cowj

[C]onfiguration [O]nly [Web] on [J]VM

[TOC]

Goal

COWJ is pronounced as Cow Jay. Reason for emphasising on COW is the configuration only web, powered by the polyglot support from underlying JVM, which is by and large one of the phenomenal piece of Engineering done, till date. Author tried doing the same with GO but not with much success - here is the link to that project: https://github.com/nmondal/goscow

Configuration-Only-Web-over-JVM

It should be very clear from the naming that:

objective is to optimize back-end development and replacing it with configurations.

This is immediately helpful in rapid development of prototypes, and in many cases, rapidly developing and deploying solutions to customers side of things. This aims to de facto replace Mule software for almost all practical purposes.

Cowj let's you build back-end systems blazing fast and has the following features:

1. You can write APIs

2. Can do Batch processing with Cron

3. Event Processing capabilities ( with RAMA & Other Plugins)

4. Data Processing

via configurations and scripts. It's backbone is written using spark-11, jetty-11 , quartz and casbin.

Very Fast Hello World

1. Clone the repository first

3. Build so that you have the cowj-jar ready with ./deps pointing to dependencies.

4. Create a directory hello

5. Inside hello create a directory static

6. Inside static folder created index.html and write down hello, world!.

7. Inside the hello folder create a hello.yaml file as follows:

# hello/hello.yaml
port: 8080

Now run the hello project as :

java -jar cowj-*.jar hello/hello.yaml

Open browser and visit localhost:8080/index.html you will see hello,world.

That is it. That is all it takes to setup a Cowj server.

What about actual service endpoints?

Just type down these into the hello.yaml :

# hello/hello.yaml
port: 8080
routes:
  get:
    /hello: _/hello.js

And restart cowj again. Hit this endpoint in curl:

curl -XGET "http://localhost:8080/hello"

You would get back Hello, World!.

What is really happening here is Cowj system is detecting an expression written using js engine - and evaluating and returning.

One can of course, for betterment move the expression from inside to outside, e.g. create a file hello.js in the hello folder, and then update the hello.yaml as follows:

And the hello.js :

// hello.js
"Hello, World!"

Now, restart cowj, run the same curl command - and voilà, you would have Hello, World again.

As you can see, code is configuration and configuration is code in Cowj.

With this note, we shall dive into the world of BED - back end development.

COWJ Setup

1. Install java 17 or newer : https://adoptium.net/temurin/releases/?version=17

2. Install gradle 8.5 or above : https://gradle.org/install/

3. Clone https://github.com/nmondal/cowj

   git clone https://github.com/nmondal/cowj # clone repo 
   cd cowj # go to cowj directory 

4. Run gradle build -x test - in the cowj directory

5. cd app/build/libs

6. Ensure ls returns java -jar cowj-0.1-SNAPSHOT.jar as well as a deps/ folder. If those two don’t exist, run gradle clean and gradle build -x test

7. java -jar cowj-0.1-SNAPSHOT.jar ../../samples/hello/hello.yaml

8. Open localhost:5003/hello/z in browser or curl.

   curl localhost:5003/hello/z
   # produces hello world
   

This should return hello world.

Back End Development

Development Today

It is pretty apparent that a lot of back-end "development" goes under:

1. Trivial CRUD over data sources
2. Reading from data sources and object massaging - via object mapper, thus mapping input to output
3. Aggregation of multiple back-end services
4. Adding random business logic to various section of the API workflows

COWJ aims to solve all these 4 problems, such that what was accomplished by 100 of developers can be done by less than 10, a ratio of 90% above in being effective.

How To Do it?

Deprecation : Service Framework

Lately, 10+ people gets allocated to maintain "how to create a service" end point. This must stop. The following must be made true:

1. Creating a service end point should be just adding configuration
2. Writing the service should be just typing in scriptable code
3. Input / Output parameters are to be assumed typeless JSON at worst
4. We can, at best do a lazy type checking to validate Schema - alternatives
   1. RAML
   2. Open API
   3. JSON Schema

"Business" Logic

There should not be any "business logic" in the code. They are susceptible to change, hence they should be hosted outside the API end points - DSL should be created to maintain.

Data Store Access

Any "Service" point requiring any "data store" access need to declare it, specifically as part of the service configuration process. Objective of the engine would be to handle the data transfer. JSON is the choice for data transfer for now.

Unit Testing - Code Coverage

If one is willing to use this, one must wonder - what Makes Anything 'Prod Ready' ? Only suitable answer is the core components must be excessively well tested, and should have really good instruction and branch coverage.

Cowj All Coverage

Thus the engine core is immensely tested, and is at per with industry standard quality.

Cowj Core Coverage

Cowj Plugin Coverage

Even plugins are reasonably tested, and are ready for production usage.

Current Implementation

Service Configuration

Here is the config:

# This shows how COWJ service is routed

# port of the server
port : 8000

# threading related stuff
threading:
  virtual: true # in case of JRE 21 it would use virtual green threads 
  min: 4 # min no of threads 
  max : 8 # max no of threads 
  timeout: 30000 # which ms to give timeouts 

# async IO configuration 
async:
   virtual: true # in case of JRE 21 it would use virtual green threads 	
   threads: 8 # for async io, if not specified, infinite in practice  
   keep: 32 # keep only 32 task results , if not specified 1024 by default 
   fail: _/scripts/js/async_task_failure_handler.js #  async task failure handler 
   retries:  # retry strategy 
      strategy: exp
      max: 4 
      interval: 1500


# routes information
routes:
  get:
    /hello : _/scripts/js/hello.js
    /param/:id  : _/scripts/js/param.js

  post:
    /hello : _/scripts/js/hello.js
    /_async_/put : _/scripts/zm/put.zm # special async push back route


# route forwarding local /users points to json_place
proxies:
  get:
    /users: json_place/users
  post:
    /users: json_place/users


# filters - before and after an URI
filters:
  before:
    "*" : _/before.zm
  after:
    "*": _/after.zm

# Web Sockets 
sockets:
   /ws : my_socket_handler.zm
   

# how to load various data sources - plugin based architecture 
plugins:
  # the package 
  cowj.plugins:
    # items from each package class::field_name
    curl: CurlWrapper::CURL
    redis: RedisWrapper::REDIS

# data store connections
data-sources:
  redis :
    type : redis
    urls: [ "localhost:6379"]

  json_place:
    type: curl
    url: https://jsonplaceholder.typicode.com

cron:
  cache:
    exec: _/cache.md
    boot: true
    retries: 
       strategy: counter
       max: 3 
       interval : 1500
    at: "*/5 * * * *"

It simply defines the routes - as well the handler script for such a route. In this way it is very similar to PHP doctrine, as well as DJango or Play.

Scripting

Is Polyglot. We support JSR-223 languages - in built support is provided right now for:

1. Groovy ( Optionally Typed )
2. JavaScript ( via Rhino Engine )
3. Kotlin Scripting ( Fully Typed )
4. Python 2.7 ( Jython )
5. ZoomBA ( Duck Typed )

And obviously any pre-existing class files just work!

Underlying we are using the specially cloned and jetty 11 migrated spark-java fork:

https://github.com/nmondal/spark-11

Also it uses ZoomBA extensively : https://gitlab.com/non.est.sacra/zoomba

Here is how it works:

1. Client calls server
2. Server creates a Request, Response pair and sends it to a handler function
3. A script is the handler function which receives the context as req,resp and can use it to extract whatever it wants.
4. DataSources are loaded and injected with _ds variable.

Here is one such example of routes being implemented:

// javascript
let x = { "id" : req.params("id") };
x;// return 

The context is defined as:

https://sparkjava.com/documentation#request

https://sparkjava.com/documentation#response

A good read about why we try to avoid type systems inside can be found here:

Clojure vs. The Static Typing World

Shared Memory

Cowj supports a global, non session oriented global memory - which can be used to double time as a poor substitute for in memory cache - and it is a ConcurrentHashMap - accessible by the binding variable _shared.

The syntax for using this would be:

v = _shared.<key_name> // groovy, zoomba 
v = _shared[<key_name>] // groovy, zoomba, js, python  

See the document "A Guide to COWJ Scripting" found here - Scripting

Threading

We can specify the min , max, and timeout for the underlying jetty threadpool. In case we specify virtual which can take either true or false - under JRE 21 it would spawn green threads - or virtual threads.

Routes

As expected routes are grouped under the HTTP verb. The idea is pretty simple, in the left side we have the virtual path of the server, while on the right side we have the real script location which should be executed to run it.

Async Routes

A special prefix is reserved _async_ , any route with this prefix would be executed asynchronously, and would return almost immediately responding with a plausible almost GUID string id for the task.

If one wants to make the server work for long running tasks, programmatically,
this is one way.

# async IO configuration 
async:
   threads: 8 # for async io, if not specified, infinite in practice  
   keep: 32 # keep only 32 task results , if not specified 1024 by default 
   fail: _/scripts/js/async_task_failure_handler.js #  async task failure handler 

Filters

These are how one can have before and after callback before and after any route pattern gets hit.

A before filter gets hit before hitting the actual route, while an after filter gets hit after returning from the route, so one can modify the response if need be.

Classic example of before filter is for auth while after filter can be used to modify response type by adding response headers.

Before

A typical use case from auth:

routes:
  get:
    /users: _/users.zm
filters:
  before:
    "*" : _/auth.zm

Now inside auth.zm :

// auth.zm
def validate_token(){ /* verification here */ }
token = req.headers("auth")
assert("Invalid Request", 403) as { validate_token(token) } 

You really do not need to do that, we have fully functional auth support, but that was just for demo.

Finally

A typical use case to json format error messages:

routes:
  get:
    /users: _/users.zm
filters:
  finally:
     "*" : _/finally.zm

Now inside finally.zm :

// finally.zm
if ( resp.status @ [200:400] ){ return }
error_msg = resp.body
resp.body( jstr( { "error" : error_msg } ) )

This ensures all errors are formatted as json.

Web Sockets

Read more on the scripting documentation. Idea is the same, various WebSocket server states are handled by the script handler. Example of websocket is in websocket app.

Cron

In the cron section one can specify the "recurring" tasks for a project. The key becomes name of the task, while:

1. exec : script that needs to be executed
2. boot : to run while system startup
3. at : cron expression to define the recurrence
4. retries : retry strategy in case the cron job at boot fails

A cron boot failure can hang the system, hence a proper retry makes sense.

Cron threadpool is not Java 21 green thread aware ( yet ) but there is a PR (quartz-scheduler/quartz#1093) submitted which if and when incorporated will be used immediately to spawn green threads for the jobs. The guarantee is needed from the core Quartz team that the system works with green threads.

Retries

They can be applied on async routes as well as cron jobs which will be used in the boot time. A typical retry looks like this:

retries:
   strategy: counter | random | exp # one of the strategy 
   max : 3 # max no of tries before it fails
   interval: 100 # delta time between two successive tries 

Strategies are counter for simple counter strategy, random for a derivative of counter, but the interval will be random distributed approximating the mean value of interval and finally exp for Exponential Backoff, where the interval is the starting point of the delta between successive tries.

Data Sources

If the idea of COWJ is to do CRUD, where it does CRUD to/from? The underlying data is provided via the data sources.

How data sources work? There is a global map for registered data sources, which gets injected in the scripts as _ds :

_ds.get("redis") // javascript 
_ds["redis"] // zoomba, groovy, python 
_ds.redis // zoomba, groovy 

Would access the data source which is registered in the name of redis .

Right now there are the following data sources supported:

JDBC

JDBC data source - anything that can connect to JDDBC drivers.

some_jdbc:
  type: jdbc # shoule have been registered before
  driver : "full-class-for-driver"
  connection: "connection-string"
  properties: # properties for connection
    x : y # all of them will be added 

CURL

External Web Service calling. This is the underlying mechanism to call Proxies to forwarding data.

some_curl:
  type: curl # shoule have been registered before
  url : "https://jsonplaceholder.typicode.com" # the url 
  headers: # headers we want to pass through
    x : y # all of them will be added 

Google Storage

Exposes Google Storage as a data source.

googl_storage:
  type: g_storage # shoule have been registered before

Redis

Exposes Redis cluster as a data source:

 redis :
   type : redis # should register before 
   urls: [ "localhost:6379"] # bunch of urls to the cluster 

Firebase Notification Service

Firebase as a data source.

Plugins

All data sources are implemented as plug and play architecture such that no code is required to change for adding plugins to the original one.

This is how one can register a plugin to the system - the following showcases all default ones:

# how to load various data sources - plugin based architecture 
plugins:
  # the package 
  cowj.plugins:
    # items from each package class::field_name
    curl: CurlWrapper::CURL
    fcm: FCMWrapper::FCM
    g_storage: GoogleStorageWrapper::STORAGE
    jdbc: JDBCWrapper::JDBC
    redis: RedisWrapper::REDIS

As one can see, we have multiple keys inside the plugins which corresponds to multiple packages - and under each package there are type of datasource we want to register it as, and the right side is the implementor_class::static_field_name.

In plain language it reads:

A static filed CURL of the class cowj.plugins.CurlWrapper implements a DataStore.Creator class and is being registered as curl type of data store creator.

Library Folder

From where plugins should be loaded? If one chose not to compile their code with COWJ - as majority of the case would be - there is a special folder in the base director defaults to lib. All jars in all directories, recursively will be loaded in system boot and would be put into class path,

One can naturally change the location of the lib folder by:

lib: _/some_other_random_dir

How to author COWJ plugins can be found in here Writing Custom Plugins

Proxies

Path/Packet forwarding. One simply creates a base host - in the data source section of type curl and then use that key as base for all forwarding.

In the Yaml example the following routing is done:

localhost:5003/users 
--> https://jsonplaceholder.typicode.com/users

System responds back with the same status as of the external web service as well as the response from the web service gets transferred back to the original caller.

Proxies can be used to transform the payload to the external server, as well as can be used to transform back the data from the external server.

First one we call "forward" transform, and the other one "reverse" transform.

Forward Transform

This is easy with the before filter. The idea is as follows:

// before.zm 
forward_payload = {
  "headers" : {
    key : value
  },
  "query" : {
    key : value
  },  
  "body" : "request body"
}
req.attribute("_proxy", forward_payload) 

As for the payload it has the following to be forwarded to the destination server :

1. headers has a mutable map of request headers
2. queries has mutable map of all query parameters
3. body as the request body

The underlying system picks up the request attribute named _proxy as present, and then forwards it to the destination server.

Reverse Transform

This is easily doable by the after filter. Just intercept the response, and we can do whatever we want to do with it.

Type System

We support json schema based input validation. To read more see Writing Input Validations

Auth

We support casbin based Authorization and pretty basic token based Authentication too. To read more see Embedding Auth

Using Stand Alone

Building

1. Clone the repo to a local directory
2. Install / Download java 17 from adoptium ( https://adoptium.net )
3. If you do no have gradle install/download gradle ( https://gradle.org/install )
4. Open command promt and run gradle, if it runs you are good.
5. Go to the local directory where cowj is cloned and issue the command gradle build
6. It would take some time and will be build.

Running

1. Build the app.
2. Go to the app/build/libs folder.
3. Run the stand-alone binary by:

java -jar cowj-0.1-SNAPSHOT.jar  <config-file>

where you can chose any yaml config file. There are variety of files in the app/samples folder. For example to run the hello app :

# you are in the folder app/build/libs 
java -jar cowj-0.1-SNAPSHOT.jar  ../../samples/hello/hello.yaml 

The jar has been created such as to have classpath property set to run as long as all the dependencies are in the deps folder.

Auto Load on Save for Development & Testing

It was suggested that for development and testing it is better to have automatic reloading of some of the files on save. In a full configuration based system this pose a problem. However, we have implemented load on save for

1. Scripts
2. JSON Schema

Any other configuration file have implication and needs to reload the entire app.

# you are in the folder app/build/libs 
java -jar cowj-0.1-SNAPSHOT.jar  ../../samples/hello/hello.yaml true # the last argument true ensures it is running in DEV mode  

And will load script and schema files automatically on save.