You are viewing a plain text version of this content. The canonical link for it is here.
Posted to cvs@httpd.apache.org by "William A. Rowe, Jr." <wr...@rowe-clan.net> on 2005/03/30 20:16:03 UTC
Re: svn commit: r159470 - in httpd/httpd/branches/2.0.x:
CHANGES STATUS include/mpm_common.h server/mpm/worker/worker.c
server/mpm_common.c
At 03:42 AM 3/30/2005, trawick@apache.org wrote:
>New Revision: 159470
>URL: http://svn.apache.org/viewcvs?view=rev&rev=159470
>
> *) Unix MPMs: Shut down the server more quickly when child processes are
> slow to exit. [Joe Orton, Jeff Trawick]
I have one concern with this backport. Have you considered
cgi processes which are database-bound or have other lengthly
cleanup operations? Some CGI processes may take a whole minute
or so to flush out the transaction they conducted. Does this
patch make it _more_ likely that the cleanups would be prematurely
terminated?
Bill
Re: svn commit: r159470 - in httpd/httpd/branches/2.0.x: CHANGES STATUS include/mpm_common.h server/mpm/worker/worker.c server/mpm_common.c
Posted by Joe Orton <jo...@redhat.com>.
On Wed, Mar 30, 2005 at 12:16:03PM -0600, William Rowe wrote:
> At 03:42 AM 3/30/2005, trawick@apache.org wrote:
> >New Revision: 159470
> >URL: http://svn.apache.org/viewcvs?view=rev&rev=159470
> >
> > *) Unix MPMs: Shut down the server more quickly when child processes are
> > slow to exit. [Joe Orton, Jeff Trawick]
>
> I have one concern with this backport. Have you considered
> cgi processes which are database-bound or have other lengthly
> cleanup operations? Some CGI processes may take a whole minute
> or so to flush out the transaction they conducted. Does this
> patch make it _more_ likely that the cleanups would be prematurely
> terminated?
This change affects only the behaviour when waiting for children at
restart and shutdown, not "cleanups" in general. In those cases, yes,
children which ignore SIGTERM will now get SIGKILLed after 8 seconds
rather than the 20 minutes or whatever the previous borked algorithm
did; that is the bug being fixed here. (This is still slower than 1.3's
~6 seconds till SIGKILL.)
joe
Re: svn commit: r159470 - in httpd/httpd/branches/2.0.x: CHANGES STATUS include/mpm_common.h server/mpm/worker/worker.c server/mpm_common.c
Posted by Jeff Trawick <tr...@gmail.com>.
On Wed, 30 Mar 2005 12:16:03 -0600, William A. Rowe, Jr.
<wr...@rowe-clan.net> wrote:
> At 03:42 AM 3/30/2005, trawick@apache.org wrote:
> >New Revision: 159470
> >URL: http://svn.apache.org/viewcvs?view=rev&rev=159470
> >
> > *) Unix MPMs: Shut down the server more quickly when child processes are
> > slow to exit. [Joe Orton, Jeff Trawick]
>
> I have one concern with this backport. Have you considered
> cgi processes which are database-bound or have other lengthly
> cleanup operations? Some CGI processes may take a whole minute
> or so to flush out the transaction they conducted. Does this
> patch make it _more_ likely that the cleanups would be prematurely
> terminated?
As Joe said: CGI processes are not affected. (On that topic, somebody
from HP had some queries about stranded CGI processes recently and we
discussed a method for cleaning those up. But I never saw a patch ;)
It seemed clear that mod_cgid at least needed to do something to clean
those up before it went away. I think your concern would be relevant
to that type of change.
Going further in the direction of trivial details: This patch didn't
change the intended sequence very much but it did avoid some major
bogosity with the prior use of a sliding timer. The timer could grow
large enough (30 or more seconds) so that even once the children had
exited, the parent would not wake up to check again for some time.