[Bug]: Nginx not find services in preview deployment

[picheidt]

Error Message and Logs

I have a PHP application using nginx as a reverse proxy. I deploy it with Coolify and Docker Compose, and the main deployment works as expected. However, the preview deployment results in the following error:

error: nginx: [emerg] host not found in upstream "php_mm2" in /etc/nginx/conf.d/default.conf:16

Below are my current files

• DockerFile.nginx

FROM nginx:stable-alpine

WORKDIR /var/www/html/

COPY ./webroot /var/www/html
COPY ./docker/nginx/default.conf /etc/nginx/conf.d/default.conf

EXPOSE 80/tcp

• nginx.conf

server {
    listen 80;
    index index.php index.html;
    server_name localhost;
    error_log /var/log/nginx/error.log;
    access_log /var/log/nginx/access.log;
    root /var/www/html;

    location / {
    	try_files $uri $uri/ /index.php?$query_string;
    }

    location ~ \.php$ {
        try_files $uri =404;
        fastcgi_split_path_info ˆ(.+\.php)(/.+)$;
        fastcgi_pass php_mm2:9000;
        fastcgi_index index.php;
        include fastcgi_params;
        fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
        fastcgi_param PATH_INFO $fastcgi_path_info;
    }
}

• docker-compose

services:
  nginx_mm2:
    build:
      context: .
      dockerfile: Dockerfile.nginx.staging
    container_name: nginx_mm2
    ports:
      - "80"
    depends_on:
      - php_mm2
    healthcheck:
      test: [ "CMD", "curl", "-f", "http://localhost:80/app/healthCheck" ]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 15s
  
  php_mm2:
    build:
      context: .
      dockerfile: Dockerfile.staging
    container_name: php_mm2
    command: sh ./run_dependencies_staging.sh
    ports:
      - "9000"

Steps to Reproduce

1. Active preview deployment
2. Use docker-compose with an application and nginx containers
3. Open a PR to trigger the preview Deployment

Example Repository URL

No response

Coolify Version

v4.0.0-beta.338

Are you using Coolify Cloud?

No (self-hosted)

Operating System and Version (self-hosted)

Ubuntu 22.04.4 LTS

Additional Information

The application and nginx containers are on the same network. I couldn’t debug nginx because it keeps restarting in a loop due to the error.

I think the reason is that during the preview deployment, the service name is changed with the addition of pr-{prId}.

[tm1000]

• Preview deployments with docker compose: how to connect between containers? #4708
• Preview deployments with docker-compose not working with depends-on #2180

[tm1000]

The solution to this problem will need to be the following in /data/coolify/applications/*/docker-compose-pr-*.yaml:

    nginx-pr-615:
        networks:
          p0g0ck8c44o4oo4w808so8w4-615: null

to

    nginx-pr-615:
        networks:
          p0g0ck8c44o4oo4w808so8w4-615:
            aliases:
              - nginx

Since this file is automatically generated a fix similar to #2204 will need to be done

I'm also not 100% sure about this because the coolify-proxy is joined into the network and so the aliases would probably clash with the primary deployment

[tm1000]

Thinking more about this the solution I provided won't conflict because the proxy addresses the container names as the full name and each deployment has a self contained network. But this has to be addressed in the code itself where it makes network: null. I know where this is at in the code but will need to setup a coolify dev instance to run it through checks

[tm1000]

Quick patch that I'm going to test this next week (that seems to work correctly)

diff --git a/bootstrap/helpers/shared.php b/bootstrap/helpers/shared.php
old mode 100644
new mode 100755
index 7aa411cd1..c7a4b9fb4
--- a/bootstrap/helpers/shared.php
+++ b/bootstrap/helpers/shared.php
@@ -3438,7 +3438,11 @@ function newParser(Application|Service $resource, int $pull_request_id = 0, ?int
                 }
             }
             foreach ($baseNetwork as $key => $network) {
-                $networks_temp->put($network, null);
+                if ($isApplication && $isPullRequest) {
+                    $networks_temp->put($network, ['aliases' => [$serviceName]]);
+                } else {
+                    $networks_temp->put($network, null);
+                }
             }
 
             if ($isApplication) {

Another way to do this is create another base network that just binds all containers together but doesn't have the coolify proxy attached. I'll wait a bit before submitting a PR to see what the best way forward is on that

[hydrandt]

@tm1000 did you have success with the suggested solution?

[rfournier1]

While this issue is still open, I share here my workaround :

I had the same problem : although refering to another container using the service name worked in a "regular" deployment, it failed in preview deployments.

In my case with the following entry in my docker-compose.yml :

services:
  db:
    image: postgres:14-alpine
    expose:
      - 5432

Refering the host db was properly resolved in my deployment (connecting to postgresql://db:5432 for example), but not in a preview deployment

I found a workaround using the "hostname" field in my docker-compose.yml : https://docs.docker.com/reference/compose-file/services/#hostname to force the container hostname on the network created by coolify, that work in both cases as it is not overwritten during the docker-compose generation.

services:
  db:
    image: postgres:14-alpine
    expose:
      - 5432
    hostname: db

[JensvandeWiel]

Any updates on this. Maybe the workaround of @rfournier1 is a good fix?

[ton77v]

Any updates on this. Maybe the workaround of @rfournier1 is a good fix?

Or at least documenting this would be really helpful, wasn't obvious at all

[hydrandt]

I confirm that setting the hostname property resolves the issue.

Another thing I had to do was getting rid of bind mounts: Even thou there seems to be some movement there (mounted folders are shown in the storage section), it did not work, and after every redeployment, they became duplicated.

Generally the functionality seems a bit shaky to me; the decision not to clone the repo second time to another folder and switch to another branch there, but append -pr-{pr_id} to file names, seems strange to me. But hey, it works now!:)

[arika0093]

After adding hostname according to the specified instructions, it still didn't work. It seems that the docker-compose file to be edited needs to be the one from the main branch, not the PR.

More precisely, the content listed in the Docker Compose Content on the General page appears to be reflected in the PR as well. In fact, when I open /data/coolify/application/(id)/docker-compose-pr-X.yaml, the changes from the PR are not reflected (at least in my case).

After editing and pushing the file from the main branch, then performing Reload Compose File and redeploying the Preview side, name resolution worked correctly. I hope this helps everyone.