A Python DSL for building REST clients
05 Jan 2015In this article, I implemented a mini-dsl in Python for describing REST APIs. This mini-dsl allows developers to intuitively build a Python HTTP client easily.
When I worked at Sense Tecnic Systems, I needed to write a Python client for our REST API. We had another guy who wrote the API up, and the code shocked me. We had a file with 2000 lines, and every function wlooked like this:
def send_data_post(self, sensor_id, data, username = None, password = None):
""" Send new data to a sensor.
:param sensor_id: Sensor ID to send data to.
:type sensor_id: str.
:param data: Data to send to this sensor.
:type data: dict
:param username: If provided with password, overrides the default login credentials supplied on initialization.
:type username: str.
:param password: Used in combination with username.
:type password: str.
:raises: WotkitException if a status code is not 200's"""
#use SenseTecnic
#log.info("Sending data to wotkit for sensor " + sensor + ": " + str(attributes))
sensor_id = str(sensor_id)
auth_credentials = self._get_login_credentials(username, password)
url = self.api_url+'/sensors/'+sensor_id+'/data'
try:
response = requests.post(url = url, auth=auth_credentials, data = data)
except Exception as e:
raise WotkitException("Error in sending new data by POST to sensor at url: " + url + ". Error: " + str(e))
if response.ok:
log.debug("Success sending POST sensor data to url: " + url)
return True
else:
raise WotkitException("Error in sending new data by POST to sensor at url: " + url + "\n Response Code: " + str(response.status_code) + "\n Response Text: " + response.text.encode(response.encoding))
Can we do better?
I abhorred this code. Was this style of coding really necessary? REST APIs have a regular pattern. I myself found it unintuitive to work with this wrapper everyday because we are bastardizing a REST API into an RPC-style one.
So, instead of this crap, can we somehow express API access REST-like. i.e.
# Equivalent to: GET http://wotkit.sensetecnic.com/api/v2/sensors/234/data
API = Base("http://wotkit.sensetecnic.com/api/v2")
data = API.sensors.id(234).data.GET()
Mapping the URLs
For developers trying to map out their REST API, it should be a simple process of mapping different URL fragments. Each fragment is represented as a RequestHandler which should:
- Invoke HTTP Methods like GET, POST, PUT, DELETE
- Link to sub-RequestHandlers.
A definition of a RequestHandler should:
- Describe the URL Fragment it represents
- Define allowed requestMethods
- Define a mapping from this RequestHandler to sub-RequestHandlers
This is what our mapped REST API looks like with this DSL.
class Base(RequestHandler):
""" Base URL Fragment which has knowledge about magical bits around it """
requestMethods = ["GET", "POST", "DELETE", "PUT"]
requestMapping = {
"users": User,
"orgs": Organization,
"organization": Organization,
"sensors": Sensor,
}
def __init__(self, api_url):
if isinstance(api_url, list):
self.fragments = api_url
else:
self.fragments = [api_url]
self.options = {}
class Organization(RequestHandler):
fragmentUrl = "orgs"
requestMapping = {"members": OrgMember}
class User(RequestHandler):
fragmentUrl = "users"
class Sensor(RequestHandler):
fragmentUrl = "sensors"
requestMapping = {
"domain_id": DomainId,
"data": SensorData,
"metrics": Metric
}
This lets you use the API like so:
API = Base("http://wotkit.sensetecnic.com/api/v2")
# This represents the path: `wotkit.sensetecnic.com/api/v2/api/sensors/foo
mySensorPath = API.sensors.id("foo")
# This is equivalent to: `GET wotkit.sensetecnic.com/api/v2/api/sensors/foo`
mySensor = mySensorPath.GET()
# This is equivalent to: `GET wotkit.sensetecnic.com/api/v2/api/sensors/foo/data`
data = mySensor.data.GET()
# Or all at once: `GET wotkit.sensetecnic.com/api/v2/api/sensors/foo/data`
API.sensors.id("foo").data.GET()
Implementation
The RequestHandler parent class does 100% of the heavy lifting. Python objects
can override __getattr__
which lets us generate attributes on the fly.
In my implementation below, __getattr__
looks up each a RequestHandler’s
requestMethods and requestMapping fields to navigate a REST API. I also
overrided __dir__
so that IPython can autocomplete auto-genenerated fields.
I needed a bit of boilerplate code to implement Id fragments. Each Id fragment needs to the requestMapping from it’s parent RequestHandler.
In the code below for example, mySensor
needs to inherit the requestMapping
from API.sensors
.
API = Base("http://wotkit.sensetecnic.com/api/v2")
mySensor = API.sensors.id("123")
Neat Trick: Enumerate an entire API
Since the API uses Python datastructures, we can easily enumerate through all the endpoints. With this, you can use it to smoke test your entire API!
def enumerateAPI(resource):
""" Enumerate the entire API as a list of fragments we want to evaluate """
currentResource = resource.id(1)
subResources = map(lambda k: currentResource.__getattr__(k), currentResource.requestMapping.keys())
totalResources = [item for sublist in [enumerateAPI(i) for i in subResources] for item in sublist]
totalResources.append(currentResource)
totalResources.append(resource)
return totalResources
API = Base("http://wotkit.sensetecnic.com/api/v2")
test_apis = enumerateAPI(base)
for api in test_apis:
# Smoke test them in here!
Source
Here’s the source code for the entire dsl + endpoints.
Example Code
import requests
import json
class RequestHandler(object):
""" Base URL Fragment which has knowledge about magical bits around it """
requestMethods = ["GET", "POST", "DELETE", "PUT"]
requestMapping = {}
def __init__(self, fragments):
self.fragments = fragments
def __getattr__(self, name):
""" Defer all the magic here: if we get a requestMethod, we return a
function which can be invoked using the requests library, otherwise we
build a map using the meta shit defined below"""
if (name == "id"):
# Prevent base nesting itself with ids
if (self.__class__ == Base):
return self
return IdFragment(self.fragments, self.__class__)
elif (name in self.requestMethods):
url = "/".join(self.fragments)
return RequestWrapper(method=name, url=url)
elif (name in self.requestMapping):
# print "attempted request map %s" % (name)
# TODO: add error handling here if requestMapping does not exist in name
nextObj = self.requestMapping[name]
new_fragments = list(self.fragments)
new_fragments.append(nextObj.fragmentUrl)
return nextObj(new_fragments)
else:
raise AttributeError("API has no subpath: %s. Allowed: %s" %
(name, ",".join(self.requestMapping.keys())))
def __str__(self):
return "/".join(self.fragments)
def __dir__(self):
return sorted(set(
dir(type(self)) + self.__dict__.keys() + self.requestMethods
+ self.requestMapping.keys()
))
def __repr__(self):
return "<" + str(self) + ">"
class RequestWrapper:
""" Wraps up a Request object so that appending options has the same API """
def __init__(self, method, url):
headers = {"content-type": "application/json"}
self.options = {"method": method, "url": url, "headers": headers}
def __call__(self, **kwargs):
""" Delegates work to the requests library below. Passes options to the requests library
with the notable extras:
- auth encodes with basic-auth. it will take either a tuple consisting of
(username, password) or convert any dictionary with username and password fields set
"""
options = dict(self.options.items() + kwargs.items())
data = options.get("data")
if (data is not None):
content_type = options["headers"]["content-type"]
if (content_type == "application/json"):
options["data"] = json.dumps(data)
r = requests.Request(**options).prepare()
s = requests.Session()
return s.send(r)
class IdFragment(RequestHandler):
""" Returns an id handler which follows the mapping of the preceding API """
def __init__(self, fragments, handlerClass):
self.fragments = fragments
self.handlerClass = handlerClass
def __call__(self, value):
""" Append this fragment to itself """
new_fragments = list(self.fragments)
new_fragments.append(str(value))
return self.handlerClass(new_fragments)
###############################################################################
# DOMAIN SPECIFIC MAPPINGS
###############################################################################
class Application(RequestHandler):
fragmentUrl = "apps"
class SensorData(RequestHandler):
fragmentUrl = "data"
class DomainId(RequestHandler):
requestMethods = ["GET", "PUT"]
fragmentUrl = "domain_id"
class Metric(RequestHandler):
fragmentUrl = "metrics"
class Sensor(RequestHandler):
fragmentUrl = "sensors"
requestMapping = {
"domain_id": DomainId,
"data": SensorData,
"metrics": Metric
}
class OrgMember(RequestHandler):
fragmentUrl = "members"
class User(RequestHandler):
fragmentUrl = "users"
class Organization(RequestHandler):
fragmentUrl = "orgs"
requestMapping = {"members": OrgMember}
class Base(RequestHandler):
""" Base URL Fragment which has knowledge about magical bits around it """
requestMethods = ["GET", "POST", "DELETE", "PUT"]
requestMapping = {
"apps": Application,
"users": User,
"orgs": Organization,
"organization": Organization,
"sensors": Sensor,
}
def __init__(self, api_url):
if isinstance(api_url, list):
self.fragments = api_url
else:
self.fragments = [api_url]
self.options = {}
def __call__(self, *args):
return self