[Web-SIG] Proposed specification: waiting for file descriptor events
cstawarz at csail.mit.edu
Wed May 21 02:42:48 CEST 2008
This is the third draft of my proposed extensions for better
supporting WSGI apps on asynchronous servers. The major changes since
the last draft are as follows:
* The title and abstract now accurately reflect the scope of the
In addition, the extensions are now in the namespace "x-
(instead of "x-wsgiorg.async").
* The proposal for an alternative, non-blocking input stream has been
dropped, since I don't see a way to add one that wouldn't break
Instead, the spec recommends that async servers pre-read the
before invoking the app (either by default or as a configurable
* The mechanism for indicating timeouts no longer requires the server to
know what environ dict the app is using (addressing one of PJE's
* The examples have been updated. The first one shows how an app can
pycurl to perform an outgoing HTTP request in a non-blocking fashion.
The updated spec is included below and is also available at
The example code and some utilities are available in a bzr repository at
Once again, I'd appreciate your comments.
This specification defines a set of extensions that allow a WSGI
application to suspend its execution until an event occurs on a
specified file descriptor.
The architecture of asynchronous (aka event driven) servers requires
all I/O operations, including both interprocess and network
communication, to be non-blocking. For a WSGI-compliant server, this
requirement extends to all applications run on the server. However,
the WSGI specification does not provide sufficient facilities for an
application to ensure that its I/O is non-blocking. Specifically, it
lacks a mechanism by which an application can suspend its execution
until an arbitrary file descriptor (such as one belonging to a socket
or pipe opened by the application) is ready for reading or writing.
This specification defines a standard interface by which servers can
provide such a mechanism to applications.
This specification introduces three new variables to the WSGI
``x-wsgiorg.fdevent.writable``, and ``x-wsgiorg.fdevent.timeout``.
The variables ``x-wsgiorg.fdevent.readable`` and
``x-wsgiorg.fdevent.writable`` are callable objects that accept two
positional arguments, one required and one optional. In the following
description, these arguments are given the names ``fd`` and
``timeout``, but they are not required to have these names, and the
application **must** invoke the callables using positional arguments.
The first argument, ``fd``, is either an integer representing a file
descriptor or an object with a ``fileno`` method that returns such an
integer. The set of acceptable file descriptors is defined to be
those accepted by ``select.select``. (Note that this set is platform
dependent: only sockets are allowed on Windows, whereas sockets,
pipes, and files are acceptable on Unix-like systems.) The second,
optional argument, ``timeout``, is either ``None`` or a floating-point
value in seconds. If omitted, it defaults to ``None``.
When called, ``x-wsgiorg.fdevent.readable`` and
``x-wsgiorg.fdevent.writable`` return the empty string (``''``), which
**must** be yielded by the application iterable to the server (passing
through any middleware). The server then suspends execution of the
application until one of the following conditions is met:
* The specified file descriptor is ready for reading (if the
application called ``x-wsgiorg.fdevent.readable``) or writing (if
the application called ``x-wsgiorg.fdevent.writable``).
* ``timeout`` seconds have elapsed without the desired file descriptor
event occurring (unless the value of ``timeout`` is ``None``, in
which case the wait will never timeout).
* The server detects an error or "exceptional" condition (such as
out-of-band data) on the file descriptor.
Put another way, if the application calls
``x-wsgiorg.fdevent.readable`` and yields the empty string, it will be
suspended until ``select.select([fd],,[fd],timeout)`` would return.
If the application calls ``x-wsgiorg.fdevent.writable`` and yields the
empty string, it will be suspended until
``select.select(,[fd],[fd],timeout)`` would return.
The variable ``x-wsgiorg.fdevent.timeout`` is an object whose truth
value can be changed by the server. (For example, it could be a
``list`` instance, whose truth value is false when empty, true
otherwise.) If ``timeout`` seconds elapse without the desired file
descriptor event occurring, ``x-wsgiorg.fdevent.timeout`` will be true
when the application resumes; otherwise, it will be false. The truth
value of ``x-wsgiorg.fdevent.timeout`` when the application is first
started or after it yields each response-body string is undefined.
The server may use any technique it desires to detect events on an
application's file descriptors. (Most likely, it will add them to the
same event loop that it uses for accepting new client connections,
receiving requests, and sending responses.)
Handling of the Input Stream
While technically outside the scope of this specification, the
application's input stream (``environ['wsgi.input']``) is another
source of potentially blocking I/O that deserves mention.
The methods provided by the input stream follow the semantics of the
corresponding methods of the ``file`` class. In particular, each of
these methods can invoke the underlying I/O function (in this case,
``recv`` on the socket connected to the client) more than once,
without giving the application the opportunity to check whether each
invocation will block. Although authors of asynchronous servers may
be tempted to provide a non-standard input stream that supports
on-demand, non-blocking reads, such an input stream would be
incompatible with WSGI middleware.
In order to avoid these problems, it is strongly recommended that
asynchronous servers pre-read the entire request body before invoking
the application, either by default or as a configurable option. Doing
so will ensure that the input stream is compatible with middleware and
that reads from it are always non-blocking.
The following application acts as a proxy to `python.org
<http://python.org/>`_. It uses a ``pycurl.CurlMulti`` instance to
perform the outgoing HTTP request in a non-blocking fashion. When the
``CurlMulti.perform`` method detects that its next I/O operation would
block, it returns control to the application, which then yields until
the file descriptor of interest becomes readable or writable as
required. If the descriptor is not ready after one second, the
application sends a ``504 Gateway Timeout`` response to the client and
def pyorg_proxy(environ, start_response):
result = StringIO()
c = pycurl.Curl()
c.setopt(pycurl.URL, 'http://python.org' + environ['PATH_INFO'])
m = pycurl.CurlMulti()
ret, num_handles = m.perform()
if ret != pycurl.E_CALL_MULTI_PERFORM:
if not num_handles:
read, write, exc = m.fdset()
yield environ['x-wsgiorg.fdevent.readable'](read, 1.0)
msg = 'The request to python.org timed out.'
start_response('504 Gateway Timeout',
start_response('200 OK', [('Content-Type', 'application/octet-
The following adapter allows an application that uses the
``x-wsgiorg.fdevent`` extensions to run on a server that does not
support them, without any modification to the application's code::
def wrapper(environ, start_response):
select_args = [None]
def readable(fd, timeout=None):
select_args = ([fd], , [fd], timeout)
def writable(fd, timeout=None):
select_args = (, [fd], [fd], timeout)
environ['x-wsgiorg.fdevent.readable'] = readable
environ['x-wsgiorg.fdevent.writable'] = writable
timeout = False
environ['x-wsgiorg.fdevent.timeout'] = TimeoutWrapper()
for result in application(environ, start_response):
if result or (not select_args):
ready = select.select(*select_args)
timeout = (ready == (, , ))
select_args = None
* The empty string yielded by an application after calling
``x-wsgiorg.fdevent.readable`` or ``x-wsgiorg.fdevent.writable``
must pass through any intervening middleware and be detected by the
server. Although WSGI explicitly requires middleware to relay such
strings to the server (see `Middleware Handling of Block Boundaries
some components may not, making them incompatible with this
* To prevent an application that does blocking I/O from blocking the
entire server, an asynchronous server could run each instance of the
application in a separate thread. However, since asynchronous
servers achieve high levels of concurrency by expressly *avoiding*
multithreading, this technique will almost always be unacceptable.
* The `greenlet <http://codespeak.net/py/dist/greenlet.html>`_ package
enables the use of cooperatively-scheduled micro-threads in Python
programs, and a WSGI server could potentially use it to pause and
resume applications around blocking I/O operations. However, such
micro-threading is not part of the Python language or standard
library, and some server authors may be unwilling or unable to make
use of it.
* Some third-party libraries (such as `PycURL
<http://pycurl.sourceforge.net/>`_) provide non-blocking interfaces
that may need to monitor multiple file descriptors for events
simultaneously. Since this specification allows an application to
wait on only one file descriptor at a time, application authors may
find it difficult or impossible to use such libraries, or they may
be limited to a subset of the libraries' capabilities.
Although this specification could be extended to include an
interface for waiting on multiple file descriptors, it is unclear
whether it would be easy (or even possible) for all servers to
implement it. Also, the appropriate behavior for a multi-descriptor
wait is not obvious. (Should the application be resumed when a
single descriptor is ready? All of them? Some minimum number?)
More information about the Web-SIG