php5-fpm，被 HUP 信号 : Main Process failed to respawn 杀死

Question

我的主 php 服务器出现问题，其中主 php5-fpm进程将被 HUP 杀死信号。在主进程被杀死后，它将无法重生。由于每个子进程只允许处理一定数量的请求，因此它们最终会死掉而不会产生任何其他子进程。这将导致服务器死机，我的用户将收到来自服务器的 502 响应。我最初能够通过一个 cron 来解决这个问题，它会检查 PHP 进程的线程数，然后如果它小于 5 则重新启动。

Sep 14 11:41:41 ubuntu kernel: [ 3699.092724] init: php5-fpm main process (3592) killed by HUP signal
Sep 14 11:41:41 ubuntu kernel: [ 3699.092740] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.160940] init: php5-fpm main process (3611) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.160954] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.216950] init: php5-fpm main process (3619) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.216966] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.283573] init: php5-fpm main process (3627) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.283590] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.337563] init: php5-fpm main process (3635) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.337579] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.385293] init: php5-fpm main process (3643) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.385305] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.430903] init: php5-fpm main process (3651) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.430913] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.482790] init: php5-fpm main process (3659) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.482800] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.532239] init: php5-fpm main process (3667) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.532249] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.595810] init: php5-fpm main process (3675) terminated with status 78
Sep 14 11:41:42 ubuntu kernel: [ 3699.595825] init: php5-fpm main process ended, respawning
Sep 14 11:41:42 ubuntu kernel: [ 3699.648253] init: php5-fpm main process (3683) terminated with status 78
Sep 14 11:41:42 ubuntu0 kernel: [ 3699.648265] init: php5-fpm respawning too fast, stopped

我的 Upstart 脚本配置

# php5-fpm - The PHP FastCGI Process Manager

description "The PHP FastCGI Process Manager"
author "Ondřej Surý <ondrej@debian.org>"

start on runlevel [2345]
stop on runlevel [016]

# Precise upstart does not support reload signal, and thus rejects the
# job. We'd rather start the daemon, instead of forcing users to
# reboot https://bugs.launchpad.net/ubuntu/+source/php5/+bug/1272788
#
#reload signal USR2

pre-start exec /usr/lib/php5/php5-fpm-checkconf

respawn
exec /usr/sbin/php5-fpm --nodaemonize --fpm-config /etc/php5/fpm/php-fpm.conf

Answer 1

在网上搜索后终于可以通过修改php5-fpm的 Upstart 脚本来解决这个问题。在 /etc/init/php5-fpm.conf

# php5-fpm - The PHP FastCGI Process Manager

description "The PHP FastCGI Process Manager"
author "Ondřej Surý <ondrej@debian.org>"

start on runlevel [2345]
stop on runlevel [016]

# Precise upstart does not support reload signal, and thus rejects the
# job. We'd rather start the daemon, instead of forcing users to
# reboot https://bugs.launchpad.net/ubuntu/+source/php5/+bug/1272788
#
#reload signal USR2

pre-start exec /usr/lib/php5/php5-fpm-checkconf
pre-start exec  /bin/bash /etc/init/php5-fpm.sh
post-start exec /bin/bash /etc/init/php5-fpm-onstart.sh

respawn
exec /usr/sbin/php5-fpm --nodaemonize --fpm-config /etc/php5/fpm/php-fpm.conf

所以添加了额外的脚本 pre-start和 post-start在 php5-fpm.conf . pre-start脚本是

#!/bin/bash
rm /var/run/php5-fpm.pid
rm /var/run/php5-fpm.sock
CHILD_PIDS_FILE="/var/run/php5-fpm-child.pid"
CHILD_PIDS=`ps -ef | grep 'php' | grep -v grep |awk '{print $2}'`
echo "$CHILD_PIDS" > "$CHILD_PIDS_FILE" 

该脚本基本上删除了 main process pid 和 sock文件。然后将子进程的 pid 写入文件，这样它们就可以在新的 php5-fpm 中被杀死。进程被创建。
post-start脚本是

#!/bin/bash
CHILD_PIDS_FILE="/var/run/php5-fpm-child.pid"
while read PID; do
   kill -9 $PID
done < $CHILD_PIDS_FILE
>$CHILD_PIDS_FILE

post-start脚本删除所有在 php5-fpm 之前运行的子 pid。重新启动。