2017-12-24 by Void Linux Contributors

The Advent of Void: Day 24: snooze

cron(8) is a nice tool, but it has some long-standing problems, among them:

A little, but flexible alternative is to use snooze(1), which essentially just waits for a particular time, and then executes a command. To get recurring jobs ala cron, we can use this together with our runit service supervision suite. If we wanted at(1) instead, we can just run snooze once.

The time for snooze is given using the options -d (for day), -m (for month), -w (for weekday), -D (for day of year), -W (for ISO week), and -H (for hour), -M (for minute), -S (for second). Each option of these can be comma-separated list of values, ranges (with -) or repetitions (with /). The default is daily at midnight, so if we wanted to run at the next full hour instead, we could run:

% snooze -n -H'*'
2017-12-24T17:00:00+0100 Sun  0d  0h 47m 33s
2017-12-24T18:00:00+0100 Sun  0d  1h 47m 33s
2017-12-24T19:00:00+0100 Sun  0d  2h 47m 33s
2017-12-24T20:00:00+0100 Sun  0d  3h 47m 33s
2017-12-24T21:00:00+0100 Sun  0d  4h 47m 33s

The -n option disables the actual execution and shows the next five matching times instead.

To run every 15 minutes, we’d use

% snooze -n -H'*' -M/15  
2017-12-24T16:15:00+0100 Sun  0d  0h  1m 31s
2017-12-24T16:30:00+0100 Sun  0d  0h 16m 31s
2017-12-24T16:45:00+0100 Sun  0d  0h 31m 31s
2017-12-24T17:00:00+0100 Sun  0d  0h 46m 31s
2017-12-24T17:15:00+0100 Sun  0d  1h  1m 31s

More complicated things are possible, for example next Friday the 13th:

% snooze -n -w5 -d13
2018-04-13T00:00:00+0200 Fri 108d  6h 45m 33s
2018-07-13T00:00:00+0200 Fri 199d  6h 45m 33s
no satisfying date found within a year.

Note that snooze bails out if it takes more than a year for the event to happen.

By default, snooze will just terminate successfully, but we can give it a command to run instead:

% snooze -H'*' -M'*' -S30 date 
Sun Dec 24 16:27:30 CET 2017

When snooze receives a SIGALRM, it immediately runs the command.

snooze is quite robust, it checks every 5 minutes the time has progressed as expected, so if you change the system time (or the timezone changes), it is noticed.

For additional robustness, you can use the timefiles option -t, which ensures a job is not started if its earlier than some modification time of a file. On success, your job can then touch this file to unlock the next iteration. Together with the slack option this can be used for anacron-style invocations that ensure a task is run, for example, every day at some point.