On Tue, 14 Aug 2001, Jonathan Cobb wrote:
> Craig R. McClanahan wrote:
>
> >I don't know of anyone working on it at the moment, although (at least for
> >Tomcat 4) it would be a fairly straightforward exercise.
> >
> I think I have seen support for this in the code (see
> org.apache.catalina.core.StandardWrapper), but I've also heard that the
> STM servlet support is not as robust as it could be. Anyone with more
> direct (codewise) experience on this care to comment?
There is infrastructure on which to build STM support if you want
it. But, in effect, the current code keeps a pool size of 1.
Anyone who wants to submit the appropriate patches to configure and manage
a cache of instances is welcome to.
>
> >Personally, I don't think STM should be encouraged -- it misleads people
> >into thinking they don't have to deal with multiple thread issues, and
> >still runs into scalability problems when you hit the maximum configured
> >number of instances for each given servlet. But I'd commit a patch to
> >enable this if someone really wanted it.
> >
> While I agree that people should think about threading issues from the
> get-go, and ideally always use multithreaded servlets, there are always
> cases where (somehow) large amounts of non-threadsafe servlet code gets
> written. When the dev team finally figures out how they've gone wrong,
> they usually don't want to rewrite the whole app, they just want it to work.
>
> Of course if you are building a high-traffic webapp you will want to
> make sure everything is multithreaded, but for many low-traffic apps,
> single-threaded models (while arguably uglier from a design standpoint)
> perform just fine.
>
I've seen lots of people (even on single user web apps) get bit by
simultaneous access to their session attributes. It's not really an app
architecture issue; it's something that the developer should be aware of
but doesn't tend to think about, because STM seems to promise that those
issues have been dealt with.
> - jonathan.
>
Craig
