What can I do if a load balancer fails a stress test?

Check the load of backend ECSs. If their CPU usage reaches 100%, applications may encounter performance bottlenecks.
Check the traffic. If the traffic exceeds the bandwidth set for the EIP of the ELB, a large number of packets will be lost and requests will fail, affecting the stress test performance.
Check the number of short connections in the time_wait state on the clients. It may be caused by insufficient client ports. To solve this problem, you can add the number of IP addresses on clients.
The listening queue backlog of the backend servers may be full. In this case, the backend ECS will not respond to syn_ack packets, and the client will time out. You can increase the upper limit of backlog by adjusting the net.core.somaxconn parameter.

How do I check if LB request traffic is being unevenly distributed?

Check whether session persistence is enabled. If session persistence is enabled and there are few clients, traffic may be unevenly distributed.
Check the health check status of backend ECSs, especially those whose health check status changes over time. If a backend ECS is Unhealthyor its health status switches between Healthy andUnhealthy, traffic is unbalanced.|
Check whether the Source IP hash algorithm is used. If it is used, requests from the same IP address are delivered to the same backend server, resulting in unbalanced traffic.
Check whether weights are set when ECSs are added to ELB backend servers. The traffic varies according to the weights.

Elastic Load Balancing

Others