Code review request: 7101658 : Backout 7082769 changes

Rob McKenna rob.mckenna at oracle.com
Mon Oct 17 22:27:56 UTC 2011


Looks good.

     -Rob

On 17/10/11 18:29, Alan Bateman wrote:
> Seán Coffey wrote:
>>
>> bug ID : http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7101658
>> (bug report not visible yet)
>>
>> changes for 7082769 fix have led to behavioral changes which can lead 
>> to native file descriptor exhaustion. Basically - file descriptors 
>> are not released until all streams referencing them are closed.
>>
>> For the short term, we should backout the 7082769 changes until a 
>> full solution which doesn't cause issue in terms of compatibility is 
>> found.
>>
>> I'm hoping to get this change into 7u2.
>>
>> webrev : http://cr.openjdk.java.net/~coffeys/webrev.7101658/
> The anti-delta looks fine.
>
> [ As background I should explain that I asked Seán off-list to back 
> out these changes because it means that closing a stream that is 
> sharing a file descriptor with another stream no longer closes the 
> underlying file as existing code expects. In order to fix 7082769 
> properly it will likely require that FileDescriptor be changed to keep 
> a reference to each of the closeables (stream and channels) that use 
> it. That way when a stream is closed then it will cause all stream and 
> channels using the file descriptor to be closed. This should be fixed 
> in 8 first and bake for a while before considering 7u. In the 
> mean-time 7u needs to be resorted to fix the current regression ].
>
> -Alan.



More information about the core-libs-dev mailing list