Docker Compose双网卡宿主机：配置Django容器访问外部PostgreSQL

Let's break down a clean, Docker-native approach to solve your problem without disrupting existing container communication or external network access.

Core Idea #

Create a custom Docker bridge network explicitly bound to your VM_D's en1 interface (the one connected to VM_P). This lets your Django-related containers access VM_P's PostgreSQL via this custom network, while still keeping the default bridge network for communication with other services (Caddy, Redis, etc.).

Step 1: Create the Custom Network #

First, create a bridge network tied to en1. Replace the subnet/gateway/IP values with ones matching your en1 network segment (run ip addr show en1 on VM_D to get these details):

docker network create \
  --driver bridge \
  --subnet 192.168.10.0/24 \  # Match en1's subnet
  --gateway 192.168.10.1 \    # Match en1's gateway
  -o "com.docker.network.bridge.host_binding_ipv4"="192.168.10.5" \  # en1's static IP on VM_D
  custom-postgres-net

Step 2: Update Your Docker Compose Config #

Modify your production.yml to include the custom network and attach it to Django/Celery services:

1. Add a networks section at the top level (under volumes):

version: '2'
volumes:
  caddy: {}
networks:
  default:  # Keep the default bridge network for internal service communication
    external: false
  custom-postgres-net:  # Reference the custom network we created
    external: true

2. Attach the custom network to your Django, Celery Worker, and Celery Beat services (since they all need database access):

django: &django
  build:
    context: .
    dockerfile: ./compose/production/django/Dockerfile
  depends_on:
    - redis
  env_file: .env
  command: /gunicorn.sh
  networks:
    - default
    - custom-postgres-net  # Add this line

celeryworker: <<: *django
  depends_on:
    - redis
  command: /start-celeryworker.sh
  networks:
    - default
    - custom-postgres-net  # Add this line

celerybeat: <<: *django
  depends_on:
    - redis
  command: /start-celerybeat.sh
  networks:
    - default
    - custom-postgres-net  # Add this line

Leave the Caddy and Redis services as-is—they don't need access to VM_P, so staying on the default network is fine.

Step 3: Configure Django's Database Connection #

Update your .env file to point to VM_P's PostgreSQL IP (the one reachable via VM_D's en1 interface):

DATABASE_URL=postgres://<db-user>:<db-password>@<vm-p-ip>:5432/<db-name>

Step 4: Verify PostgreSQL Access on VM_P #

Ensure VM_P's PostgreSQL allows connections from VM_D's en1 subnet:

1. Edit pg_hba.conf on VM_P and add:

host all all 192.168.10.0/24 md5  # Replace with your en1 subnet

2. Restart PostgreSQL to apply changes:

sudo systemctl restart postgresql  # Adjust command based on your OS

Why This Works #

• No network conflicts: Your Django containers use two networks—default bridge for internal services, custom network for VM_P access. Docker supports multiple networks per container, so this avoids the "only one host network" error from your first attempt.
• Preserves external access: VM_D's en0 interface remains unaffected, so your containers can still pull images, access external APIs, etc.
• Native Docker tooling: No third-party tools like Pipework needed—this uses Docker's built-in network capabilities.

Troubleshooting Tips #

• Confirm the custom network is bound correctly: Run docker network inspect custom-postgres-net and check the Options section for the host binding IP.
• Test connectivity from the Django container:

docker exec -it <django-container-name> ping <vm-p-ip>
docker exec -it <django-container-name> psql -h <vm-p-ip> -U <db-user> -d <db-name>

• Ensure en1 on VM_D has a static IP (avoid DHCP changes breaking the network binding).

内容的提问来源于stack exchange，提问作者AlEmerich