Monitor php-fpm max processes count script - php

Question

We have an issue in which on production server, some bug in our system locks/hangs a php-fpm process and is not being released, this causes over a period of 10-15 minutes to more processes to lock (probably trying to access a shared resource which is not released) and after a while the server cannot serve any new users as no free php-fpm processes are available.
Parallel to trying and find what is creating that dead-lock, we were thinking of creating a simple cron job , which runs every 1-2 minutes and if it sees max processes above X it will either kill all php-fpm processes or restart the php-fpm .
What do you think of that simple temporary fix for the problem ?
Simple php script ,
$processCount = shell_exec("ps aux|grep php-fpm|grep USERNAME -c");
$killAll = $processCount >=60;
if($killAll){
echo "killing all processes";
try{
shell_exec("kill -9 $(lsof -t -i:9056)");
}catch(Exception $e1){
}
shell_exec("sudo service php56u-php-fpm restart");
$processCount = shell_exec("ps aux|grep php-fpm|grep USERNAME -c"); //check how much now
}

Answer 1

Killing all php processes doesn't seem like a good solution to your problem. It would also kill legitimate processes and return errors to visitors, and generally just hide the problem deeper. You may also introduce data inconsistencies, corrupt files and other problems killing processes indiscriminately.
Maybe it would be better to set some timeout, so the process would be killed if it takes too long to execute.
You could add something like this to php-fpm pool config:
request_terminate_timeout = 3m
and/or max_execution_time in php.ini
You can also enable logging in php-fpm config:
slowlog = /var/log/phpslow.log
request_slowlog_timeout = 2m
This will log slow requests and may help you find the cultprit of your issue.

Answer 2

it's not a good solution to kill PHP processes. in your PHP-fpm config file (/etc/php5/pool.d/www.conf)
set pm.max_requests=100 so after 100 requests the process will close and another process will start for the rest of requests.
also maybe there's a problem with your code, please make sure the request is ending.
So if the problem with your script try request_terminate_timeout=2m
The timeout for serving a single request after which the worker process will
; be killed. This option should be used when the 'max_execution_time' ini option
; does not stop script execution for some reason. A value of '0' means 'off'.
; Available units: s(econds)(default), m(inutes), h(ours), or d(ays)
; Default Value: 0
;request_terminate_timeout = 0
Please note that if you are doing some long polling, this may affect your code.