Success - New Multisite Install on Dedicated server using ServerPilot, Nginx and Apache

PopsRocker · April 29, 2015, 8:37pm

Here’s what I did to install a successful installation of Discourse for multiple instances on a dedicated server that uses Nginx and Apache.

I decided to create a new topic since my installation seems to be a little unique from what I have seen others having to go through. I am very new to Nginx webservers and thought this might help others who may be struggling to connect the dots. I’m also at a beginner level when it comes to using the terminal to access my server, I still have to look up commands for most of the stuff I want to do, but I am finding it so much faster and powerful vs FTP. I still use FTP for a visual representation to locate files and text editing (I don’t like vi or nano editing in the terminal) I use Coda (on a Mac) as my text editor so I have direct access to the server. Which means I can live edit my files just as one would do in the terminal.

I’m using ServerPilot as the control panel to manage my websites on an Ubuntu 14.04 installation. ServerPilot uses Nginx as a reverse proxy in front of the Apache webserver. At first I was a bit confused with this, and basically what they have done is use Nginx as the hub where website requests come into the Nginx server and get routed to Apache to be executed and then back through Nginx to be displayed. This proved to cause some trouble with my installation as I had to figure out where all the config files and such were located.

I started with this tutorial for installing the first instance of Discourse as a dev version …

https://github.com/discourse/discourse/blob/master/docs/INSTALL-cloud.md

Pay special attention to the TCP/IP ports as that was the first place I stumbled, and I see a lot of other posts in the forums from other people who have done the same. If you leave them at the default “80:80” you may get an error stating that port is already in use.

Change the first half of the ports to a port that is not being used, I used 85

This got me through the installation with no errors, but when I went to the url I setup for the forum nothing was showing up. I could connect to the forum using the IP with the port number. Since the forum doesn’t need Apache, I found the conf file that was directing the site to the Apache server with a local IP. I changed this IP to the forum IP with port number.

The path to that section (if you’re also using ServerPilot)

/etc/nginx-sp/vhosts.d/main.conf
(ServerPilot recommends that you change the name of the main.conf file so it is not overwritten on updates)

Now I had a working instance of Discourse, and here’s what I did to set up the multiple instances.

I started with this tutorial, but I was having trouble understanding how the structure was supposed to be setup

If you scroll down to this section …

https://meta.discourse.org/t/multisite-configuration-with-docker/14084/18

This is where I based my installation from, however I did not use the hook settings in the original post. Instead, I duplicated the app.yml file for each site I was creating (only two sites) and renamed them for site specific use. I’ll use the example - site1 and site2.

So in the /var/discourse/containers the files there are as such …

app.yml
site1.yml
site2.yml

(these are your separate containers, which was a term that I didn’t understand at first)

For each file you need to make sure to change the following areas to be specific to your instance …

## which TCP/IP ports should this container expose?
expose:
  - "127.0.0.1:4000:80"   # fwd host port 80   to container port 80 (http)
  - "2222:22" # fwd host port 2222 to container port 22 (ssh)

Notice I changed the first port on both, otherwise you will get port conflict errors.

  ## TODO: List of comma delimited emails that will be made admin and developer
  ## on initial signup example 'user1@example.com,user2@example.com'
  DISCOURSE_DEVELOPER_EMAILS: 'changeme@site1.com'

change the developer email to the email you will use for the admin account

In the mailserver section that follows, be sure to change any settings that are specific to this site such as the username and password. I ended up using Mandrill so the smtp and port numbers were the same for all instances, but I created two accounts. (one for each website)

This next section to edit is key to you getting a separate database for each site. Be sure to change the ‘yoursite’ to something unique to each site.

## These containers are stateless, all data is stored in /shared
volumes:
  - volume:
      host: /var/discourse/shared/yoursite
      guest: /shared
  - volume:
      host: /var/discourse/shared/yoursite/log/var-log
      guest: /var/log

The next thing to do is make sure you have the Nginx proxy URL (the one that is pointing to Apache) changed to match the port you configured in the TCP/IP section. So using the port I used as an example your file should look like this and as I mentioned above, ServerPilot recommends changing the name of the file if you MUST edit (that’s why they give the do not edit warning)

###############################################################################
# DO NOT EDIT THIS FILE.
#
# Your changes to this file will be overwritten by ServerPilot.
#
# For information on how to customize nginx settings, see
# https://serverpilot.io/community/articles/customize-nginx-settings.html
###############################################################################

# Send all requests to apache.
location / {
    proxy_pass      http://localhost:4000;

Be sure to restart Nginx if you make changes to the .conf file (I forgot to do this) and with ServerPilot the restart command is a little different than I had seen elsewhere, this is what did it for me …

service nginx-sp restart

Then start bootstrapping your containers. (do this for each site your configured) Enter these commands in the terminal

./launcher bootstrap site1
(after that is done)
./launcher start site1

You should now be able to load the URL you have for the forum and see your Discourse forum.

I ended up with one of the sites not recognizing the Admin account when I registered with the developer email, and I found I had a stray character at the end of the URL that caused the email to be different. I then went into the /var/discourse/shared/ directory and deleted the folder matching that site. Then I used ./launcher destroy site2 command followed by ./launcher bootstrap site2 and then ./launcher start site2 I’m not sure a rebuild would work since you want the database to be recreated.

I know I probably over-simplified this is some areas, but from what I have seen in other topics/threads, some people need instructions that are overly simplified

Please let me know if I left any holes, or there is any need for clarification as long as you understand I can only speak for what worked for me since I barely understand this myself HA!!!

One more thing … does any one see any issue with duplicating the app.yml file to be used in this way? It seems this would work perfectly for updating as well.

Best regards … Pops

elberet · April 29, 2015, 9:28pm

Actually, ports <1024 are considered “privileged”, using them without assignment is… not technically wrong, but ugly. Also, you should bind Discourse to the local interface:

expose:
  - "127.0.0.1:4000:80"

Change your nginx config accordingly to proxy_pass http://localhost:4000;

Overall, nice write-up, but do turn your YAML and other config file snippets into proper fenced code blocks (``` on a line by itself before and after the code block). Especially for YAML, indentation is really important and it gets lost in simple quote blocks.

PopsRocker · April 29, 2015, 10:10pm

That makes complete sense, I made the changes suggested to my server and they work perfectly. I also changed them in the above instructions as well as used the code markdown instead of the quote markdown (thanks for pointing that out)

… Pops

PopsRocker · April 29, 2015, 10:30pm

@elberet hmmm, something isn’t quite right when changing the ports on the production sites

Changing the ports as you suggested on the app.yml file worked great, but after changing the two additional container files with the appropriate changes I get a 502 bad gateway error.

I made the changes to the exposed port, as well as the proxy_pass, I then rebuilt the the instance using ./launcher rebuild as well as restarted the Nginx server.

Any idea what’s wrong?

… Pops

EDIT: very weird … I was issuing the restart command for Nginx, but when I stopped it, and then started instead of “restart” everything is working fine

riking · April 30, 2015, 3:04am

You can also stop exporting ports entirely by using web.socketed.template.yml:

PopsRocker · April 30, 2015, 7:11pm

I saw that post, but had a difficult time understanding how to implement it in the way my server is set up since Nginx is already installed and running in front of Apache.

It had to do with the last step, the path you referenced for sites-enabled an sites-available doesn’t exist and if I just needed to create them, I didn’t know how the files as a whole were to be set. (just a lack of knowledge on my part)

Gromeen · June 26, 2015, 9:19pm

@PopsRocker Hi Pops!

Like you I’m trying to set up Discourse on DigitalOcean with ServerPilot, already running another website in a ServerPilot app (app named ‘test’). I only need a single Discourse site, so my setup will be simpler than yours.

Along with app#1 (test) hosting a website, I’ve got a 2nd ServerPilot app (named ‘discourse’) set up for purposes of pointing to the Discourse install. This gives me the /etc/nginx-sp/vhosts.d/discourse.conf and /etc/nginx-sp/vhosts.d/discourse.d/main.conf files.

So my question: what modifications did you make to your .conf files to get it to forward requests to discourse? I’ve got Discourse running and can access it from discourse.example.com:8888. What is it I need to do in the nginx conf files so that I can see Discourse at discourse.example.com, without appending the port number?

Thx

Gromeen · June 26, 2015, 9:30pm

Sorry, replied too soon and then found my answer above:

in app.yml:

expose:
  - "127.0.0.1:8888:80" # fwd host port 80 to container port 8888 (http)
  - "2222:22" # fwd host port 22 to container port 2222 (ssh)

in /etc/nginx-sp/vhosts.d/discourse.d/main.conf (renamed to main.custom.conf as suggested):

location / {
    proxy_pass $backend_protocol://$backend_host:8888;
}

tannoy · March 31, 2016, 12:23am

I’m still a little confused with this Serverpilot / Discourse setup. When you created your “discourse” app in Serverpilot, what did you put in the apps/discourse/public/ directory? I have my current discourse installation in the /var/discourse directory but am not sure how Serverpilot is suppose to recognize that?

Gromeen · March 31, 2016, 1:42pm

You don’t actually have to put anything in the apps/discourse/public/ directory. Instead, you want requests to that Serverpilot app forwarded to your Discourse install living in /var/.

This is accomplished via the /var/discourse/containers/app.yml and /etc/nginx-sp/vhosts.d/discourse.d/main.custom.conf modifications referenced in my previous post.

tannoy · March 31, 2016, 4:33pm

Thanks for your help! I was able to get it up and running.

Dibbyo456 · June 28, 2017, 8:41am

I’m using ServerPilot too. Glad I found this article. I will let you know, how it’s worked up for me.
Thank you.

MostHated · February 18, 2018, 5:21am

Thanks for the info on this. I know it has been a while, does anyone happen to know what kind of increase in resources on the system end up being consumed when you have a second one?

Alex3 · May 29, 2019, 4:52pm

Maybe it’s because this post is from 2015 … but now I can not find the nginx-sp directory in my installation (I did it with the docker automatically).

I have all the second.yml completely changed, but I don’t know where it’s the nginx file to do the proxy thing.

Hifihedgehog · April 17, 2020, 9:42pm

(post withdrawn by author, will be automatically deleted in 24 hours unless flagged)

riking · July 27, 2020, 10:18pm

This path should contain /standalone/ in it for the first site, no?

Benjamin_D · July 27, 2020, 10:23pm

I wonder if there isn’t a confusion between multisite and WP+discourse in dual container setup?

neounix · July 28, 2020, 5:43am

We run multiple Discourse apps easily (on the same server) with one container for the data and one container for the app, for each “domain instance” we run.

In addition, for each “domain instance” we run two applications containers. This means there is zero downtime when we rebuild a container.

Why?

Because when “container a” is running, we rebuild “container b”; and after “container b” has rebuilt, we wait about 1 minute and we reconfigure the reverse proxy to proxy to the unix socket for “container b” and run (in the apache2 case) service apache2 restart.

This means there is basically zero downtime (maybe a few millseconds) when switching containers.

I have been following this thread in awe at how difficult (or complicated) many people are making something which is basically very simple.

Production Discourse runs in a Docker container; and the good Discourse team has provided the core templates to expose Discourse either a TCP/IP socket or UNIX domain socket (per application instance). In addition, the good meta team has provided the core template to run the data (PostgreSQL) container separately from the application (Rails, JS) container.

The exact same concept applies for any node.js application which runs in a container. These configurations (reverse proxies to containers) are basic Docker tasks which apply to any web applications which runs in a container, Discourse just happens to be one of them, and one of the best as we all know

Sometimes I think the excellent team at meta Discourse makes spinning up their container so easy with their well done wrapper scripts, that users fail to understand that the underlying fundamentals of running a web-based Docker application behind a reverse proxy is really quite simple.

We can run just about “as many Discourse applications” as we like, as virtual hosts, no different than a webmaster would run 20 “old fashioned LAMP web apps” in the same server. The only difference is that instead of a virtual host to a directory of PHP files, there is a virtual host to (in our case) a UNIX domain docket. Docker makes this very efficient by sharing core libs and overlays between containers!

In addition, because networking between containers is built-in to Docker, it is very easy to set up Discourse (as an application container) to work with a PostgreSQL (data) container; which is provided for all users as a example template by the good meta team, for those who want to enjoy running Discourse in more “interesting” ways (configurations).

My advice, for what little it is worth, is to understand that Discourse is simply a web app running in a container behind a reverse proxy, like any web app. We also run our private Docker registry (more than one actually) in the same manner as well; a Docker application behind the same reverse proxy to port 5000 (the default Docker registry port).

Discourse, in their wisdom to make life easy for users (and Discourse accessible to non-sys admin gurus), provides the capability to expose the web applications directly (without a reverse proxy) in their OOTB configuration. In addition, setting up a reverse proxy outside the Docker container is the same for all of these Dockerized web applications. The meta team provides those templates as well (such great support if you think about it!).

In closing, it is very easy to run one, twenty, or more dockerized applications behind a reverse proxy. They are just containers (and all of them could be a different Discourse forum) and the choice of how to expose the containers are up the user (the sys admin) and so are the choices of how to set up persistant storage volumes including databases.

I think that this is so simple and the meta team has made it so simple for everyone; but one of the side-effects by making it so simple, is that people forgot (or do not learn) the fundamentals of how easy and elegant the dockerized configuration (behind a reverse proxy) actually is (independent of Discourse).

Hope this helps in some small way.

All the best!

neounix · July 28, 2020, 7:20am

It’s really simple…

Each application has its own shared volume and that volume is where the unix domain socket runs (and where all the other persistent data is like the uploads, etc.)

You do not need to work with docker-compose directly, special pup directives, or change anything at all (only set up the basic yml files correctly for the containers, and run launcher).

If one of your Discourse sites / apps is called “farmer_forums”, then your unix socket is located (by default), for example:

/var/discourse/shared/farmer_forums/nginx.http.socket

If your next Discourse site / app is “race_car_forums”, then your unix socket is (for example) here:

/var/discourse/shared/race_car_forums/nginx.http.socket

You can do this for 100s of forums, if you wish; and each one is just an entry in a reverse proxy to that socket. Just name your container yml files appropriately.

Each one of those forums will have their own data container as well. You can use any naming convention you like for those containers for example

farmer_forums_data
race_car_forums_data

Of course, each one of those will have their own shared volume for their database(s).

All you need to do is to create one working “two container solution” and replicate it behind a reverse proxy. Creating one site, or 100 sites, it is all the same if you follow this simple method.

PS: I have heard (read) some people try to have one database container for many different forums. I would never recommend that, personally. That configuration creates a single point of failure for all forums and is not how I like to do things “keep it simple and stupid” (KISS). But, I like KISS in sys admin, because I have fat fingered things over the decades more than once and so I don’t like breaking more than I have too when I make a mistake (which we all do, from time to time).

So, if you are struggling with this; just get one site up and running with two containers, the app and the data container, behind a reverse proxy.

Then just replicate that, as many times as you want, with an app and a data container for each site. All your container names must be unique and should follow some easy to remember naming convention like in the farmers and race car samples above.

I’m not trying to sound repetitive nor preachy., but it is almost trivial to set up many Discourse sites behind a reverse proxy in this manner.

Hope this further explanation helps, in some small way.

In a nutshell, for me “multisite” simply means doing the “two container” solution behind a reverse proxy, multiple times, as many times as you wish! One or ten, it’s all the same configuration, basically, only the names of the containers are different, and there is some “easy to understand” container naming configuration (like in the two examples above). Docker will share the libs and overlays appropriately between images, so there is “nothing to do here” because “that is what docker does” (sets up all the shared libs, builds the docker network, etc) all by itself, which makes this “efficient” due to how docker shares libs, between images, etc.

Benjamin_D · July 28, 2020, 7:47am

Maybe we should call this setup something else than multisite (which in my mind shares the data container cf the howto topic closed by sam earlier last year)
Parallel site? i don’t know…
If i’m not mistaken, since you have an .yml file per data/web couple containers, you can specify different plugins set and smtp per domain?