I added a cassandra CQL service which loadbalances connections to the CQL ports of all Cassandra nodes.
But on further investigation, it seems that clients (such as GoCQL) will discover all the C* nodes automatically, given a connection to one of them.
And the loadbalancer IP of the CQL service might even confuse this process, since it's not an IP address advertised by any of the nodes.
Figure out the best practice for connecting to Cassandra and document it and perhaps add an example to the documentation and to the E2E tests.
I added a cassandra CQL service which loadbalances connections to the CQL ports of all Cassandra nodes.
But on further investigation, it seems that clients (such as GoCQL) will discover all the C* nodes automatically, given a connection to one of them. And the loadbalancer IP of the CQL service might even confuse this process, since it's not an IP address advertised by any of the nodes.
Figure out the best practice for connecting to Cassandra and document it and perhaps add an example to the documentation and to the E2E tests.
Get rid of the CQL service if it is found to be unnecessary.
/kind feature