Issue with Windows IOCP not ***really*** writing data

[Alan Bateman]

Joel Buckley wrote:
> More details sent offline.  Workload that replicated the issue had 1x 
> 8k, 1x 16k, and multiple x 128k IO Sizes.  First file started (with 
> either 8k or 16k) resulted in issue detection.
Just to follow-up on this. There is/was indeed a bug here and under load 
(only duplicated me for on an 8-core system) 
AsynchronousFileChannel.write could complete immediately and invoke the 
completion handler with the incorrect number of bytes. This has been 
fixed for jdk7 b81, which is due 1/28 [1]. Thanks for bug report and the 
information needed to duplicate this.

-Alan.

[1] http://openjdk.java.net/projects/jdk7/calendar/