How's your Greek? Did you know that Χρόνος (chronos) means 'time'? Sure you did! And in Greek mythology, Chronos was the god of the ages, the personification of time. Hence we have words like chronometer and chronology. Given that programmers have never been famous for being that great at spelling, we also have Cron – a Linux service that arranges for actions to take place at specific times.
In this tutorial you'll learn how to configure Cron to schedule your own jobs, and how to make sense of the Cron configuration that comes with your Linux distribution. Cron is great for machines that are left running continuously, but in the second half of the tutorial we'll look at its younger sister, Anacron, which may be more appropriate for personal computers that spend a lot of their time switched off.
Some uses for Cron
- Rotating log files (for more details about this, see the Log File Lumberjack feature on page 56 of LXF92, May 2007).
- Summarising log files using tools such as Logwatch and Webalizer.
- Automatically checking distro package repositories for updates.
- Rebuilding the slocate database. (This is a database of all the filenames in the system, used to support the slocate command.)
- Writing backups of the filesystem (just remember to load a tape into the machine before you go home!).
- Weeding ancient files from directories such as /tmp.
- Running intrusion detection scans using tools such as Tripwire.
- Launching overnight software rebuilds.
Cron has been around a long time (though not quite so long as Chronos). Even my bedraggled copy of the manual for Edition Six Unix, dated 1974, has an entry for it. It has been rewritten several times since then, and the version found in Linux distributions today is usually known as Vixie Cron, written by Paul Vixie.
Simply, Cron is a daemon that arranges for commands to be run at specific times of day, and/or specific days of the week. It works by waking up once every minute and consulting its configuration files – known as crontabs – to see if anything is scheduled to run that minute. If so, it executes the appropriate shell command. Then it goes back to sleep for another minute. Cron itself is normally started at boot time.
We use Cron to schedule a variety of activities; some examples are listed in the Some Uses For Cron box, below. Most Linux distributions usually provide Cron preconfigured to perform some of these, but it is not hard to add your own, once you know how to drive it.
Indeed, people get quite inventive about using Cron. A colleague of mine once wrote a script that gathered 'busyness' statistics from the machine. Using Cron, he ran the script on every machine on the network at 0, 15, 30 and 45 minutes past each hour and copied the results to a central machine.
Another script, run by Cron at 5, 20, 35 and 50 minutes past each hour on the central machine, massaged the data further to produce HTML-formatted reports that the managers and engineers could look at. He also created bar charts showing how the load varied throughout the day.
Another colleague used Cron to download a weather satellite image from the Met Office every hour; he did this for two years then stitched the collection of images together into a movie.
I know a manager who used Cron to turn off execute permission on all the games at 9:00 each morning and turn it back on again each evening, and a parent who used it to turn off internet access (with iptables) late each evening to prevent his kids staying up all night surfing, then turned it back on in the morning. All very god-like.
Cron is told what to do – and when to do it – by a series of configuration files called crontabs. First, there's /etc/crontab, sometimes called the system crontab. The anatomy of a single entry in this file is shown in the diagram opposite. In each of the five time fields you can put a:
- *, meaning 'any value'.
- Single value.
- Comma-separated list of values such as 15,25,40.
- Range such as 9-18.
- Period, usually used following *. For example, */5 in the minutes field means "every five minutes".
Here are a couple of examples:
*/10 * * 5,7,11 * root command1
0 9-18 * * 1-5 root command2
These lines will run command1 every ten minutes during May, July and November, and command2 once an hour during the working day (09:00 to 18:00) Monday to Friday. You can see that there's a lot of flexibility here! By the way, the sixth field in the crontab, which is root in our examples, specifies the user identity under which the script will run.
You nearly always see root here, though a security-savvy administrator might apply the principle of least privilege and choose a less privileged identity unless the activity really does need root permission.
The anatomy of a crontab entry
Although it's perfectly permissible to run your commands directly from the system crontab in this way, the current fashion among Linux distros is to interpose an intermediate script, usually called run-parts. For example, here's the system crontab from Red Hat and Fedora:
01 * * * * root run-parts /etc/cron.hourly
02 4 * * * root run-parts /etc/cron.daily
22 4 * * 0 root run-parts /etc/cron.weekly
42 4 1 * * root run-parts /etc/cron.monthly
The run-parts script is simple enough: it just runs all the executables in the specified directory. There are, as you see, four such directories, corresponding to four frequencies of execution.
This scheme makes life more convenient, in the sense that you only have to drop your new script into the appropriate directory (depending on how often you want it to run) and Cron will automatically take care of running it, but you lose the flexibility to run your daily jobs at different times of the day, and everything runs as root.
And of course this scheme can't handle operations such as the "disable the games in the morning and re-enable them in the evening" example mentioned earlier.
A glance at /etc/crontab in OpenSUSE 10.2 suggests that its authors have almost entirely abandoned the system crontab. There's just one line in /etc/crontab that runs the script /usr/lib/cron/run-crons every 15 minutes. This script, which is undocumented and largely inscrutable, apparently takes over running the scripts in the cron.hourly, cron.daily, cron.weekly and cron.monthly directories, though the logic it applies to decide what to run and when is not obvious.
Cron for mortals
In addition to the system-wide /etc/crontab, Vixie Cron lets each user have their own crontab. These are stored in files named after the user's login under /var/spool/cron/crontab (the exact location varies between distros). These files have the same format as the system-wide Cron with one exception – there is no user ID column, because individual users are only allowed to run Cron jobs as themselves.
You are not allowed to edit these files directly; you need to run crontab -e, which will open the file for you and drop you into the editor of your choice (actually, the editor specified by the EDITOR environment variable) to edit it. Why the crontab -e wrapper?
Well, for one thing, it checks the syntax of your new entries before installing them, and for another, it jiggles the timestamp on the spool directory to tell Cron that one of the Cron files has been changed. (Unusually for a Linux daemon, there is no need to signal Cron after you've edited the config file, it will automatically notice the next time it wakes up.)
Administrators of multi-user systems can control who is allowed to use cron by placing user names in one of two files, /etc/cron.allow and /etc/cron.deny. The Can Fred Use Cron example in the box overpage shows the logic used to make this decision. This facility was probably more useful back in the days of true multi-user machines with many accounts; it is less relevant on machines that support only one or two users.
Can Fred use cron?
A cloudy project
One aspect of Cron I've ignored up to now is the environment that Cron jobs run in. A regular user's Cron jobs run with their current directory set to that user's home directory, and as we've already noted, they run with that user's identity. The number of environment variables defined will be minimal.
Cron does not read a user's login scripts to establish user-specific settings, and in particular the search path contains only /usr/bin and /bin, so you need to take care when running commands from your crontab – in fact, it's common practice to refer to commands by their full path name to be sure they will be found.
It's also common practice to redirect the standard output and standard error of a Cron job to a file, so you'll often see entries in a crontab along these lines:
15 6 * * * /usr/bin/someprog > /tmp/someprog.out 2>&1
If a Cron job generates output that is not redirected, that output will be mailed to the job's owner.
You can define your own environment variables in your crontab if you want. For example, you could specify which shell will be used to interpret the command, and extend your search path with lines like this:
If, like me, you learn by experimentation, you will probably want to try out a few Cron entries of your own at this stage, but if you need some suggestions, try an experiment something like this:
The first crontab entry in our experiment simply confirms the environment that a crontab job runs with. The second entry shows that if a crontab job generates output that is not redirected to a file, that output is mailed to the user. And the third entry demonstrates how to collect an infrared satellite image every hour.
Stitching the image sequence together into a movie is left as an exercise to the reader, as they say in all the best text books.
Experimenting with Anacron
Cron is great, but it has one weakness: it is designed for machines that are left running continuously. If your machine is turned off when a job falls due, it simply won't be run. So, if in the interests of reducing your carbon footprint you are careful to turn your computer off at night, those daily Cron jobs (which run at 04:02 on Red Hat/Fedora and 06:25 on Ubuntu) will never happen.
To handle this situation better, a new tool called Anacron was written. Returning to our language lesson for a moment, 'anachronistic' means 'chronologically misplaced'. For example, a documentary about Samuel Pepys that showed him writing his diary on a laptop might raise a few eyebrows.
Anyway... Anacron's mission is to guarantee that those daily, weekly and monthly activities do actually occur, assuming only that the computer was switched on at some point during the day (or week, or month). Anacron is not concerned with ensuring that activities occur at some precise time.
Anacron has a rather simple configuration file, /etc/anacrontab, whose syntax is shown in the diagram below. You'll see from this figure that the shortest period of time that Anacron concerns itself with is a day; consequently, Anacron is not a replacement for Cron.
You cannot tell Anacron to run commands at a specific hour or minute, so, to return to one of my earlier Cron examples, you couldn't use Anacron to turn off internet access in the evening, and restore it the following morning. To keep track of what has been run, and when, Anacron writes timestamps to files in /var/spool/anacron.
The files are named after the job ID in the anacrontab file. For the example shown in the figure, the timestamp file would be /var/spool/anacron/cron.weekly. The timestamp records only the day, month and year that the job was last run. It does not record hours, minutes or seconds.
Now, Anacron is not a daemon in the usual sense of the word. That is, it does not start at boot time and lurk around waiting for something to do. Each time it is started, Anacron figures out what, if anything, needs to be run, waits for the specified delay to run it, then exits.
So it is appropriate to ask: "When does Anacron actually get run?" A close look at the config files in Red Hat Enterprise Linux (also Fedora) reveals that Anacron is run at boot time, via the script /etc/init.d/anacron. At this time, Anacron reads the anacrontab file, checks the timestamp files in /var/spool/anacron to see when each of the jobs was last run, and figures out what needs to get run.
For each of the tasks specified in anacrontab, Anacron waits for the specified delay, then runs the command. The delays are there to stop Anacron trying to do everything at once as soon as the machine has booted, at which time the user is, presumably, looking forward to getting some work done and would prefer not to have their machine fully occupied performing menial updates or, as Douglas Adams might have it, figuring out how to make a nice cup of tea.
Taken by itself, Anacron is not a difficult beast to understand. However, it gets more interesting when we try to figure out the interaction between Anacron and Cron, especially if the same activities are being managed by both. In the Red Hat/Fedora configuration, anacrontab looks like this:
1 65 cron.daily run-parts /etc/cron.daily
7 70 cron.weekly run-parts /etc/cron.weekly
30 75 cron.monthly run-parts /etc/cron.monthly
From this, we see that Anacron is being asked to run precisely the same things as Cron; that is, at the appropriate interval (1, 7, or 30 days) it uses the run-parts script to run all the scripts in /etc/cron.daily, /etc/cron.weekly or /etc/cron.monthly.
You'll notice that Anacron does not run the cron.hourly scripts. As I've said, it really isn't suitable for that, because it only records the timestamps of the most recently executed tasks to the nearest day.
Anatomy of anacrontab
Don't tread on my toes
When the same activities are being managed by both Cron and Anacron, there is a danger that Cron will run a job that Anacron has already run, or vice versa. Different distros use different tricks to prevent this.
If you examine the Red Hat/Fedora configuration closely, you'll discover that the cron.daily directory includes a script called 0anacron that runs Anacron, but with the -u option. So, we see that Anacron is run by Cron every day. However, the -u option tells Anacron merely to update its timestamps and not to actually run the commands.
Essentially, Cron is saying to Anacron "please note that I have run these commands today (or this week, or this month ...) so you don't need to run them".
Ubuntu uses a different trick. Its system crontab looks something like this:
17 * * * * root run-parts /etc/cron.hourly
25 6 * * * root test -x /usr/sbin/anacron || run-parts /etc/cron.daily
47 6 * * 7 root test -x /usr/sbin/anacron || run-parts /etc/cron.weekly
52 6 1 * * root test -x /usr/sbin/anacron || run-parts /etc/cron.monthly
The hourly entry is straightforward; it looks more or less like Fedora's. The daily, weekly and monthly entries employ a cunning shell programming trick to ensure that run-parts is only run if Anacron is not present. (See The Magic Run-parts Trick, below, for an explanation of how this works.) We can summarise the difference between the Red Hat and Ubuntu approaches by saying, "In Red Hat, Cron wins. In Ubuntu, Anacron wins."
The magic run-parts script
The shell's logical OR operator, ||, combines the true or false values of its two operands to yield a true/false result. So for example, a || b is true, if a is true or b is true. (That is to say, if either the command a or the command b returns zero, exit status.)
Now – bear with me here, there is a point to this! – if a returns true, the value of a || b will be true regardless of the value of b. Yes? And in fact, in this case the command b won't even get run. This seemingly trivial aspect of the shell's behaviour is actually the whole point of what we see in Ubuntu's crontab. The shell command
test -x /usr/sbin/anacron || run-parts /etc/cron.daily
will run two commands, test and (maybe) run-parts. The test command is actually built into the shell; test -x returns true if the specified file exists and is executable. So, what this line says is, "if /usr/sbin/anacron exists and is executable, then this entire expression is 'true' and there is no need to run run-parts to determine this", or to put it another way, "only run run-parts if Anacron is not present". Quod erat demonstrandum, as the Romans said.
The default configurations of Cron and Anacron provided by the various Linux distributions can be a little confusing, and there is, I suppose, a message to be delivered on top of all of this: you do not have to use Cron and Anacron in precisely the same way that your Linux distributor has opted to use them. If you understand the operation of these two services, and the syntax of their configuration files, you can make them do whatever you like.
One-off tasks with at
We have just enough space to squeeze in a mention of the at command. Unlike Cron and Anacron, which schedule regularly repeated activities, at is used to schedule one-off activities in the future. Suppose you would like your computer to rebuild your Linux kernel this evening while you're at the pub. You might proceed like this:
$ at 20.30
warning: commands will be executed using /bin/sh
at> cd /usr/src/linux
at> make all
job 2 at Wed Feb 14 19:30:00 2007
As you'll see from this, at collects the commands you want to run from its standard input, up to an EOT character (^D by default). Your list of queued jobs can be examined using atq:
2 Wed Feb 14 19:30:00 2007 a chris
The at daemon (atd) will start the job at the specified time, leaving you free to do something more interesting. Notice, however, that at will not actually arrange to get you to the pub at the appointed hour, nor will it offer to buy a round of drinks. You might, however, show it a little consideration and buy it a packet of crisps.
The details of the jobs you queue are placed in a spool directory (typically /var/spool/at). If you queue a job or two then examine the contents of this directory, you'll notice that, unlike Cron, at goes to some trouble to establish the same environment for the job that you'd have if you had just logged in. The daemon atd is responsible for actually running the jobs.
If you change your mind about an at job you previously submitted, you can delete a job from the queue using atrm:
$ atrm 2
The argument to atrm is the job number (2 in this example).
The thing about at that fascinates me most is the wide range of notations it understands for specifying the time at which the command is to be run. The following are all valid:
$ at 14.30 tomorrow
$ at noon Friday
$ at 4pm + 3 days
My favourite is the use of the keyword teatime as a synonym for 4pm:
$ at teatime tomorrow
You should follow us on Identi.ca or Twitter