23 Aug 2018

[Blog] How to build a Twitch-alike service with MistServer

Hey all! First of all, our apologies for the lack of blog posts recently - we've been very busy with getting the new 2.14 release out to everyone. Expect more posts here so that we can catch back up to our regular posting pace!

Anyway - hi 👋! This is Jaron, not with a technical background article (next time, I promise!) but with a how-to on how you can build your own social streaming service (like Twitch or YouTube live) using MistServer. We have more and more customers running these kind of implementations lately, and I figured it would be a good idea to outline the steps needed for a functional integration for future users.

A social streaming service, usually has several common components:

  • A login system with users
  • The ability to push (usually RTMP) to an "origin" server (e.g. sending your stream to the service)
  • A link between those incoming pushes and the login system (so the service knows which stream belongs to which user)
  • A check to see if a viewer is allowed to watch a specific stream (e.g. paid streams, password-protected streams, age restricted stream, etc)
  • The ability to record streams and play them back later as Video on Demand

Now, MistServer can't help you with the login system - but you probably don't want it to, either. You'll likely already have a login system in place and want to keep that and its existing database. It's not MistServer's job to keep track of your users anyway. The Unix philosophy is to do one thing and do it well, and Mist does streaming; nothing else.

How to support infinite streams without configuring them all

When you're running a social streaming service, you need to support effectively infinite streams. MistServer allows you to configure streams over the API, but that is not ideal: Mist start to slow down after a few hundred streams are configured, and the configuration becomes a mess of old streams.

Luckily, MistServer has a feature that allows you to configure once, and use that stream config infinite times at once: wildcard streams. There's no need to do anything special to activate wildcard mode: all live streams automatically have it enabled. It works by placing a plus symbol (+) behind the stream name, followed by any unique text identifier. For example, if you configured a stream called "test" you could broadcast to the stream "test", but also to "test+1" and "test+2" and "test+foobar". All of them will use the configuration of "test", but use separate buffers and have separate on/off states and can be requested as if they are fully separate streams.

So, a sensible way to set things up is to use for example the name "streams" as stream name, and then put a plus symbol and the username behind it to create the infinite separate streams. For example, user "John" could have the stream "streams+John".

Receiving RTMP streams in the commonly accepted format for social streaming

Usually, social streaming uses RTMP URLs following a format similar to rtmp:// However, MistServer uses the slightly different format It's inconvenient for users to have to comply with Mist's native RTMP URL format, so it makes sense to tweak the config so they are able to use a more common format instead.

The ideal method for this is using the RTMP_PUSH_REWRITE trigger. This trigger will call an executable/script or retrieve an URL, with as payload the RTMP URL and the IP address of the user attempting to push, before MistServer does any parsing on it whatsoever. Whatever your script or URL returns back to MistServer is then parsed by MistServer as-if it was the actual RTMP URL, and processing continues as normal afterwards. Blanking the returned URL results in the push attempt being rejected and disconnected. Check MistServer's manual (under "Integration", subchapter "Triggers") for the documentation of this trigger.

An example in PHP could look like this:

//Retrieve the data from Mist
$payload = file_get_contents('php://input');
//Split payload into lines
$lines = explode("\n", $payload);
//Now $lines[0] contains the URL, $lines[1] contains the IP address.

//This function is something you would implement to make this trigger script "work"
$user = parseUser($lines[0], $lines[1]);
if ($user != ""){
  echo "rtmp://".$user;
  echo ""; //Empty response, to disconnect the user
//Take care not to print anything else after the response, not even any newlines! MistServer expects a single line as response and nothing more.

The idea is that the parseUser function looks up the stream key from the RTMP URL in a database of users, and returns the username attached to that stream key. The script then returns the new RTMP URL as rtmp://, effectively allowing the push as well as directing it to a unique stream for the authorized user. Problem solved!

How to know when a user starts/stops broadcasting

This one is pretty easy with triggers as well: the STREAM_BUFFER trigger is ideal for this purpose. The STREAM_BUFFER trigger will go off every time the buffer changes state, meaning that it goes off whenever it fills, empties, goes into "unstable" mode or "stable" mode. Effectively, MistServer will let you know when the stream goes online and offline, but also when the stream settings aren't ideal for the user's connection and when they go back to being good again. All in real-time! Simply set up the trigger and store the user's stream status into your own local database to keep track. Check MistServer's manual (under "Integration", subchapter "Triggers") for the documentation of this trigger.

Access control

Now, you may not want every stream accessible for every user. Limiting this access in any way, is a concept usually referred to as "access control". My colleague Carina already wrote an excellent blog post on this subject last year, and I suggest you give it a read for more on how to set up access control with MistServer.

Recording streams and playing them back later

The last piece of the puzzle: recording and Video on Demand. To record streams, you can use our push functionality. This sounds a little out of place, until you wrap your head around the idea that MistServer considers recording to be a "push to file". A neat little trick is that configuring an automatic push for the stream "stream+" will automatically activate this push for every single wildcard instance of the stream "stream"! Combined with our support for text replacements (detailed in the manual in the chapter 'Target URLs and settings'), you can have this automatically record to file. For example, a nice target URL could be: /mnt/recordings/$wildcard/$datetime.mkv. That URL will sort recordings into folders per username and name the files after the date and time the stream started. This example records in Matroska (MKV) format (more on that format in my next blog post, by the way!), but you could also record in FLV or TS format simply by changing the extension.

If you want to know when a recording has finished, how long it is, and what filename it has... you guessed it, we have a trigger for that purpose too. Specifically, RECORDING_END. This trigger fires off whenever a recording finishes writing to file, and the payload contains all relevant details on the new recording. As with the previous triggers, the manual, under "Integration", subchapter "Triggers", has all relevant details.

There is nothing special you need to do to make the recordings playable through MistServer as well - they can simply be set up like any other VoD stream. But ideally, you'll want to use something similar to what was described in another of our blogposts last year, on how to efficiently access a large media library. Give it a read here, if you're interested.

In conclusion

Hopefully that was all you needed to get started with using MistServer for social streaming! As always, contact us if you have any questions or feedback!