Web frameworks - separating UI from logic

I’m starting to look again at available web application frameworks for
Ruby

  • it’s been a few years since I last did. What I’m looking for is to be
    able
    structure my application like this:

      HTTP+
      HTML            RPC(*)                  SQL
    

    <--------> user <--------> application <--------> database
    interface logic

(*) e.g. DRb, or just direct invocation of a separate object

The idea here is to be able to bolt multiple front-ends onto the same
application logic. For example, you might start by having a web form for
entering new customer orders, but then you could add a SOAP interface so
they can be submitted automatically, or a batch-upload interface which
accepts CSV files.

    HTTP+
    HTML             RPC                    SQL
 <--------> user <--------> application <--------> database
          interface      ->    logic
             1           |
    SOAP                 |
 <--------> user <--------
          interface
             2

The important thing is that the business logic for (say) validating
orders,
or performing a series of actions in response to submission of a valid
order, exists exactly once and so is the same regardless of how the
request
came in (*).

In ‘MVC’ terminology, I think the “application logic” box is a
Controller
with a Model sitting behind it. It will also need to provide an
interface to
query the state of the Model, unless the Model directly exposes a
read-only
view to the UIs.

My first port of call for looking at modern frameworks is Rails. Looking
through the Rails tutorials, I get the impression that a Rails
“controller”
is responsible for three things:

  1. Parse the HTTP request, e.g. extract relevant params()

  2. Validate the request and perform any database updates directly

  3. Decide which view to send back to the browser (which in turn also
    queries the database directly)

Is that a fair summary?

Would it also be true to say that in Rails, the de-facto model is just a
bunch of database tables, with a separate ActiveRecord facade sitting in
front of each table? For example, if a particular action requires four
tables to be updated, would the controller typically get four
ActiveRecords,
modify them and write them all back? If so, this seems to bundle the
‘user
interface’ and ‘application logic’ parts together more closely than I’d
like.

I expect I can write my own application logic server, and use Rails as a
front-end to submit RPC requests to it. This means that the controllers
would do very little though, just:

  • bundle up the paramters and submit an RPC request for an action
  • look at the response (success or fail)
  • submit an RPC request to get model state
  • generate a suitable HTML page response

It looks like I’d hardly be using any of Rails - just the HTML
templating
and perhaps session tracking. Maybe then a more lightweight framework
would
be better suited? (**)

I wrote a tiered application along these lines a few years ago for a
different employer. I ended up writing a framework to handle the FCGI
main
loop; compile Amrita templates on demand and cache them; and handle
unexpected exceptions, displaying a suitable page back to the browser.

Perhaps there’s a simple framework I can use which will avoid having to
re-write this from scratch?

Thanks for your ideas.

Regards,

Brian.

(*) This could be quite complex logic. For example, “create order for
customer” might perform a bunch of validation, do a credit check in some
circumstances, add an order line into a database, send a message to a
billing system, and send messages to service provisioning platforms.

(**) I can imagine cases where it might make sense not to stick rigidly
to
the tiered approach. As mentioned above, maybe the UI could be allowed
to
read the model state directly; this suggests

    HTML             RPC                    SQL
 <--------> user ---------> application <--------> database
          interface            logic                  |
             ^                                      .
              ` . . . . . . . . . . . . . . . . . .
                       read-only query path

In that case, if the application logic was already using ActiveRecord as
its
model, then Rails would probably be a good choice for the UI. You’d
share
the ActiveRecord model definition, and give the UI read-only DB
credentials.

However in my case I may need to force all data accesses via the
“application logic” layer, in particular to provide a central point of
enforcing access control policies - i.e. who can read what.

I can also see that the UI might legitimately need R/W access to a
database
for its own private purposes (for example, the UI could be responsible
for
assembling a shopping cart, before submitting a completed order). In
this
case the database is just a UI scratchpad, entirely separate from the
business model. Therefore, you might want something like ActiveRecord
available even with the tiered approach.

On Fri, 23 Feb 2007, Brian C. wrote:

            1           |

In ‘MVC’ terminology, I think the “application logic” box is a Controller
2. Validate the request and perform any database updates directly
modify them and write them all back? If so, this seems to bundle the ‘user
interface’ and ‘application logic’ parts together more closely than I’d
like.

hi brian-

i think you can flex on #3 above to get what you want. basically, write
your
entire logic stack, in it’s entirety, as Controller private methods.
these
private methods will deal with ruby values only - no html, xml, etc…

then you can test this. when that’s complete, you can bolt on views as
html,
xml, soap, etc. in otherwords, you can simply add on layer of
abstraction via
method calls inside your controller, another controller class, etc. in
otherwords simply defer parsing input from params, sessions, etc and
also
avoid generating view components. simply write all logic with ruby
values
in/out.

in the end you’ll have two tiers of controllers

logic-centric controllers

 1) perform logic based on ruby arguments.  returns ruby values.

view-centric controllers

 1) parse params, etc

 2) use a logic-controller

 3) decide what output to send

of course there are other ways to slice this, such as a very thick model

  • but
    that’s not very mvc.

regards.

-a

On Fri, 23 Feb 2007, Brian C. wrote:

            1           |
   SOAP                 |
<--------> user <--------
         interface
            2

The important thing is that the business logic for (say) validating orders,
or performing a series of actions in response to submission of a valid
order, exists exactly once and so is the same regardless of how the request
came in (*).

This makes sense, though I don’t know that you need an RPC layer in
there.

In ‘MVC’ terminology, I think the “application logic” box is a Controller
with a Model sitting behind it. It will also need to provide an interface to
query the state of the Model, unless the Model directly exposes a read-only
view to the UIs.

MVC is a reinvigorated buzzword when it comes to web applications, even
when what a web application is implementing is something a bit different
from the classical definition of MVC. MVC is just a more marketing
friendly term than what is really important – a separation of concerns.
So don’t sweat the details of what is or isn’t MVC too much. What you
really want is just a logical separation of concerns.

Would it also be true to say that in Rails, the de-facto model is just a
bunch of database tables, with a separate ActiveRecord facade sitting in
front of each table? For example, if a particular action requires four
tables to be updated, would the controller typically get four ActiveRecords,
modify them and write them all back? If so, this seems to bundle the ‘user
interface’ and ‘application logic’ parts together more closely than I’d
like.

You’ll find that for most simple web apps, this level of separation is
the
common case. For that majority of applications, it is a sweet spot
between having enough separation to make code management easier and
cleaner, while not having so much separation as to introduce additional
complexities. This doesn’t preclude having greater separation, though.

It looks like I’d hardly be using any of Rails - just the HTML templating
and perhaps session tracking. Maybe then a more lightweight framework would
be better suited? (**)

Perhaps there’s a simple framework I can use which will avoid having to
re-write this from scratch?

There are a few you could consider. Nitro has been around for a while.
It is lighter weight than Rails, for sure, and is a capable system.
It’s
currently undergoing a massive rewrite which should make the Nitro world
a
much prettier one, too. Unfortunately, I have had some serious problems
getting the latest release, 0.41, to work (at least from a gem
install) because of a dependency issue.

Camping is a very lightweight framework that you might consider.

Likewise, for a Rails-lite that might be right up your alley, look at
Merb. It’s designed to be fast and light for things that don’t need all
that Rails has built in.

You might also consider my framework, IOWA. The 1.0 release of it is
now
imminent (after nearly 5 years, wooo hooo!). Feel free to drop in on my
little freenode IRC channel, #iowa, or email me directly
([email protected]) if you have any questions.

Now, regarding your overall application design, I don’t understand why
you’d need to implement an RPC server for the application logic, so my
first impression is that you would be best served by writing a class
library that encapsulates the application logic.

Once you have that, whether you are writing a web interface or a SOAP
interface or a command line tool, they all have access to the same
functionality by using your class library, without the added
complication/overhead of maintaining an RPC service.

Kirk H.

On Fri, Feb 23, 2007 at 11:50:17PM +0900, [email protected] wrote:

common case. For that majority of applications, it is a sweet spot
between having enough separation to make code management easier and
cleaner, while not having so much separation as to introduce additional
complexities. This doesn’t preclude having greater separation, though.

OK.

Perhaps what confused me about Rails is that there seem to be a lot of
default controls - the CRUD stuff - which act on a single table at a
time.

If I could hide those sort of controls, and expose only higher-level
controls to the user which perform business logic, that would be part of
my
problem solved.

The other part is ensuring that my application isn’t tied only to
interactive HTTP use. But then I should be able to build (say) a SOAP
facade
which maps onto the same controls, yes?

Then all I have to do is ensure that no database updates are possible
except
via these controllers - for example, by keeping the database write
password
secret.

Now, regarding your overall application design, I don’t understand why
you’d need to implement an RPC server for the application logic, so my
first impression is that you would be best served by writing a class
library that encapsulates the application logic.

Well, “RPC” was also assuming that by the time this was deployed, there
would need to be multiple application servers for load-sharing and
resilience. But equally, “RPC” could just be direct object invocation -
i.e.
LPC :slight_smile:

Thanks,

Brian.

On Sat, 24 Feb 2007, Brian C. wrote:

You’ll find that for most simple web apps, this level of separation is the
common case. For that majority of applications, it is a sweet spot
between having enough separation to make code management easier and
cleaner, while not having so much separation as to introduce additional
complexities. This doesn’t preclude having greater separation, though.

The other part is ensuring that my application isn’t tied only to
interactive HTTP use. But then I should be able to build (say) a SOAP facade
which maps onto the same controls, yes?

Then all I have to do is ensure that no database updates are possible except
via these controllers - for example, by keeping the database write password
secret.

I do think that the easiest way to do this is to put all of your common
code into a class lib, including everything that interacts with your
backend database. Then, whether you use Rails or something else to
create
your HTTP interface, you guarantee that nothing goes directly to your
database simply by forcing all interaction to go through your API, yes.

Well, “RPC” was also assuming that by the time this was deployed, there
would need to be multiple application servers for load-sharing and
resilience. But equally, “RPC” could just be direct object invocation - i.e.
LPC :slight_smile:

I understand. Makes sense.

I’m not going to steer you specifically away from Rails or towards my
framework or someone else’s, specifically. But I do think you are
probably right in thinking that a lot of what Rails provides regarding
interaction with your model you won’t be using, so Rails may indeed be
heavier than what you need. If I were you I’d pick a few frameworks and
ask some questions about them. The answers will probably steer you
towards something that matches your needs and your way of thinking about
web applications.

Kirk H.

This forum is not affiliated to the Ruby language, Ruby on Rails framework, nor any Ruby applications discussed here.

| Privacy Policy | Terms of Service | Remote Ruby Jobs