This is the project that I used to deploy ElixirConf.com.
It is a collection of scripts and instructions for setting up a cloud web server, along with DNS, SSL, and deploying a Phoenix web application using Distillery and eDeliver.
The scripts (and these instructions) were written to assist me in setting up new web servers quickly. With them I can create a VM, setup a SSL website, and deploy a local Phoenix app in under 10 minutes.
This guide should be of some help to anyone wanting to setup a website with Phoenix, even if you are not using FreeBSD or GCE. Just note, that some scripts will need to be customized to your situation. I will try to point that out wherever possible.
To use the scripts provided here, you will need to have a GCE account and have the gcloud SDK installed on your computer.
Also, you will need to download this project to get access to some of the gcloud helper scripts to run on your local computer.
git clone [email protected]:jfreeze/gce-freebsd.git
The setup includes instructions on creating a FreeBSD instance on Google Compute Engine, setting up DNS using DNSimple, free SSL cert generation using Letsencrypt for https transport, and Nginx to handle web services.
I should point out that the deployment configuration contained herein does not yet cover DB setup and deployment, and for now is a simple configuration where the build and deploy machines are the same machine. However, it is a simple task to to extend the examples shown here to have a separate build machine and deploy machine.
At some future date I will extend the examples and cover deployment where databases and data migration are involved.
- Create a GCE Instance
- Configure nginx
- Install and configure Distillery
- Install and configure eDeliver
- Deploy Phoenix app to web server
- Create SSL cert from Letsencrypt
- Configure nginx for SSL
The first order of business is to create a VM instance on the Google Compute Engine cloud service.
If you don't already have a Google Compute Engine account, you will need to get one. I originally started these scripts for AWS, but switched to GCE because of their better (read faster) network, and scaling to faster CPU's with more RAM was more economical.
Once in GCE, you will need to create a project to run your Google Compute Engine VM's under, if you don't already have one. Google allows up to five projects.
Since this is a seldom done task and is easy enough to do, I am not providing and console oriented way of creating a project.
Simply open up your GCE console
in a web browser, click the three horizontal bars (the hamburger menu) in the upper
left of the page, select IAM & Admin, then click on
All Projects. You should see a + CREATE PROJECT
link in the top center of the page to create a new project.
I prefer a customized network so I don't have unused firewall rules in it
and I like GCE's target tags that allow firewall rules to be applied to selected
machines only. For this reason, I delete the default network and firewall rules
on new projects. Notice that the examples shown here are for the GCE project
named elixirconf. You will need to change elixirconf to
the name of your project.
# delete default firewall rules
./gcloud-compute-firewall-rules-delete.sh elixirconf
# delete default network
gcloud compute networks delete "default" \
--project elixirconf
Next, create a network. I used the name elixirconf-net.
./gcloud-compute-network-create.sh elixirconf elixirconf-net
The gcloud-compute-firewall-rules-create.sh script creates
firewall rules for private cloud access, http, https, ping and ssh.
./gcloud-compute-firewall-rules-create.sh elixirconf elixirconf-net
The gcloud-compute-instance-create.sh script creates
a FreeBSD 11.0-RELEASE machine with the smallest configuration of
CPU and RAM possible. It runs about $5 per month. I chose the name
elixirconf-www-01 for the name of this web server VM.
./gcloud-compute-instance-create.sh elixirconf elixirconf-www-01 elixirconf-net
If you want, you can get a reserved static IP for your new VM instance. Since this is a one time task, it is simple enough to do it from the GCE web console.
In the console, click on the instance name, click edit at the top of the page,
scroll down to External IP and change the address from
ephemeral to New static IP address....
You will need to create a name for your static IP.
I used elixirconf-com.
Save the changes.
I find the gcloud compute ssh command useful, but clunky,
so I always setup ssh access to my servers.
When you first installed gcloud and ran gcloud init,
it created an ssh key for you called google_compute_engine.
ls -l ~/.ssh/
total 64
-rw-r--r-- 1 jimfreeze staff 404 Dec 13 12:28 config
-rw------- 1 jimfreeze staff 1675 Nov 22 16:30 google_compute_engine
-rw-r--r-- 1 jimfreeze staff 402 Nov 22 16:30 google_compute_engine.pub
-rw-r--r-- 1 jimfreeze staff 1700 Dec 12 21:26 google_compute_known_hosts
Connecting to your server one time with gcloud will update project ssh metadata on
your server. In other words, it will push your public key to your new server, so
run the command below to do an initial login to your new server. You will need to
change the project and server name to the ones you used for your VM instance.
You may also be able to skip the --zone part of this command.
gcloud compute ssh "elixirconf-www-01" \
--project "elixirconf" \
--zone "us-central1-a"
Until we set a domain for this machine, we can configure ~/.ssh/config
for easy access and we must also tell ssh to use the
google_compute_engine key since it has a non-standard name.
Edit your ~/.ssh/config file and add
Host <shortname>
HostName <host-IP>
User <username>
IdentityFile ~/.ssh/google_compute_engine
Host <server-IP>
HostName <server-IP>
IdentityFile ~/.ssh/google_compute_engine
With this setup, you can run ssh <shortname> to connect
to the server. It's very convenient.
You will also need the server-IP version for deployment
later on since eDeliver will not connect unless Host
is an actual DNS name or an IP address.
Depending on when you create your new VM, there may be updates ready for the machine.
FreeBSD makes it very easy to keep a machine up-to-date and secure. Simply run
the following on your machine after connecting via ssh.
sudo freebsd-update fetch
sudo freebsd-update install
sudo shutdown -r now # only required if an update was found
Note that the intances on GCE cannot be logged into using a password. This is the default setting.
Also, you cannot log into the server as root. However, root does
not have a password set by default, so the sudo commands above can be executed
in batch.
There are several apps that need to be installed for the new server to be a build host and
a web host. The differences between the two is that a web host needs nginx and erlang,
but not elixir. And the build host needs elixir.
The scripts provided here install elixir and provide the script for nginx
installation, but requires it to be manually installed at the end. So, even though elixir
is not needed for a web host, the scripts here could still be used for that purpose without any change.
The server setup script prepares this host as a build and web server.
Run the command below:
\curl -sSL https://raw.githubusercontent.com/jfreeze/gce-freebsd/master/freebsd-setup.sh | bash
This script installs several apps and also places some code in /tmp for you to run manually
as needed. The base installs should not take too long to run.
This script sets the user environment with user-env.sh,
installs the base applications with base-installs.sh,
installs the web applications with web-installs.sh,
and installs and sets up the config files for nginx (but prompts the user that installation must be run manually).
FreeBSD by default uses sh as a default shell. We need to change that to bash
to support eDeliver.
# Update the shell for edeliver
sh /tmp/chg-shell-to-bash.sh
exit # logout and log back in
If this host is to be used as a web server,
edit the install script first at /tmp/nginx-setup.sh
and set the DOMAIN variable for your project.
# Edit DOMAIN to point to hostname
DOMAIN=.ElixirConf.com
Then run the nginx setup script
/tmp/nginx-setup.sh
Finally, feel free to review the nginx config file
with the command
sudo vim /usr/local/etc/nginx/nginx.conf
If not rebooted yet, you can start nginx with
sudo service nginx start
That's it for the remote server.
You are now ready to update your Phoenix project by adding
Distillery and Edeliver deps to mix.exs.
Switch to your Phoenix project and edit mix.exs.
# in defp deps add
{:distillery, "~> 1.0" },
{:edeliver, "~> 1.4.0"},
Also, add :edeliver to your list of applications in mix.exs.
It should look something like this:
applications: [:phoenix, :phoenix_pubsub, :phoenix_html, :cowboy, :logger, :gettext,
:phoenix_ecto, :postgrex, :edeliver]]
Now update your project
mix deps.get
and create the Distillery release directory rel
mix release.init
The default .gitignore file generated by Phoenix ignores the contents of the priv/static
directory, which means that when deploying your project, the contents of that directory must be
regenerated on the build machine. The default Phoenix method for doing this involves brunch and
can be done inside a shell script, however Distillery offers a plugin API which allows us to execute Elixir
code as part of the release process, which is the approach we'll use below.
You can add that plugin to populate the priv/static directory during deployments
with the following script. Replace <ProjectName> with the name of your project
and run
\curl -sSL https://raw.githubusercontent.com/jfreeze/gce-freebsd/master/distillery-plugin.sh | bash -s <ProjectName>
This script will create the plugin script in rel/plugins/digest_plugin.exs, and import it in rel/config.exs.
Now edit rel/config.exs and change the default environment to :prod
# Change default build environment to :prod
# This sets the default environment used by `mix release`
default_environment: :prod
Add the newly added plugin to the :prod environment in rel/config.exs.
Remember to reference the name of your plugin instead of Elixirconf.
environment :prod do
plugin Elixirconf.PhoenixDigestTask
set output_dir: "/app/deploys/elixirconf" # for v1.0
...
Remember to change the plugin name ElixirConf to the name of your plugin.
While you are editing this file, you can go ahead and add the output_dir
directory for the build machine. This is the directory where deploys are located.
Next we setup for eDeliver by creating .deliver directory
and adding a .deliver/config file.
# Add .deliver/config with settings changed for your project.
APP="elixirconf"
AUTO_VERSION=commit-count+branch-unless-master
BUILD_CMD=mix
RELEASE_CMD=mix
USING_DISTILLERY=true
BUILD_HOST="130.211.190.72" # change this when DNS is set
# Needs to be an actual DNS or IP
# address. Won't accept a .ssh/config
# alias
BUILD_USER="jimfreeze"
BUILD_AT="/app/builds/elixirconf"
#STAGING_HOSTS=""
#STAGING_USER="jimfreeze"
PRODUCTION_HOSTS="130.211.190.72" # deploy / production hosts separated by space
PRODUCTION_USER="jimfreeze" # local user at deploy hosts
DELIVER_TO="/app/deploys" # writes releases to $DELIVER_TO/$APP
# set to agree with output_dir in rel/config.exs
RELEASE_DIR="/app/deploys/elixirconf" # use same as output_dir if needed
# For *Phoenix* projects, symlink prod.secret.exs to our tmp source
pre_erlang_get_and_update_deps() {
local _prod_secret_path="/app/builds/secret/prod.secret.exs"
if [ "$TARGET_MIX_ENV" = "prod" ]; then
__sync_remote "
ln -sfn '$_prod_secret_path' '$BUILD_AT/config/prod.secret.exs'
"
fi
}
In .deliver/config you need to set your project name.
This is the same project name as set in mix.exs.
APP="elixirconf"
I have also added AUTO_VERSION to my config file to
prevent having to specify it in scripts or from the command line.
There are various options to setting the version. See the docs
for more options.
AUTO_VERSION=commit-count+branch-unless-master
The BUILD_AT directory is where your local mix project directory
gets recreated on the build machine.
Also, new with Distillery 1.0, you don't need RELEASE_DIR, but
you will need the DELIVER_TO directory. I made this match
the output_dir from rel/config.exs.
Finally, you need to copy your config/prod.secret.exs file to the
build server and specify that location in the
pre_erlang_get_and_update_deps() part of the script.
I prefer to keep my build and deploy directories outside of a users directory,
and have chosen /app to place my builds and releases.
Create directories on the build server as specified in .deliver/config if needed.
ssh ecw "sudo mkdir /app; sudo chown jimfreeze:jimfreeze /app"
# Create the needed directories on the build server and copy prod.secret.exs to the build server
ssh ecw "mkdir -p /app/builds/secret"
scp config/prod.secret.exs jimfreeze@ecw:/app/builds/secret
The ecw is an ssh alias for my build/deploy machine.
Ok, not quite yet. We have a few more changes to make to our local project before we can deploy.
Since we haven't setup the database in this deployment, you will
need to comment out the Repo in lib/elixirconf.ex.
# supervisor(Elixirconf.Repo, []),
AND, setup the production config file config/prod.exs,
config :elixirconf, Elixirconf.Endpoint,
#http: [port: {:system, "PORT"}],
http: [ip: {127,0,0,1}, port: 4000],
server: true,
url: [host: "elixirconf.com", port: 80],
version: Mix.Project.config[:version],
cache_static_manifest: "priv/static/manifest.json"
Pay particular attention to
http: [ip: {127,0,0,1}, port: 4000],
server: true,
Since we are running this server through nginx, the deployed Phoenix
server only needs to listen on the local machine -- and this coincides with how
we setup our proxy in nginx.conf. This will prevent outside access
to your Phoenix server on port 4000. Also note that instead of host:
the address is specified with ip:, AND that the address is a tuple.
Finally, note the line server: true. YOUR PHOENIX SERVER WILL NOT RUN
WITHOUT THIS AND IT IS NOT INCLUDED BY DEFAULT. Don't forget to add this line.
If you do, a telltale result is that you will see the server listening on random
ports yet not responding to any attempted connections.
Before deploying, make sure that your project files are up to date with the Distillery and eDeliver configs checked into git.
Start the deploy process by building the release
mix edeliver build release
Finish the deploy and restart the service in one command with
mix edeliver deploy release to production --start-deploy
Add the --verbose flag if needed to debug any issues.
You can also skip the --start-deploy flag and manually start
and stop the server with the commands
mix edeliver stop production
mix edeliver start production
mix edeliver restart production
Now that you have verified that deployment works, you can setup a reboot cronjob that starts up the web server should the server be rebooted. On the deployment server, add the following to cron (with your appropriate changes)
echo "@reboot /app/deploys/elixirconf/bin/elixirconf start" | crontab -
As you will find out, deployments start to enter the wild and wooly west at this point.
As you do multiple deployments, you will see that mix edeliver build release
will fail if multiple releases are present on the build machine unless
the RELEASE_VERSION environment variable is set when mix is run.
You will also notice that if multiple releases exist on the local machine, that
mix edeliver deploy release to production will prompt for which
version to copy to the deploy machine.
To keep things simple, and so deployments can run without human input, you can clear out releases on the local and remote machines.
Here is one possible script for auto deployment. The ecw is an ssh alias
for my build/deploy machine.
#!/bin/sh
# remove releases on the build server
ssh ecw "rm -rf /tmp/releases; mv -f /app/deploys/elixirconf/releases /tmp; rm -rf /app/deploys/elixirconf/releases"
# remove local releases
rm .deliver/releases/*gz
mix edeliver build release
mix edeliver deploy release to production --start-deploy
For SSL configuration of the website we are going to obtain a certificate from
Letsencrypt. Adding this certificate in
the standalone method used here means that we will need to stop the production
webserver. The process only takes a few seconds, so if you have your
new nginx.conf file ready, you can limit your downtime.
You will also need to configure DNS before setting up SSL. An "A" record and a "CNAME" are required to complete SSL authentication.
Type Name Content
A elixirconf.com 1.2.3.4
CNAME www.elixirconf.com elixirconf.com
If you are setting your DNS server for the first time, point your browser to your new domain to verify DNS is pointing to your new server.
On the web server, run the following to install the cert script
sudo pkg install -y py27-certbot
And stop nginx so you can authenticate with the cert server.
sudo service nginx stop
And run the following script to automatically obtain your certs. Don't forget to update for your project.
sudo certbot certonly --standalone --rsa-key-size 4096 --email <[email protected]> -d <mydomain.com>
If all goes well, you should see output like
IMPORTANT NOTES:
- Congratulations! Your certificate and chain have been saved at
/usr/local/etc/letsencrypt/live/elixirconf.com/fullchain.pem. Your
cert will expire on 2017-03-14. To obtain a new or tweaked version
of this certificate in the future, simply run certbot again. To
non-interactively renew *all* of your certificates, run "certbot
renew"
Update your nginx.conf file to be similar to
load_module /usr/local/libexec/nginx/ngx_mail_module.so;
load_module /usr/local/libexec/nginx/ngx_stream_module.so;
worker_processes 1;
events {
worker_connections 256;
}
http {
include mime.types;
default_type application/octet-stream;
sendfile on;
keepalive_timeout 65;
#gzip on;
server {
listen 80;
server_name elixirconf.com;
return 301 https://$server_name$request_uri;
}
server {
listen 443 ssl;
server_name .elixirconf.com;
ssl_certificate /usr/local/etc/letsencrypt/live/elixirconf.com/fullchain.pem;
ssl_certificate_key /usr/local/etc/letsencrypt/live/elixirconf.com/privkey.pem;
location / {
proxy_pass http://127.0.0.1:4000;
}
}
}
and restart nginx
sudo service nginx start
With Letsencrypt, SSL certs only last a maximum of three months. According to the docs, it is advised that you setup a cronjob to run twice a day to renew the certs. It won't hurt to run the renew script before the certs expire, the request will just be ignored, but Letsencrypt reserves the right to cancel SSL certs at anytime, so running a renew twice a day will minimize any downtime should the cert be recalled early.
Simply add to your root user cronjobs by running the following on the deploy machine. The docs also request that you pick a random minute with which to run the cronjob.
sudo su -
crontab -e
# add the following to root's cron
14 11,23 * * * /usr/local/bin/certbot renew
One final step in the SSL configuration is to edit config/prod.exs to have Phoenix use the https scheme
for URLs and force_ssl for rejecting http connections and setting headers that force ssl (thanks ericmj)
#url: [host: "elixirconf.com", port: 80],
url: [host: "elixirconf.com", scheme: "https", port: 443], # static_url will now return https:// addresses
force_ssl: [hsts: true, host: nil, rewrite_on: [:x_forwarded_proto]],
I have had some issues with sites redirecting to https://127.0.0.1 when using the force_ssl line in prod.exs.
If you experience problems, comment out the line. Or, better yet, if you experience problems and have a fix, a pull request will be much appreciated!
You can also go back now and edit .deliver/config and update the server
variables with your new DNS name.
...
BUILD_HOST="elixirconf.com"
...
PRODUCTION_HOSTS="elixirconf.com"
...
It's interesting to note that these simple steps allow the following URLs to be handled:
http://elixirconf.com
http://www.elixirconf.com
https://elixirconf.com
https://www.elixirconf.com
They all redirect to https://elixirconf.com.