The current scheduling policy prefer to keep on polling the Netty I/O (including scheduled and async tasks) although there is little to do.
The same won't happen with NIO, which can safely unmount the Netty I/O event loop once there is little to do, and just drain any pending virtual thread tasks.
This is a conservative mechanism to allow Netty I/O and timers to be handled in time, but if a Netty transport have little to do (eg it has no handled I/O events for the last N milliseconds) we could save polling it that frequently, at the cost of some queuing of Netty I/O related events.
The current scheduling policy prefer to keep on polling the Netty I/O (including scheduled and async tasks) although there is little to do.
The same won't happen with NIO, which can safely unmount the Netty I/O event loop once there is little to do, and just drain any pending virtual thread tasks.
This is a conservative mechanism to allow Netty I/O and timers to be handled in time, but if a Netty transport have little to do (eg it has no handled I/O events for the last N milliseconds) we could save polling it that frequently, at the cost of some queuing of Netty I/O related events.