Using cassandra, we hit a situation where the thrift server refused to accept more connections even though we killed all the clients.
I noticed that we were not calling transport close() from our client applications. Those processes terminate normally but never call close(). Is this a problem? My coworker did a test where he connected with a bunch of clients and then terminated them while monitoring the server ... and he saw the server releasing threads when the processes terminated. So presumably it is doing some level of cleanup when it detects that the client is gone. But I am wondering if this is still a bad practice?
