How does one make use of connection pooling when using a replicated database with one entrypoint? Since we have an internal loadbalancer in our cluster, the entrypoint to the replicated instances is one and the same, so when our Restful api issues a connection, the same replicated database instance is returned for all subsequent calls.

By disabling connection pooling, we can retrieve different read slaves as the loadbalancer determines, but wouldn't this be a big performance loss?