Process Control
Slackware systems often run hundreds or thousands of programs, each of
which is refered to as a process. Managing these processes is an
important part of system administration. So how exactly do we handle
all of these seperate processes?
ps
The first step in managing processes is figuring out what processes are
currently running. The most popular and powerful tool for this is
ps(1). Without any arguments,
ps won't tell you much information. By
default, it only tells you what processes are running in your currently
active shell. If we want more information, we'll need to look deeper.
darkstar:~$ ps
PID TTY TIME CMD
12220 pts/4 00:00:00 bash
12236 pts/4 00:00:00 ps
Here you can see what processes you are running in your currently
active shell or terminal and only some information is included. The
PID is the "Process ID"; every process is assigned a unique number. The
TTY tells you what terminal device the process is attached to.
Naturally, CMD is the command that was run. You might be a little
confused by TIME though, since it seems to move so slowly. This isn't
the amount of real time the process has been running, but rather the
amount of CPU time the process has consumed. An idle process uses
virtually no CPU time, so this value may not increase quickly.
Viewing only our own processes isn't very much fun, so let's take a
look at all the processes on the system with the -e
argument.
darkstar:~$ ps -e
PID TTY TIME CMD
1 ? 00:00:00 init
2 ? 00:00:00 kthreadd
3 ? 00:00:00 migration/0
4 ? 00:00:00 ksoftirqd/0
7 ? 00:00:11 events/0
9 ? 00:00:01 work_on_cpu/0
11 ? 00:00:00 khelper
102 ? 00:00:02 kblockd/0
105 ? 00:01:19 kacpid
106 ? 00:00:01 kacpi_notify
... many more lines ommitted ...
The above example uses the standard ps
syntax, but much more information can be discovered if we use BSD
syntax. In order to do so, we must use the aux argument.
This is distinct from the -aux argument, but in most cases
the two arguments are equivilant. This is a decades-old relic. For more
information, see the man page for ps.
darkstar:~$ ps aux
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 1 0.0 0.0 3928 632 ? Ss Apr05 0:00 init [3]
root 2 0.0 0.0 0 0 ? S Apr05 0:00 [kthreadd]
root 3 0.0 0.0 0 0 ? S Apr05 0:00 [migration/0]
root 4 0.0 0.0 0 0 ? S Apr05 0:00 [ksoftirqd/0]
root 7 0.0 0.0 0 0 ? S Apr05 0:11 [events/0]
root 9 0.0 0.0 0 0 ? S Apr05 0:01 [work_on_cpu/0]
root 11 0.0 0.0 0 0 ? S Apr05 0:00 [khelper]
... many more lines ommitted ....
As you can see, BSD syntax offers much more information, including what
user controls the process and what percentage of RAM and CPU the process
is consuming when ps is run.
Finally, ps can also create a process tree.
This shows you which processes have children processes. Ending the
parent of a child process also ends the child. We do this with the
-ejH argument.
darkstar:~$ ps -ejH
... many lines ommitted ...
3660 3660 3660 tty1 00:00:00 bash
29947 29947 3660 tty1 00:00:00 startx
29963 29947 3660 tty1 00:00:00 xinit
29964 29964 29964 tty7 00:27:11 X
29972 29972 3660 tty1 00:00:00 sh
29977 29972 3660 tty1 00:00:05 xscreensaver
29988 29972 3660 tty1 00:00:04 xfce4-session
29997 29972 3660 tty1 00:00:16 xfwm4
29999 29972 3660 tty1 00:00:02 Thunar
... many more lines ommitted ...
As you can see, ps is an incredibly powerful
tool for determining not only what processes are currently active on
your system, but also for learning lots of important information about
them.
kill and killall
Managing processes isn't only about knowing which ones are running, but
also about communicating with them to change their behavior. The most
common way of managing a program is to terminate it. Thus, the tool for
the job is named kill(1). Despite the name,
kill doesn't actually terminate processes,
but sends signals to them. The most common signal is a SIGTERM, which
tells the process to finish up what it is doing and terminate. There
are a variety of other signals that can be sent, but the three most
common are SIGTERM, SIGHUP, and SIGKILL.
What a process does when it receives a signal varies. Most programs
will terminate (or attempt to terminate) whenever they receive any
signal, but there are a few important differences. For starters, the
SIGTERM signal informs the process that it should terminate itself at
its earliest convenience. This gives the process time to finish up any
important activities, such as writing information to the disk, before
it closes. In contrast, the SIGKILL signal tells the process to
terminate itself immediately, no questions asked. This is most useful
for killing processes that are not responding and is sometimes called
the "silver bullet". Some processes (particularly daemons) capture the
SIGHUP signal and reload their configuration files whenever they
receive it.
In order to signal a process, we first need to know it's PID. You can
get this easily with ps as we discused. In
order to send different signals to a running process, you simply pass
the signal number and -s as an argument. The -l
argument lists all the signals you can chose and their number. You can
also send signals by their name with -s.
darkstar:~$ kill -l
1) SIGHUP 2) SIGINT 3) SIGQUIT 4) SIGILL
5) SIGTRAP 6) SIGABRT 7) SIGBUS 8) SIGFPE
9) SIGKILL 10) SIGUSR1 11) SIGSEGV 12) SIGUSR2
13) SIGPIPE 14) SIGALRM 15) SIGTERM 16) SIGSTKFLT
... many more lines ommitted ...
darkstar:~$ kill 1234 # SIGTERM
darkstar:~$ kill -s 9 1234 # SIGKILL
darkstar:~$ kill -s 1 1234 # SIGHUP
darkstar:~$ kill -s HUP 1234 # SIGHUP
Sometimes you may wish to terminate all running processes with a
certain name. You can kill processes by name with
killall(1). Just pass the same arguments to
killall that you would pass to
kill.
darkstar:~$ killall bash # SIGTERM
darkstar:~$ killall -s 9 bash # SIGKILL
darkstar:~$ killall -s 1 bash # SIGHUP
darkstar:~$ killall -s HUP bash # SIGHUP
top
So far we've learned how to look at the active processes for a moment
in time, but what if we want to monitor them for an extended period?
top(1) allows us to do just that. It
displays an ordered list of the processes on your system, along with
vital information about them, and updates periodically. By default,
processes are ordered by their CPU percentage and updates occur every
three seconds.
darkstar:~$ top
top - 16:44:15 up 26 days, 5:53, 5 users, load average: 0.08, 0.03, 0.03
Tasks: 122 total, 1 running, 119 sleeping, 0 stopped, 2 zombie
Cpu(s): 3.4%us, 0.7%sy, 0.0%ni, 95.5%id, 0.1%wa, 0.0%hi, 0.2%si, 0.0%st
Mem: 3058360k total, 2853780k used, 204580k free, 154956k buffers
Swap: 0k total, 0k used, 0k free, 2082652k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1 root 20 0 3928 632 544 S 0 0.0 0:00.99 init
2 root 15 -5 0 0 0 S 0 0.0 0:00.00 kthreadd
3 root RT -5 0 0 0 S 0 0.0 0:00.82 migration/0
4 root 15 -5 0 0 0 S 0 0.0 0:00.01 ksoftirqd/0
7 root 15 -5 0 0 0 S 0 0.0 0:11.22 events/0
9 root 15 -5 0 0 0 S 0 0.0 0:01.19 work_on_cpu/0
11 root 15 -5 0 0 0 S 0 0.0 0:00.01 khelper
102 root 15 -5 0 0 0 S 0 0.0 0:02.04 kblockd/0
105 root 15 -5 0 0 0 S 0 0.0 1:20.08 kacpid
106 root 15 -5 0 0 0 S 0 0.0 0:01.92 kacpi_notify
175 root 15 -5 0 0 0 S 0 0.0 0:00.00 ata/0
177 root 15 -5 0 0 0 S 0 0.0 0:00.00 ata_aux
178 root 15 -5 0 0 0 S 0 0.0 0:00.00 ksuspend_usbd
184 root 15 -5 0 0 0 S 0 0.0 0:00.02 khubd
187 root 15 -5 0 0 0 S 0 0.0 0:00.00 kseriod
242 root 20 0 0 0 0 S 0 0.0 0:03.37 pdflush
243 root 15 -5 0 0 0 S 0 0.0 0:02.65 kswapd0
The man page has helpful details on how to interact with
top such as changing its delay interval, the
order processes are displayed, and even how to terminate processes
right from within top itself.
cron
Ok, so we've learned many different ways of viewing the active
processes on our system and means of signalling them, but what if we
want to run a process periodically? Fortunately, Slackware includes
just the thing, crond(8). cron runs
processes for every user on the schedule that user demands. This makes
it very useful for processes that need to be run periodically, but
don't require full daemonization, such as backup scripts. Every user
gets their own entry in the cron database, so non-root users can
periodically run processes too.
In order to run programs from cron, you'll need to use the
crontab(1). The man page lists a variety of
ways to do this, but the most common method is to pass the
-e argument. This will lock the user's entry in the cron
database (to prevent it from being overwritten by another program),
then open that entry with whatever text editor is specified by the
VISUAL environment variable. On Slackware systems, this is typically
the vi editor. You may need to refer to the
chapter on vi before continuing.
The cron database entries may seem a little archaic at first, but they
are highly flexible. Each uncommented line is processed by
crond and the command specified is run if
all the time conditions match.
darkstar:~$ crontab -e
# Keep current with slackware
30 02 * * * /usr/local/bin/rsync-slackware64.sh 1>/dev/null 2>&1
As mentioned before, the syntax for cron entries is a little difficult
to understand at first, so let's look at each part individually. From
left to right, the different sections are: Minute, Hour, Day, Month,
Week Day, and Command. Any asterisk * entry matches
every minute, hour, day, and so on. So from the example above, the
command is "/usr/local/bin/rsync-slackware64.sh 1>/dev/null 2>&1", and
it runs every weekday or every week of every month at 2:30 a.m.
crond will also e-mail the local user with
any output the command generates. For this reason, many tasks have
their output redirected to /dev/null, a special
device file that immediately discards everything it receives. In order
to make it easier for you to remember these rules, you might wish to
paste the following commented text at the top of your own cron entries.
# Redirect everything to /dev/null with:
# 1>/dev/null 2>&1
#
# MIN HOUR DAY MONTH WEEKDAY COMMAND
By default, Slackware includes a number of entries and comments in
root's crontab. These entries make it easier to setup periodic system
tasks by creating a number of directories in /etc
corrosponding to how often the tasks should run. Any script placed
within these directories will be run hourly, daily, weekly, or monthly.
The names should be self-explainatory:
/etc/cron.hourly,
/etc/cron.daily,
/etc/cron.weekly, and
/etc/cron.monthly.