Hi! Andy Wingo <wi...@pobox.com> skribis:
> The web server is single-threaded and uses blocking IO (though it does > poll(2) for keepalive). As such, any slow writer or slow reader can > block the process. Using non-blocking I/O is too difficult, for now. > So, threads. > > I'd like to create a pool of threads for I/O. Some threads would pop > ports off of the "to-read" queue, read request headers and bodies, then > push the requests onto a "to-process" queue. Something (currently the > main thread) would process requests, and push them onto the "to-write" > queue. IO threads would pop data (or closures) off of the to-write > queue, and write them to clients, possibly pushing the ports back on a > "to-keepalive" queue, which the poll loop would notice and add those fds > back to the poll set. OK, I see. > I'd also like to consider creating a separate pool of threads for > computation. Obviously the size of these thread pools would be > different. We could use futures for that, I suppose, but I'd like to > also be able to stop those threads, forcefully if needed, when the web > server stops. What do you think of adding a ‘cancel’ primitive to futures? Thanks, Ludo’.
