I find it a bit confusing how Google Cloud Auto Scaler, Managed Instance Group, Target Pools and Network Load Balancers work together. The documentation is a bit sparse and there is no explicit connection between how to use a NLB (we need this since we require HTTPS, so the current Http Load Balancer won't cut it) and an Auto Scaler. I would think that the Auto Scaler main use case would be to manage instances behind a NLB.

Problem 1: it doesn't seem that once we assign a target pool to a managed instance group (via the [set-target-pool][1]) the already exiting instances get assigned to that target pool (may be our mistake).

Problem 2: it doesn't seem that managed group instances get into rotation on the NLB unless they have a public (albeit ephemeral) IP. Could be our mistake once again, but shouldn't a private IP be sufficient?

Problem 3: if an instance on the managed instance group becomes unhealthy (say.. it stops responding to HTTP) the NLB removes it from the rotation (via the health check) but it seems that the Auto Scaler stays happy and considers it to be a valid running instance.

What are we missing?


HTTPS load balancing is now available in Alpha and may now represent the best solution for you.

Problem 1: Instance Group Manager expects it's groups to be completely homogeneous. Therefore when a managed instance group is configured to add it's instances to a target pool it is natural for it to add the existing ones as well as future ones. If you need to split so that some instances that are added to the target pool and some which are not, you should have to separate instance groups.

Problem 2: It is possible to use network load balancing with instances that have private IP addresses only. However the forwarding rule must have an external IP address.

Problem 3: You are correct, Autoscaler does not do health checks, and will not automatically destroy an instance that fails load balancing health checks. If you want this currently you will do so manually (I recommend using the Instance Group Manager "recreate" call).

  • Hi @David. I think you misunderstood our Problem 1. We expected the already existing instances in the managed instance group to be added to the target pool once we set it for the instance group manager, but that didn't happen. – miguelv Mar 28 '15 at 17:42
  • HTTPS load balancing is invite only. Any idea how we can get invited? Problem 2: our instances never got any traffic when using private IP addresses. Not sure what we did wrong. Problem 3: don't understand how it is possible to have auto-scaling working decently if no health checks are employed. If all our instances die, we then have zero capacity and the auto scaler won't kick in? – miguelv Mar 28 '15 at 17:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.