Ridiculously fast and easy module-to-API transformations. Learn in minutes, implement in seconds. Batteries included.
Raw developing speed and ease of use, that's why. asymmetric
is based on starlette ✨! While starlette
is a powerful tool to have, getting it to work from scratch can be a bit of a pain, especially if you have never used it before. The idea behind asymmetric
is to be able to take any module already written and transform it into a working API in a matter of minutes, instead of having to design the module ground-up to work with starlette
(it can also be used to build an API from scratch really fast). With asymmetric
, you will also get some neat features, namely:
- Auto logging (configure logs with the
LOG_FILE
environmental variable). - Server-side error detection and exception handling.
- Asynchronous callback endpoints to make a request, terminate the request immediately and then have the server make a request to a callback endpoint with the results! ✨
- Auto-generated
/docs
and/redoc
endpoint for your API with interactive documentation. - Auto-generated OpenAPI Specification documentation files for your API.
asymmetric
is the spiritual successor to symmetric
. In fact, the base API is identical (you could probably search and replace symmetric
for asymmetric
in legacy code), but asymmetric
uses a more modern underlying technology to achieve better performance and some nifty tricks. Its internal design is also significantly changed! 🎉
Install using pip!
pip install asymmetric
asymmetric
has its own batteries included out of the box! To start your development API server, just run:
asymmetric run <module> --reload
Where <module>
is your module name (in the examples, we will be writing in a file named module.py
, so the module name will be just module
). A starlette
instance will be spawned immediately and can be reached at http://127.0.0.1:8000 by default. We don't have any endpoints yet, so we'll add some later. All of the flags for uvicorn
are accepted as flags for asymmetric run
.
Sidenote: asymmetric
is fully compatible with any ASGI
server! If you don't know what that is, don't worry! You don't need to. But if you want to use another ASGI
server, you can! Just make sure to target the asymmetric
object! In fact, the asymmetric run <module>
command is practically an alias for uvicorn <module>:asymmetric
. So yes, by default the server running will be uvicorn
, but, if you want to, you can also use daphne
or hypercorn
!
The module consists of a main object called asymmetric
, which includes an important element: the router
decorator. Let's analyze it:
from asymmetric import asymmetric
@asymmetric.router("/some-route", methods=["post"], response_code=200, callback=False)
The decorator recieves 4 arguments: the route
argument (the endpoint of the API to which the decorated function will map), the methods
argument (a list of the methods accepted to connect to that endpoint, defaults in only POST
requests), the response_code
argument (the response code of the endpoint if everything goes according to the plan. Defaults to 200
) and the callback
argument (a boolean or an object specifying the request style for that endpoint, defaults to False
, generating normal endpoint behaviour). The callback
attribute will have its own section below, for now we will use the default callback=False
.
Now let's imagine that we have the following method:
def some_function():
"""Greets the world."""
return "Hello World!"
To transform that method into an API endpoint, all you need to do is add one line:
@asymmetric.router("/sample", methods=["get"])
def some_function():
"""Greets the world."""
return "Hello World!"
Run asymmetric run module --reload
and send a GET
request to http://127.0.0.1:8000/sample
. You should get a Hello World!
in response! (To try it with a browser, make sure to run the above command and click this link).
But what about methods with arguments? Of course they can be API'd too! Let's now say that you have the following function:
def another_function(a, b=372):
"""
Adds :a and :b and returns the result of
that operation.
"""
return a + b
To transform that method into an API endpoint, all you need to do, again, is add one line:
@asymmetric.router("/add")
def another_function(a, b=372):
"""
Adds :a and :b and returns the result of
that operation.
"""
return a + b
To give parameters to a function, all we need to do is send a json
body with the names of the parameters as keys. Let's see how! Run asymmetric run module --reload
and send a POST
request (the default HTTP
method) to http://127.0.0.1:8000/add
, now using the httpx
module.
import httpx
payload = {
"a": 48,
"b": 21
}
response = httpx.post("http://127.0.0.1:8000/add", json=payload)
print(response.json())
We got a 69
response! (48 + 21 = 69
). Of course, you can return dictionaries from your methods and those will get returned as a json
body in the response object automagically!
With this in mind, you can transform any existing project into a usable API very quickly!
Given that the underlying framework is starlette
, you can use async
to define your methods, no problem! Here's an example:
@asymmetric.router("/another-add")
async def another_async_function(a, b=372):
"""
Adds :a and :b asynchronously and returns the
result of that operation.
"""
return a + b
Don't you hate it when people don't call you back after a date? We all have lived that annoying experience. But don't worry! asymmetric
will call you back!
Some functions may be too heavy to be executed to respond to an HTTP
request. Maybe your function is a predictor of some sort, and it requires an hour of processing time to spit out results. Here's when the callback
parameter of the asymmetric
decorator comes into play! You can ask asymmetric
to terminate the HTTP
request immediately, keep processing stuff and then, once it finishes, execute a request to a specified endpoint with the results. Let's imagine that we have a predict
endpoint that we want to transform into an API
:
def predict(data):
values = Model.predict(data)
# One hour later...
return values
Just add the asymmetric
decorator and you're good to go!
@asymmetric.router("/predict", callback=True)
def predict(data):
values = Model.predict(data)
# One hour later...
return values
Of course, if you rely on some async
sorcery for your operations, asymmetric
can handle it!
@asymmetric.router("/predict", callback=True)
async def predict(data):
values = await Model.predict(data)
# One hour later...
return values
Start the server with asymmetric run module --reload
and now you are able to call the endpoint using the following snippet:
import httpx
response = httpx.post(
"http://127.0.0.1:8000/predict",
json={"data": mydata},
headers={
"Asymmetric-Callback-URL": "http://callback.url/receive/predictions",
"Asymmetric-Callback-Method": "post",
}
)
print(response)
Wow... What?! You just witnessed the magic of asymmetric
. The response will be available immediately with a 202
status code. Meanwhile, the server will keep processing the request. When it finishes, it will make a POST
request to the endpoint specified in the headers with the content of the method's return value. Cool, right? But what if I want to send the content of the method's return value inside a json
, as the value of a predictions
key? Well, that's easy! Just change the headers!
import httpx
response = httpx.post(
"http://127.0.0.1:8000/predict",
json={"data": mydata},
headers={
"Asymmetric-Callback-URL": "http://callback.url/receive/predictions",
"Asymmetric-Callback-Method": "post",
"Asymmetric-Custom-Callback-Key": "predictions",
}
)
print(response)
That will send a json
with one element, with predictions
as a key and the result of the function as the value. The key here are the headers. They specify what to do with the result of your function. You can also change the required headers, if you want to!
callback_parameters = {
"callback_url_header": "Send-Me-Here",
"callback_method_header": "Use-Me",
"custom_callback_key_header": "Put-Me-In-Here",
}
@asymmetric.router("/predict", callback=callback_parameters)
async def predict(data):
values = await Model.predict(data)
# One hour later...
return values
Now, to achieve the same result as before, the requests must change their headers!
import httpx
response = httpx.post(
"http://127.0.0.1:8000/predict",
json={"data": mydata},
headers={
"Send-Me-Here": "http://callback.url/receive/predictions",
"Use-Me": "post",
"Put-Me-In-Here": "predictions",
}
)
print(response)
As you probably imagine by now, the callback
parameter can be a boolean or a dictionary with the following pseudo-schema:
{
"callback_url_header": {
"required": False,
"type": str,
},
"callback_method_header": {
"required": False,
"type": str,
},
"custom_callback_key_header": {
"required": False,
"type": str,
},
}
If no HTTP
method is specified, the server will POST
the information to the callback URL
.
By default, you can GET
the /docs
or the /redoc
endpoints (using a browser) to access to interactive auto-generated documentation about your API. It will include request bodies for each endpoint, response codes, headers required, default values, and much more!
Tip: Given that the ReDoc Documentation and the SwaggerUI Documentation are based on the OpenAPI standard, using type annotations in your code will result in a more detailed interactive documentation. Instead of the parameters being allowed to be any type, they will be forced into the type declared in your code. Cool, right?
You can also get a json
file containing the OpenAPI spec of your API by running the following command:
asymmetric docs module
You can also use the flag --filename
to specify the name of the output file. It defaults to openapi.json
.
- Parse callback
URL
s to make sure that they are validURL
s, and fail if they aren't.
Clone the repository:
git clone https://github.com/daleal/asymmetric.git
cd asymmetric
Recreate environment:
make get-poetry
make venv-with-dependencies
Run the linters:
make linters
Run the tests:
make tests
This project wouldn't be possible without Encode, as literally all hard dependencies (starlette
and httpx
) are developed by them. Even uvicorn
is theirs. Thanks guys! asymmetric
also takes inspiration from FastAPI
on some implementation details, so thank you too for such an awesome project. Also, I want to thank every person who has given me some kind of advice or that has had to listen to me going on and on and on about this project and the reasons that make it great. Finally, thank you! Yes, you! If you are reading this, there is a good chance that you have at least some interest in asymmetric
. If you have feedback, if you find a bug, if you want to ask for a feature or even if you just want to talk, I'd be glad to hear from you! You're the reason this project exists ♥.