© Tony Lawrence, aplawrence.com
Cron, Batch and At
These three commands are used to run commands at some other time. They differ in their usage, their environment, and their default actions, so are sometimes a source of confusion. If you are having trouble with cron, you might want to read Cron is not working first.
If you don't know how to create the script that cron or at will call, see New To Unix
The "batch" command really just calls "at" with special flags set:
at -q b -m now
but "cron" is truly a totally separate command.
One of the common problems posted to the Unix newsgroups goes something like:
I have a command script. If I run it from the command line or with "at", it works, but if I run it with "cron" it fails. Why?
Not only is this a common question, but amazingly enough, it usually generates three or four wrong answers every time it appears. The correct answer is that it fails because "cron" runs with a different environment than what you have. You have a certain PATH, you have other environment variables set, and "at" deliberately notices all that and makes sure that when your command runs, all those things will be in place. The "cron" utility does not: it has its own environment, probably very different from yours.
Note: this article covers both SCO and Linux cron. I keep the SCO stuff here for those unfortunate folks who haven't been able to move to Linux yet. There's lots here to help you do that: SCO/Linux Transition Guide is a start.
Very often, it's just the PATH that is different. For example, root's environment usually has
but cron (in SCO OSR5, see below for Linux which has a much smarter cron system) sets its environment to
Note: the man page for SCO crontab says that the PATH will be "/bin:/usr/bin:" but the above PATH is what actually happens on my 5.0.5 system. The difference is especially significant because of the placement of the lone ":", which adds "." to cron's PATH. According to the manual, the "." would end up at end of the path, and thus would be the last place searched. It actually ends up at the beginning. As the first command found is the command executed, this can cause unexpected results. Cron does cd to the home directory of the user whose cron job is being run, so "." will always refer to that directory- unless the command script itself changes directories and then issues another command.
The other likely cause for failure is a missing environment variable. Cron only has:
HOME=/ (or the home directory of the user whose job is running) HZ=100 IFS= <TAB><LF> (not actually set to this; edited for viewing) LOGNAME=root (or whomever's job is running) MAILCHECK=600 OPTIND=1 PATH=:/bin:/usr/bin:/usr/lbin SHELL=/bin/sh TZ=EST5EDT
If your command script requires anything variables not in cron's normal environment and not otherwise set (in the script or in a Linux crontab file), it will fail. If your script required Korn shell features, it would also fail (if cron notices that you are not running /bin/sh when you change a crontab, it will remind you that it plans to use /bin/sh. On Linux, it uses Bash).
A simple way to fix the environment issue is to use "at" to set up your environment. You can, for example, type:
at now + 1 hour <ENTER> fakecommand <ENTER> <CTRL-D>
"at" will spit back a job number. Change directories to /usr/spool/cron/atjobs and you'll see that same number listed. Notice that it is owned by root with group of cron, and that the permissions are: ---Sr-S---. Copy that file somewhere else, then rm the file in the spool area, and take a look at the copy you made. It will look something like this:
: at job export _; _='/usr/bin/at' export HZ; HZ='100' export PATH; PATH=':/bin:/etc:/usr/bin:/tcb/bin:/usr/local/bin:/usr/local/etc' export HUSHLOGIN; HUSHLOGIN='FALSE' export LOGNAME; LOGNAME='root' export MAIL; MAIL='/usr/spool/mail/root' export SHELL; SHELL='/bin/sh' export HOME; HOME='/' export TERM; TERM='scoansi' export PWD; PWD='/tmp' export TZ; TZ='EST5EDT' export ENV; ENV='/.kshrc' : # @(#) proto 23.2 91/08/29 # # Copyright (C) 1988-1991 The Santa Cruz Operation, Inc. # All Rights Reserved. # The information in this file is provided for the exclusive use of # the licensees of The Santa Cruz Operation, Inc. Such users have the # right to use, modify, and incorporate this code into other products # for purposes authorized by the license agreement provided they include # this notice and the associated copyright notice with any such product. # The information in this file is provided "AS IS" without warranty. # #ident "@(#)adm:.proto 1.2" cd /tmp ulimit 4194303 umask 22 fakecommand
Notice what great pains "at" has gone to to match your environment. Not only do you have all your environment variables, but your ulimit and umask have been matched and it has even cd'd you to the directory you were in when you issued the command. So now use that script from crontab rather than "fakecommand" directly. That is, if you copied the at script to /usr/local/bin/runfake, it is /usr/local/bin/runfake that you would invoke from cron, not "fakecommand".
Recently someone asked me why their crontab wasn't working. They understood that they needed to set their environment, but what they did was something like this:
17 5 * * * ./setmyenv.cmd;domystuff.cmd
That'll never work, and wouldn't work from the shell either. You would need to add a ". ./setmyenv.cmd" inside "domystuff" (that's dot space dot slash setnyenv.cmd).
You probably know that you do not edit the crontab files directly (on Linux that's fine, but not on most Unixes). Some people use "crontab -e", but a more safe procedure is:
crontab -l > /tmp/mycrontab vi /tmp/mycrontab (make your changes) crontab /tmp/mycrontab
Or, if for another user:
crontab -u john -l > /tmp/mycrontab vi /tmp/mycrontab (make your changes) crontab -u john /tmp/mycrontab
Two reasons that is safer: first, if your EDITOR variable is accidentally set to something you don't know (as it may very well be on an unfamiliar system), you may accidentally wipe out the crontab getting out of it. Another reason for avoiding -e is the proximity of e and r on your keyboard - if you type -r, that crontab is gone instantly.
By the way, some old systems wouldn't work if you did (for example) "export EDITOR=vi" rather than "export EDITOR=/usr/bin/vi". This was simply a matter of whether they used execlp instead of execl (execlp searches $PATH).
Be careful with crontabs set to other users. Remember that cron cd's to the user's directory when it starts up your job. If the user's directory doesn't exist cron fails and sends mail to that user. That doesn't sound too horrible, does it? Well, on the older 3.2v4.2 release, there was some bug somewhere that sometimes caused /usr/sys to be removed. The "sys" user still existed, and still had its crontab file, but when cron tried to run, it couldn't cd to the non-existent home directory. This caused it to start sending mail to "sys" complaining. That wouldn't have been too bad, but who reads "the "sys" user's mail? Usually nobody, so the mail file would build up larger and larger. Eventually it would start to affect the performance of mmdf, and mmdf would get backed up- it couldn't clear out its own spool directories quite as fast as they were growing, which meant that the size of the directories in /usr/spool/mmdf started growing. The larger a directory is, the more time it takes to search it, so this made mmdf run even more slowly, which caused it to get more behind, which, of course, caused the directories to grow larger yet... and to add insult to injury, mmdf would itself start generating messages about mail it couldn't deliver (now that's dumb!), and those only added to the problem. Eventually this would get bad enough to affect performance, because mmdf was spending every spare cpu cycle available trying to deliver mail. So sar would show 0 idle time, the disk would be thrashing as mmdf tried to catch up, and performanced nose dived. What a mess, and all because of a missing directory.
The "at" and "batch" programs are much less complex. Batch takes no arguments; it just runs your program. The details of these are covered in the man pages.
Note for some SCO 5.0.6 users
Some versions of 5.0.6 cron had a bug where a cron job could be run twice. The fix is to get the proper version of cron or (if that's not possible) be sure your script can handle that possibility (set some flag that says it has run or is running and exit if the other instance sees that).
(OLDER) <- More Stuff -> (NEWER) (NEWEST)
Printer Friendly Version