There may be the following reasons:

• The weight settings of back-end servers are different

When the weight settings of back-end servers are inconsistent, the number of load balanced connections must be inconsistent.

• TCP Keepalived

If some backend servers enable TCP Keepalive to maintain long connections, while others do not, or the set long connection timeout is different, the connections will pile up on the backend servers that maintain long connections, causing imbalance.

• The backend server service status fluctuates

If the backend server is occasionally unavailable, the load balancer will check it and the client request will be allocated to other backend servers.

Troubleshooting

• Check whether the weights of back-end ECS are the same;
• Check the long connection settings of the back-end server;
• Check whether the load balancing back-end server service is normal.