On Thu, 24 Feb 2011 10:14:45 +0200 Or Gerlitz <ogerlitz@xxxxxxxxxxxx> wrote: > FUJITA Tomonori wrote: > > The description looks fine but it would be better to put the fd to > > epoll (that is, using tgt_event_add)? > > yes, tgt_event_add would be more elegant way to handle that. > > One thing which exists with the current patch but wouldn't be possible > by just using tgt_event_add is --limiting-- the time tgt is willing > to allow the cluster management software to actually return something > on the redirect callback invocation. > > So my plan is to use tgt_event_add so the redirect pipe fd is added to > the epoll set, and also set a limit on the time tgt will be willing to > wait for that reply by using the new timer-based work scheduling. > > so the patch would like > > 1. instead of the possibly endless waitpid code > --------------------------------------- > add_work(child pipe expiry handler) > tgt_event_add(child pipe fd, child pipe read handler) > > > 2. child pipe expiry handler > ------------------------ > terminate child pid > tgt_event_del(child pipe fd) > return err on login request > > 3. child pipe read handler > ----------------------- > del_work() > waitpid on child pid > read from pipe etc > return redirect info to login request Sounds good. > I guess we're fine from possible locking issues between del_work to > tgt_event_del as tgt is single threaded? Yeah. We need to think about it for the multi-thread design in the near future. -- To unsubscribe from this list: send the line "unsubscribe stgt" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
