It is far easier to just patch the threaded MPM to use a single listener model, and
then you
can use the t0 worker slot as I posted earlier today. If we apply this patch, and
then apply
the single listener patch, then we should remove the quiescing field.
I also dislike adding a field to the scoreboard that is not going to be used by
non-threaded
MPMs.
Whatever though. +0
Ryan
On Friday 27 July 2001 15:31, Greg Ames wrote:
> This patch prevents multiple processes which are starting up from
> grabbing the same process slot in the scoreboard. Other processes may
> still share that process slot while they are quiescing (due to
> MaxRequestsPerChild, perform_idle_server_maintenance, or restarts).
>
> Since only one process per slot will be starting new threads at any
> given time, it eliminates the race conditions where multiple processes
> both see what looks like a vacant worker slot simultaneously.
>
> This works for me. I would appreciate comments, and would love to see
> other folks bang on it.
>
> Greg
----------------------------------------
Content-Type: text/plain; charset="us-ascii"; name="threaded.patch"
Content-Transfer-Encoding: 7bit
Content-Description:
----------------------------------------
--
_____________________________________________________________________________
Ryan Bloom [EMAIL PROTECTED]
Covalent Technologies [EMAIL PROTECTED]
-----------------------------------------------------------------------------
