What it takes (IMHO) to be a system admin

Maybe it is naive of me, but it seems that one noble aspiration for Linux users is to get a 'real job' as a system administrator (SA). The question has come up in some discussions recently 'What is involved in moving from Linux user to system administrator?'. Accordingly, here are some of my thoughts.

DISCUSSION

Change your mindset

As a competent Linux user on your own hardware you are God. Even an incompetent Linux user is God on his own box. If you want to reboot, so what? If you 'rpm -e <something important>', you may be up half the night, but that's what caffeine is for. System admins have no such luxury.
The true SA is a combination caretaker, security guard, and baby sitter. It is his/her job to keep the machine running at almost all costs. In many cases corporate profits depend directly on the SA's ability to protect and maintain the system.
For instance, on one of the systems I administer, a reboot takes half an hour (memory and cpu checking) and idles approximately 600 people spread across 3 states. There is no room for 'oops'.
Paranoia is not only good, it is indispensable.

Learn new toolsets

If one wants to move into the corporate arena, then you must be able to take whatever tools are laying around and get the job done. This means the ability to learn new tools and to use old tools in new ways.
For instance, 'locate' is nice, but it quickly becomes apparent that you have to know 'find'. As in 'find all of the files that are owned by joe, larger than 1M, haven't been accessed in over 30 days and delete them'
It won't be long before a compile/install breaks. You have to be able to work through a make file to find the missing include; to interpret the 'command not found' message; to debug some vendor's install script that is expecting the default shell to be tcsh instead of sh and a dozen other things.
One of the Perl scripts below mentions that I had to fix an error in a system level include file to get the Log::Syslog module to work correctly.
Use your work as an excuse to learn more. I am using this 'project' as an excuse to learn about txt2html.pl. This is a Perl script for changing plain text files into basic html pages. As I learn something here, I apply it to several other projects for various clients.

Learn to handle pressure

With your personal Linux box, you can always power down and walk away if it gets too intense. Not so as an SA.
Expect to have to work with others looking over your shoulder. It adds a new level of pressure to have a Senior VP of a billion dollar company watching you type! (Or just the guy who signs your paycheck.)
Forget about lusers. They are users and they are why you exist. Whether they have a clue or not. By definition, they come to you when something is 'wrong' in their world. It is your job to make it right (and then to keep it from happening again).

Never start from scratch. Find something close and modify

For scripting, start with the boot up scripts (/sbin/init.d, /etc/rc.d, etc)
For Perl, Perl.org
You can always see farther standing on someone else's shoulders

Hang out with experts

Unix Guru Universe
Your local Linux group
Don't be afraid of appearing ignorant. Fear staying ignorant

Practice good debugging habits

Understand it the way it is (broken) before you try to fix it
Document what you have found (they're called backups)
Change one thing and retest
Debug ahead of time by trying to understand everything you can

Learn manually, then codify

Figure out the commands manually
Remember the commands by writing a script for them and commenting the script.
One of the beauties of Unix is that you don't have to type it again if you don't want to. Write a script
Change scripting languages just for fun (and learning)
Keep a notebook of tips and gems you've found

Document what you do

Comment your scripts liberally. The best comments (IMHO) are the ones that explain 'Why?'.
Use a standard README file in directories that you modify. That way when they call you at night and say "Where is my file?" you have a prayer of finding it.
Documentation will score points with the next SA as well as increasing your value in the community.
Set personal standards (like creating README files). It is amazing what you can forget in a week.

Learn to share

Many people will be very interested in what you are doing. Both from a business perspective and from a technical one. You must be able to discern how much to share with whom.
Share what you've learned with others (that's why I'm doing this page)
Find somewhere online to hangout. Listen, learn and contribute. Mailings lists or irc can be good starting points.

Remember to have fun

I like to add humor to my scripts. Note the printed comments for the kill scripts. I figure if I can make the night operator chuckle once or twice, then he is more likely to help me.
Make Unix your passion, not just your job. Don't consider becoming a SA if it isn't your passion. There was recently a discussion on the COLUG list about the meaning of a vanity license plate on an SUV '8MYCSH'. The Unix die hards interpreted that as 'ate my c-shell'. More likely it was meant as 'ate my cash'.
When things get too intense, go visit Simon, the Bastard Operator from Hell (or other system admin related humor sites)

Example scripts, killing things

Here are some examples of scripts that I have written. I offer them as a way to see inside one of the problems that an SA faces. The essential problem is how to allow inexperienced operators reset certain processes without giving them the root password. The obvious first step is to install sudo and add the kill command to the allowed commands. But this is too broad, and too narrow at the same time. Too broad in that they can kill things that they shouldn't (system processes, etc), too narrow in that they may need to kill hundreds of similar processes quickly. (All of these scripts are written in Perl 5.005 and will work on an HP-UX 10.20 system. I've made no effort to adapt them to Linux, because that is not the point. All of the kill scripts require root access or sudo.)

In following my own advice in #2 above, I was reteaching myself Perl while writing these scripts. They could also have been written using sh, or awk.

killp.pl script

To address the first problem, I wrote the following Perl script. It allows a person with sudo access to this script (not to kill in general) to kill processes belonging to a person. System processes are protected. killp.pl

killu.pl script

To address the second problem (killing batches of users) I created the killu.pl script. It takes a user prefix and kills all of the users whose login ids start with that prefix. It allows for some special ids to be protected so you don't kill supervisors, etc. Some of the protections can be overridden by specifying a longer prefix.

Examples, other things

du.pl script

On larger systems (AIX, HPUX, etc) you will encounter the idea of volume groups and logical volumes. This is a great idea which allows you to create virtual disks from collections of smaller or larger disks. Unfortunately, it spreads out the information so it is no longer easy to answer the question 'How much space is left?'. I wrote this du.pl script to combine the output of two commands (vgdisplay and bdf)

sortu script

Once you get over a screenful of users logged in, managing them, finding the idle ones, etc becomes a problem. This script sortu.pl combines the hectic output from who into a more readable format as well as allowing a summary to be generated.

Author

Jim Wildman has been mucking around Unix since 1985. After a few years on Suns, he changed jobs and started working on HP's. Then he added Linux in 1994 or 95 and a dabbling of AIX in '98. Those jobs have included stints in electronics manufacturing, healthcare, the online sales industry and Internet/ecommerce consulting. He is currently employed as a senior consultant by divine, Inc.

These pages were produced using some combination of RedHat Linux, Quanta, txt2html, and of course vim.

Questions and comments can be sent to Jim at jim@rossberry.com. All trademarks are the property of their respective holders. Last update Feb 26, 2000.