Steve, Personally, I don't mind having a watchdog to make sure that something somewhere doesn't get hung. But DefaultTimeScheduler is not designed to be reset on a fine grained basis. If we keep a watchdog (I understand that not everyone thinks we need one), we want to keep track of a limited number of things (basically just make sure that the thread doesn't hang), and we want to be able to reset it for free.
Besides, I hadn't suggested (yet) removing the watchdog without having something to back it up. But that doesn't mean we should not set the socket timeout when we have a known implementation issue. --- Noel -----Original Message----- From: Steve Short [mailto:[EMAIL PROTECTED]] Sent: Sunday, August 11, 2002 21:22 To: James Developers List Subject: RE: handling connection vs sheduler problem Agreed, I think using the socket timeout option is the way to go. I don't see any advantage in implementing a manual timeout solution. One less moving part to go wrong and less overhead because we remove the SchedulerNotifyInputStream. Regards Steve > -----Original Message----- > From: Noel J. Bergman [mailto:[EMAIL PROTECTED]] > Sent: Sunday, August 11, 2002 3:59 PM > To: James Developers List > Subject: RE: handling connection vs sheduler problem > > > Andrei, > > What issue do you have with using socket timeout? Is it not > reliable enough for you? In the case of his problem, calling > it fixes a platform specific problem, and should be neutral > for other platforms. > > I've looked at replacing the DefaultTimeScheduler with > something more appropriate for James, but I haven't had time > to do it yet. > > --- Noel > > -----Original Message----- > From: Andrei Ivanov [mailto:[EMAIL PROTECTED]] > Sent: Sunday, August 11, 2002 18:10 > To: James Developers List > Subject: handling connection vs sheduler problem > > > Hi > I don't think that setting socket timeout will help to solve > scheduler problem. It is not possible to control connection > by setting socket timeout. As it was several times mentioned > before scheduler isn't designed to be used as in James > servers. It is nice and elegant solution, I mean > ShedulerNotifyînputStream, but it results in overhead we know > about. The solution is in cornerstone connection classes > which create connection handlers for James. My idea is to > adapt (modify) cornerstone connection classes by adding > connection control facilities. > > About cornerstone. I've been using different cornerstone > blocks a lot for my project. I can say that cornerstone is > well written and reliable library. On the other hand for > complex and particular phoenix based applications (like > James) cornerstone shall be considered more as guideline than > ready to use solution. If we really want to improve James we > shall adapt cornerstone for James (but not James for > cornerstone as it is now) > > Andrei > > ----- Original Message ----- > From: "Noel J. Bergman" <[EMAIL PROTECTED]> > To: "James Users List" <[EMAIL PROTECTED]> > Sent: Monday, August 12, 2002 12:04 AM > Subject: RE: Admin problem > > > > I have no problem with the socket.setSoTimeout(timeout) call. We > > should start making more use of setSoTimeout elsewere, and wean > > ourselves off of the scheduler. > > > > With respect to the logging, I think that only the one > > where we echo the timeout is necessary, not the one > > where we log it for each connection. > > > > Would you please submit a [PATCH] to James-Dev? > > > > --- Noel -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>