Cron executing jobs at innacurate times (observing "jitter" of up to 1 hour)

Michael Ross gmx at ross.cx
Sat Nov 8 22:55:56 UTC 2014


On Sat, 08 Nov 2014 23:26:57 +0100, Darren Pilgrim  
<list_freebsd at bluerosetech.com> wrote:

> I recently upgraded a pair of servers from 9.3 to 10.0 via  
> freebsd-update.  After the upgrade, I noticed my logs were not rotating  
> on the hour, but at some random time within the hour following the  
> scheduled time (the logs in question were set * for size and $D0 or @T00  
> for when in newsyslog.conf).
>

I remembered reading this bug, which is close to what you describe:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=194236

In short: 10.0-RELEASE on i386 runs cron jobs at wrong time.

Reported solved in 10.1, so maybe try this.

Michael


> Some testing revealed that cron is no longer executing tasks in  
> /etc/crontab on the specified time.  For example, a job scheduled to run  
> hourly at the zeroth minute would actually execute sometime within an  
> hour, not on the hour as it used to.
>
> I created a simple test to see this behaviour.  I have six servers:
>
> - catnip (amd64 9.3-p3, freebsd-update only)
> - chombo (amd64 9.1-P6, source updates only)
> - pug (amd64 10.0-p10, freebsd-update only)
> - poodle (amd64 10.0-p10, freebsd-update only)
> - luigi (i386 10.0-p10, freebsd-update only, Xen VPS)
> - mario (i386 10.0-p10, freebsd-update only, Xen VPS)
>
> The behaviour test:
>
> All six have a job in /etc/crontab as follows:
>
> * * * * * root date >>/var/log/test/0000 2>&1
>
> That is, log the output of date every minute.
>
> For catnip, chombo, pug, and poodle, the logs from 13:50 to 14:05 PST  
> today look like this:
>
> Sat Nov  8 13:50:00 PST 2014
> Sat Nov  8 13:51:00 PST 2014
> Sat Nov  8 13:52:00 PST 2014
> Sat Nov  8 13:53:00 PST 2014
> Sat Nov  8 13:54:00 PST 2014
> Sat Nov  8 13:55:00 PST 2014
> Sat Nov  8 13:56:00 PST 2014
> Sat Nov  8 13:57:00 PST 2014
> Sat Nov  8 13:58:00 PST 2014
> Sat Nov  8 13:59:00 PST 2014
> Sat Nov  8 14:00:00 PST 2014
> Sat Nov  8 14:01:00 PST 2014
> Sat Nov  8 14:02:00 PST 2014
> Sat Nov  8 14:03:00 PST 2014
> Sat Nov  8 14:04:00 PST 2014
> Sat Nov  8 14:05:00 PST 2014
>
> This is luigi's:
>
> Sat Nov  8 21:51:00 UTC 2014
> Sat Nov  8 21:51:55 UTC 2014
> Sat Nov  8 21:53:47 UTC 2014
> Sat Nov  8 21:54:38 UTC 2014
> Sat Nov  8 21:54:59 UTC 2014
> Sat Nov  8 21:56:34 UTC 2014
> Sat Nov  8 21:57:34 UTC 2014
> Sat Nov  8 21:58:29 UTC 2014
> Sat Nov  8 22:00:38 UTC 2014
> Sat Nov  8 22:00:59 UTC 2014
> Sat Nov  8 22:02:47 UTC 2014
> Sat Nov  8 22:03:38 UTC 2014
> Sat Nov  8 22:04:34 UTC 2014
>
> And this is mario's:
>
> Sat Nov  8 21:50:34 UTC 2014
> Sat Nov  8 21:51:29 UTC 2014
> Sat Nov  8 21:53:34 UTC 2014
> Sat Nov  8 21:54:29 UTC 2014
> Sat Nov  8 21:56:34 UTC 2014
> Sat Nov  8 21:57:34 UTC 2014
> Sat Nov  8 21:58:25 UTC 2014
> Sat Nov  8 21:59:55 UTC 2014
> Sat Nov  8 22:01:51 UTC 2014
> Sat Nov  8 22:02:12 UTC 2014
> Sat Nov  8 22:03:12 UTC 2014
> Sat Nov  8 22:05:00 UTC 2014
>
> Backing off the frequency to every 5 minutes see the variance increase  
> to a couple of minutes.  For the production jobs running hourly,  
> variance is 0 to 59 minutes late.  This is particularly problematic for  
> things like newsyslog.  All machines have ntpd running and synchronized.  
>   All of the machines except luigi run cron with the defaults.  For  
> luigi, the following is in /etc/rc.conf:
>
> cron_dst="NO"
> cron_flags="-J 1 -o"
>
> My HV provider hasn't heard of anything causing this.  Before I go  
> reloading things back to 9.3, has anyone observed this?
> _______________________________________________
> freebsd-questions at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-questions
> To unsubscribe, send any mail to  
> "freebsd-questions-unsubscribe at freebsd.org"


More information about the freebsd-questions mailing list