Skip to main content

WSGI server implemented in Rust.

Project description

Pyruvate WSGI server

Pyruvate is a reasonably fast, multithreaded, non-blocking WSGI server implemented in Rust.



If you are on Linux and use a recent Python version,

$ pip install pyruvate

is probably all you need to do.

Manylinux2010 binary wheels

Manylinux2010 wheels are available for active Python 3 versions (currently 3.6-3.9). Pip supports manylinux2010 wheels since version 19.0. Setuptools (used by e.g. zc.buildout) supports manylinux2010 wheels since version 42.0.0. So if you are on Linux and the Pyruvate source distribution is preferred over the binary package try upgrading pip and/or setuptools first.

Source installation

On macOS or if for any other reason you want to install the source tarball (e.g. using pip install –no-binary) you will need to install Rust first.

Development Installation

  • Install Rust

  • Install and activate a Python 3 (>= 3.6) virtualenv

  • Install setuptools_rust using pip:

    $ pip install setuptools_rust
  • Install pyruvate, e.g. using pip:

    $ pip install -e git+[test]

Using Pyruvate in your WSGI application

From Python

A hello world WSGI application using pyruvate listening on and using 2 worker threads looks like this:

import pyruvate

def application(environ, start_response):
    """Simplest possible application object"""
    status = '200 OK'
    response_headers = [('Content-type', 'text/plain')]
    start_response(status, response_headers, None)
    return [b"Hello world!\n"]

pyruvate.serve(application, "", 2)

Using PasteDeploy

Again listening on and using 2 worker threads:

use = egg:pyruvate#main
socket =
workers = 2

Configuration Options

Required: The TCP socket Pyruvate should bind to. pyruvate also supports systemd socket activation If you specify None as the socket value, pyruvate will try to acquire a socket bound by systemd.
Required: Number of worker threads to use.
Optional: Use a blocking connection for writing. Pyruvate currently supports two types of workers: The default worker will write in a non-blocking manner, registering WSGI responses for later processing if the socket isn’t available for writing immediately. By setting this option to True you can enable a worker that will instead set the connection into blocking mode for writing. Defaults to False.
Optional: Maximum number of request headers that will be parsed. If a request contains more headers than configured, request processing will stop with an error indicating an incomplete request. The default is 24 headers
Optional: Log asynchronously using a dedicated thread. Defaults to True.
Optional: Specify how often to reuse an existing connection. Setting this parameter to 0 will effectively disable keep-alive connections. This is the default.
Optional: Specify a timeout in integer seconds for keepalive connection. The persistent connection will be closed after the timeout expires. Defaults to 60 seconds.
Optional: Whether to use chunked transfer encoding if no Content-Length header is present. Defaults to False.


Pyruvate uses the standard Python logging facility. The logger name is pyruvate. See the Python documentation (logging, logging.config) for configuration options.

Example Configurations

Django 2

After installing Pyruvate in your Django virtualenv, create or modify your file (one worker listening on

import os
import pyruvate

from django.core.wsgi import get_wsgi_application

os.environ.setdefault("DJANGO_SETTINGS_MODULE", "your_django_application.settings")

application = get_wsgi_application()

pyruvate.serve(application, "", 1)

You can now start Django + Pyruvate with:

$ python

Override settings by using the DJANGO_SETTINGS_MODULE environment variable when appropriate. Tested with Django 2.2.x.


First create a basic WSGI configuration following the MapProxy deployment documentation. Then modify so it is using Pyruvate (2 workers listening on

from logging.config import fileConfig
import os.path
import pyruvate
fileConfig(r'/path/to/mapproxy/log.ini', {'here': os.path.dirname(__file__)})

from mapproxy.wsgiapp import make_wsgi_app
application = make_wsgi_app(r'/path/to/mapproxy/mapproxy.yml')

pyruvate.serve(application, "", 2)

Start from your virtualenv:

$ python

Tested with Mapproxy 1.12.x.

Plone 5.2

Using zc.buildout and plone.recipe.zope2instance you can define an instance part using Pyruvate’s PasteDeploy <> _entry point:

recipe = plone.recipe.zope2instance
http-address =
eggs =
wsgi-ini-template = ${buildout:directory}/templates/

The server section of the template provided with the wsgi-ini-template option should look like this (3 workers listening on http-address as specified in the buildout [instance] part):

use = egg:pyruvate#main
socket = %(http_address)s
workers = 3

There is a minimal buildout example configuration for Plone 5.2 in the examples directory of the package.

Tested with Plone 5.2.x.


Install Pyruvate in your Pyramid virtualenv using pip:

$ pip install pyruvate

Modify the server section in your .ini file to use Pyruvate’s PasteDeploy <> _entry point (listening on and using 5 workers):

use = egg:pyruvate#main
socket =
workers = 5

Start your application as usual using pserve:

$ pserve path/to/your/configfile.ini

Tested with Pyramid 1.10.x.

Nginx settings

Like other WSGI servers pyruvate should be used behind a reverse proxy, e.g. Nginx:

location / {
    proxy_pass http://localhost:7878;

Nginx doesn’t use keepalive connections by default so you will need to modify your configuration if you want persistent connections.


1.0.3 (2021-06-05)

  • HEAD request: Do not complain about content length mismatch (#4)
  • More appropriate log level for client side connection termination (#5)
  • Simplify request parsing

1.0.2 (2021-05-02)

  • Close connection and log an error in the case where the actual content length is less than the Content-Length header provided by the application
  • Fix readme

1.0.1 (2021-04-28)

  • Fix decoding of URLs that contain non-ascii characters
  • Raise Python exception when response contains objects other than bytestrings instead of simply logging the error.

1.0.0 (2021-03-24)

  • Improve query string handling

0.9.2 (2021-01-30)

  • Better support for HTTP 1.1 Expect/Continue
  • Improve documentation

0.9.1 (2021-01-13)

  • Improve GIL handling
  • Propagate worker thread name to Python logging
  • Do not report broken pipe as error
  • PasteDeploy entry point: fix option handling

0.9.0 (2021-01-06)

  • Reusable connections
  • Chunked transfer-encoding
  • Support macOS

0.8.4 (2020-12-12)

  • Lower CPU usage

0.8.3 (2020-11-26)

  • Clean wheel build directories
  • Fix some test isolation problems
  • Remove a println

0.8.2 (2020-11-17)

  • Fix blocksize handling for sendfile case
  • Format unix stream peer address
  • Use latest mio

0.8.1 (2020-11-10)

  • Receiver in non-blocking worker must not block when channel is empty

0.8.0 (2020-11-07)

  • Logging overhaul
  • New async_logging option
  • Some performance improvements
  • Support Python 3.9
  • Switch to manylinux2010 platform tag

0.7.1 (2020-09-16)

  • Raise Python exception when socket is unavailable
  • Add Pyramid configuration example in readme

0.7.0 (2020-08-30)

  • Use Python logging
  • Display server info on startup
  • Fix socket activation for unix domain sockets

0.6.2 (2020-08-12)

  • Improved logging
  • PasteDeploy entry point now also uses at most 24 headers by default

0.6.1 (2020-08-10)

  • Improve request parsing
  • Increase default maximum number of headers to 24

0.6.0 (2020-07-29)

  • Support unix domain sockets
  • Improve sendfile usage

0.5.3 (2020-07-15)

  • Fix testing for completed sendfile call in case of EAGAIN

0.5.2 (2020-07-15)

  • Fix testing for completed response in case of EAGAIN
  • Cargo update

0.5.1 (2020-07-07)

  • Fix handling of read events
  • Fix changelog
  • Cargo update
  • ‘Interrupted’ error is not a todo
  • Remove unused code

0.5.0 (2020-06-07)

  • Add support for systemd socket activation

0.4.0 (2020-06-29)

  • Add a new worker that does nonblocking write
  • Add default arguments
  • Add option to configure maximum number of request headers
  • Add Via header

0.3.0 (2020-06-16)

  • Switch to rust-cpython
  • Fix passing of tcp connections to worker threads

0.2.0 (2020-03-10)

  • Added some Python tests (using py.test and tox)
  • Improve handling of HTTP headers
  • Respect content length header when using sendfile

0.1.0 (2020-02-10)

  • Initial release

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for pyruvate, version 1.0.3
Filename, size File type Python version Upload date Hashes
Filename, size pyruvate-1.0.3-cp36-cp36m-manylinux2010_x86_64.whl (1.1 MB) File type Wheel Python version cp36 Upload date Hashes View
Filename, size pyruvate-1.0.3-cp37-cp37m-manylinux2010_x86_64.whl (1.1 MB) File type Wheel Python version cp37 Upload date Hashes View
Filename, size pyruvate-1.0.3-cp38-cp38-manylinux2010_x86_64.whl (1.1 MB) File type Wheel Python version cp38 Upload date Hashes View
Filename, size pyruvate-1.0.3-cp39-cp39-manylinux2010_x86_64.whl (1.1 MB) File type Wheel Python version cp39 Upload date Hashes View
Filename, size pyruvate-1.0.3.tar.gz (65.2 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring DigiCert DigiCert EV certificate Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page