Quickstart
Ready to write your first API client with Uplink? This guide will walk you through what you'll need to know to get started.
First, make sure you've installed (or updated) Uplink <install>:
pip install -U uplink
Defining an API Client
Writing a structured API client with Uplink is very simple.
To start, create a subclass of uplink.Consumer. For example, here's
the beginning of our GitHub client (we'll add some methods to this class
soon):
When creating an instance of this consumer, we can use the base_url
constructor argument to identify the target service. In our case, it's
GitHub's public API:
Note
base_url is especially useful for creating clients that target
separate services with similar APIs; for example, we could use this
GitHub consumer to also create clients for any GitHub Enterprise
instance for projects hosted outside of the public
GitHub.com service. Another example is creating
separate clients for a company's production and staging environments,
which are typically hosted on separate domains but expose the same API.
So far, this class looks like any other Python class. The real magic happens when you define methods to interact with the webservice using Uplink's HTTP method decorators, which we cover next.
Making a Request
With Uplink, making a request to a webservice is as simple as invoking a method.
Any method of a Consumer subclass can be decorated with one of
Uplink's HTTP method decorators: @get <uplink.get>,
@post <uplink.post>, @put <uplink.put>, @patch <uplink.patch>,
@head <uplink.head>, and @delete <uplink.delete>:
class GitHub(Consumer):
@get("repositories")
def get_repos(self):
"""List all public repositories."""
As shown above, the method's body can be left empty.
The decorator's first argument is the resource endpoint: i.e., the
relative path from base_url, which we covered above:
Note
To build a request's absolute URL, Uplink resolves the relative path
against the Consumer's base url using urljoin
<urllib.parse.urljoin>, which implements the RFC
3986 standards. For a
simplified overview of these standards, see these recommendations and
examples
from Retrofit's documentation.
You can also specify query parameters:
Finally, invoke the method to send a request:
>>> github = GitHub(base_url="https://api.github.com/")
>>> github.get_repos()
<Response [200]>
>>> _.url
https://api.github.com/repositories
By default, uplink uses
Requests, so the response we get
back from GitHub is wrapped inside a requests.Response instance. (If
you want, you can swap out <swap_default_http_client> Requests for a
different backing HTTP client, such as aiohttp <sync_vs_async>.)
Path Parameters
Resource endpoints can include URI template
parameters that depend on method
arguments. A simple URI parameter is an alphanumeric string surrounded
by { and }.
To match the parameter with a method argument, either match the argument's name with the alphanumeric string, like so:
or use the uplink.Path annotation.
Query Parameters
Query parameters can be added dynamically using the uplink.Query
argument annotation.
Setting a default value for the query parameter works like you'd expect it to:
To make the query parameter optional, set the argument's default value
to None. Then, if the argument is not specified at runtime, the
parameter will not appear in the request.
Useful for "catch-all" or complex query parameter combinations, the
uplink.QueryMap annotation accepts a mapping of query parameters:
You can set static query parameters for a method using the
@params <uplink.params> decorator.
@params({"client_id": "my-client", "client_secret": "****"})
@get("users/{username}")
def get_user(self, username): pass
@params <uplink.params> can be used as a class decorator for query
parameters that need to be included with every request:
Request Headers
You can set static headers for a method using the
@headers <uplink.headers> decorator.
@headers({
"Accept": "application/vnd.github.v3.full+json",
"User-Agent": "Uplink-Sample-App"
})
@get("users/{username}")
def get_user(self, username): pass
@headers <uplink.headers> can be used as a class decorator for headers
that need to be added to every request:
@headers({
"Accept": "application/vnd.github.v3.full+json",
"User-Agent": "Uplink-Sample-App"
})
class GitHub(Consumer):
...
A request header can depend on the value of a method argument by using
the uplink.Header function parameter annotation:
@get("user")
def get_user(self, authorization: Header("Authorization"):
"""Get an authenticated user."""
Request Body
The uplink.Body annotation identifies a method argument as the HTTP
request body:
This annotation works well with the keyword arguments parameter
(denoted by the ** prefix):
Moreover, this annotation is useful when using supported serialization
formats, such as JSON <json> and Protocol
Buffers. Take a look at
this guide <serialization> for more about serialization with Uplink.
Form Encoded, Multipart, and JSON Requests
Methods can also be declared to send form-encoded, multipart, and JSON data.
Form-encoded data is sent when
@form_url_encoded <uplink.form_url_encoded> decorates the method. Each
key-value pair is annotated with a uplink.Field annotation:
Multipart requests are used when @multipart <uplink.multipart>
decorates the method. Parts are declared using the uplink.Part
annotation:
JSON data is sent when @json <uplink.json> decorates the method. The
uplink.Body annotation declares the JSON payload:
@json
@patch("user")
def update_user(self, **user_info: uplink.Body):
"""Update an authenticated user."""
Alternatively, the uplink.Field annotation declares a JSON field:
@json
@patch("user")
def update_user_bio(self, bio: Field):
"""Update the authenticated user's profile bio."""
Handling JSON Responses
Many modern public APIs serve JSON responses to their clients.
If your uplink.Consumer subclass accesses a JSON API, you can
decorate any method with @returns.json <uplink.returns.json> to
directly return the JSON response, instead of a response object, when
invoked:
class GitHub(Consumer):
@returns.json
@get("users/{username}")
def get_user(self, username):
"""Get a single user."""
>>> github = GitHub("https://api.github.com")
>>> github.get_user("prkumar")
{'login': 'prkumar', 'id': 10181244, ...
You can also target a specific field of the JSON response by using the
decorator's key argument to select the target JSON field name:
class GitHub(Consumer):
@returns.json(key="blog")
@get("users/{username}")
def get_blog_url(self, username):
"""Get the user's blog URL."""
Note
JSON responses may represent existing Python classes in your application
(for example, a GitHubUser). Uplink supports this kind of conversion
(i.e., deserialization), and we detail this support in
the next guide <serialization>.
Persistence Across Requests from a Consumer
The session property of a uplink.Consumer instance exposes the
instance's configuration and allows for the persistence of certain
properties across requests sent from that instance.
You can provide default headers and query parameters for requests sent
from a consumer instance through its session property, like so:
class GitHub(Consumer):
def __init__(self, base_url, username, password):
super(GitHub, self).__init__(base_url=base_url)
# Creates the API token for this user
api_key = create_api_key(username, password)
# Send the API token as a query parameter with each request.
self.session.params["access_token"] = api_key
@get("user/repos")
def get_user_repos(self, sort_by: Query("sort")):
"""Lists public repositories for the authenticated user."""
Headers and query parameters added through the session are applied to
all requests sent from the consumer instance.
github = GitHub("prkumar", "****")
# Both `access_token` and `sort` are sent with the request.
github.get_user_repos(sort_by="created")
Notably, in case of conflicts, the method-level headers and parameters override the session-level, but the method-level properties are not persisted across requests.
Response and Error Handling
Sometimes, you need to validate a response before it is returned or even calculate a new return value from the response. Or, you may need to handle errors from the underlying client before they reach your users.
With Uplink, you can address these concerns by registering a callback
with one of these decorators:
@response_handler <uplink.response_handler> and
@error_handler <uplink.error_handler>.
@response_handler <uplink.response_handler> registers a callback to
intercept responses before they are returned (or deserialized):
def raise_for_status(response):
"""Checks whether or not the response was successful."""
if 200 <= response.status_code < 300:
# Pass through the response.
return response
raise UnsuccessfulRequest(response.url)
class GitHub(Consumer):
@response_handler(raise_for_status)
@post("user/repo")
def create_repo(self, name: Field):
"""Create a new repository."""
@error_handler <uplink.error_handler> registers a callback to handle
an exception thrown by the underlying HTTP client (e.g.,
requests.Timeout):
def raise_api_error(exc_type, exc_val, exc_tb):
"""Wraps client error with custom API error"""
raise MyApiError(exc_val)
class GitHub(Consumer):
@error_handler(raise_api_error)
@post("user/repo")
def create_repo(self, name: Field):
"""Create a new repository."""
To apply a handler onto all methods of a uplink.Consumer subclass,
you can simply decorate the class itself:
Notably, the decorators can be stacked on top of one another to chain their behaviors:
@response_handler(check_expected_headers) # Second, check headers
@response_handler(raise_for_status) # First, check success
class GitHub(Consumer):
...
Lastly, both decorators support the optional argument
requires_consumer. When this option is set to True, the registered
callback should accept a reference to the Consumer instance as its
leading argument:
@error_handler(requires_consumer=True)
def raise_api_error(consumer, exc_type, exc_val, exc_tb):
"""Wraps client error with custom API error"""
...
class GitHub(Consumer):
@raise_api_error
@post("user/repo")
def create_repo(self, name: Field):
"""Create a new repository."""
Retrying
Networks are
unreliable.
Requests can fail for various reasons. In some cases, such as after a
connection timeout, simply retrying a failed request is appropriate. The
@retry <uplink.retry> decorator can handle this for you:
from uplink import retry, Consumer, get
class GitHub(Consumer):
@retry
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
Without any further configuration, the decorator will retry requests
that fail for any reasons. To constrain which exceptions should prompt
a retry attempt, use the on_exception argument:
from uplink import retry, Consumer, get
class GitHub(Consumer):
# Retry only on failure to connect to the remote server.
@retry(on_exception=retry.CONNECTION_TIMEOUT)
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
Further, as long as the expected exception is thrown, the decorator will
repeatedly retry until a response is rendered. If you'd like to cease
retrying after a specific number of attempts, use the max_attempts
argument:
from uplink import retry, Consumer, get
class GitHub(Consumer):
# Try four times, then fail hard if no response.
@retry(max_attempts=4)
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
The @retry <uplink.retry> decorators offers a bunch of other features!
Below is a contrived example... checkout the
API documentation <retry_api> for more:
from uplink import retry, Consumer, get
class GitHub(Consumer):
@retry(
# Retry on 503 response status code or any exception.
when=retry.when.status(503) | retry.when.raises(Exception)
# Stop after 5 attempts or when backoff exceeds 10 seconds.
stop=retry.stop.after_attempt(5) | retry.stop.after_delay(10)
# Use exponential backoff with added randomness.
backoff=retry.backoff.jittered(multiplier=0.5)
)
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
Finally, like other Uplink decorators, you can decorate a Consumer
subclass with @retry <uplink.retry> to add retry support to all
methods of that class <decorate_consumer>.
Note
Response and error handlers (see here <custom response
handler>) are invoked after the retry condition breaks or after all
retry attempts are exhausted, whatever comes first. These callbacks will
receive the first response/exception that triggers the retry's stop
condition or doesn't match its when filter.
Client-Side Rate Limiting
Often, an organization may enforce a strict limit on the number of requests a client can make to their public API within a fixed time period (e.g., 15 calls every 15 minutes) to help prevent denial-of-service (DoS) attacks and other issues caused by misbehaving clients. On the client-side, we can avoid exceeding these server-side limits by imposing our own rate limit.
The @ratelimit decorator enforces a constraint of X calls every Y
seconds:
from uplink import ratelimit, Consumer, get
class GitHub(Consumer):
@ratelimit(calls=15, period=900) # 15 calls every 15 minutes.
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
When the consumer reaches the limit, it will wait until the next period
before executing any subsequent requests. For blocking HTTP clients,
such as Requests, this means the main thread is blocked until then. On
the other hand, using a non-blocking client <sync_vs_async>, such as
aiohttp, enables you to continue making progress elsewhere while the
consumer waits for the current period to lapse.
Alternatively, you can fail fast when the limit is exceeded by setting
the raise_on_limit argument:
class GitHub(Consumer):
# Raise Exception when the client exceeds the rate limit.
@ratelimit(calls=15, period=900, raise_on_limit=Exception)
@get("user/{username}")
def get_user(self, username):
"""Get user by username."""
Like other Uplink decorators, you can decorate a Consumer subclass
with @ratelimit <uplink.ratelimit> to
add rate limiting to all methods of that class <decorate_consumer>.