On Thu, Dec 1, 2016 at 10:06 AM Eric Covener <cove...@gmail.com> wrote:
> Trying to look at the event stuff and the strict stuff. > > If any more experienced event folks are lurking, help on the review > would be great (And thanks to sf!) > This lurker isn't up to doing a full on line-by-line review. But I will take the bait and make some observations: * I am a huge sf fan. Wish he had been around in the 2.0 beta and early event development days. * What I see here http://apache.org/server-status and here https://bz.apache.org/bugzilla/show_bug.cgi?id=53555 reinforces the above. * The scoreboard "squatting" code was one of the ugliest warts I ever had the displeasure to try to band-aid. Good riddance. * Test cases rock! Hopefully we've got some that can run big workloads with both normal and very long running requests, plus MPM tuning parameters to intentionally make it thrash processes. How did squatting come about? The 2.0 threaded MPM simply took 1.3's one dimensional array data structure for the scoreboard and turned it into a two dimensional flat array. Ooooops! Design discussions of doing a more sophisticated scoreboard data structure were squashed by someone with a proprietary module which walked the scoreboard. The threaded MPM also in inherited 1.3's management of numbers of workers, and just translated it from worker processes into worker threads. Then we started hitting "scoreboard full" errors with mixtures of fast and sloooow requests, a la PR 53555, and squatting was born. The threaded MPM became the worker MPM which evolved into event, and you know the rest. * What about worker? It's got a scoreboard full / squatting problem too. I'd say either patch it too,or just phase it out. Maybe y'all have already decided on the latter. If not, it is trivial to make event behave like worker, controlled by a config directive if you feel that's necessary. I believe the code is in ap_process_async_connection. I *really* regret not doing that in the first event patch I posted to this list. Que sera, sera. Enough ranting. Greg Ames <cloak on>
