"Linux Gazette...making Linux just a little more fun! "


SLEW: Space Low Early Warning

By James T. Dennis jim@starshine.org


One of the worst things you can do to your linux or other Unix-like system is to allow any of the filesystems to get full.

System performance and stability will suffer noticeable degradation when you pass about 95% and programs will begin failing and dying at %100 percent. Processes that are run as 'root' (like sendmail and syslog) will actually fill their filesystem past 100% since the kernel will allocate some of the reserved space for them.

(Yes, you read that right -- when you format a file system a bit of space is reserved for root's exclusive use -- read the mke2fs, e2fsck, tune2fs for more on that).

Considering the importance of this issue you might think that our sophisticated distributions would come with a pre-configured way to warn you long before there was a real problem.

Sadly this is one of those things that is "too easy" to bother with. Any professional Unix developer, system administrator or consultant would estimate a total time for writing, installing and testing such an application at about 15 minutes (I took my time and spent an hour on it).

Here's the script:


	#! /bin/bash
		## SLEW: Space Low Early Warning
		## 	by James T. Dennis, 
		##	Starshine Technical Services
		##	
		## Warns if any filesystem in df's output
		## is over a certain percentage full --
		## mails a short report -- listing just 
		## "full" filesystem.
		## Additions can be made to specify
		## *which* host is affected for 
		## admins that manage multiple hosts

	adminmail="root"
		## who to mail the report to


	threshold=${1:?"Specify a numeric argument"}
		## a percentage -- *just the digits*

	# first catch the output in a variable
	fsstat=`/bin/df`

	echo "$fsstat" \
		| gawk '$5 + 0 > '$threshold' {exit 1}'  \
	   || echo "$fsstat" \
		| { echo -e "\n\n\t Warning: some of your" \
			"filesystems are almost full \n\n" ;
			gawk  '$5 + 0 > '${threshold}' + 0 { print $NF, $5, $4 }' } \
		| /bin/mail -s "SLEW Alert" $adminmail

	exit


That's twelve lines of code and a mess of comments (counting each of the "continued" lines as a separate line).

Here's my crontab entry:

	## jtd: antares /etc/crontab
	## SLEW: Space Low Early Warning
	## 	Warn me of any filesystems that fill past 90%
	30 1 * * * nobody /root/bin/slew 90

Note that the only parameter is a 1 to 3 digit percentage. slew will silently fail (ignore without error) any parameter(s) that don't "look like" numbers to gawk.

Here's some typical output from the 'df' command:

Filesystem         1024-blocks  Used Available Capacity Mounted on
/dev/sda5              96136   25684    65488     28%   /
/dev/sda6              32695      93    30914      0%   /tmp
/dev/sda10            485747  353709   106951     77%   /usr
/dev/sda7              48563   39150     6905     85%   /var
/dev/sda8              96152    1886    89301      2%   /var/log
/dev/sda9             254736  218333    23246     90%   /var/spool
/dev/sdb5             510443  229519   254557     47%   /usr/local

Note that I will be getting a message from slew tomorrow if my news expire doesn't clean off some space on /var/spool. The rest of my filesystems are fine.

Obviously you can set the cron job to run more or less often and at any time you like -- this script takes almost no time memory or resources.

The message generated can be easily modified -- just add more "continuation" lines after the 'echo -e' command like:


	   || echo "$fsstat" \
		| { echo -e "\n\n\t Warning: some of your" \
			"filesystems are almost full \n\n" \
			"You should add your custom text here.\n"\
			"Don't forget to move the ';' semicolon "\
			"and don't put whitespace\n" \
			"after the backslash at the ends of these lines\n\n";

Note how the first echo feeds into the grouping (enclosed by the braces) so that the contents of $fsstat are appended after the message. This is a trick that might not work under other shells.

Also, if you plan on writing similar shell scripts, note that the double quotes around the variable names (like "$fsstat") preserve the linefeeds in their values. If you leave the quotes out your text will look ugly.

The obvious limitation to this script is that you can only specify one threshold value for all of the file systems. While it would be possible (and probably quite easy) to do this some other way -- it doesn't matter to 90% of us. I suspect that almost anyone who does install this script will set the threshold to 85, 90 or 95 and forget about it.

One could also extend this script to do some groping (using various complex find commands) to list things like:

I chose to keep this script very simple and will develope specific scripts to suggest file migrations and deletions as needed.

As you can see it is possible to do quite a bit in Linux/Unix using high level tools and very terse commands. Certainly the hardest part of writing a script like this is figuring out minor details about quoting and syntax (when to enclose blocks in braces or parentheses) and in determining how to massage the text that's flowing through your pipes.

The first time I wrote slew was while standing in a bookstore a couple of years ago. A woman near me was perusing Unix books in my favorite section of the store and I asked if I could help her find something in particular. She described the problem as it was presented to her in a job interview. I suggested a 'df | grep && df | mail' type of approach and later, at home, fleshed it in and got it working.

Over the years I lost the original (which was a one-liner) and eventually had one of the systems I was working with hiccup. That made me re-write this. I've left it on all of my systems ever since.

I'd like to encourage anyone who developes or maintains a distribution (Linux, FreeBSD, or whatever) to add this or something like it to the default configuration on your systems. Naturally it is free for any use (it's too trivial to copyright in my personal opinion; so, that there is no doubt, I hereby place SLEW (comments and code) into the public domain).


Copyright © 1997, James T. Dennis, Starshine Technical Services
Published in Issue 13 of the Linux Gazette


[ TABLE OF CONTENTS ] [ FRONT PAGE ]  Back  Next