SOLUTION: WebSockets, Django Channels, MacOS 10.12, Server 5.2, Apache 2.4.23

Fri 13, January 2017


I have WebSocket connections working via Apache in MacOS Server 5.2.

As this caused a lot of head/heartache over the last few weeks I thought I’d better share what I experienced in case someone finds themselves in the same boat!

And the solution is simple.

The short version:

I use MacOS Server 5.2 (ships with Apache 2.4.23) to run a python Django application via the mod_wsgi module.

I had been trying to setup proxypass and wstunnel in MacOS 10.12 & Server 5.2 to handle websocket connections via an ASGI interface server called Daphne running on localhost on port 8001.

I wanted to reverse proxy any WebSocket connection to wss://myapp.local/chat/stream/ to ws://localhost:8001/chat/stream/

From what I had read on all the forums and mailing lists that I had scoured was to simply make some proxypass definitions in the appropriate virtual host and make sure that the mod_proxy and mod_proxy_wstunnel modules were loaded and it would work.

Long story short – from what I understand all of this trouble came down to MacOS Server 5 and one major change:
“A single instance of httpd runs as a reverse proxy, called the Service Proxy, and several additional instances of httpd run behind that proxy to support specific HTTP-based services, including an instance for the Websites service.”

All I needed to do to proxy the websocket connection was the following:


Add (around line 297 (in the section about user websites, webdav):
ProxyPass / http://localhost:8001/
ProxyPassReverse / http://localhost:8001/

RewriteEngine on
RewriteCond %{HTTP:UPGRADE} ^WebSocket$ [NC]
RewriteCond %{HTTP:CONNECTION} ^Upgrade$ [NC]
RewriteRule .* ws://localhost:8001%{REQUEST_URI} [P]

I then kicked over the service proxy:

sudo launchctl unload -w /Applications/Server.app/Contents/ServerRoot/System/Library/LaunchDaemons/com.apple.serviceproxy.plist
sudo launchctl load -w /Applications/Server.app/Contents/ServerRoot/System/Library/LaunchDaemons/com.apple.serviceproxy.plist

And the web socket connections were instantly working!

The long version:

For many weeks I have been trying to get WebSocket connections functioning with Apache and Andrew Godwin’s Django Channels project in an app I am developing.

Django Channels is “a project to make Django able to handle more than just plain HTTP requests, including WebSockets and HTTP2, as well as the ability to run code after a response has been sent for things like thumbnailing or background calculation.”

My interest in Django Channels came from my requirement of a chat system in my webapp. After watching a several of Andrew’s demos on youtube and having read through the docs and eventually installing Andrew’s demo django channels project I figured I would be able to get this working in production on our MacOS Server.

The current release of MacOS 10.12 and Server 5.2 ships with Apache 2.4.23. This comes with the necessary mod_proxy_wstunnel module to be able to proxy WebSocket connections (ws:// and secure wss://) in Apache and is already loaded in the server config file:


Daphne is Andrew’s ASGI interface server that supports WebSockets & long-poll HTTP requests. WSGI does not.

With Daphne running on localhost on a port that MacOS isn’t occupying (i went with 8001) the idea was to get Apache to reverse proxy certain requests to Daphne.

Daphne can be run on a specified port (8001 in tis example) like so (-v2 for more feedback):
daphne -p 8001 yourapp -v2

I wanted Daphne to handle only the web socket connections (as I use currently depend on some apache modules for serving media such as mod_xsendfile). In my case the websocket connection was via /chat/stream/ based on Andrew’s demo project.

From what I had read in MacOS Server’s implementation of Apache the idea is to declare these proxypass commands inside the virtual host files of your “sites” in:


In config files such as:

I did also read that any customisation for web apps running on MacOS Server should be made to the plist file for the required web app in:
In a plist file such as:

I edited the 0000_127.0.0.1_34543_.conf file adding:

ProxyPass /chat/stream/ ws://localhost:8001/
ProxyPassReverse /chat/stream/ ws://localhost:8001/

Eager to test out my first web socket chat connection I refreshed the page only to see an error printed in the apache log:

No protocol handler was valid for the URL /chat/stream/. If you are using a DSO version of mod_proxy, make sure the proxy submodules are included in the configuration using LoadModule.

I had read of many people finding a solution at least with Apache on Ubuntu or a custom install on MacOS.
I even tried installing Apache using Brew and when that didn’t work I almost proceeded to install nginx.

After countless hours/days of googling I reached to the Apache mailing list for some help with this error. Yann Ylavic was very generous with his time and offered me various ideas on how to get it going. After trying the following:

SetEnvIf Request_URI ^/chat/stream/ is_websocket
RequestHeader set Upgrade WebSocket env=is_websocket
ProxyPass /chat/stream/ ws://myserver.local:8001/chat/stream/

I noticed that the interface server on port 8001 Daphne was starting to receive ws connections!
However in the client browser it was logging:

“Error during WebSocket handshake: ‘Upgrade’ header is missing”

From what I could see mod_dumpio was logging that the “Connection: Upgrade” and “Upgrade: WebSocket” headers were being sent as part of the web socket handshake:

mod_dumpio: dumpio_in (data-HEAP): HTTP/1.1 101 Switching Protocols\r\nServer: AutobahnPython/0.17.1\r\nUpgrade: WebSocket\r\nConnection: Upgrade\r\nSec-WebSocket-Accept: 17WYrMeMS8a4ImHpU0gS3/k0+Cg=\r\n\r\n
mod_dumpio.c(164): [client] mod_dumpio: dumpio_out
mod_dumpio.c(58): [client] mod_dumpio: dumpio_out (data-TRANSIENT): 160 bytes
mod_dumpio.c(100): [client] mod_dumpio: dumpio_out (data-TRANSIENT): HTTP/1.1 101 Switching Protocols\r\nServer: AutobahnPython/0.17.1\r\nUpgrade: WebSocket\r\nConnection: Upgrade\r\nSec-WebSocket-Accept: 17WYrMeMS8a4ImHpU0gS3/k0+Cg=\r\n\r\n

However the client browser showed nothing in the response headers.

I was more stumped than ever.

I explored the client side jQuery framework as well as the the Django channels & autobahn module to see if perhaps something was amiss, and then revised my own app and various combinations of suggestions about Apache and it’s module. But nothing stood out to me.

Then I reread the ReadMe.txt inside the apache2 dir:


“Special notes about the web proxy architecture in Server application 5.0:

This version of Server application contains a revised architecture for all HTTP-based services. In previous versions there was a single instance of httpd acting as a reverse proxy for Wiki, Profile, and Calendar/Address services, and also scting as the Websites service.º With this version, there is a major change: A single instance of httpd runs as a reverse proxy, called the Service Proxy, and several additional instances of httpd run behind that proxy to support specific HTTP-based services, including an instance for the Websites service.

Since the httpd instance for the Websites service is now behind a reverse proxy, or Service Proxy, note the following:

It is only the external Service Proxy httpd instance that listens on TCP ports 80 and 443; it proxies HTTP requests and responses to Websites and other HTTP-based services.

I wondered if thus ServiceProxy had something to do with it. I had a look over:


and noticed a comment – “# The user websites, and webdav”.
I figured it wouldn’t hurt to try adding the proxypass definitions & rewrite rules that people had suggested on the forums as their solution.

ProxyPass / http://localhost:8001/
ProxyPassReverse / http://localhost:8001/

RewriteEngine on
RewriteCond %{HTTP:UPGRADE} ^WebSocket$ [NC]
RewriteCond %{HTTP:CONNECTION} ^Upgrade$ [NC]
RewriteRule .* ws://localhost:8001%{REQUEST_URI} [P]

Sure enough after restarting the ServiceProxy it all started to work!


Auto mount SMB/AFP/* volumes in Mac OS X

Wed 7, December 2016

Many moons ago I posted some findings on how to get network volumes to auto mount in Mac OS X.

Recently I needed to do this again – but things have changed a little since OS X 10.5 (currently running 10.11).

All the info I found was in this great post:



edit this file:


add something along the lines of:

/mnt auto_smb -nosuid

(this example creates a mount point on your machine “/mnt” based on the file “auto_smb” that we need to create next.

create a file such as:


add mount points to this auto_smb file:
(this example creates an smb mount point called “projects” which connects to an IP address of as guest and mounts the  “allthestuff” directory from that machine)

projects -fstype=smbfs,soft ://guest@

kick it over with:

sudo automount -vc


Up until now it has been working well!


fcpxImageExporter and my 5 minutes of coding fame

Mon 2, December 2013

A little while ago at Orangutan we had a client ask us to export several hundred frames from a 1 hour video we had produced for them.

We worked in fcpx and manually exporting frames can be quite a task.

So I spent a late afternoon putting together a little app to read fcpxml and use it to export still images based on the markers we had put in the timeline.

I polished it a little more and shared it on the fcp.co forums and over the week the little app had been downloaded over 700 times.

It even got a nice review from Larry Jordan and several retweets from other fcpx pros like Richard Taylor and sites like premiumbeat.com. 

fcpxImageExporter is available over at Orangutan.


Recursive rsync with only specific file types

Mon 28, May 2012

I had never worked out how to use rsync to create intermediate directories when syncing over certain filetypes (in my case nuke’s .nk files). So the other day I googled around and found exactly what i was looking for.

A great way to sync directories and certain filetypes recursively:

rsync -pavr –progress –include “+ */” -include=”*.nk” -exclude=”- *” /my/source/directory /my/destination/

And all the thanks needs to go to Mike:

NSSlider, Mouse Down & NSThread

Tue 2, August 2011

I hope the following makes some sense and is of some use to newbie Mac Developers out there.

For the past week I have been working on Sub It to try get it ready for first release on the Mac App Store.

One of the tools I have put in is a slider to adjust the movie playback rate – similar to the scrub tool in Final Cut Pro.

There were a few hurdles I had to overcome to get this to work.

Initially I realised I would have to subclass NSSlider to be able to detect/grab the MouseDown and MouseUp NSEvents (I wanted the slider to reset to 0 after the user had released the mouse – putting the movie in pause/stop) – googling revealed that NSSlider runs inside its own event loop in the main thread.

Hooking up a delegate to NSSlider subclass allowed me to call actions/methods in the main controller on those Mouse events – eventually I got this to work with the following code (and via Interface Builder I hooked up the delegate property to the my main Controller):


#import <Cocoa/Cocoa.h>
#import <Foundation/Foundation.h>
@interface TimeScrubberSlider : NSSlider {
	IBOutlet id delegate;
@property (readwrite,retain) id delegate;


#import "TimeScrubberSlider.h"
#import "Controller.h"
@implementation TimeScrubberSlider
@synthesize delegate;
- (void)mouseDown:(NSEvent *)theEvent  {
	[delegate doScrubTime:nil];
	[super mouseDown:theEvent];
	[self mouseUp:theEvent];
- (void) mouseUp: (NSEvent*) theEvent
	[self setIntValue:0];
	[delegate doPlay:nil];

This was all working, but the problem was that the time display of the movie would not constantly update – only when the slider was moved would the current time update – not continuously.

So I decided to try attach the updating of the current time display to a separate thread – using NSThread. My thought was that no matter what happened in the main controllers’ thread, this other thread would always do it’s own thing and hopefully not be interrupted.

To get this going I created a method (called ‘launchTimer’) that is called when a file is loaded, and this calls the ‘updateCurrentTime’ method, and this ‘launchTimer’ method is run under the seperate thread. Umm i think that kinda makes sense.

Something along these lines:

snippets from the Controller.m

- (void)setupGUIFromFileRead{
	//just a snippet...
	//begin timer to update the current frame/time display - as separate thread so nsslider "scrub" can play nicely
	[NSThread detachNewThreadSelector:@selector(launchTimer) 
    NSAutoreleasePool *pool = [[NSAutoreleasePool alloc] init];
    [NSThread setThreadPriority:1.0];
    updateTimerToggle = YES;
    while (updateTimerToggle) {  // loop until cancelled
        [self doUpdateCurrentTimeDisplay:nil];
        [NSThread sleepForTimeInterval:.1];
    [pool release];

Multiply – little xcode project

Mon 28, September 2009

Yesterday afternoon I was sitting with Erin & Hugh seeing if we could do 2 digit multiplications in our head, such as 34 * 94.

It turned out I am pretty bad at this so I figured I’d make a little mac app/game to practice with and get back into coding.

Looks like it’s a 64 bit app, and i think it only runs in 10.6.


I might rebuild it for the iPhone.

multiply app (151)

xcode project:

multiply xcode project .zip (163)

distributed rendering using Apple Qmaster 3 – success!!!

Wed 27, February 2008


Thanks to a fair amount of googling and a few days on and off of testing I seem to have a working setup for doing distributed processing using Apple Qmaster – for Shake, Maya and Compressor.

Here is my effort at explaining what I did to get it working – big apologies, it’s poorly written. I’ll revisit this post soon.

some notes on our workflow

  • Our current workflow for dealing with files/assets evolves around a directory structure that breaks each project up into shot numbers and the various departments we have etc..
  • Our xserve is the centralised area for our assets and the QMaster render manager.
  • Each artist’s local machine has a directory somewhere that mimics the xServe’s project directory structure for working locally (to keep network traffic down over gigabit ethernet).
  • I set up $JOBS_LOCAL and $JOBS_SERVER env variables on each machine and the xserve – these variables point to the relevant local project directory on the artist’s mac and the project directory on the server.
  • I created a python script that does a find and replace of the 2 variables and writes our a new shake script renamed “*_SERVER.shk”, or “*_LOCAL.shk”
  • (See further down for setting up the ENV variables.)

    Centralised $NR_INCLUDE_PATH
    I setup an env variable for $NR_INCLUDE_PATH for all Shake machines and the xserve – to look at a sharepoint (the nreal folder) on the xServe and automatically mount it – so all the Shake machines would be using the same macros/plugins and settings. I setup a new user on the xserve “shake” that can only mount the nreal directory.

    After some googling around I found a way to automount volumes:

    OS X 10.5 (fstab)
    /etc/fstab – to automount sharepoint on xServe
    LINK: http://blogs.sun.com/lowbit/entry/easy_afp_autmount_on_os

    # ————————————————————————————-
    # Mount AFP share from xServe via AFP
    # ————————————————————————————-
    ourserver.local:/nreal /Network/nreal url auto,url==afp://nreal:password@ourserver.local/nreal 0 0
    # ————————————————————————————-

    how to refresh automount(login as root/su):

    sudo automount -cv

    OS X 10.4 (netinfo manager)
    LINK: http://ask.metafilter.com/54223/Help-me-automount-my-smb-share-in-Apple-OS-X-reward-inside
    in terminal.app:

    nidump fstab . > /tmp/fstab
    echo “ourservernreal /Network/nreal url url==cifs://nreal:password@ourserver/nreal 0 0” >> /tmp/fstab
    sudo niload fstab . < /tmp/fstab sudo killall -HUP automount

    how to refresh automount(login as root/su):

    sudo killall -HUP automount

    I had tried for some time to get shake scripts to render over our network using Qmaster but it just wouldn’t work. The QMaster logs were where I found all my errors. ‘frame 0021 could not be found’, ‘UNIX error 3 file could not be found’

    things to check

  • shake / maya / qmaster 3 node is installed on all render machines
  • the location of all your media can be accessed by all machines
  • What seemed strange was that if I logged into the xServe and executed a script via terminal with shake (to render just on the xserve) the render would complete successfully. Then it clicked that maybe the environment variables I was using in my scripts ($xxx) might not be getting recognised by Qmaster or the way Qmaster launches Shake??

    The big tip-off
    I googled for the errors I kept seeing and luckily enough this forum post popped up:

    “I have pinned this down to at least one reason – that the shake qmaster isn’t picking up the NR_INCLUDE_PATH environment variable. Does anyone know where you need to set this up on a cluster node (I can get the qmasterd to pick it up but that doesn’t solve the problem!) “

    If you are trying to use Qmaster, and need to set environment variables, then you need to create a wrapper script that sets the variables and then calls the appropriate version of shake.

    For example, (this was from Apple)

    NR_INCLUDE_PATH=/Network_Applications/Shake Plugins/nreal/include;export NR_INCLUDE_PATH
    NR_ICON_PATH=/Network_Applications/Shake Plugins/nreal/icons;export NR_ICON_PATH

    umask 000

    /Applications/Shake3.50/shake.app/Contents/MacOS/shake $*
    if [ $status -ne 0 ]
    exit $status

    Then when using Qmaster, you run the application using this
    script (saved for example as /usr/bin/shakeWrapper) which must be
    installed on all nodes in the cluster.


    Cheers Nell!

    I took Nell’s script, added a few lines to it and stuck it in /usr/bin/

    /usr/bin/shakeWrapper – create a file to launch Shake respecting ENV variables – later alias ‘shake’ to this file

    echo “Shake 4 running through a wrapper script – /usr/bin/shakeWrapper”

    umask 000

    /Applications/Shake/shake.app/Contents/MacOS/shake $*
    if [ $status -ne 0 ]
    exit $status

    I added the first few lines there so when I later made an alias to this script the Shake user would have some idea what is going on when launching Shake via the terminal.

    Setting up the ENV (environment) variables…

    /etc/profile – to declare system-wide Environment variables / aliases
    (alias shake to use a wrapper to make it launch respecting/recognising the env variables)

    # System-wide .profile for sh(1)

    if [ -x /usr/libexec/path_helper ]; then
    eval `/usr/libexec/path_helper -s`

    if [ “${BASH-no}” != “no” ]; then
    [ -r /etc/bashrc ] && . /etc/bashrc

    export JOBS_LOCAL

    export JOBS_SERVER

    export NR_INCLUDE_PATH

    alias shake=”/usr/bin/shakeWrapper”

    *remember to enter into terminal:
    source /etc/profile