Autoscaled instance group behind an HTTP(S) load balancer restarts VMs every minute; instances have no public IP and curl from an instance returns the correct site. What should you check to fix the backend?

The problem describes instances restarting every minute, which is characteristic of an autoscaling group terminating unhealthy instances. Since curl from an instance works, the application itself is likely running correctly. The most common reason for this behavior when a load balancer is involved is that the load balancer's health checks are failing. These health checks need specific firewall rules to allow them to reach the instances. Without these rules, the load balancer perceives the instances as unhealthy and replaces them, leading to a continuous restart cycle. Incorrect options: Allowing source HTTP/HTTPS traffic to the load balancer is standard but doesn't address the backend health check issue. Assigning public IPs to instances is unnecessary and often undesirable for security; the load balancer communicates with private IPs. Tagging instances with the load balancer name and using it as a source in a firewall rule is not how load balancer health check firewall rules are typically configured; health checks originate from specific Google-managed IP ranges.