On Tue, Jan 06, 2009 at 12:10:25PM -0600, William Rowe wrote: > Would folks comment on Nathan's, Joe's and Stefan's work on > > https://issues.apache.org/bugzilla/show_bug.cgi?id=42829 > > and offer any comments on why this patch; > > https://issues.apache.org/bugzilla/attachment.cgi?id=22822 > > should not be applied to trunk/ and branches/2.2.x/ in time for > the next releases?
I never found the time to commit this for a bunch of reasons: 1) I never convinced myself that adding a bunch of complexity to prefork, per Stefan's patch, was the right way to address this 2) I never worked out what impact the lack of check on the poll descriptor event type (POLLERR etc) had on this, but that needs resolving too 3) I hadn't checked whether worker et al had similar issues 4) I think that a simpler solution to this problem would be to add a timeout (e.g. 30s) on the child pollset_poll() call so they are guaranteed to pop at some point, but I haven't had time to try this or work out whether that will suck in some other way. (Also, my stuff to dup2() against the listener was demonstrably useless, per the thread Jeff referenced, so that should definitely not be committed, regardless.) Regards, Joe
