Provide rapid failover away from unhealth and/or removed VMs from the Load Balancer backend pool
Presently, the Standard SKU Load Balancer takes up to several minutes to stop sending traffic to backend VMs which have been identified as unhealthy by Health probes and/or have been manually removed from a backend pool through a configuration change.
This delay prevents using the Load Balancer as an SLA/availability solution and is counter-intuitive. A preferable design would be to immediately cease sending any additional traffic to an unhealthy VM once it has been marked as unhealthy (unless it is the only VM in the backend pool.)
I’m sorry to hear this and not how the product should be operating. Please reach out to me or open a support case to investigate.