On Thursday, April 11, 2002, at 04:45 AM, Jim Jagielski wrote:
> Roy T. Fielding wrote: >> >> If Brian needs to create a new MPM to fix this kind of problem, then >> I will remove the worker MPM from the code base as soon as the new one >> is committed. I don't believe in leaving code in httpd that is not >> robust enough for normal use. >> > > I wonder how perchild has survived all this time :) > Because all previous releases were betas. At least perchild tries to provide functionality that is different from the other MPMs. A leader-follower MPM (or any other multithreaded MPM that is simply trading off performance) has a very simple judgment value: if using it results in a server that fails to respond under high load, where some other MPM would be fine, then it should not be distributed with the core code. I'd prefer that someone fix the problem in worker. But, if they don't and some more robust alternative shows up, the other MPM is gone. The purpose of an MPM is to be robust under high loads. There is absolutely NO circumstance in which "worrying about stability of the code base" can be preferred over shipping a broken MPM. BTW, if nobody is maintaining perchild then it should not have been included in the release. It is the responsibility of the committers on the project to rip out any code that isn't being maintained. Somebody just has to have the balls to do it. If one person wants that functionality, then they can bloody well maintain it in their local tree and re-commit when it is once again maintainable. ....Roy
