Load test Dragonfly for processing pub sub commands; When the request volume reaches 40W TPS, a performance bottleneck issue occurs.

[Memorydoc123]

PID: 71740 CPU at full capacity

Continue to track the program with 100% CPU usage: PID: 7176

It was found that a large number of write response handle operations were carried out, causing the single thread to be completely blocked.

Could you please take a look at how Dragonfly can meet the TPS performance requirement of 40W for the pub/sub commands?

Explanation: Drafgonfly is deployed based on Docker and is installed on a 64C 128 ECS machine.

docker run -d --name dragonfly_db --ulimit memlock=-1 docker.dragonflydb.io/dragonflydb/dragonfly --proactor_threads 64 --maxmemory 64G --multi_exec_mode 2 --default_lua_flags=allow-undeclared-keys --lock_on_hashtags

[romange]

what's a 40W?