IssueTracker

Can you give me examples of application(s) which has been impacted by this behavior.

One affected application is TeamSpeak, but this is valid for any situation in which a server-side application makes the (relatively reasonable) assumption that the IP address it's explicitly bound to is the same address clients will connect to.  

I have forwarded this request to the engineering team. We will update this issue with any progress updates and a resolution.

Adding the external IP to a local interface is just as you said:

  ip addr add <IPADDRESS>/32 dev eth0 label eth0:shortlabel

To remove it, replace 'add' with 'del'.

For figuring out the external IP of the network load balancer you can take a look at the code in https://github.com/GoogleCloudPlatform/compute-image-packages/blob/master/google-daemon/usr/share/google/google_daemon/address_manager.py.

According to that script, you can get the network load balancer IPs configured on the VM by running (there can be more than one):

  ip route ls table local type local dev eth0 scope host proto 66

What I do is to run a cronjob every minute that synchronizes the IPs from 'ip route ls' and the IPs configured on eth0 that have the eth0:shortlabel label.

2gonza...
It didn't help.
But it's strange.
Before I did
 ip addr add <IPADDRESS>/32 dev eth0 label eth0:shortlabel
I could telnet my port and after I couldn't.
But I could telnet 22 port.
So:
1) Before  ip addr add <IPADDRESS>/32 dev eth0 label eth0:shortlabel
telnet 104.155.66.4 22  - OK
telnet 104.155.66.4 9042 - OK
After  ip addr add <IPADDRESS>/32 dev eth0 label eth0:shortlabel
telnet 104.155.66.4 22  - OK
telnet 104.155.66.4 9042 - Connection refused

Have there been any known updates to this topic in another issue?

Any update?

Any update?

Amy update?

I think this can basically be achieved on Linux on GCE via un-nat-ting in input (prerouting) and re-nat-ting in output (postrouting).  I've been using something which is roughly the following:

#!/bin/bash

main() {
  case "$(hostname)" in
    your-vm-hostname)                  # This is your VM's hostname (allows use of same script on multiple VMs)
      local -r GW_IP=10.1.0.1         # This is the gateway
      local -r MY_IP=10.1.0.2          # This is the VM internal static IP
      local -r EXT_IP=123.4.5.6      # This is the VM external static IP
      ;;
    *)
      echo "Unknown hostname $(hostname)." 1>&2
      return 1
      ;;
  esac

  echo "Hostname($(hostname)) MY_IP[${MY_IP}] GW_IP[${GW_IP}] EXT_IP[${EXT_IP}]"

  ip addr add "${EXT_IP}/32" dev eth0

  iptables -t nat -F PREROUTING
  iptables -t nat -A PREROUTING ! -i eth0 -j ACCEPT
  iptables -t nat -A PREROUTING -s      0.0.0.0/8  -j ACCEPT
  iptables -t nat -A PREROUTING -s     10.0.0.0/8  -j ACCEPT
  iptables -t nat -A PREROUTING -s   100.64.0.0/10 -j ACCEPT
  iptables -t nat -A PREROUTING -s    127.0.0.0/8  -j ACCEPT
  iptables -t nat -A PREROUTING -s  169.254.0.0/16 -j ACCEPT
  iptables -t nat -A PREROUTING -s   172.16.0.0/12 -j ACCEPT
  iptables -t nat -A PREROUTING -s  192.168.0.0/16 -j ACCEPT
  iptables -t nat -A PREROUTING -s    224.0.0.0/3  -j ACCEPT
  iptables -t nat -A PREROUTING -d "${MY_IP}" -j DNAT --to-destination "${EXT_IP}"

  iptables -t nat -F POSTROUTING
  iptables -t nat -A POSTROUTING -o eth0 -s "${EXT_IP}" -j SNAT --to-source "${MY_IP}"

  ip -4 route del default
  ip -4 route del "${GW_IP}" dev eth0 scope link
  ip -4 route add "${GW_IP}" dev eth0 scope link src "${MY_IP}"
  for SUBNET in 10.0.0.0/8 100.64.0.0/10 169.254.0.0/16 172.16.0.0/12 192.168.0.0/16; do
    ip -4 route del "${SUBNET}"
    ip -4 route add "${SUBNET}" via "${GW_IP}" dev eth0 src "${MY_IP}"
  done
  ip -4 route add default via "${GW_IP}" dev eth0 src "${EXT_IP}"
}

main "$@"; exit

Thank you for this hack. So you rewrite Public IP with Private IP on the kernel level (iptables) and the end application does not notice anything.

Might be Google will support in the future simpler solution as other Cloud Provider.
Just provides two interfaces one for public and another one for private network.

Alex