CancelledKeyException in SPY memcached-2.0-pre5

[Dustin Sallings]

   Alright.  If the JVM fix isn't readily available, then I pretty  
much have to try to work around it.

   I've submitted a patched version for testing.  It's mostly a small  
impact where I'd expect this case to be detected and initiate a  
reconnection sequence for the bad descriptor.

-- 
Dustin Sallings (mobile)

On Dec 12, 2007, at 0:11, Andrew Miller <amiller at like.com> wrote:

>>
>>    This appears to be a bug in java on Linux.  It's a bit difficult  
>> for me to reproduce myself, but here's the information that's been  
>> given to me so far:
>>
>>    http://forum.java.sun.com/thread.jspa?forumID=535&threadID=5135128
>>
>>    It may be the following bug:
>>
>>    http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=2147719
>>
>>    It's claimed to be fixed in a later release, but I don't know in  
>> which version.
>>
>>    Would you be willing to work with me on confirming the problem  
>> goes away for you?
>>
>>
>
> Dustin,
>
> Thanks for the reply and those links -- that sounds like the exact  
> problem we're having.  Upgrading a minor JVM revision (i.e. within  
> the 1.5 tree) is probably feasible for us, though it'd be a tough  
> sell for me to require all the apps to go to java6 to fix this  
> problem -- We've discussed it but it'd take a release/QA cycle or  
> two. Also a lot of our developers are on macs, where there is still  
> not a supported java6 implementation.  Obviously that doesn't  
> prevent us from running java6 in production but I'd rather keep the  
> 'It worked on MY machine, it must be the JVM' chatter to a minimum  
> if possible.   :-)
>
> If there are some code based workarounds that didn't cruft up your  
> code too much, I'd be more than happy to test - I have the new  
> implementation on a branch and have a development environment I can  
> abuse for a little while.  I can also get away with upgrading the  
> JVM in that environment to see if that fixes the problem, but may  
> not be possible for me to get said JVM into production if my only  
> choice is the java6 early access build.  (Apparently the fixes are  
> not yet in jdk6u3).   I downloaded 1.6.0_10b08 (latest early access  
> build) and I'll see if I can reproduce the problem tomorrow.
>
> Thanks!
> -Andy
>
>