Chuck Esterbrook wrote:
> At 03:21 PM 6/6/00 -0400, Geoff Talvola wrote:
> >A question for Chuck, Jay, and anyone else who wants to chime in:
> >How close do you think WebKit is to being usable for a real web
> >deployment? I'm
> >not talking about a high-volume thing, but maybe for a relatively low-volume
> >db-driven application where performance isn't all that important?
> I like the way you think, Geoff. :-)
> My gut feeling says 2 months for "early adopters". There are lots of
> refinements to be made and as we develop sites with Webware we're going to
> discover things that we didn't think of.
> With regards to fault-tolerance, my first thought in this area was to have
> the adaptors launch the app server if it's not around. That seemed like an
> easy, quick hit.
Good idea, and this actually eliminates the need for WebKit to include a Windows
NT service version of the appserver, which is good since the less
Windows-specific baggage the better. It slightly increases the response time for
the first request after a reboot -- I think I can live with that :-)
> Another thought is that some app servers (such as WebObjects) put a
> lifetime on their application instances, like say 12 hours. At the end of
> their lifetime, they get no new sessions and when all sessions are expired
> they quit. While the an app is dying, new sessions are going to other
> application instances. This helps protect against stability problems such
> as memory leaks and dangling pointers.
This also sounds good. I guess this is just a little additional logic inside the
CGI wrapper. It does mean that you can't restart an appserver or reboot the
machine without losing session data, but I think I could live with that.
I'd be happy to help out with either of these tasks, but first I need to dive
into the existing code and try to understand it.
> And then there's the watchdog technique you mentioned.
Maybe not needed if the CGI wrapper has the smarts to kill the old appserver and
start a new one if the existing one isn't responding. In effect, the CGI wrapper
becomes the watchdog. I think this could be a very robust solution.
I've seen some complaints on the Zope mailing list about Zope locking up under
various circumstances, and when it happens, since Zope is limited to a single
server process (unless you are using ZEO), you're stuck and you have to manually
restart the server. With this scheme where CGIWrapper automatically kills a
malfunctioning appserver and starts a new one, WebKit may be less vulnerable to
malfunctions -- you might lose some sessions, but the site as a whole should stay
> However, I don't have a very explicit list in general, because I don't
> think we'll think of everything now and a lot of what's required will
> depend on the person deploying.
> In related news, I started developing a site with Webware this Sunday. I
> won't let the cat out of the bag just yet, but it's intended for consumers,
> is entirely dynamic and it's subject has nothing to do with computers or
> technology. I'm shooting for an alpha at the end of this month and a beta
> at the end of next month.
I'm looking forward to seeing it.
- Geoff Talvola