FastAPI

Why fastAPI

• a free UI to test your rest api - interactive API documentation systems, OpenAPI (Swagger) and JSON Schema standards
• Supposedly security and authentication is integrated
• Supposedly fast
• Editor support, useful for code editor autocompletion

Technologies

• REST
  • how you form your API
  • Representational State Transfer
  • Has an HTTP verb, a URL which related to a bit of data, and a parameterization of that
• Async
  • How you deliver it
  • Co-operative multitasking
  • Puts the onus on you to decide when the rest of your computer should be able to do stuff
  • Easy to get wrong
  • Easy to write code that will block the whole web application rather than make just one request slow
  • Excels in that it's able to deal with thousands and thousands of network sockets
• Dependency Injection (DI)

URL mapping

app = FastAPI()

@app.get( "/" )
def root( ... ):
   return {"greeting": "Hello world"}

app.include_router( events.router, prefix="/events"...)

router = APIRouter()

@router.post( "/", ..., status=201)
def create_object( ... ):
   ...

Getting stuff from requests: Path variables

• given this info
  • GET /events/1234?detail=full
  • Authorization: Token ABCD1234
• implicit way

@router.get( "/event/{id}" )
def get_object( id: int ):
  ...

• more explicit way - make things be required

from fastapi import Path
from pydantic import Required

@router.get( "/event/{id}" )
def get_object( id: int = Path(Required) ):
  ...

Getting stuff from URL requests: Query parameters

• given this info
  • GET /search/?text=something
  • Authorization: Token ABCD1234
• implicit way

@router.get( "/search" )
def search( text: str = None, offset: int = 0, limit: int = 100 ):
  ...

• more explicit way - make things be required

from fastapi import Query

@router.get( "/search" )
def search(
  text: str = Query( default=None, max_length=20 ),
  offset: int = Query( 0 ),
  limit: int = Query( 100 )
):

Data Validation: Pydantic

• Is this valid data??
• Throws helpful error messages if the data isn't formatted correctly
• "This field is required, but you didn't specify it."

from datatime import date as DateType
from enum import Enum
from pydantic import BaseModel

class Types( Enum ):
  done = 'DONE'
  cancelled = 'CANCELLED'

class Event( BaseModel ):
  date: DateType
  type: Types
  text: str

Extract blob of JSON that came in from request

from pydantic import Required

@router.post( "/events/" )
def create_object( event: Event = Required ):
   ...

Creating responses

• If you need to retun something, just return simple Python objects. The framework renders them to JSON for you

@app.get( "/" )
def root( ... ):
  return {"greeting": "Hello World" }

@app.post( "/events/". response_model=Event, status_code=201 )
def create_object( event: Event = Required ):
  ...
  return {
    "date": "2019-06-02",
    "type": "DONE",
    "text": "some stuff got done"
  }

Implied and explicit Query, Path and Body Parameters

Can use standard python types, or more funky python types like datetime.timedelta

Query

• Implicit way: params given in function args are by default query params. Can declare type and default value/optionality
• If you need additional validation, set arg default value as type fastapi.Query
  • min/max length
  • regex
• Declare more metadata
  • title
  • description
  • alias
  • deprecated=True
• set default as first arg to fastapi.Query()
• make required by using elipsis as first arg
• Can allow user to reuse query key by declaring type as typing.List[<type>]
  • Can define list of multiple values as default

Path

• Implicit way: Path paramters are always required as it has to be part of the path
  • Just put the path parameter inside the curly braces on the path argument, e.g., "@app.get("/items/{item_id}")"
• Explicit way: Include the parameter explicitly as a function argument of type fastapi.Path()
• order of the different parameters doesn't matter
  • Can have first argument of function by * to let python know that all arguments following it should be taken as kwargs, which will allow function arg reordering.
• validations
  • gt/lt, ge/le

Body

• subclass the pydantic.BaseModel object
• declare class attributes
• In function args declare type as your new subclass
• fastapi.Body - Use to declare the entire unit of body params in the subclass to be optional, or to setup additional validation, metadata.
  • Additional optional args to Body: example
• pydantic.Field - Use Field for additional validation on individual members of the BaseModel subclass
  • Model attributes/validation/metadata declared via Field() function args include title, max_length, description, gt, example="Foo", etc...
• can nest list of Body Parameters withing body parameters using typing.List, typing.Set (converts collections with duplicate data into set of unique items), typing.Dict[int, float] for example, etc...
• include a member class Config within subclass class attributes
  • schema_extra to provide documentation example

Form

• pass

File

• File inherits from Form
• Files are uploaded as "Form data". Form data has special encoding that;s different from JSON
  • Thus the header multipart/form-data
• UploadFile has following attributes
  • filename
  • content_type
  • file - an object with a file-like interface
  • async methods that need to be awaited if your function is async
    • contents = await myfile.read()
    • write, seek, and close

Response models

• Declare a response model for validation and documentation purposes. Pydantic will also filter out any members not declared in the model.
  • Can also leave out unset params in response using arg response_model_exclude_unset=True
• Typical to have multiple related models.
  • Can dump contents of model A's data instance directly into the constructor for modelB's instance another using ModelB( **model_a.dict() ). The filtering will also happen at this step.
  • Can have SUB-subclasses to reduce code duplication.

Using the Request Directly (GET IP ADDRESS)

• fastapi.Request is just alias to starlette.Request
• Include request: Request in the function args
• This is how you get access to unvarnished, unvalidated request as it comes in from client
• request attributes
  • request.method
  • request.url, request.url.path, request.url.port, request.url.scheme
  • request.headers['content-type'] (e.g.)
  • request.query_params['search'] (e.g.)
  • request.path_params['username'] (e.g.)
  • request.client, *HOSTNAME OR IP ADDRESS* request.client.host,
  • request.cookies.get('mycookie')

Background Tasks

• Send email notifications, process data, write to logs, etc AFTER returning a response

1. Import fastapi.BackgroundTasks
2. Define a parameter background_tasks in the path operation function with type declaration of BackgroundTasks
3. Use background_tasks.add_task( function, args, kwargs )

• Use dependency injection to hang a background task on the queue anywhere on the call stack (not just within the path operation function)

Other stuff I need to learn

• fastapi.Cookie
• the role of header parameters