FastCGI and HTTP handlers for Python's
socketserverclasses
FastCGI is a way for front-end servers to talk to back-end workers in a (somewhat) efficient and (somewhat) simple way. Although it's been around since 1996, it is not very widely appreciated, except in the PHP community, where it is very commonly used.
It can be a great approach for hosting Python scripts, avoiding the overhead of creating a new Python process for every request (as standard CGI would otherwise require) and without requiring large dependencies, complex C projects, or fiddly deployments. fastcgi has no dependencies other than fastcore.
There's no new frameworks or concepts to learn. Just call send to send anything you like back to the client, read the parameters from params, and the input from the client from stdin.
fastcgi requires a front-end web server. If you don't already have one set up, we recommend Caddy. To forward all requests to example.com to a fastcgi server listening on port 1234 create a file called Caddyfile with the following contents, and then caddy run:
example.com
reverse_proxy localhost:1234 { transport fastcgi }
This library also provides an HTTP handler that can be used in an identical way, except remove { transport fastcgi } from the above Caddyfile example. Python's standard library already includes an HTTP handler (in http.server), however the documentation warns that that module should not be used in production code. The HTTP handler provided here is trimmed down to a minimal implementation (just 40 lines of code) so that it can easily be studied and extended. It uses the same basic API as Python's other socketserver classes (and the same as FcgiHandler here) so there's fewer new concepts to understand.
pip install fastcgi or conda install -c fastai fastcgi
See the full docs pages for each class for details. Quick overviews of each approach are shown below.
Using the fastcgi decorator you can use CGI scripts with minimal changes. Just add the decorator above a function used for CGI, and it converts that script automatically into a FastCGI server, e.g if you save this as server.py:
@fastcgi()
def hello():
query = os.environ["QUERY_STRING"]
content = sys.stdin.read()
sys.stdout.write(f"Content-type: text/html\r\n\r\n<html>{content} ; ")
sys.stdout.write(f"{query}</html>\r\n")...then if you run python server.py it will make a unix socket available as fcgi.sock in the current directory.
FcgiHandler is used in much the same way as Python's BaseRequestHandler. Here's an example:
class TestHandler(FcgiHandler):
def handle(self):
print('query:', self.environ['QUERY_STRING'])
print('content type:', self.environ['HTTP_CONTENT_TYPE'])
print('stdin:', self['stdin'].read())
self['stdout'].write(b"Content-type: text/html\r\n\r\n<html>foobar</html>\r\n")You can run this using any of Python's socketserver classes, e.g to listen on localhost port 1234:
with TCPServer(('localhost',1234), TestHandler) as srv:
srv.handle_request()See the API docs for FcgiHandler for an end-to-end example.
You can also create a forking or threading server by using Python's mixins or predefined classes.
In your handle method, you can use the stdin, stdout, and stderr attributes, which each contain a BytesIO stream.
fastcgi also comes with the MinimalHTTPHandler class, which provides very similar functionality to FcgiHandler, but using the HTTP protocol instead of the FastCGI protocol. Here's an example:
class _TestHandler(MinimalHTTPHandler):
def handle(self):
print(f'Command/path/version: {self.command} {self.path} {self.request_version}')
print(self.headers)
self.send_response(200)
self.send_header("Content-Type", "text/plain")
self.send_header('Content-Length', '2')
self.end_headers()
self.wfile.write(b'ok')You can run it with a socketserver server in the same way shown above for FcgiHandler.