registrator container ip address, not target container ip address, in etcd about registrator HOT 7 CLOSED

I have a similar problem about the registrator's IP is filled into the container's IP.

from registrator.

For etcd, the workaround at the moment is to use the -internal flag. While the documentation states it will use the docker0 interface IP, it actually registers the container IP address.

from registrator.

is there a workaround if you're using consul? I only want to allow my containers to discover the service on another container within the same host. I only have 1 host

from registrator.

I also tested the Registrator on consul, and see the same issue. The following is the string rendered from a haproxy template file for servers:

baseStr="server ci05:mad_archimedes:8085 172.17.2.121:49226 check"

The ip address 172.17.2.121 is the docker container internal ip address, not the native host, which is ci05, IP address. So I created a python script to parse this line, use dig to query the IP address for ci05, and replace the IP address to create the following new line for the above line

server ci05:mad_archimedes:8085 172.20.2.147:49226 check

Using consul-template, once the consul template file get rendered, run a script like the following to fix the IP address in the Haproxy cfg file, before restarting Haproxy.

Use the following regex to test haproxy server config lines

serverRegex=re.compile('\ server . .* check\ *')

Use the following loop to scan the haproxy cfg file, and fix the IP address in the server setting line.

with open(haproxy.cfg) as f:
for line in f:
       line=line[:-1]
   m=serverRegex.match(line)
   if m:
    newline=fix_addr(line)
    print newline
   else:
    print line

The server setting lines get fixed, printed without the leading spaces as in the original haproxy.cfg file.

This "plugin script to fix config files" after template rendering are applicable for fixing similar issues, not limited to the ip address issue reported here for Registrator.

from registrator.

Does someone has test the -internal parameter with consul registry since 856084a ?

It seems it doesn't work as intended, I'm still getting the host IP instead of container IP.

from registrator.

Same here.

I'm running a CoreOS cluster with Flannel Networklayer and want Registrator to push all ContainerIPs to Consul respecting all Ports exposed in the Dockerfile.

Because of the internal virtual network which connects all containers i use -p only to expose Ports to the public World.

from registrator.

The latest version resolves this when using -internal for both etcd and Consul.

from registrator.