webish: reduce POST memory footprint by overriding http.Request

The original twisted.web.http.Request class has a requestReceived method
that parses the form body (in the request's .content filehandle) using
the stdlib cgi.parse_multipart() function. parse_multipart() consumes a
lot of memory when handling large file uploads, because it holds the
arguments entirely in RAM. Nevow's subclass of Request uses cgi.FieldStorage
instead, which is much more memory-efficient.

This patch uses a local subclass of Request and a modified copy of the
requestReceived method. It disables the cgi.parse_multipart parsing and
instead relies upon Nevow's use of FieldStorage. Our code must look for
form elements (which come from the body of the POST request) in req.fields,
rather than assuming that they will be copied into req.args (which now
only contains the query arguments that appear in the URL).

As a result, memory usage is no longer inflated by the size of the file
being uploaded in a POST upload request. Note that cgi.FieldStorage uses
temporary files (tempfile.TemporaryFile) to hold the data.

This closes #29.

diff --git a/src/allmydata/webish.py b/src/allmydata/webish.py
index 8da27ffc09cb028e38b4f63f85c519d2c5e44c68..62a973b52ba95c1c63340bd1a6e09069bc5f95fb 100644 (file)
--- a/src/allmydata/webish.py
+++ b/src/allmydata/webish.py
@@ -20,6 +20,76 @@ def getxmlfile(name):
 class IClient(Interface):
     pass
 
+
+# we must override twisted.web.http.Request.requestReceived with a version
+# that doesn't use cgi.parse_multipart() . Since we actually use Nevow, we
+# override the nevow-specific subclass, nevow.appserver.NevowRequest . This
+# is an exact copy of twisted.web.http.Request (from SVN HEAD on 10-Aug-2007)
+# that modifies the way form arguments are parsed. Note that this sort of
+# surgery may induce a dependency upon a particular version of twisted.web
+
+parse_qs = http.parse_qs
+class MyRequest(appserver.NevowRequest):
+    def requestReceived(self, command, path, version):
+        """Called by channel when all data has been received.
+
+        This method is not intended for users.
+        """
+        self.content.seek(0,0)
+        self.args = {}
+        self.stack = []
+
+        self.method, self.uri = command, path
+        self.clientproto = version
+        x = self.uri.split('?', 1)
+
+        if len(x) == 1:
+            self.path = self.uri
+        else:
+            self.path, argstring = x
+            self.args = parse_qs(argstring, 1)
+
+        # cache the client and server information, we'll need this later to be
+        # serialized and sent with the request so CGIs will work remotely
+        self.client = self.channel.transport.getPeer()
+        self.host = self.channel.transport.getHost()
+
+        # Argument processing.
+
+##      The original twisted.web.http.Request.requestReceived code parsed the
+##      content and added the form fields it found there to self.args . It
+##      did this with cgi.parse_multipart, which holds the arguments in RAM
+##      and is thus unsuitable for large file uploads. The Nevow subclass
+##      (nevow.appserver.NevowRequest) uses cgi.FieldStorage instead (putting
+##      the results in self.fields), which is much more memory-efficient.
+##      Since we know we're using Nevow, we can anticipate these arguments
+##      appearing in self.fields instead of self.args, and thus skip the
+##      parse-content-into-self.args step.
+
+##      args = self.args
+##      ctype = self.getHeader('content-type')
+##      if self.method == "POST" and ctype:
+##          mfd = 'multipart/form-data'
+##          key, pdict = cgi.parse_header(ctype)
+##          if key == 'application/x-www-form-urlencoded':
+##              args.update(parse_qs(self.content.read(), 1))
+##          elif key == mfd:
+##              try:
+##                  args.update(cgi.parse_multipart(self.content, pdict))
+##              except KeyError, e:
+##                  if e.args[0] == 'content-disposition':
+##                      # Parse_multipart can't cope with missing
+##                      # content-dispostion headers in multipart/form-data
+##                      # parts, so we catch the exception and tell the client
+##                      # it was a bad request.
+##                      self.channel.transport.write(
+##                              "HTTP/1.1 400 Bad Request\r\n\r\n")
+##                      self.channel.transport.loseConnection()
+##                      return
+##                  raise
+
+        self.process()
+
 class Directory(rend.Page):
     addSlash = True
     docFactory = getxmlfile("directory.xhtml")
@@ -969,6 +1039,7 @@ class WebishServer(service.MultiService):
         service.MultiService.__init__(self)
         self.root = Root()
         self.site = site = appserver.NevowSite(self.root)
+        self.site.requestFactory = MyRequest
         s = strports.service(webport, site)
         s.setServiceParent(self)
         self.listener = s # stash it so the tests can query for the portnum