Geoffrey Young wrote:
Well, I think touching requests_this_child is not really what we'd want an API for. Since changing that value isn't really what we are after. We are looking for an API to allow clean child termination, so I would suggest more precisely to implement something like ap_mpm_child_terminate() API that could certainly be implemented by setting requests_this_child to zero. But the intent of the api is/should be child termination, allowing various MPMs to do it differently (if at all (thinking win32 mpm)).
right, that's pretty much what I meant - offer an appropriate ap_ API (like ap_mpm_child_terminate that you suggested) but have it toggle the appropriate switches to allow for an elegant process cleanup.
as for names, I'd suggest staying away from "child"-based names, since it probably doesn't make sense for mpms outside of prefork. maybe ap_process_terminate or ap_mpm_process_terminate (for process versus thread, instead of child versus parent). whatever we choose it should be intuitive in name for each mpm.
ap_mpm_process_terminate() would probably be my choice, since that's really what it's doing.
if the requests_this_child approach isn't sufficient, forget this email )
It's not quite sufficitent, as in worker, for instance, you'd need to set requests_this_child < 0 and also make MaxRequestPerChild > 0.
right, but the idea is the same - allow termination via standard apache processes that are (currently) private but already in place.
Yup.
Attached is a possible patch for worker and prefork, the other mpms are left to do.
hmm, well that's certainly a nice tidy solution :) but I thought there were problems with just sending signals? (though I admit that I got a bit lost in all the to-and-fro between you and stas here).
Well, I didn't have much time to spend on my original proof-of-concept patch, so basically, while writing this patch, I was thinking along the lines :
sending SIGTERM is the most generic way to do it, even if not quite elegant. Make it the default works-for-all default, and implement specific better alternatives for each MPM
--Geoff
-- -------------------------------------------------------------------------------- Philippe M. Chiasson m/gozer\@(apache|cpan|ectoplasm)\.org/ GPG KeyID : 88C3A5A5 http://gozer.ectoplasm.org/ F9BF E0C2 480E 7680 1AE5 3631 CB32 A107 88C3A5A5
signature.asc
Description: OpenPGP digital signature
