php sleep function odd behaviour - php

Question

We've inherited a platform that has a crobjob that every minute curls a local php script three times with different parameters (curl -s -o --url https://localhost/myscript.php?option=XYZ -k). That script runs for about 1 minute and its possible multiple instances with the same option overlap for a bit of time. The script logs in a different file per option given and each log starts with the timestamp when the script started so it acts as an instance identifier.
The script has this skeleton:
<?php
$option=XYZ;
$scriptId = time();
$file = "log_$option.txt";
file_put_contents($file,"\n$scriptId: Start\n",FILE_APPEND);
session_start();
$expires = time()+60;
file_put_contents($file,"\n$scriptId: Expires at $expires\n",FILE_APPEND);
while(time()<$expires){
file_put_contents($file,"\n$scriptId: Not expired at ".time()."\n",FILE_APPEND);
switch($option){
case X:
do_db_stuff();
break;
...
}
file_put_contents($file,"\n$scriptId: Will sleep at ".time()."\n",FILE_APPEND);
sleep(13);
file_put_contents($file,"\n$scriptId: Woke up at ".time()."\n",FILE_APPEND);
}
file_put_contents($file,"\n$scriptId: Finished at ".time()."\n",FILE_APPEND);
Normally this script runs fine (even if they overlap when instance A is sleeping for the last time and instance B starts) but on occasion we have two issues that we can confirm with the logs:
sometimes it sleeps for less than 13 seconds (a
variable amount of time and always less than 13);
sometimes the script stops (no more logging after the "Will sleep" one, and we can verify that no db stuff is being done). [Update on this in Edit 2]
We've looked into the possible causes but couldn't find any:
php max_execution_time is set to 240 seconds and the script never
takes more than one and a half minutes;
sleep documentation says it is per session but curl isn't using cookies so it should be different sessions in every instance (and also if it was using the same it would always block since we always execute three script instances, which it doesn't);
the hosting tech team says there are no errors neither in the server
error log nor in php error log in the timestamp where these issues
happen.
I can't reproduce the issues at will, but they happen at least once a day.
What I'd like to know is what can be interfering with the sleep behaviour? How can I detect or fix it?
Additional information:
linux system
mysql 5.5
apache
php 5.3
php max_execution_time set to 240
Edit 1: Just to clarify: actually we have 3 options, so it writes to 3 log files, one for each option. At any given time there can be up to two instances per option running (each instance of the same option overlaps a small amount of time).
Edit2: As per #Jan suggestion, I added log to the sleep function result. The script stopped once with that log already:
[2016-01-05, 13:11:01] Will sleep at 2016-01-05, 13:11:29
[2016-01-05, 13:11:01] Woke up at 2016-01-05, 13:11:37 with sleep return 5
[2016-01-05, 13:11:01] Not expired at 2016-01-05, 13:11:37
[2016-01-05, 13:11:01] Will sleep at 2016-01-05, 13:11:37
[2016-01-05, 13:11:01] Woke up at 2016-01-05, 13:11:38 with sleep return 13
... no more log from instance [2016-01-05, 13:11:01] ...
[2016-01-05, 13:12:01] Start
According to the sleep documentation:
If the call was interrupted by a signal, sleep() returns a non-zero value. On Windows, this value will always be 192 (the value of the WAIT_IO_COMPLETION constant within the Windows API). On other platforms, the return value will be the number of seconds left to sleep.
So according to the documentation and the log it seems that the sleep is being cut short due to an interrupt.
How can I know what interrupt caused this (pcntl_signal?), where did it come from and is there any way to avoid it?
Edit3: I've added code to handle signals with pcntl_signal (trying to register from signal 1 till 255) and log them, the issue still happens but the log is empty still.

Answer 1

You can define signal handlers with pcntl_signal.
With those handlers you can log when a interruption happens. But AFAIK you can't detect where it comes from.
Also you can use pcntl_alarm for your delayed jobs.
Check PHP Manual - PCNTL Alarm

Answer 2

Given your stack, the interrupt signal could come from apache. Depending on how Apache is communicating with PHP on your stack, there are several timeout options in Apache configuration that could infer with the script execution.
If your cron calls a curl on localhost, maybe it could directly call the PHP file instead, and so avoid using apache processes to keep the request alive? You'll have to edit your dedicated CLI php.ini file with the max_execution_time value.