I took this thread private in order to allow testing to be
performed without publicly posting large quantities of debugging
data.   The end result of much searching was the discovery of a
deadlock condition that was being triggered on Frank's machine
but not on any of the machines that I have access to.

With the deadlock removed, saving files from Microsoft Office
applications in Frank's environment does not result in a delayed
write error.  Given a sufficiently large and complex file
with a lossy network it is possible to still produce "delayed write"
errors.  However, I do not believe that the majority of users are
storing documents consisting of hundreds of embedded objects and
thousands of pages.

This fix will appear in 1.4.2 and 1.5.2.

Jeffrey Altman


Jeffrey Altman wrote:
> As I indicated on Tuesday, the cache manager does not serialize
> the write requests.  Examining the code used for managing the
> sleeping threads showed that the queues were being treated as LIFO
> instead of FIFO.  This encouraged starvation.  Fixing this does not
> appear to solve the problem but it does significantly improve the
> situation.  Using the default CIFS session timeout of 45 seconds I
> am now able to write files from Word up to 28mb in size.  Your
> mileage may vary.
> 
> It seems that the long term solution to this problem is going to
> require that the age of the CIFS request be taken into account
> when scheduling waking up blocked threads.
> 
> Jeffrey Altman

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to