apache/rocketmq image It seems unable to gracefully shut down

[syrm-ll]

I tried two versions of Apache/RocketMQ: 5.3.2 and Apache/RocketMQ: 5.3.3 located on Dockerhub, and found that the process with PID 1 started by the container is a shell script rather than a JVM
This violates the best practice of containers. If the shutdown signal forwarding is not manually handled, the JVM should be made a PID1 process to ensure that the JVM shutdown hook is executed correctly, rather than forcing a shutdown after timeout

---

我尝试了位于 dockerhub 的 apache/rocketmq:5.3.2 apache/rocketmq:5.3.3 两个版本, 发现容器启动的 PID 为 1 的进程都是 shell 脚本而非 JVM

这违反了容器的最佳实践, 如果没有手动处理关机信号转发, 应该让 JVM 成为PID1进程, 以确保JVM关机挂钩正确执行, 而非超时后强制关闭

参考: https://docs.docker.com/build/building/best-practices/#entrypoint

容器内的进程:

rocketmq       1       0  0 23:38 ?        00:00:00 sh mqbroker
rocketmq       7       1  0 23:38 ?        00:00:00 sh /home/rocketmq/rocketmq-5.3.3/bin/runbroker.sh -Drmq.logback.configurationFile=/home/rocketmq/rocketmq-5.3.3/conf/rmq.broker.logback.xml org.apache.rocketmq.broker.BrokerStartup
rocketmq      14       7 17 23:38 ?        00:00:12 /opt/java/openjdk/bin/java -server -Xms2g -Xmx2g -Xmn1g -XX:MaxDirectMemorySize=1g -XX:+UseG1GC -XX:G1HeapRegionSize=16m -XX:G1ReservePercent=25 -XX:InitiatingHeapOccupancyPercent=30 -XX:SoftRefLRU
rocketmq     171       0  0 23:39 pts/0    00:00:00 /bin/sh -c [ -e /bin/bash ] && (/bin/bash || exit 0) || /bin/sh
rocketmq     177     171  0 23:39 pts/0    00:00:00 /bin/sh -c [ -e /bin/bash ] && (/bin/bash || exit 0) || /bin/sh
rocketmq     178     177  0 23:39 pts/0    00:00:00 /bin/bash

mqbroker 启动脚本部分内容:
sh 开启了子进程

if [ "$enable_proxy" = true ]; then
  args_for_proxy=$other_args" -pm local"
  if [ "$broker_config" != "" ]; then
      args_for_proxy=${args_for_proxy}" -bc "${broker_config}
  fi
  sh ${ROCKETMQ_HOME}/bin/runbroker.sh -Drmq.logback.configurationFile=$ROCKETMQ_HOME/conf/rmq.proxy.logback.xml org.apache.rocketmq.proxy.ProxyStartup ${args_for_proxy}
else
  args_for_broker=$other_args
  if [ "$broker_config" != "" ]; then
      args_for_broker=${args_for_broker}" -c "${broker_config}
  fi
  sh ${ROCKETMQ_HOME}/bin/runbroker.sh -Drmq.logback.configurationFile=$ROCKETMQ_HOME/conf/rmq.broker.logback.xml org.apache.rocketmq.broker.BrokerStartup ${args_for_broker}
fi

目前在 docker 环境中部署时, docker compose down 卸载或 docker compose restart 重启能明显感觉到迟缓, 因为没有人处理关机信号, docker 在超时后强制结束容器, 所以始终有固定的延迟

[syrm-ll]

更新这个问题, JVM 也不适合作为 PID 1 刚发现 JFR 对此有警告

Process runs with PID 1.

Java processes should not run with PID 1 as this PID is reserved for the init process in Linux, which is responsible for signal handling > and process management. Java processes do not natively handle signals like SIGTERM correctly, leading to issues such as unclean > shutdowns and orphaned child processes.

Enforce the process does not run with PID 1. In Docker, this can be achieved by using the lightweight init system tini with the '--init' > flag. For an example with Docker Compose refer to the documentation from Docker Compose.

但是总好过现在这样连关机信号都没有程序处理

更好的方案是使用 tini 或类似工具