John Peacock wrote:
> Any time files are created and destroyed in rapid succession (like the
> temporary files in .svn/tmp), there is a significant risk that the
> following will happen:
The significance is 100% here :-| since all developers have the same
virus scanner installed. It's 100% reproducable with McAfee. Switch the
service off -> everything works. Switch the service on -> modified WCs
get wedged when trying to delete them.
>
> 1) Process A (svn) creates a temporary file
> 2) Process B (virus scanner) opens the file to scan it
> 3) Process A attempts to delete the temporary file
> 4) Process B blocks #3
I wonder, why this only happens, if the WC contains modifications. If
the WC is unmodified, it is never wedged.
> There is already code in svn to retry certain operations under Win32
> because of this. The CVSNT project has also come up against this
> behavior and made similar changes. Ultimately, it is poor design at the
> NTFS level, but all we can do is work around it.
I wonder, if the workaround in subversion only works around the problem,
when the WC is not modified. Maybe there is a small oversight, that in
some cases svn does not retry this operation.
> Depending on the load
> on the system, the only acceptable way to completely prevent the problem
> is to completely uninstall the virus scanner from the server (subversion
Client in my case. Our Servers run under Linux. ;-)
Cheers & thanks
-Lübbe
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Dec 12 15:46:56 2003