Fixed Crons stopped running

[digitalpoint]

Under 1.2.0, my crons would stop every couple days (mainly because of some Twitter API stuff I was doing and when the Twitter API failed, it would cause an error while running the cron).

This was fixed in 1.2.1, but I still ran into an issue where the cron tasks just stopped working again (only has happened once).

I did some checking around in the code, and I suspect it's the
XenForo_Model_Cron::getMinimumNextRunTime() method that can cause crons to stop working every once in a blue moon.

If the query it runs fails for whatever reason, cron will effectively stop running (maybe there's a locking issue and the query times out... maybe you are in the middle of restarting MySQL so it went away for a split second while that query ran, etc.)

If that query returns nothing, the system sets the next cron run time to January 19, 2038.

In theory that cron should *never* return nothing unless all crons are disabled.

It probably would be a better idea to fall back to a default next run time of something more reasonable... like 15 minutes in the future, rather than never checking again for crons to run until 2038.

 

[Mike]

The fallback is probably reasonable, but I think it's very unlikely that is the cause -- it's likely still a potential race condition that's wiping out the deferred process. I have already tweaked this again for 1.2.2 as I realized that there was a situation where a fatal error could still trigger a problem, particularly on a busy site. I think that fits you...

If it happens again, before you fix it, look at your xf_deferred table. If you don't see a "Cron" entry, then you got hit by the race condition.

 

[digitalpoint]

Not sure what the race condition that was tweaked in 1.2.2 was, but would it apply to a setup that the deferred.php only runs once a minute? We removed the system that triggers it from page views and instead trigger deferred.php every minute from a single server via crontab.

* * * * *       root    /usr/local/bin/php /home/sites/digitalpoint.com/web/deferred.php > /dev/null 2>&1

 

[digitalpoint]

Haven't upgraded to 1.2.2 yet, but happened again today. The xf_deferred table has no records in it... so I hopefully it's whatever was fixed in 1.2.2. There also wasn't a MailQueue entry in there until we disabled/reenabled a random cron.