perf: hibernate main loop

[JimMoen]

Compared 0.4.26 and this branch with same benchmark command
(both using docker image local built by command make docker)

bench compose file:

Connecting to a 2-nodes-cluster

services:

  node1-bench1:
    image: emqx/emqtt_bench:${EMQTT_BENCH_TAG}
    restart: always
    container_name: node1-bench1
    command: "conn -i 2 -h 172.19.0.2 -u admin -P public123 -c ${BENCH_CONN_COUNT_PER_NODE:-25000} --prefix node1-bench1"
    hostname: node1-bench1
    networks:
      emqx-net:
        ipv4_address: 172.19.0.31

  node1-bench2:
    image: emqx/emqtt_bench:${EMQTT_BENCH_TAG}
    restart: always
    container_name: node1-bench2
    command: "conn -i 2 -h 172.19.0.2 -u admin -P public123 -c ${BENCH_CONN_COUNT_PER_NODE:-25000} --prefix node1-bench2"
    hostname: node1-bench2
    networks:
      emqx-net:
        ipv4_address: 172.19.0.32

  node2-bench1:
    image: emqx/emqtt_bench:${EMQTT_BENCH_TAG}
    restart: always
    container_name: node2-bench1
    command: "conn -i 2 -h 172.19.0.3 -u admin -P public123 -c ${BENCH_CONN_COUNT_PER_NODE:-25000} --prefix node2-bench1"
    hostname: node2-bench1
    networks:
      emqx-net:
        ipv4_address: 172.19.0.33

  node2-bench2:
    image: emqx/emqtt_bench:${EMQTT_BENCH_TAG}
    restart: always
    container_name: node2-bench2
    command: "conn -i 2 -h 172.19.0.3 -u admin -P public123 -c ${BENCH_CONN_COUNT_PER_NODE:-25000} --prefix node2-bench2"
    hostname: node2-bench2
    networks:
      emqx-net:
        ipv4_address: 172.19.0.34

networks:
  emqx-net:
    ipam:
      config:
        - subnet: 172.19.0.0/16
    driver: bridge

CPU usage when the number of connections is about 14,000, (about 20 seconds after starting)

0.4.26:

this branch:

[qzhuyan]

I think the 'perf improvements' is mostly by changing it from 500ms to 50ms and it is case by case, we see reduced CPU usage and lowered memeory usage but latency become higher I guess.

Could you make this configurable instead rather than hardcode it?

[qzhuyan]

same here.