Lately I’ve been working to deploy a suite of PHP microservices using Docker containers. One of the problems is that our PHP applications are set up to work with PHP-FPM and Nginx (instead of the admittedly simpler Apache/PHP setup covered here), so each PHP microservice needs two containers (and by extension, two Docker images): a PHP-FPM container and an NGinx container.
The app runs over half a dozen services, so if you multiply that out, it will be 30 or so containers between dev and prod. Rather than build unique NGinx images for each one, I decided to build a single NGinx Docker image that would take the PHP-FPM host name as an environmental variable and run a unique configuration file with it.
In this blog post, I’ll outline my journey for going from Method 1 to Method 2 above, and finally end it with the solution I came to using a new custom Docker image. I’ve made this image open source, so if this is a problem you’re familiar with, feel free to check that out.
Using NGinx with PHP-FPM can yield better performance for PHP apps, but the downside is that the PHP-FPM Docker image isn’t bundled with NGinx by default like the PHP Apache image is. If you want to connect an NGinx container to a PHP-FPM backend, you need to add the DNS record for that backend into your NGinx config. For example, if the PHP-FPM container were running as a container named php-fpm-api
, then your NGinx configuration file should have this in it:
location ~ \.php$ {
fastcgi_split_path_info ^(.+\.php)(/.+)$;
# This line passes requests through to the PHP-FPM container
fastcgi_pass php-fpm-api:9000;
fastcgi_index index.php;
include fastcgi_params;
fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
fastcgi_param PATH_INFO $fastcgi_path_info;
}
Hard-writing the name of the container into your NGinx config is fine if you’re only serving a single NGinx container, but as I mentioned above, we need to run several NGinx containers - one for each of our PHP services. Creating a new NGinx image (that we later have to maintain and upgrade) would be a pain, and even managing a bunch of different volumes seemed like a lot of work for changing a single variable name.
Initially, I thought this would be easy. There’s a great little section about how to do this using envsubst
in the Docker docs, but unfortunately, this didn’t work for my NGinx config file:
server {
listen 80;
index index.php index.html;
root /var/www/public;
client_max_body_size 32M;
location / {
try_files $uri /index.php?$args;
}
location ~ \.php$ {
fastcgi_split_path_info ^(.+\.php)(/.+)$;
fastcgi_pass ${NGINX_HOST}:9000;
fastcgi_index index.php;
include fastcgi_params;
fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
fastcgi_param PATH_INFO $fastcgi_path_info;
}
}
My vhosts.conf
file takes advantage of several built in NGinx variables, so when I ran the Docker command outlined in the docs (/bin/bash -c "envsubst < /etc/nginx/conf.d/mysite.template > /etc/nginx/conf.d/default.conf && nginx -g 'daemon off;'"
), I got errors complaining that $uri
and $fastcgi_script_name
weren’t defined. These variables are normally passed in by NGinx, so I couldn’t easily figure out what they were and pass them in myself, plus that would make my container less dynamic.
Next, I went searching for a different NGinx base image. Two came up, but neither had been updated in a couple years. I started with martin/nginx, and tried to see if I could get a prototype working.
Martin’s image is a little bit unconventional because it requires a specific folder structure. In the root, I added a Dockerfile
:
FROM martin/nginx
Next, I added an app/
directory that was empty and a conf/
directory with a single file called vhosts.conf
:
server {
listen 80;
index index.php index.html;
root /var/www/public;
client_max_body_size 32M;
location / {
try_files $uri /index.php?$args;
}
location ~ \.php$ {
fastcgi_split_path_info ^(.+\.php)(/.+)$;
fastcgi_pass $ENV{"NGINX_HOST"}:9000;
fastcgi_index index.php;
include fastcgi_params;
fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
fastcgi_param PATH_INFO $fastcgi_path_info;
}
}
This is pretty much the same as my original configuration file with this one line changed: fastcgi_pass $ENV{"NGINX_HOST"}:9000;
. Now when I wanted to start an NGinx container with a PHP container named php-fpm-api
, I could build a new image and run it with an environmental variable:
docker build -t shiphp/nginx-env:test .
docker run -it --rm -e NGINX_HOST=php-fpm-api shiphp/nginx-env:test
It worked! But, two things bothered me about this method:
/app
directory seems like an unnecessary requirement plus my files will be stored in a different directory.I figured Martin’s image would be a good place to start for a custom solution, so I forked the project and built a new NGinx base image that fixes the two issues. Now, if you want to run a dynamically-named backend with an NGinx container, you can simply do this:
# Pull down the latest from Docker Hub
docker pull shiphp/nginx-env:latest
# Run a PHP container named "php-fpm-api"
docker run --name php-fpm-api -v $(pwd):/var/www php:fpm
# Start this NGinx container linked to the PHP-FPM container
docker run --link php-fpm-api -e NGINX_HOST=php-fpm-api shiphp/nginx-env
If you want to customize the image by adding your own files or NGinx configuration files, then just extend it with a Dockerfile:
FROM shiphp/nginx-env
ONBUILD ADD <PATH_TO_YOUR_CONFIGS> /etc/nginx/conf.d/
...
Now all my PHP-FPM containers use their own instance of a single Docker image, making my life much easier when it comes time to upgrade NGinx, change permissions, or tweak something.
All the code is available on Github, so if you see any problems or want to suggest improvements, feel free to create an issue. If you have questions with this or anything else Docker-related find me on Twitter to continue the conversation.
In this book, PHP developers will learn everything they need to know to start building their applications on Docker, including:
You can buy this book on Leanpub today.