Docker bridge IP being reported to Consul

[getstek]

Hi,

I have an Ansible build which uses Registrator for container registration with Consul. As of yesterday new builds started using the Docker bridge IP and not the host IP of the Mesos slave.

I run Registrator with the following syntax (this has not changed in months)

docker run -d --name=registrator --net=host --volume=/var/run/weave.sock:/tmp/docker.sock glider labs/registrator:latest consul://x.x.x.x:xx

When I dig the domain I get:

[root@dc01-worker-01 ~]# dig @192.168.100.249 -p 8600 postgresql.service.dc01.han.solo

; <<>> DiG 9.9.4-RedHat-9.9.4-29.el7_2.3 <<>> @192.168.100.249 -p 8600 postgresql.service.dc01.han.solo
; (1 server found)
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 16591
;; flags: qr aa rd; QUERY: 1, ANSWER: 1, AUTHORITY: 0, ADDITIONAL: 0
;; WARNING: recursion requested but not available

;; QUESTION SECTION:
;postgresql.service.dc01.han.solo. IN   A

;; ANSWER SECTION:
postgresql.service.dc01.han.solo. 0 IN  A       172.17.0.2

I am not sure why this is happening, as I have not set the -internal flag.

[Mobius5150]

I can confirm this problem as well - registrator is registering the local docker bridge IP rather than the host IP. By moving to registrator:v6 and using the exact same configuration it began working correctly.

Obviously it would be optimal to use the latest version.

[robvelor]

I have sort of the same issue, it might be related. I am using docker 1.10.3, consul-server 0.6, consul-agent 0.6, and registrator v7. I have web service tied to a defined network. When I use registrator v7 the service is registered using its internal network ip address instead of the host address. This causes a registered HTTP check to use the web service internal ip (the registrator & agents are not part of this network) instead of the host address so the check fails. If I use registrator v6 the container is added with the host address where the service is running (albeit localhost) so the check works.

Consul server compose

version: '2'
services:
  consul-server:
    image: gliderlabs/consul-server:0.6
    restart: always
    ports:
      - "8300:8300"
      - "8301-8302:8301-8302"
      - "8301-8302:8301-8302/udp"
      - "8400:8400"
      - "8500:8500"
      - "8600:53/udp"
    command: "-bootstrap -advertise $NODE_IP -node $NODE_NAME"
    network_mode: "host"

Consul agent compose:

version: '2'
services:
  consul-agent:
    image: gliderlabs/consul-agent:0.6
    restart: always
    ports:
      - "8300:8300"
      - "8301-8302:8301-8302"
      - "8301-8302:8301-8302/udp"
      - "8400:8400"
      - "8500:8500"
      - "8600:53/udp"
    command: "-join $CONSUL_SERVER_IP -advertise $NODE_IP -node $NODE_NAME"
    network_mode: "host"
    labels:
      - "SERVICE_8300_NAME=consul-agent"
      - "SERVICE_53_IGNORE=true"
      - "SERVICE_8301_IGNORE=true"
      - "SERVICE_8302_IGNORE=true"
      - "SERVICE_8400_IGNORE=true"
      - "SERVICE_8500_IGNORE=true"
      - "SERVICE_TAGS=agent"

Registrator compose

version: '2'
services:
  registrator:
    image: gliderlabs/registrator:v7
    restart: unless-stopped
    volumes:
      - /var/run/docker.sock:/tmp/docker.sock
    network_mode: "host"
    command: ["consul://localhost:8500"]

Web service compose

version: '2'
services:
  web:
    image: webservice
    ports:
      - "3000:3000"
    networks:
      - front_tier
      - back_tier
    labels:
      - "SERVICE_3000_NAME=webservice"
      - "SERVICE_TAGS=www"
      - "SERVICE_3000_CHECK_HTTP=/"
      - "SERVICE_3000_CHECK_INTERVAL=15s"
      - "SERVICE_3000_CHECK_TIMEOUT=1s"

# other services go here

networks:
  front_tier:
  back_tier:

If I query for the service:
v7
http://192.168.99.101:8500/v1/catalog/service/webservice = "ServiceAddress":"10.0.1.2"

v6
http://192.168.99.101:8500/v1/catalog/service/webservice = "ServiceAddress":"127.0.0.1"

"Address" holds the correct node ip on both counts.

[mattatcha]

@getstek Can you see this is resolved in :master?

[electic]

I am having this issue too with v7. I am on etcd and the port being written is the internal docker port, not the one outside. Did anyone figure this one out?

[robvelor]

@electic No word yet, I downgraded to v6 and it works. Looks like support for etcd was introduced in v6 so you should be good with that one.

[josegonzalez]

@electic @robvelor can you two try the patch in #441 and see if it fixes your issue?

[josegonzalez]

Seems like the creator of that patch doesn't think it resolves this issue, so I closed the pr for now.