Currently, every registration (and now metric push) initiates a new TCP/IP connection to the orchestrator. I wonder if it would be more efficient if each node set the keep-alive setting on the http connection and reused their socket connection for each interaction with the orchestrator. It is possible this will have no effect or cause other issues, but it is at least worth exploring.
Currently, every registration (and now metric push) initiates a new TCP/IP connection to the orchestrator. I wonder if it would be more efficient if each node set the keep-alive setting on the http connection and reused their socket connection for each interaction with the orchestrator. It is possible this will have no effect or cause other issues, but it is at least worth exploring.