Real max_execution_time for PHP on linux - php

Question

According to the documentation:
max_execution_time only affect the execution time of the script itself.
Any time spent on activity that happens outside the execution of the script
such as system calls using system(), stream operations, database queries, etc.
is not included when determining the maximum time that the script has been running.
This is not true on Windows where the measured time is real.
This is confirmed by testing:
Will not time out
<?php
set_time_limit(5);
$sql = mysqli_connect('localhost','root','root','mysql');
$query = "SELECT SLEEP(10) FROM mysql.user;";
$sql->query($query) or die($query.'<br />'.$sql->error);
echo "You got the page";
Will time out
<?php
set_time_limit(5);
while (true) {
// do nothing
}
echo "You got the page";
Our problem is that we really would like PHP to timeout, regardless of what it is doing, after a given amount of time (as we don't want to keep resources busy if we know we've failed delivering a page in an acceptable amount of time, like 10 seconds). We know we can play with settings such as the MySQL wait_timeout for the SQL queries, but the page timeout will depend on the number of queries that are executed.
Some people have tried to come up with workarounds and it doesn't seem implementable.
Q: Is there an easy way to get a real PHP max_execution_time on linux, or are we better timing out elsewhere, such as Apache level?

Answer 1

This is quite a tricky advice, but it will definitely do what you want, if you are willing to modify and recompile PHP.
Take a look at the PHP source code at https://github.com/php/php-src/blob/master/Zend/zend_execute_API.c (the file is Zend/zend_execute_API.c), at function zend_set_timeout. This is the function that implements time limit. Here's how it works on different platforms:
on Windows, create a new thread, start a timer on it, and when it finishes, set a global variable called timed_out to 1, the PHP execution core checks this variable for every instruction, then exits (very simplified)
on Cygwin, use itimer with ITIMER_REAL, which measures real time, including any sleep, wait, whatever, then raise a signal that will interrupt any processing and stop processing
on other unix systems, use itimer with ITIMER_PROF, which only measures CPU time spent by the current process (but both in user-mode and kernel-mode). This means waiting for other processes (like MySQL) doesn't count into this.
Now what you want to do is to change the itimer on your Linux from ITIMER_PROF to ITIMER_REAL, which of course you need to do manually, recompile, install etc. The other difference between these two is that they also use different signal when the timer runs out. So my suggestion is to change the ifdef:
# ifdef __CYGWIN__
into
# if 1
so that you set both ITIMER_REAL and the signal that PHP waits for to SIGALRM.
Anyway this whole idea is untested (I use it for some very specific system, where ITIMER_PROF is broken, and it seems to work), unsupported, etc. Use it at your own risk. It may work with PHP itself, but it could break other modules, in PHP and in Apache, if they for whatever reason, use the SIGALRM signal or other timer.

Answer 2

This is an old and answered question. But for the sake of helping others, I wanted to point out the request_terminate_timeout php-fpm option. If you're using PHP-FPM, it is most likely what you need.
If set, this option allows you to tell PHP-FPM to kill a request after N seconds, regardless of what PHP does.
See http://php.net/manual/en/install.fpm.configuration.php#request-terminate-timeout for details.

Answer 3

From httpd.conf:
Timeout: The number of seconds before receives and sends time out
Timeout 300