stop pulling a connection from the accept queue. This leads to a back pressure on a client side
Pull and immediately close. This usually leads to retries on the client side
This PR implements the second option, which seems like a bad idea, but in some cases is necessary. For example, when we failover to a differrent host, we want a client to retry until its DNS cache is updated with the new host ip and it reconnects to a differrent node. This mode essentially simulates the behavior of a closed port.
There are two ways to stop accepting:
This PR implements the second option, which seems like a bad idea, but in some cases is necessary. For example, when we failover to a differrent host, we want a client to retry until its DNS cache is updated with the new host ip and it reconnects to a differrent node. This mode essentially simulates the behavior of a closed port.