Why would a specific docker network pick a weird range?

Question

I have two "user" networks: the LAN on 192.168.10.0/24 and Wireguard on 192.168.20.0/24. I also host on 192.168.10.2 a docker engine that powers a dozen of containers and a few networks that all neatly land on 172.XX.0.0/16. Everything works great.

I added a set of new containers to host a Graylog service. I used the docker-compose.yml provided by the repo, with a few minor changes: a connection to a docker srv network that allows web containers to be connected to a caddy reverse proxy. The full docker-compose.yaml is as follows:

version: '2'
services:
  # MongoDB: https://hub.docker.com/_/mongo/
  mongodb:
    image: mongo:4.2
    volumes:
      - mongo_data:/data/db
    networks:
      - default
  # Elasticsearch: https://www.elastic.co/guide/en/elasticsearch/reference/7.10/docker.html
  elasticsearch:
    image: docker.elastic.co/elasticsearch/elasticsearch-oss:7.10.2
    volumes:
      - es_data:/usr/share/elasticsearch/data
    environment:
      - http.host=0.0.0.0
      - transport.host=localhost
      - network.host=0.0.0.0
      - "ES_JAVA_OPTS=-Xms512m -Xmx512m"
    ulimits:
      memlock:
        soft: -1
        hard: -1
    mem_limit: 1g
    networks:
      - default
  # Graylog: https://hub.docker.com/r/graylog/graylog/
  graylog:
    image: graylog/graylog:4.3
    volumes:
      - graylog_data:/usr/share/graylog/data
    environment:
      # CHANGE ME (must be at least 16 characters)!
      - GRAYLOG_PASSWORD_SECRET=XXX
      # Password: admin
      - GRAYLOG_ROOT_PASSWORD_SHA2=XXXXX
      - GRAYLOG_HTTP_EXTERNAL_URI=https://graylog.XXXX/
    entrypoint: /usr/bin/tini -- wait-for-it elasticsearch:9200 --  /docker-entrypoint.sh
    links:
      - mongodb:mongo
      - elasticsearch
    restart: always
    depends_on:
      - mongodb
      - elasticsearch
    ports:
      # Graylog web interface and REST API
      #- 9000:9000
      # Syslog TCP
      - 1514:1514
      # Syslog UDP
      - 1514:1514/udp
      # GELF TCP
      - 12201:12201
      # GELF UDP
      - 12201:12201/udp
    networks:
      - default
      - srv
# Volumes for persisting data, see https://docs.docker.com/engine/admin/volumes/volumes/
volumes:
  mongo_data:
    driver: local
  es_data:
    driver: local
  graylog_data:
    driver: local

networks:
  default:
  srv:
    external: true
    name: srv

For some reason the docker network that was automatically created was on 192.168.16.0/20 and it took me quite a lot of time to understand why Wireguard was not working anymore: this network contains 192.168.20.0/24 and the wireguard traffic was routed there...

Whe shutting down Graylog and manually deleting 192.168.16.0/20 everything is back to normal.

My question: why would a docker composition pick a wildly different network from the ones other containers use? How is this choice driven if there is nothing in docker-compose.yml that would hint at the network to use?

Answer 1

After manually deleting the docker network (per the information in the lat part of the question) I restarted the Graylog containers and, behold, the created network follows the 172.XX.0.0/16` template.

My wild guess would be that there was a network forced on 192.168..., then Graylog decided to follow this new template, and the former network was deleted.