Autohealing in Compute Engine is a feature of managed instance groups (MIGs), not of load balancers. You create a MIG and configure its Autohealing health check policy, specifying parameters such as the check interval, timeout, and the unhealthy threshold (number of consecutive failed attempts) before the MIG recreates an instance. Setting the health check to a healthy (HTTP) check with 3 attempts of 10 seconds each directly implements the required autohealing behavior with the fewest steps, since it doesn't require setting up a separate Application Load Balancer (which controls traffic distribution, not instance recreation) and is unrelated to autoscaling, which manages the number of instances based on load rather than instance health.
Community Discussion