On 20 Mar 2015, at 3:19 PM, Stefan Eissing <stefan.eiss...@greenbytes.de> wrote:
> Looking at the code in httpd, mpm_worker and mpm_event seem to create own > alloctors for the pools they use and those allocators are not protected by a > mutex. And need not, since normally a connection and all its sub-requests run > in the same thread, right? As I understand it yes, or more specifically “a connection and all its subrequests run in one thread at a time”. An idea that I have been kicking around for a while is to separate the outer connection and the inner requests with pipes. Not sure what this would mean for zero-copy of RAM, and atomic writes, but it is something worth considering (unless someone tried it and rejected it). In other words the inner requests would for all intents and purposes look like discrete connections in the classic sense, while in reality they’d be communicating with another thread that would talk to the world. Regards, Graham —