unit tests try to talk to themselves via rpc

[avikivity]

running any cql_query_test shows (may need debug mode to slow it down)

ERROR 2023-06-25 20:33:52,036 [shard 0] rpc - client 127.0.0.1:7000: client connection dropped: connection is closed
ERROR 2023-06-25 20:33:52,142 [shard 0] rpc - client 127.0.0.1:7000: client connection dropped: connection is closed
ERROR 2023-06-25 20:33:52,241 [shard 0] rpc - client 127.0.0.1:7000: client connection dropped: connection is closed
ERROR 2023-06-25 20:33:52,342 [shard 0] rpc - client 127.0.0.1:7000: client connection dropped: connection is closed
ERROR 2023-06-25 20:33:52,439 [shard 0] rpc - client 127.0.0.1:7000: client connection dropped: connection is closed

this is wrong, cql_test_env shouldn't try to talk to anything.

[avikivity]

@kbr-scylla @kostja

[kostja]

@margdoc I thinik this is something you could look at, CC @kbr-scylla

[nyh]

@avikivity I'm doing "wireshark" over an actual single-node Scylla run (not a unit test), and it also seems to be "talking to itself" - sending a message to TCP port 7000 (rpc) to 127.0.0.1 (this node itself), roughly once every 0.1 seconds, very regularly. I'm not sure it's the same issue that you saw in unit tests, but maybe? Why would Scylla need to send itself messages over TCP?

[kostja]

Direct failure detector currently doesn't make a distinction between self and other nodes in the cluster. This can be fixed now after @kbr-scylla refactoring of CQL query processor, we can load host id early and detect that we're trying to ping self in the failure detector.

[avikivity]

Minor, not backporting.