[elephant-devel] Advice on multiple deploying multiple applications/repositories

[John]

Thanks for the prompt response. We hadn't actually looked at any web
frameworks per se. Our intention is not to develop our own. Instead have
read up a little on weblocks, ucw, kpax, kanamit, and hunchentoot.

Our goal is to set up two different layers on Lisp code (maybe different
machines for each layer). The lower layer will be a Lisp image offering a
web-service-like API to the middle-layer. The middle-layer will contain the
actual web framework that will interact with Apache and the lower layer. I
suppose the lower layer could work with something as "simple" as hunchentoot
and both layers will communicate via internal HTTP. The middle layer will
have the business-front logic and presentation code and this is where we
liked weblocks for. We also liked ucw but feel weblocks seems to have a more
active community.

I guess we could just as well use weblocks for the lower layer, but that may
just be too much for what's really needed off it.

I could ask which you would recommend but that probably doesn't belong in
this list and it also probably depends on the actual goal and design of the
application. It may sound as if this whole setup seems like an overkill for
such simple application, but the applications I mentioned were just mere
examples. Our intention is to develop much more "mission-critical"
applications... but we still have a lot to learn.

So, assuming we're only concentrating on the lower-layer, I think the
recommendation to dynamically bind *store-controller* for each
request/thread depending on the app may sound like the best approach.

Your other suggestion of deploying them on multiple machines is very valid.
The only reason we don't consider that is strictly cost reasons. I know you
can get a shared hosting account where you can install Lisp for about
$9/month, but our corporate requirements for IT systems require us to use
much more expensive environments, and so we are trying to re-use as much as
possible the resources given.

Thanks again,
JD

On Fri, Feb 6, 2009 at 5:08 PM, Ian Eslick <eslick at media.mit.edu> wrote:

> The answer to this depends on the web framework you are using.
>
> A simple answer is to dynamically bind *store-controller* for each app
> for each request/thread.  Since each thread will have its own binding,
> each is tied to a particular application.
>
> If you put all your database IO into a with-transaction body you can
> use the :store-controller argument to pass in the current store which
> will then be bound to *store-controller* in the body of the with-
> transaction form.
>
> If you use a higher level web framework like weblocks, it already
> directs through its own store abstraction and supports multiple
> webapps, each with a different store which will then handle all the
> *store-controller* stuff for you if you don't violate the abstraction
> (which you will probably want to do, however).
>
> If you are rolling your own web framework, then I'd just implement
> abstractions around the database operations which know what app is
> running and directly supplies the appropriate store-controller to the
> functions requiring/supporting :sc or :store-controller arguments.
>
> However, if each of your apps is that independent, why not just put
> one per server and avoid the complexity?
>
> Ian
>
> On Feb 6, 2009, at 4:31 PM, John wrote:
>
> > Since our interest is to use Lisp and Elephant to create web
> > applications, I'd like to request some deployment advise for Lisp
> > and/or Elephant.
> >
> > Our intention is to have Apache in front of a Lisp image running
> > some web server, where Apache will simply allow us to proxy/load-
> > balance traffic. Maybe we could have Lisp running in a large machine
> > "fronted" by multiple Apache servers.
> >
> > Imagine we want to offer to separate services. One would allow
> > people to maintain a list of friends and contacts (AKA a contact
> > database) and the other will allow people to maintain a To-Do List.
> >
> > We thought we could have a single Lisp instance listen for requests
> > based on URL or entry points. Each application would have its own
> > data store. By default, Elephant functions use *store-controller* to
> > connect to the data store. However, since we'll have multiple data
> > stores, the question is: should we be rebinding *store-controller*
> > for each application, pass a specific :sc argument to all our
> > function calls, or simply have multiple Lisp instances running and
> > listening on different ports for each application?
> >
> > I thought we could just have a single Lisp instance for all the
> > applications we wanted to deploy, but am confused as to how a single
> > Elephant instance could handle multiple stores, one for each
> > application, or better yet, what would be the most advisable
> > deployment model for something like this.
> >
> > Thanks in advanced,
> > JD
> > _______________________________________________
> > elephant-devel site list
> > elephant-devel at common-lisp.net
> > http://common-lisp.net/mailman/listinfo/elephant-devel
>
>
> _______________________________________________
> elephant-devel site list
> elephant-devel at common-lisp.net
> http://common-lisp.net/mailman/listinfo/elephant-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.common-lisp.net/pipermail/elephant-devel/attachments/20090206/73652317/attachment.html>