Astyanax marks the entire (Cassandra) host as down when a TTransportException (Thrift exception) is thrown because max Thrift message size (thrift_max_message_length_in_mb, Cassandra knob) is exceeded.
The host is marked as down because Astyanax wraps TTransportException (Thrift exception) in a TransportException. (That exception implements IsDeadConnectionException).
A better approach would be to close that single tcp connection (the connection we received the thrift frame size error from)
This behavior was implemented in version 1.56.12 and up.
I review the code ThriftSyncConnectionFactoryImpl.java it seems no config to set this
transport = new TFramedTransport(socket);