Route handlers#
Route handlers are the core of Litestar. They are constructed by decorating a function or class method with one of the handler decorators exported from Litestar.
For example:
from litestar import get
@get("/")
def greet() -> str:
return "hello world"
In the above example, the decorator includes all the information required to define the endpoint operation for
the combination of the path "/"
and the HTTP verb GET
. In this case it will be a HTTP response with a
Content-Type
header of text/plain
.
Synchronous and asynchronous callables
Both synchronous and asynchronous callables are supported. One important aspect of this is that using a synchronous function which perform blocking operations, such as I/O or computationally intensive tasks, can potentially block the main thread running the event loop, and in turn block the whole application.
To mitigate this, the sync_to_thread
parameter can be set to True
, which
will result in the function being run in a thread pool. Should the function be
non-blocking, sync_to_thread
should be set to False
instead.
If a synchronous function is passed, without setting an explicit sync_to_thread
value, a warning will be raised.
See also
Declaring paths#
All route handler decorators accept an optional path argument.
This argument can be declared as a kwarg using the
path
parameter:
from litestar import get
@get(path="/some-path")
async def my_route_handler() -> None: ...
It can also be passed as an argument without the keyword:
from litestar import get
@get("/some-path")
async def my_route_handler() -> None: ...
And the value for this argument can be either a string path, as in the above examples, or a list
of
string
paths:
from litestar import get
@get(["/some-path", "/some-other-path"])
async def my_route_handler() -> None: ...
This is particularly useful when you want to have optional path parameters:
from litestar import get
@get(
["/some-path", "/some-path/{some_id:int}"],
)
async def my_route_handler(some_id: int = 1) -> None: ...
“reserved” keyword arguments#
Route handler functions or methods access various data by declaring these as annotated function kwargs. The annotated kwargs are inspected by Litestar and then injected into the request handler.
The following sources can be accessed using annotated function kwargs:
Additionally, you can specify the following special kwargs, (known as “reserved keywords”):
cookies
: injects the requestcookies
as a parseddictionary
.headers
: injects the request headers as a parseddictionary
.query
: injects the requestquery_params
as a parseddictionary
.request
: injects theRequest
instance. Available only for HTTP route handlersscope
: injects the ASGI scopedictionary
.socket
: injects theWebSocket
instance. Available only for websocket route handlersstate
: injects a copy of the applicationState
.body
: the raw request body. Available only for HTTP route handlers
Note that if your parameters collide with any of the reserved keyword arguments above, you can provide an alternative name.
For example:
from typing import Any, Dict
from litestar import Request, get
from litestar.datastructures import State
@get(path="/")
async def my_request_handler(
state: State,
request: Request,
headers: Dict[str, str],
query: Dict[str, Any],
cookies: Dict[str, Any],
) -> None: ...
from typing import Any
from litestar import Request, get
from litestar.datastructures import State
@get(path="/")
async def my_request_handler(
state: State,
request: Request,
headers: dict[str, str],
query: dict[str, Any],
cookies: dict[str, Any],
) -> None: ...
Tip
You can define a custom typing for your application state and then use it as a type instead of just using the
State
class from Litestar
Type annotations#
Litestar enforces strict type annotations. Functions decorated by a route handler must have all their arguments and return value type annotated.
If a type annotation is missing, an ImproperlyConfiguredException
will be raised during the
application boot-up process.
There are several reasons for why this limitation is enforced:
To ensure best practices
To ensure consistent OpenAPI schema generation
To allow Litestar to compute the arguments required by a function during application bootstrap
HTTP route handlers#
The most commonly used route handlers are those that handle HTTP requests and responses.
These route handlers all inherit from the HTTPRouteHandler
class, which is aliased as the
decorator called route()
:
from litestar import HttpMethod, route
@route(path="/some-path", http_method=[HttpMethod.GET, HttpMethod.POST])
async def my_endpoint() -> None: ...
As mentioned above, @route()
is merely an alias for HTTPRouteHandler
,
thus the below code is equivalent to the one above:
from litestar import HttpMethod
from litestar.handlers.http_handlers import HTTPRouteHandler
@HTTPRouteHandler(path="/some-path", http_method=[HttpMethod.GET, HttpMethod.POST])
async def my_endpoint() -> None: ...
Semantic handler decorators#
Litestar also includes “semantic” decorators, that is, decorators the pre-set
the http_method
kwarg to a specific HTTP verb,
which correlates with their name:
These are used exactly like @route()
with the sole exception that you cannot configure the
http_method
kwarg:
Click to see the predefined route handlers
from litestar import delete, get, patch, post, put, head
from litestar.dto import DTOConfig, DTOData
from litestar.contrib.pydantic import PydanticDTO
from pydantic import BaseModel
class Resource(BaseModel): ...
class PartialResourceDTO(PydanticDTO[Resource]):
config = DTOConfig(partial=True)
@get(path="/resources")
async def list_resources() -> list[Resource]: ...
@post(path="/resources")
async def create_resource(data: Resource) -> Resource: ...
@get(path="/resources/{pk:int}")
async def retrieve_resource(pk: int) -> Resource: ...
@head(path="/resources/{pk:int}")
async def retrieve_resource_head(pk: int) -> None: ...
@put(path="/resources/{pk:int}")
async def update_resource(data: Resource, pk: int) -> Resource: ...
@patch(path="/resources/{pk:int}", dto=PartialResourceDTO)
async def partially_update_resource(
data: DTOData[PartialResourceDTO], pk: int
) -> Resource: ...
@delete(path="/resources/{pk:int}")
async def delete_resource(pk: int) -> None: ...
Although these decorators are merely subclasses of HTTPRouteHandler
that pre-set
the http_method
, using @get()
,
@patch()
, @put()
, @delete()
, or
@post()
instead of @route()
makes the code clearer and simpler.
Furthermore, in the OpenAPI specification each unique combination of HTTP verb (e.g. GET
, POST
, etc.) and path
is regarded as a distinct operation, and each
operation should be distinguished by a unique operation_id
and optimally
also have a summary
and
description
sections.
As such, using the @route()
decorator is discouraged.
Instead, the preferred pattern is to share code using secondary class methods or by abstracting code to reusable
functions.
Websocket route handlers#
A WebSocket connection can be handled with a @websocket()
route handler.
Note
The websocket handler is a low level approach, requiring to handle the socket directly, and dealing with keeping it open, exceptions, client disconnects, and content negotiation.
For a more high level approach to handling WebSockets, see WebSockets
from litestar import WebSocket, websocket
@websocket(path="/socket")
async def my_websocket_handler(socket: WebSocket) -> None:
await socket.accept()
await socket.send_json({...})
await socket.close()
The @websocket()
decorator is an alias of the
WebsocketRouteHandler
class. Thus, the below code is equivalent to the one above:
from litestar import WebSocket
from litestar.handlers.websocket_handlers import WebsocketRouteHandler
@WebsocketRouteHandler(path="/socket")
async def my_websocket_handler(socket: WebSocket) -> None:
await socket.accept()
await socket.send_json({...})
await socket.close()
In difference to HTTP routes handlers, websocket handlers have the following requirements:
They must declare a
socket
kwarg.They must have a return annotation of
None
.They must be async functions.
These requirements are enforced using inspection, and if any of them is unfulfilled an informative exception will be raised.
OpenAPI currently does not support websockets. As such no schema will be generated for these route handlers.
See also
ASGI route handlers#
If you need to write your own ASGI application, you can do so using the @asgi()
decorator:
from litestar.types import Scope, Receive, Send
from litestar.status_codes import HTTP_400_BAD_REQUEST
from litestar import Response, asgi
@asgi(path="/my-asgi-app")
async def my_asgi_app(scope: Scope, receive: Receive, send: Send) -> None:
if scope["type"] == "http":
if scope["method"] == "GET":
response = Response({"hello": "world"})
await response(scope=scope, receive=receive, send=send)
return
response = Response(
{"detail": "unsupported request"}, status_code=HTTP_400_BAD_REQUEST
)
await response(scope=scope, receive=receive, send=send)
Like other route handlers, the @asgi()
decorator is an alias of the
ASGIRouteHandler
class. Thus, the code below is equivalent to the one above:
from litestar import Response
from litestar.handlers.asgi_handlers import ASGIRouteHandler
from litestar.status_codes import HTTP_400_BAD_REQUEST
from litestar.types import Scope, Receive, Send
@ASGIRouteHandler(path="/my-asgi-app")
async def my_asgi_app(scope: Scope, receive: Receive, send: Send) -> None:
if scope["type"] == "http":
if scope["method"] == "GET":
response = Response({"hello": "world"})
await response(scope=scope, receive=receive, send=send)
return
response = Response(
{"detail": "unsupported request"}, status_code=HTTP_400_BAD_REQUEST
)
await response(scope=scope, receive=receive, send=send)
Limitations of ASGI route handlers#
In difference to the other route handlers, the @asgi()
route handler accepts only three
kwargs that must be defined:
scope
, a mapping of values describing the ASGI connection. It always includes atype
key, with the values being eitherhttp
orwebsocket
, and apath
key. If the type ishttp
, the scope dictionary will also include amethod
key with the value being one ofDELETE
,GET
,POST
,PATCH
,PUT
,HEAD
.receive
, an injected function by which the ASGI application receives messages.send
, an injected function by which the ASGI application sends messages.
You can read more about these in the ASGI specification.
Additionally, ASGI route handler functions must be async functions. This is enforced using inspection, and if the function is not an async functions, an informative exception will be raised.
See the ASGIRouteHandler API reference documentation
for full
details on the @asgi()
decorator and the kwargs it accepts.
Route handler indexing#
You can provide a name
kwarg in all route handler
decorators. The value for this kwarg must be unique, otherwise
ImproperlyConfiguredException
exception will be raised.
The default value for name
is value returned by the handler’s
__str__()
method which should be the full dotted path to the handler
(e.g., app.controllers.projects.list
for list
function residing in app/controllers/projects.py
file).
name
can be used to dynamically retrieve (i.e. during runtime) a mapping
containing the route handler instance and paths, also it can be used to build a URL path for that handler:
from litestar import Litestar, Request, get
from litestar.exceptions import NotFoundException
from litestar.response import Redirect
@get("/abc", name="one")
def handler_one() -> None:
pass
@get("/xyz", name="two")
def handler_two() -> None:
pass
@get("/def/{param:int}", name="three")
def handler_three(param: int) -> None:
pass
@get("/{handler_name:str}", name="four")
def handler_four(request: Request, name: str) -> Redirect:
handler_index = request.app.get_handler_index_by_name(name)
if not handler_index:
raise NotFoundException(f"no handler matching the name {name} was found")
# handler_index == { "paths": ["/"], "handler": ..., "qualname": ... }
# do something with the handler index below, e.g. send a redirect response to the handler, or access
# handler.opt and some values stored there etc.
return Redirect(path=handler_index[0])
@get("/redirect/{param_value:int}", name="five")
def handler_five(request: Request, param_value: int) -> Redirect:
path = request.app.route_reverse("three", param=param_value)
return Redirect(path=path)
app = Litestar(route_handlers=[handler_one, handler_two, handler_three])
route_reverse()
will raise NoRouteMatchFoundException
if route with given
name was not found or if any of path parameters is missing or if any of passed path
parameters types do not match types in the respective route declaration.
However, str
is accepted in place of datetime
, date
,
time
, timedelta
, float
, and Path
parameters, so you can apply custom formatting and pass the result to route_reverse()
.
If handler has multiple paths attached to it route_reverse()
will return the path that consumes
the most number of keyword arguments passed to the function.
from litestar import get, Request
@get(
["/some-path", "/some-path/{id:int}", "/some-path/{id:int}/{val:str}"],
name="handler_name",
)
def handler(id: int = 1, val: str = "default") -> None: ...
@get("/path-info")
def path_info(request: Request) -> str:
path_optional = request.app.route_reverse("handler_name")
# /some-path`
path_partial = request.app.route_reverse("handler_name", id=100)
# /some-path/100
path_full = request.app.route_reverse("handler_name", id=100, val="value")
# /some-path/100/value`
return f"{path_optional} {path_partial} {path_full}"
When a handler is associated with multiple routes having identical path parameters
(e.g., an indexed handler registered across multiple routers), the output of route_reverse()
is
unpredictable. This callable will return a formatted path; however, its selection may appear arbitrary.
Therefore, reversing URLs under these conditions is strongly advised against.
If you have access to Request
instance you can make reverse lookups using
url_for()
method which is similar to route_reverse()
but
returns an absolute URL.
Adding arbitrary metadata to handlers#
All route handler decorators accept a key called opt
which accepts a dictionary
of arbitrary values, e.g.,
from litestar import get
@get("/", opt={"my_key": "some-value"})
def handler() -> None: ...
This dictionary can be accessed by a route guard, or by accessing the
route_handler
property on a Request
object,
or using the ASGI scope
object directly.
Building on opt
, you can pass any arbitrary kwarg to the route handler decorator,
and it will be automatically set as a key in the opt
dictionary:
from litestar import get
@get("/", my_key="some-value")
def handler() -> None: ...
assert handler.opt["my_key"] == "some-value"
You can specify the opt
dictionary at all layers of your application.
On specific route handlers, on a controller, a router, and even on the app instance itself as described in
layered architecture
The resulting dictionary is constructed by merging opt
dictionaries of all layers.
If multiple layers define the same key, the value from the closest layer to the response handler will take precedence.
Signature namespace#
Litestar produces a model of the arguments to any handler or dependency function, called a “signature model” which is used for parsing and validation of raw data to be injected into the function.
Building the model requires inspection of the names and types of the signature parameters at runtime, and so it is
necessary for the types to be available within the scope of the module - something that linting tools such as ruff
or flake8-type-checking
will actively monitor, and suggest against.
For example, the name Model
is not available at runtime in the following snippet:
from __future__ import annotations
from typing import TYPE_CHECKING
from litestar import Controller, post
if TYPE_CHECKING:
from domain import Model
class MyController(Controller):
@post()
def create_item(data: Model) -> Model:
return data
In this example, Litestar will be unable to generate the signature model because the type Model
does not exist in
the module scope at runtime. We can address this on a case-by-case basis by silencing our linters, for example:
from __future__ import annotations
from typing import TYPE_CHECKING
from litestar import Controller, post
# Choose the appropriate noqa directive according to your linter
from domain import Model # noqa: TCH002
However, this approach can get tedious; as an alternative, Litestar accepts a signature_types
sequence at
every layer of the application, as demonstrated in the following example:
from __future__ import annotations
from dataclasses import dataclass
@dataclass
class Model:
a: int
b: str
This module defines our controller, note that we do not import Model
into the runtime namespace,
nor do we require any directives to control behavior of linters.
from __future__ import annotations
from typing import TYPE_CHECKING
from litestar import Controller, post
if TYPE_CHECKING:
from .domain import Model
class MyController(Controller):
@post(sync_to_thread=False)
def post_handler(self, data: Model) -> Model:
return data
Finally, we ensure that our application knows that when it encounters the name “Model” when parsing signatures, that it
should reference our domain Model
type.
from __future__ import annotations
from litestar import Litestar
from .controller import MyController
from .domain import Model
app = Litestar(route_handlers=[MyController], signature_types=[Model])
Tip
If you want to map your type to a name that is different from its __name__
attribute,
you can use the signature_namespace
parameter,
e.g., app = Litestar(signature_namespace={"FooModel": Model})
.
This enables import patterns like from domain.foo import Model as FooModel
inside if TYPE_CHECKING
blocks.
Default signature namespace#
Litestar automatically adds some names to the signature namespace when parsing signature models in order to support injection of the “reserved” keyword arguments.
These names are:
Headers
ImmutableState
Receive
Request
Scope
Send
State
WebSocket
WebSocketScope
The import of any of these names can be safely left inside an if TYPE_CHECKING:
block without any configuration
required.