httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bill Stoddard" <stodd...@raleigh.ibm.com>
Subject Re: cvs commit: apache-2.0/src/modules/mpm/winnt winnt.c
Date Fri, 12 May 2000 14:06:14 GMT
The problem is a race condition between TransmitFile and the client side closing the
connection (I think...). Sometimes Tf does the right thing but the socket is still not
usable. I only see it at the tail end of apachebench runs when the connections are being
taken down. So no, we should not flag it as not working after the first failure.

Bill

> > From: stoddard@locus.apache.org [mailto:stoddard@locus.apache.org]
> > Sent: Thursday, May 11, 2000 7:44 PM
> >
> > stoddard    00/05/11 17:43:43
> >
> >   Modified:    src/modules/mpm/winnt winnt.c
> >   Log:
> >   Win32: Some cleanup of winnt.c. Retry AcceptEx on failure
> > in preparation for
> >   socket reuse patch. Sometimes the recycled accept socket
> > will not work; need
> >   to handle this case by closing it and obtaining a new one.
>
> Is it consistent?  Is there a pattern?
>
> I can see (based on docs) where one vendor would implement valid
> unbound accept sockets after SendDisconnect, where another wouldn't.
> Can we flag it after the very first try, and leave that state
> switch set?


Mime
View raw message