Compute Engine questions on the Professional Cloud Architect exam often combine a Managed Instance Group with a load balancer, and the wording of those questions assumes you already understand how the two pieces fit together. I want to walk through that pairing in the same shape it tends to appear on the exam.
Picture incoming traffic on the left, a Cloud Load Balancer in the middle, and a Managed Instance Group of VMs on the right. The load balancer takes user requests, whether that is website traffic or API calls from a mobile app, and spreads them across the VMs in the group.
The MIG itself is autoscaling. As demand rises, more VMs spin up. As demand falls, VMs are removed. The load balancer keeps redistributing traffic across whatever instances currently exist in the group. No single VM gets overwhelmed, and the group stays sized to actual demand.
That is the core pattern: autoscaling handles capacity, the load balancer handles distribution. They are designed to work together, and the exam treats them as a default pairing for any compute workload that needs to scale horizontally.
For most workloads, a Global Load Balancer is the right answer. In GCP, Global Load Balancer essentially means an HTTPS load balancer, and it gives users low-latency access from anywhere in the world.
The exception is TCP workloads. If the question describes raw TCP traffic rather than HTTP or HTTPS, the Network Load Balancer is the correct choice because it is built for those protocols. When you see a question naming the protocol, let the protocol pick the load balancer.
Load balancers run health checks against the VMs in the MIG to confirm each instance is actually serving traffic. If a firewall rule blocks those health check probes, the load balancer concludes the VMs are unhealthy. That can trigger unnecessary instance replacements or pull instances out of rotation when nothing is actually wrong.
You have to allow health checks in your firewall rules. This is one of those small details that shows up in scenario questions where the symptom looks like instability or churn in the MIG, and the actual fix is a firewall rule.
Cloud Load Balancers do not require a Managed Instance Group. You can put one in front of individual virtual machines or in front of an unmanaged instance group. That is useful when horizontal scaling is not the goal, for example with stateful workloads or an active-standby configuration where one VM serves traffic and another stands by.
So the mental model is: MIGs give you scalability, load balancers give you distribution and health-aware traffic routing, and the two are commonly combined but not strictly bound to each other.
For the Professional Cloud Architect exam, anchor a few points. Cloud Load Balancing distributes traffic across a MIG and works with the autoscaler so the group can grow and shrink without manual intervention. Global HTTPS load balancers are the default for web traffic. Network Load Balancers are the answer for TCP workloads. Health checks need to be allowed in firewall rules. And load balancers can sit in front of single VMs or unmanaged groups when scaling is not what you need.
My Professional Cloud Architect course covers MIGs and load balancing alongside the rest of the compute material.
