Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
0.37.0
-
None
Description
Currently the protocol specific processing is handled in its own single threaded executor which creates a performance drop as reads and writes are queued into Netty for handling. We can achieve a significant performance boost by handling all the protocol work inside the Netty event loop and not hopping between threads as we do now.
This requires some refactoring of connect and shutdown logic and some safeguards around all callbacks in the transport to ensure that we always operate on the event loop and not one on a client thread.