docker - 尽管设置了 `-t 120`，Docker仍几乎立即停止关闭容器

Question

这就是我构建docker镜像的方式:

docker build -f Dockerfile -t python-trapterm .

这就是我的运行方式:

 docker run -t python-trapterm:latest

我试图阻止它

 docker stop -t 120 sharp_jones

我希望在关闭容器之前需要2分钟的运行时间。但是它几乎立即关闭了吗？为什么？

这是容器过程的输出:

> docker run -t python-trapterm:latest
Setting traps...
 * Starting NTP server ntpd                                              [ OK ] 
2017-09-10 06:43:02,501 pid 26 tid 140693944227584 INFO     greatapp Registering signal handler
2017-09-10 06:43:02,502 pid 26 tid 140693944227584 INFO     greatapp Hello!
2017-09-10 06:43:04,507 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
2017-09-10 06:43:06,510 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
2017-09-10 06:43:08,512 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
[snip]
2017-09-10 06:44:12,629 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
2017-09-10 06:44:14,631 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
Sending term to 26...
2017-09-10 06:44:16,537 pid 26 tid 140693944227584 INFO     greatapp Handling shutdown...15
2017-09-10 06:44:16,538 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...
2017-09-10 06:44:16,539 pid 26 tid 140693944227584 INFO     greatapp Handling shutdown...10
2017-09-10 06:44:16,539 pid 26 tid 140693944227584 INFO     greatapp Checking every 2s...

您可以看到在06:44:16左右收到TERM信号，但是作业在06:44:14左右已经关闭，但是该过程在06:44:16之后终止。没有10秒的等待时间。

这是我的docker文件:

FROM ubuntu:16.04

RUN apt-get update --fix-missing && apt-get install -y \
  git \
  python \
  python-pip \
  ntp \
  curl vim screen lsof

RUN mkdir -p /app

COPY ./app /app

CMD ["/app/start-app.sh"]

这是我的 app目录中的两个文件:

应用程序/start-app.sh

#!/bin/bash
set -eo pipefail

echo "Setting traps..."
trap 'cleanup_term' TERM
trap 'cleanup_kill' KILL

cleanup_term() {
    echo "Sending term to `jobs -p`..."
    kill -TERM `jobs -p`
    kill -USR1 `jobs -p`
}

cleanup_kill() {
    echo "Received kill..."
    kill -TERM `jobs -p`
}


service ntp start

cd /app
python -m greatapp &

wait

和

应用程序/greatapp.py

import os
import time
import logging
import signal


LOG_FORMAT = "%(asctime)-15s pid %(process)d tid %(thread)4d %(levelname)-8s" \
    " %(module)-8s %(message)s"

logging.basicConfig(format=LOG_FORMAT)
LOG = logging.getLogger(__file__)
LOG.setLevel(logging.DEBUG)

def docker_shutdown_handler(_signum, _frame):
    LOG.info("Handling shutdown...%s", _signum)
    # sys.exit()


LOG.info('Registering signal handler')
signal.signal(signal.SIGTERM, docker_shutdown_handler)
signal.signal(signal.SIGUSR1, docker_shutdown_handler)


def main():
    LOG.info("Hello!")
    sleep_time = 2
    while True:
        time.sleep(sleep_time)
        LOG.info("Checking every %ds...", sleep_time)


if __name__ == '__main__':
    main()

Answer 1

您的Shell脚本正在退出。发生这种情况时，由于是pid 1，该容器将立即退出。如果将以下内容添加到脚本顶部，则可以看到此信息:

trap 'echo Script exiting' EXIT

这很可能是bash中SIGKILL的默认操作，因为在最后一次等待之后它没有运行任何命令。解决此问题的快速方法是在清除函数中添加一个等待，以便信号处理程序挂起:

cleanup_term() {
    echo "Sending term to `jobs -p`..."
    kill -TERM `jobs -p`
    kill -USR1 `jobs -p`
    echo "waiting again"
    wait
    echo "cleanup_term done"
}

或者我建议只执行python脚本，以便它直接处理信号，并消除pid 1中的bash。start-app.sh如下所示:

#!/bin/bash
set -eo pipefail

service ntp start

cd /app
exec python -m greatapp