← Back to team overview

ladon-dev-team team mailing list archive

Re: REST implementation

 

I knew this already :-) And that's not nessesarilly a bad thing, depending
on what you are developing. I just think we need to think through how to
implement it in Ladon. Since you have enjoyed using Ladon until now, you
probably already know that I am a hugh fan of end-user simplicity.

And since REST is strictly object-oriented, I really think we should be
looking at decorating the LadonType instead of methods. I haven't thought
it all the way through yet, but I'm almost sure that would be the place to
do it.

/ Jakob


2012/10/23 Mykhailo Stadnyk <mikhus@xxxxxxxxx>

> Exacly! REST is a constraint to objects by design! You've caught it. But
> it's good principal for data-oriented services with no overhead by any
> top-level protocol. It's a principal of building RESTful services. And will
> be good if Ladon will support it
>
> 2012/10/23 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>
>> But, I think I know what you mean. It's the fact that it utilizes all
>> HTTP methods that makes you say that REST is in the HTTP protocol. I just
>> don't nessesarily find that this is very revolutionary. It's more of a
>> constraint, cause you are limited to objects by design.
>>
>> / Jakob
>>
>>
>> 2012/10/23 Mykhailo Stadnyk <mikhus@xxxxxxxxx>
>>
>>> For REST HTTP is not underlying level but the protocol itself. In terms
>>> of REST design of service is known, so all you need is just to describe a
>>> list of objects. And then for each object there is 4 operations you are
>>> able to perform Create (HTTP PUT), Read (HTTP GET), Update (HTTP POST) and
>>> Delete (HTTP DELETE). That's why in terms of REST you do not need the
>>> description. You have only URLs each representing the object you can do
>>> direct HTTP requests and you have no overhead of having some top-level
>>> protocol, etc. That's why I said that REST is HTTP, because there is
>>> nothing on top of HTTP in REST. And there is nothing to describe, only the
>>> object URLs.
>>>
>>> It means that RESTful service architecture is not as flexible as SOAP,
>>> so if you, as developer decided to create RESTful service it means that you
>>> already chosen the service architecture. It also means that not every SOAP
>>> service can be RESTful, but RESTful service could be delivered with SOAP.
>>> So you, as developer MUST decide first if you are building a RESTful
>>> service, if so - go ahead, make it and Ladon give you benefit to deliver
>>> your RESTful service throug varius different supported protocols.
>>>
>>> Again. You CAN deliver RESTful service throug SOAP/XML-RPC/JSON-WSP/etc.
>>> But you CAN NOT deliver abstract SOAP/XML-RPC/JSON-WSP service through REST
>>> if it wasn't designed as RESTful
>>>
>>> In my implementation I've tried to make everything with maximum
>>> flexibility, so each RESTFul service could have more than 4 CRUD methods,
>>> for each HTTP request method you can have as many service methods as you
>>> need if the service methods have different signatures (arguments)
>>>
>>> like
>>>
>>> class Service:
>>>     @restfulize(method="GET")
>>>     def get_one_record( r_id):
>>>          pass
>>>     @restfulize(method="GET")
>>>     def get_last_record( r_time):
>>>          pass
>>>     @restfulize(method="GET")
>>>     def get_all_records():
>>>          pass
>>>     @restfulize(method="GET")
>>>     def get_filtered_records( filters):
>>>          pass
>>>
>>> As you see in this example 4 service methods are bind to HTTP GET
>>> method. Depending of parameters bypassed through HTTP I try to determine
>>> which method to call, but from terms of protocol you always do
>>>
>>> GET /Service/rest?filters['name']=*bla&filters['another_key']=somevalue
>>> HTTP/1.1
>>> Host: yourservice.com
>>> ...
>>>
>>> The same for other HTTP methods. You can have 2 different methods for
>>> POST, for example for User service example you may have split User data
>>> update and user password update, etc..
>>>
>>> Or maybe your service may not provide some methods. Like you create the
>>> public read-only service, so you don't provide PUT/POST/DELETE
>>>
>>> P.S. REST does not provide the way do bypass which service method to
>>> call by name. REST is not RPC. You have only the Resource, it's URL and 4
>>> operations. It is REST.
>>>
>>> Best regards,
>>> Mike
>>>
>>>
>>> 2012/10/23 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>
>>>> Everything in Ladon is HTTP :-)
>>>>
>>>> But HTTP is just the underlying protocol, and under that is a transport
>>>> protocol etc.
>>>>
>>>> When I am talking about protocol in Ladon I mean service protocols like
>>>> soap, json-wsp and soon json-rpc which are also protocols. I would put REST
>>>> at this level aswell.
>>>>
>>>> I haven't really used REST before, but from what I can see it is
>>>> nothing more than another service protocol which utilizes the HTTP standard
>>>> a bit more, like using PUT,POST,GET,DELETE methods. Also it looks like
>>>> there is not so much constraint on how parameters are passed so they can be
>>>> JSON or XML or query-string.
>>>>
>>>> I don't like that there is no service description format for REST
>>>> services. That is kind of a key feature in Ladon, that all supported
>>>> protocols should be able to describe themselves.
>>>>
>>>>
>>>> / Jakob
>>>>
>>>> 2012/10/22 Mykhailo Stadnyk <mikhus@xxxxxxxxx>
>>>>
>>>>> What do you mean?
>>>>>
>>>>> Actually REST is HTTP in terms of protocol, so what do I need to
>>>>> implement? Maybe you can explain me more detailed your thought.
>>>>>
>>>>> Best regards,
>>>>> Mike
>>>>>
>>>>> 2012/10/22 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>>>
>>>>>> My first question about your RESTful implementation is, why are you
>>>>>> not implementing it as a protocol?
>>>>>>
>>>>>> / Jakob
>>>>>>
>>>>>> 2012/10/22 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>>>>
>>>>>>> Just a remark about the package name tracepyd. That might be an
>>>>>>> unfortunate package name, cause on that odd platform called Windows
>>>>>>> pyd-files are binary python modules. Just a remark :-)
>>>>>>>
>>>>>>>
>>>>>>> 2012/10/22 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>>>>>
>>>>>>>> Hi Mike.
>>>>>>>>
>>>>>>>> Let's talk about tracepyd first, does it solve this bug:
>>>>>>>>
>>>>>>>> https://bugs.launchpad.net/ladon/+bug/877727
>>>>>>>>
>>>>>>>> rpclib has a terribly ugly solution for this, as you can see in the
>>>>>>>> bug, and we want to be better than rpclib :-)
>>>>>>>>
>>>>>>>> / Jakob
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> 2012/10/22 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>>>>>>
>>>>>>>>> Hi Mike
>>>>>>>>>
>>>>>>>>> Looking into this now, you will have my reply today :-)
>>>>>>>>>
>>>>>>>>> / Jakob
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 2012/10/22 Mykhailo Stadnyk <mikhus@xxxxxxxxx>
>>>>>>>>>
>>>>>>>>>> Currently it's an experimental feature which is not ready to be
>>>>>>>>>> integrated. I didn't test it at all (keep working on it).
>>>>>>>>>>
>>>>>>>>>> So here is my personal road-map for this feature:
>>>>>>>>>>   1. Resolve problem with decorators dependencies (something I
>>>>>>>>>> described in previous message)
>>>>>>>>>>   2. Write tests and test the functionality
>>>>>>>>>>   3. I've not checked how does it deal with requests containing
>>>>>>>>>> multipart boundaries and binary data
>>>>>>>>>>   4. Not sure what to do with service description (currently I've
>>>>>>>>>> just removed, but have an idea we can use it to describe service any way,
>>>>>>>>>> even as long as REST does not specify this)
>>>>>>>>>>   5. Write documentation/examples
>>>>>>>>>>
>>>>>>>>>> But I'm stuck on issue #1. I want to know:
>>>>>>>>>>  * do you agree to add dependency in Ladon to tracepyd module or
>>>>>>>>>> not?
>>>>>>>>>>  * If not - do you agree to include tracepyd's implementation
>>>>>>>>>> into Ladon (not as dependency, but to place it somewhere in tools)?
>>>>>>>>>>  * If not - do you agree to modify @ladonize decorator
>>>>>>>>>> functionality to add extra-params for REST (for me its ugly).
>>>>>>>>>>  * If not - which ideas do you have then? :)
>>>>>>>>>>
>>>>>>>>>> I'd like to suggest to add dependency to tracepyd. Currently I
>>>>>>>>>> have a version locally which works fine, so I can publish this solution to
>>>>>>>>>> my branch to let you see.
>>>>>>>>>>
>>>>>>>>>> Thank you in advance for your time and advice,
>>>>>>>>>> Mike
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> 2012/10/22 Jakob Simon-Gaarde <jakobsg@xxxxxxxxx>
>>>>>>>>>>
>>>>>>>>>>> Interesting :-)
>>>>>>>>>>>
>>>>>>>>>>> I'll look at it later - we must make sure that it doesn't break
>>>>>>>>>>> anything.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> 2012/10/22 Mykhailo Stadnyk <mikhus@xxxxxxxxx>
>>>>>>>>>>>
>>>>>>>>>>>> Hi Jacob.
>>>>>>>>>>>>
>>>>>>>>>>>> As I told I'm trying to implement RESTful services support by
>>>>>>>>>>>> Ladon. The basic functionality is implemented - you can check my branch:
>>>>>>>>>>>> lp:~mikhus/ladon/rest<https://code.launchpad.net/~mikhus/ladon/rest> to
>>>>>>>>>>>> see how does it work. Here is an example of RESTful service you may run to
>>>>>>>>>>>> see:
>>>>>>>>>>>>
>>>>>>>>>>>> from ladon.ladonizer import ladonize, restfulize
>>>>>>>>>>>> from ladon.compat import PORTABLE_STRING
>>>>>>>>>>>> from ladon.types.ladontype import LadonType
>>>>>>>>>>>>
>>>>>>>>>>>> class User(LadonType):
>>>>>>>>>>>>  name = PORTABLE_STRING
>>>>>>>>>>>> email = PORTABLE_STRING
>>>>>>>>>>>> passwd = PORTABLE_STRING
>>>>>>>>>>>>  def __init__():
>>>>>>>>>>>>  self.name = ""
>>>>>>>>>>>>  self.email = ""
>>>>>>>>>>>>  self.passwd = ""
>>>>>>>>>>>>
>>>>>>>>>>>> class UserService( object) :
>>>>>>>>>>>>  @restfulize(method="PUT")
>>>>>>>>>>>>  @ladonize(str, str, str, rtype=User)
>>>>>>>>>>>> def newUser( self, u_name, u_email, u_passwd) :
>>>>>>>>>>>>  user = User() # where User is LadonType object
>>>>>>>>>>>> user.name   = u_name
>>>>>>>>>>>>  user.email  = u_email
>>>>>>>>>>>> user.passwd = u_passwd
>>>>>>>>>>>> return user
>>>>>>>>>>>>  @restfulize(method="GET")
>>>>>>>>>>>> @ladonize(int, rtype=[User])
>>>>>>>>>>>>  def getUser(self, u_id) :
>>>>>>>>>>>> return User()
>>>>>>>>>>>>  @restfulize(method="GET")
>>>>>>>>>>>> @ladonize(int, rtype=[User])
>>>>>>>>>>>> def getUsers(self) :
>>>>>>>>>>>>  return [User(), User(), User()]
>>>>>>>>>>>>  @restfulize(method="POST")
>>>>>>>>>>>>  @ladonize(int, str, str, str, rtype=User)
>>>>>>>>>>>> def updUser( self, u_id, u_name, u_email, u_passwd):
>>>>>>>>>>>>  user = User()
>>>>>>>>>>>> user.name   = u_name
>>>>>>>>>>>> user.email  = u_email
>>>>>>>>>>>>  user.passwd = u_passwd
>>>>>>>>>>>> return user
>>>>>>>>>>>>  @restfulize(method="DELETE")
>>>>>>>>>>>> @ladonize(int, rtype=bool)
>>>>>>>>>>>> def delUser( self, u_id) :
>>>>>>>>>>>>  return True
>>>>>>>>>>>>
>>>>>>>>>>>> By the way, there is at least one issue I don't know how to
>>>>>>>>>>>> solve better, so I want your advice.
>>>>>>>>>>>>
>>>>>>>>>>>> As you see I've implemented specific decorator to mark service
>>>>>>>>>>>> methods as RESTful. It's @restfulize decorator. The problem that it is
>>>>>>>>>>>> implemented now as:
>>>>>>>>>>>>
>>>>>>>>>>>> def restfulize( *dargs, **dkwargs):
>>>>>>>>>>>>  def decorator(fn):
>>>>>>>>>>>> * minfo = fn._ladon_method_info*
>>>>>>>>>>>> if 'method' in dkwargs:
>>>>>>>>>>>>  if not minfo.restfulize( **dkwargs):
>>>>>>>>>>>> raise RestfulMethodConfigMismatch( fn.func_name)
>>>>>>>>>>>>
>>>>>>>>>>>> return fn # returns ladonized method
>>>>>>>>>>>> return decorator
>>>>>>>>>>>>
>>>>>>>>>>>> The problem is highlighted with bold. Will be good to make this
>>>>>>>>>>>> decorator independent of what @ladonize defined in it's private props. But,
>>>>>>>>>>>> as far as "fn" here is a reference to ladon's "injector" function I have no
>>>>>>>>>>>> way to determine correctly which real method in which class is decorated.
>>>>>>>>>>>> It means that in such case I have no way to extract LadonMethodInfo
>>>>>>>>>>>> associated to "fn" in other way but implemented for the moment.
>>>>>>>>>>>>
>>>>>>>>>>>> The real problem is when I want to override @ladonize
>>>>>>>>>>>> decorator. In this case @resftulize becomes nonoperational because I loose
>>>>>>>>>>>> reference to "_ladon_method_info" and means that I should take care to
>>>>>>>>>>>> bypass all the existing private properties in 3-d party override
>>>>>>>>>>>> implementation, which is not good at all. So I want to have @restfulize and
>>>>>>>>>>>> @ladonize implemented without hard dependency between them.
>>>>>>>>>>>> Via global_service_collection() it's possible to do but you
>>>>>>>>>>>> need to determine real method function in some way.
>>>>>>>>>>>>
>>>>>>>>>>>> So I see the following ways to fix it:
>>>>>>>>>>>>
>>>>>>>>>>>> 1. Via decorator registry (Here we solved such kind of problem
>>>>>>>>>>>> by creating DecoratorRegistry (I've published it as
>>>>>>>>>>>> http://pypi.python.org/pypi/tracepyd))
>>>>>>>>>>>> 2. By adding REST functionality to @ladonize decorator itself,
>>>>>>>>>>>> to have something like:
>>>>>>>>>>>>     @ladonize(PORTABLE_STRING, rtype=[PORTABLE_STRING], *rest={
>>>>>>>>>>>> "method" : "GET", "produces": "XML" }*)
>>>>>>>>>>>>
>>>>>>>>>>>> Maybe you can point me to some other ideas you know how to
>>>>>>>>>>>> resolve such kind of problem.
>>>>>>>>>>>>
>>>>>>>>>>>> Best regards,
>>>>>>>>>>>> Mike
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> --
>>>>>>>>>>> Med venlig hilsen / Best regards
>>>>>>>>>>> Jakob Simon-Gaarde
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Med venlig hilsen / Best regards
>>>>>>>>> Jakob Simon-Gaarde
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Med venlig hilsen / Best regards
>>>>>>>> Jakob Simon-Gaarde
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Med venlig hilsen / Best regards
>>>>>>> Jakob Simon-Gaarde
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Med venlig hilsen / Best regards
>>>>>> Jakob Simon-Gaarde
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Med venlig hilsen / Best regards
>>>> Jakob Simon-Gaarde
>>>>
>>>
>>>
>>
>>
>> --
>> Med venlig hilsen / Best regards
>> Jakob Simon-Gaarde
>>
>> --
>> Mailing list: https://launchpad.net/~ladon-dev-team
>> Post to     : ladon-dev-team@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~ladon-dev-team
>> More help   : https://help.launchpad.net/ListHelp
>>
>>
>


-- 
Med venlig hilsen / Best regards
Jakob Simon-Gaarde

Follow ups

References