> On Jan 11, 2017, at 12:28 PM, Yann Ylavic <ylavic....@gmail.com> wrote: > > On Wed, Jan 11, 2017 at 6:17 PM, Jim Jagielski <j...@jagunet.com> wrote: >> >>> On Jan 11, 2017, at 12:12 PM, Joe Orton <jor...@redhat.com> wrote: >>> >>>> The only reason why I can see why the orig idea to use s->process->pool >>>> was to make watchdog run independent of any restarts of httpd >>>> itself... that is, a truly independent watchdog. But that would >>>> imply that you can't reconfig watchdog on restarts which >>>> doesn't seem quite right... > > Some (third party) modules may depend on this, no? > >>> >>> Since the callbacks registered with the watchdog come from modules, and >>> modules have lifetime of pconf, it seems right to me to use pconf. (I >>> would guess that without this fix, if at restart you unloaded a module >>> which had registered a watchdog, it would segfault the server?) > > Modules could "mimic" mod_watchdog behaviour (by using process->pool > and recover their context at restart)... > > But I don't know any such module, just noting about the behaviour > change if we use pconf here. >
I don't think the intent was such... looking at everything else, the design of watchdog is consistent with a typical pconf/clean on restarts and the initial use of s->process->pool was a bug.