Skip to content

straszheimjeffrey/The-Kiln

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The Kiln

The Kiln is an evaluation strategy for insanely complex functions. It was designed based on my experience with managing several large, complex, ever-changing web applications in Clojure, including with tools like Compojure and Ring. Those tool are nice. They are simple and well thought out. However, they do little to address the real complexities of developing big commercial system.

The Kiln can help.

Usage

If you use Leiningen, add this to your project.clj:

[kiln "1.2.0"]

Kilns fire clays. A clay is a very simple sort of object. Here is one:

(defclay example)
    :value 4)

This clay is somethign with a value of 4. To get its value, we fire it in a kiln.

(fire (new-kiln) example)
-> 4

Not very exciting. Let's make another clay:

(defclay another
    :value (+ 3 (?? example)))

We'll fire it:

(fire (new-kiln) another)
-> 7

So the (?? ...) syntax lets us fire that other clay also. When a clay is fired in a kiln, it can ask for the value of another clay inside that same kiln.

But so what, you ask? Let's make a clay like this:

(defclay random
    :value (rand-int 1000))

And let's fire it a few times:

(fire (new-kiln) random)
-> 704
(fire (new-kiln) random)
-> 671
(fire (new-kiln) random)
-> 443

Looks random enough. But what about this:

(def kiln (new-kiln))
(fire kiln random)
-> 226
(fire kiln random)
-> 226
(fire kiln random)
-> 226

Within the same kiln, a clay remembers its value.

This is the point of kilns and clays. Within a single kiln, a clay is some particular value that was computed. Normally, its value will depend on the other clays in the kiln. Plus the coals.

A coal is like a clay, except with a coal, you do not compute its value in the kiln. Instead, you set its value. We call that "stoking" the coal.

(defcoal some-coal)
(defclay some-clay
    :value (+ 1 (?? some-coal)))
(defclay another-clay
    :value (+ (?? some-clay) (?? some-coal)))

(def kiln (new-kiln))
(stoke-coal kiln some-coal 5)
(fire kiln another-clay)
-> 11

(def kiln-2 (new-kiln))
(stoke-coal kiln-2 some-coal 6)
(fire kiln-2 another-clay)
-> 13

Nice, you say, but what is the point?

The point is this: in an application such as a web server, you compute a bunch of data for each request, such as the request uri, the request cookies, the current user, his session, the dispatch data, the search results, the page header, and so on and so on. During that request,

  • The data does not change;

  • The data is useful in many places;

  • But it is difficult to pass around and manage.

Enter clays and kilns. The kiln gathers all this data into one scoped mechanism where it is visible but controlled.

For a more detailed example, including how clays cleanup after themselves (such as your database connection clay, which knows to close) and how clays can be wrapped by glaze, which provides features similar to middleware/aspects/etc., go [here] (http://github.com/straszheimjeffrey/The-Kiln/wiki/Worked_Example).

Clays with arguments

The basic idea of a clay is that it is a single named value that is available throughout your computation (e.g. a web request). However, sometimes you will want to define a clay whose behavior can vary on arguments. Doing this will make a clay behave somewhat like a memoized function, except the memoization is local to the specific kiln.

It looks like this:

(defclay a-clay-with-arguments
   :args [a b]
   :value (+ a b)
   :cleanup (do-something ?self a b))

You can use it like this:

(fire some-kiln a-clay-with-arguments 1 2)

Which will return 3. Also at cleanup time (do-something 3 1 2) will be called. Note, this will only happen once. If you call it again with those same arguments, the same value is returned, but it is not recomputed. The cleanup is only called once.

On the other hand, if you do this

(do
  (fire some-kiln a-clay-with-arguments 1 2)
  (fire some-kiln a-clay-with-arguments 2 3))

the computation will happen twice, as well as the cleanup.

Note that the kiln must remember each invocation, so if you call a kiln a very large number of times with different arguments, all of those values (along with the arguments) are stored in the kiln. If you must do this, you may be better off with a function.

Glazes can also take arguments. The log glaze in the worked example shows how.

Clays and Threads

Each Kiln should only be used within the thread where it was created. There is no support for sharing kilns between threads. Note this does not apply to clays. Clays are entirely stateless. They can be shared. Many threads can create their own kilns, and use them to process the same set of clays.

This, of course, is exactly how they should be used for web applications. Each request creates a kiln. They share your clays.

License

Copyright (C) 2012 Jeffrey Straszheim

Distributed under the Eclipse Public License, the same as Clojure.

About

A computational model for insanely complex functions

Resources

Stars

Watchers

Forks

Packages

No packages published