httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Greg Stein <gst...@lyra.org>
Subject RE: Review then commit.
Date Tue, 30 May 2000 00:19:56 GMT
Add these items to STATUS.


On Mon, 29 May 2000, William A. Rowe, Jr. wrote:

> Ok... since this is (now) the STATUS discussion thread :-)
> 
> I have what I believe are two showstoppers.
> 
> One is specific to the WinNT MPM, but others beware, they
> may suffer from the same fate!
> 
> The top pool (pconf) passed to the MPM is perm only between 
> restarts.  The WinNT MPM abused this pool to presume it would
> live for the duration of execution, and may have dropped
> some stuff in there that can't be so volitale.  I'll know
> more before I commit the services patch, but all MPM coders
> should be aware of this kind of abuse.  You can steal the
> process->pool from the rewrite_args hook, but that's the
> only pool that lives across restarts unless the MPM has
> created it's own.
> 
> The second issue is a thorough review of exit();  We are
> abusing this all over the place, in modules, in apr, and
> wherever.
> 
> The only place we should perform a die(); is in the case
> of memory allocation failure, and even that may not be
> fatal enough not to unwind the critical processes (MPMs.)
> 
> I'd like to see a well implemented ap_server_fault() and
> ap_server_quit(), the former doing an unwind of the proc
> that must not rely on memory or resource allocation, the
> other doing a clean exit.  But even at that, we shouldn't
> be quitting from everywhere.
> 
> We have ap to ap_terminate, pools to purge, cleanups to
> call, I just don't like the state of stability here.
> 
> Bill
> 

-- 
Greg Stein, http://www.lyra.org/


Mime
View raw message