spiderweb/docs/middleware/custom_middleware.md

8.3 KiB

writing your own middleware

Sometimes you want to run the same code on every request or every response (or both!). Lots of processing happens in the middleware layer, and if you want to write your own, all you have to do is write a quick class and put it in a place that Spiderweb can find it. A piece of middleware only needs two things to be successful:

  • it must be a class that inherits from SpiderwebMiddleware
  • it must handle either requests, responses, or both!

That's really all there is to it. Here's a template you can copy:

from spiderweb.middleware import SpiderwebMiddleware
from spiderweb.request import Request
from spiderweb.response import HttpResponse


class TestMiddleware(SpiderwebMiddleware):
    def process_request(self, request: Request) -> None:
        # example of a middleware that sets a flag on the request
        request.spiderweb = True

    def process_response(self, request: Request, response: HttpResponse) -> None:
        # example of a middleware that sets a header on the resp
        if hasattr(request, "spiderweb"):
            response.headers["X-Spiderweb"] = "true"

Middleware is run twice: once for the incoming request and once for the outgoing response. You only need to include whichever function is required for the functionality you need.

process_request(self, request: Request) -> Optional[HttpResponse]:

process_request is called before the view is reached in the execution order. You will receive the assembled Request object, and any middleware declared above this one will have already run. Because the request is the single instantiation of a class, you can modify it in-place without returning anything and your changes will stick.

This function also has a special ability; it can stop execution before the view is called by returning a response. If a response is returned, Spiderweb will immediately skip to applying the response middleware and sending the response back to the client. Here's an example of what that might look like:

class JohnMiddleware(SpiderwebMiddleware):
    def process_request(self, request: Request) -> Optional[HttpResponse]:
        if (
            hasattr(request, "user")
            and user.name == "John"
            and request.path.startswith("/admin")
        ):
            return HttpResponse("Go away, John!", status_code=403)

In this case, if the user John tries to access any route that starts with "/admin", he'll immediately get denied and the view will never be called. If the request does not have a user attached to it (or the user is not John), then the middleware will return None and Spiderweb will continue processing.

process_response(self, request: Request, response: HttpResponse) -> None:

This function is called after the view has run and returned a response. You will receive the request object and the response object; like with the request object, the response is also a single instantiation of a class, so any changes you make will stick automatically.

Unlike process_request, returning a value here doesn't change anything. We're already processing a request, and there are opportunities to turn away requests / change the response at both the process_request layer and the view layer, so Spiderweb assumes that whatever it is working on here is what you mean to return to the user. The response object that you receive in the middleware is still prerendered, so any changes you make to it will take effect after it finishes the middleware and renders the response.

on_error(self, request: Request, triggered_exception: Exception) -> Optional[HttpResponse]:

This is a helper function that is available for you to override; it's not often used by middleware, but there are some (like the pydantic middleware) that call on_error when there is a validation failure.

post_process(self, request: Request, rendered_response: str) -> str:

New in 1.3.0!

After process_request and process_response run, the response is rendered out into the raw text that is going to be sent to the client. Right before that happens, post_process is called on each middleware in the same order as process_response (so the closer something is to the beginning of the middleware list, the more important it is).

Note that this function must return something. Each invocation of post_process overwrites the entire output of the response, so make sure to return everything that you want to send. For example, here's a middleware that breaks adjusts the capitalization of the response and also demonstrates passing variables into the middleware:

import random

from spiderweb.request import Request
from spiderweb.middleware import SpiderwebMiddleware
from spiderweb.exceptions import ConfigError


class CaseTransformMiddleware(SpiderwebMiddleware):
    # this breaks everything, but it's hilarious so it's worth it.
    # Blame Sam.
    def post_process(self, request: Request, rendered_response: str) -> str:
        valid_options = ["spongebob", "random"]
        # grab the value from the extra data passed into the server object
        # during instantiation
        method = self.server.extra_data.get("case_transform_middleware_type", "spongebob")
        if method not in valid_options:
            raise ConfigError(
                f"Invalid method '{method}' for CaseTransformMiddleware."
                f" Valid options are {', '.join(valid_options)}"
            )

        if method == "spongebob":
            return "".join(
                char.upper() 
                if i % 2 == 0 
                else char.lower() for i, char in enumerate(rendered_response)
            )
        else:
            return "".join(
                char.upper() 
                if random.random() > 0.5 
                else char for char in rendered_response
            )

# usage:

app = SpiderwebRouter(
    middleware=["CaseTransformMiddleware"],
    case_transform_middleware_type="random",
)

checks

If you want to have runtime verifications that ensure that everything is running smoothly, you can take advantage of Spiderweb's checks feature.

Tip

If you just want to run startup checks, you can also tie this in with the UnusedMiddleware exception, as it'll trigger after the checks run.

A startup check looks like this:

from spiderweb.exceptions import ConfigError
from spiderweb.server_checks import ServerCheck


class MyCheck(ServerCheck):
    # You don't have to extract the message out into a top-level
    # variable, but it does make testing your middleware easier.
    MYMESSAGE = "Something has gone wrong!"

    # The function must be called `check` and it takes no args.
    def check(self):
        if self.server.extra_args.get("mykeyword") != "propervalue":
            # Note that we are returning an exception instead of
            # raising it. All config errors are collected and then
            # raised as a single group of all the errors that
            # happened on startup.
            # If everything looks good, don't return anything.
            return ConfigError(self.MYMESSAGE)

Tip

You should have one check class per actual check that you want to run, as it will make identifying issues much easier.

You can have as many checks as you'd like, and the base Spiderweb instance is available at self.server. All checks must return an exception (not raising it!), as they will all be raised at the same time as part of an ExceptionGroup called StartupErrors.

To enable your checks, link them to your middleware like this:

class MyMiddleware(SpiderwebMiddleware):
    
    checks = [MyCheck, ADifferentCheck]

    def process_request(self, request):
        ...

List as many checks as you need there, and the server will run all of them during startup.

UnusedMiddleware

from spiderweb.exceptions import UnusedMiddleware

If you don't want your middleware to run for some reason, process_request, process_response and post_process can all raise the UnusedMiddleware exception. If this happens, Spiderweb will kick your middleware out of the processing order for the rest of the life of the server. Note that this applies to the middleware as a whole, so all functions in the middleware will not be run if an UnusedMiddleware is raised. This is a great way to mark debug middleware that shouldn't run or create time-delay middleware that runs until a certain condition is met!