Configuring nginx with php-fpm in kubernetes and the "File not found." error

nginx, php, docker, kubernetes

30 May 2023

For a course at university I wanted to setup two different containers:

One should run a PHP application using php-fpm
Another one should run nginx

This sounds rather easy, but unfortunately it took me quite some time to get this working. Mostly because I could not find enough information on the internet about it. The biggest issue was how to setup FastCGI for nginx, so that it passes the requests correctly to the container running php-fpm.

But let me start with the setup of the docker image itself. I’ve already had an application written using Symfony, and the repository contained a Dockerfile in the root of the project directory:

FROM php:8.2-fpm
WORKDIR /var/www
RUN apt-get update
RUN apt-get install -y libpq-dev libzip-dev zip zlib1g-dev
RUN docker-php-ext-install pdo_pgsql
RUN docker-php-ext-install zip
RUN pecl install redis-5.3.7 && docker-php-ext-enable redis
COPY --from=composer /usr/bin/composer /usr/bin/composer
COPY . .
RUN composer install

This might not be perfect and I might still run into stuff using that, but since that’s not the point of this article I am just going to ignore that part. In order to build the docker image and push it to a registry I executed the following commands (which I am only publishing in this blog post with some placeholders):

docker build -t <project-name> .
docker tag <project-name> <username>/<project-name>
docker push <username>/<project-name>

Afterwards I created a YAML file to describe a Kubernetes deployment. Since the software I was building was about administering something, I called this file administration.yaml.

apiVersion: apps/v1
kind: Deployment
metadata:
    name: administration-deployment
spec:
    replicas: 1
    selector:
        matchLabels:
            app: administration
    template:
        metadata:
            labels:
                app: administration
        spec:
            containers:
                - name: administration
                  image: <username>/<project-name>:latest
                  envFrom:
                      - secretRef:
                            name: administration

This was good enough to have my docker image using php-fpm running. However, a web server was still missing. Therefore I wanted to create a Kubernetes deployment running a nginx web server. I stored the following YAML configuration in nginx-deployment.yaml:

apiVersion: apps/v1
kind: Deployment
metadata:
    name: nginx-deployment
spec:
    replicas: 1
    selector:
        matchLabels:
            app: nginx
    template:
        metadata:
            labels:
                app: nginx
        spec:
            containers:
                - name: nginx
                  image: nginx:1.24
                  volumeMounts:
                      - name: nginx-configmap
                        mountPath: /etc/nginx/conf.d
            volumes:
                - name: nginx-configmap
                  configMap:
                      name: nginx-configmap

This starts an nginx web server container, and maps a ConfigMap to the file system of the container at /etc/nginx/conf.d, which happens to be the folder for nginx configuration files. Now the crucial part to get that setup working: The ConfigMap mounted in the previous file:

apiVersion: v1
kind: ConfigMap
metadata:
    name: nginx-configmap
data:
    default.conf: |
        server {
            location / {
                try_files $uri /index.php$is_args$args;
            }

            location ~ ^/index\.php(/|$) {
                fastcgi_split_path_info ^(.+\.php)(/.*)$;

                fastcgi_pass administration-service:9000;
                fastcgi_param SCRIPT_FILENAME /var/www/public/index.php;
                fastcgi_param REQUEST_METHOD $request_method;
                fastcgi_param REQUEST_URI $request_uri;

                internal;
            }

            location ~ \.php$ {
                return 404;
            }
        }

The crucial part is setting the correct SCRIPT_FILENAME as fastcgi_param. And the problem is that I hardly found any information about this on the world-wide web (which motivates me to write this blog post). I’ve took a look at the examples I found on the nginx website, but none of them seemed to run nginx and php-fpm in a different container. Including the default configuration file fastcgi_params that comes with many distributions did not work. The problem was that they set the SCRIPT_FILENAME to $document_root$fastcgi_script_name and $documentRoot is a path within the nginx machine, but in the php-fpm container there was another folder structure. This made the php-fpm container always respond with a page saying “File not found.”, and since it did not even say which file, this was rather hard to debug.

I am sure there are different ways to solve it, e.g. by mirroring the folder structure in both container, but I found it easier (and also cleaner) to simply adjust the fastci_param SCRIPT_FILENAME and set it to the absolute path of the public/index.php file of the Symfony application on the php-fpm container.

So all left to do is to execute the kubectl apply -f <directory> command and enjoy a working Symfony application.