processing.
 Even if it did, the buffers are associated with the socket, not the event, so that if a READ
event did push out data, it would be in order; when the WRITE event finally executed, it would
fall out, doing nothing.

Maybe I'm missing something.  Could you provide a thread trace of the problem?


Regards,
Alan
---------------------------------------------------------------------
View this comment:
  http://issues.apache.org/jira/browse/GERONIMO-288?page=comments#action_37437

---------------------------------------------------------------------
View the issue:
  http://issues.apache.org/jira/browse/GERONIMO-288

Here is an overview of the issue:
---------------------------------------------------------------------
        Key: GERONIMO-288
    Summary: NIO Network code might send unordered messages
       Type: Bug

     Status: Open
   Priority: Major

    Project: Apache Geronimo
 Components: 
             core

   Assignee: Alan Cabrera
   Reporter: Bulent Erdemir

    Created: Tue, 24 Aug 2004 8:15 AM
    Updated: Tue, 24 Aug 2004 11:33 AM

Description:
Hi, 
Geronimo network code might deliver messages out of order. 

To be more specific, the network code tries to write a buffer and if remaining()>0, registers
an OP_WRITE interest in order to drain the buffer contents later (when the channel is available
for write). When the server is loaded, things can get hairy and we might receive another write
event which might get scheduled to run before the OP_WRITE is processed. 

More specifically, SocketProtocol.serviceWrite reads:

 long count = socketChannel.write(sendBuffer);
            log.trace("Wrote " + count);
if (sendBuffer[i].hasRemaining()) {
                    // not all was delivered in this call setup selector
                    // so we setup to finish sending async.
                    log.trace("+OP_WRITE " + selectionKey);
                    selectorManager.addInterestOps(selectionKey, SelectionKey.OP_WRITE);

                    return;
                }

Since there's no synchronization, in a situation where the selector returns with OP_WRITE
(in SelectorManager) but not yet finds the chance to process the event, another thread loaded
with a read event might sneak in and call the above lines, the result of which would be sending
data out of order. 

Regards,
Bulent Erdemir


---------------------------------------------------------------------
JIRA INFORMATION:
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

If you want more information on JIRA, or have a bug to report see:
   http://www.atlassian.com/software/jira


Mime
View raw message