我假设您想使用 Pyramid 应用程序构建 HTTP 代理,尽管您不完全清楚为什么要使用 Pyramid 执行此操作,或者它是这里的最佳解决方案。
Pyramid 和大多数现代 Python Web 应用程序使用 WSGI 标准来解码和处理 HTTP 请求。下面我提出了一个适用于所有 WSGI 服务器的解决方案。这是一个稍作修改的wsgi-proxy application。该示例的完整源代码和实现可在in pyramid_notebook 获得。此实现仅允许代理到不同的localhost 端口,但可以对其进行更新以允许任意 Intranet 源 IP。
您可以编写一个 Pyramid 视图来捕获对某个路径的所有请求,例如/map-to-internal-server:
def proxy_it(request, port):
"""Proxy HTTP request to upstream server."""
proxy_app = WSGIProxyApplication(port)
return request.get_response(proxy_app)
然后你可以拥有 WSGIProxyApplication 来执行连接上游服务器并将响应向下流式传输到客户端的逻辑:
# Courtesy of https://bitbucket.org/dahlia/wsgi-proxy/raw/02ab0dfa8e0078add268e91426e1cc1a52664cf5/wsgi_proxy/__init__.py
import http.client
import logging
from urllib.parse import urlparse, urlunsplit, urlunparse, unquote_plus
#: (:class:`frozenset`) The set of hop-by-hop headers. All header names
#: all normalized to lowercase.
HOPPISH_HEADERS = frozenset([
'keep-alive', 'proxy-authenticate',
'proxy-authorization', 'te', 'trailers', 'transfer-encoding',
'proxy-connection'
# "upgrade", "connection"
])
def is_hop_by_hop(header):
"""Returns :const:`True` if the given ``header`` is hop by hop.
:param header: the header name
:type header: :class:`basestring`
:returns: whether the given ``header`` is hop by hop or not
:rtype: :class:`bool`
"""
return header.lower() in HOPPISH_HEADERS
def reconstruct_url(environ, port):
"""Reconstruct the remote url from the given WSGI ``environ`` dictionary.
:param environ: the WSGI environment
:type environ: :class:`collections.MutableMapping`
:returns: the remote url to proxy
:rtype: :class:`basestring`
"""
# From WSGI spec, PEP 333
url = environ.get('PATH_INFO', '')
if not url.startswith(('http://', 'https://')):
url = '%s://%s%s' % (
environ['wsgi.url_scheme'],
environ['HTTP_HOST'],
url
)
# Fix ;arg=value in url
if '%3B' in url:
url, arg = url.split('%3B', 1)
url = ';'.join([url, arg.replace('%3D', '=')])
# Stick query string back in
try:
query_string = environ['QUERY_STRING']
except KeyError:
pass
else:
url += '?' + query_string
parsed = urlparse(url)
replaced = parsed._replace(netloc="localhost:{}".format(port))
url = urlunparse(replaced)
environ['reconstructed_url'] = url
return url
class WSGIProxyApplication:
"""WSGI application to handle requests that need to be proxied.
You have to instantiate the class before using it as WSGI app::
from wsgiref.simple_server import make_server
app = WSGIProxyApplication()
make_server('', 8080, app).serve_forever()
"""
#: (:class:`types.ClassType`) The connection class of :mod:`httplib` module.
#: It should be a subtype of :class:`httplib.HTTPConnection`.
#: Default is :class:`httplib.HTTPConnection`.
connection_class = http.client.HTTPConnection
def __init__(self, port):
# Target port where we proxy IPython Notebook
self.port = port
def handler(self, environ, start_response):
"""Proxy for requests to the actual http server"""
logger = logging.getLogger(__name__ + '.WSGIProxyApplication.handler')
url = urlparse(reconstruct_url(environ, self.port))
# Create connection object
try:
connection = self.connection_class(url.netloc)
# Build path
path = url.geturl().replace('%s://%s' % (url.scheme, url.netloc),
'')
except Exception:
start_response('501 Gateway Error', [('Content-Type', 'text/html')])
logger.exception('Could not Connect')
yield '<H1>Could not connect</H1>'
return
# Read in request body if it exists
body = length = None
try:
length = int(environ['CONTENT_LENGTH'])
except (KeyError, ValueError):
# This is a situation where client HTTP POST is missing content-length.
# This is also situation where (WebOb?) may screw up encoding and isert extranous = in the body.
# https://github.com/ipython/ipython/issues/8416
if environ["REQUEST_METHOD"] == "POST":
if environ.get("CONTENT_TYPE") == 'application/x-www-form-urlencoded; charset=UTF-8':
body = environ['wsgi.input'].read()
try:
body = unquote_plus(body.decode("utf-8"))
# Fix extra = at end of JSON payload
if body.startswith("{") and body.endswith("}="):
body = body[0:len(body)-1]
except Exception as e:
logger.exception(e)
logger.error("Could not decode body: %s", body)
length = len(body)
else:
body = environ['wsgi.input'].read(length)
# Build headers
logger.debug('environ = %r', environ)
headers = dict(
(key, value)
for key, value in (
# This is a hacky way of getting the header names right
(key[5:].lower().replace('_', '-'), value)
for key, value in environ.items()
# Keys that start with HTTP_ are all headers
if key.startswith('HTTP_')
)
if not is_hop_by_hop(key)
)
# Handler headers that aren't HTTP_ in environ
try:
headers['content-type'] = environ['CONTENT_TYPE']
except KeyError:
pass
# Add our host if one isn't defined
if 'host' not in headers:
headers['host'] = environ['SERVER_NAME']
# Make the remote request
try:
logger.debug('%s %s %r',
environ['REQUEST_METHOD'], path, headers)
connection.request(environ['REQUEST_METHOD'], path,
body=body, headers=headers)
except Exception as e:
# We need extra exception handling in the case the server fails
# in mid connection, it's an edge case but I've seen it
logger.exception(e)
start_response('501 Gateway Error', [('Content-Type', 'text/html')])
yield '<H1>Could not proxy IPython Notebook running localhost:{}</H1>'.format(self.port).encode("utf-8")
return
response = connection.getresponse()
hopped_headers = response.getheaders()
headers = [(key, value)
for key, value in hopped_headers
if not is_hop_by_hop(key)]
start_response('{0.status} {0.reason}'.format(response), headers)
while True:
chunk = response.read(4096)
if chunk:
yield chunk
else:
break
def __call__(self, environ, start_response):
return self.handler(environ, start_response)