cpu

#!@@GOODSH@@
#
: <<=cut

=head1 NAME

cpu - Plugin to monitor CPU usage.

=head1 APPLICABLE SYSTEMS

All Linux systems

=head1 CONFIGURATION

The following is default configuration

  [cpu]
    env.HZ          100
    env.scaleto100  no

"scaleto100" may be "yes" or "no". With "yes" all CPU-related values
are scaled for a limit of 100.  With "no" (the default) all CPU-related
values are scaled for a limit of n * 100 ("n" being the number of CPUs).

The "ignore_fields" environment variable (empty by default) may be used
for listing fields (space separated) that should not be visualized.
For example hiding the "idle" field may improve the visualization of the
combined cpu usage.

See "BUGS" for an explanation of the "HZ" setting.

=head2 EXAMPLE WARNING AND CRITICAL SETTINGS

You can also set warning and critical levels for each of the data
series the plugin reports.  The following environment variables are
used as default for all fields:

  env.warning
  env.critical

But each field can be controlled separately:

  env.system_warning
  env.system_critical

  env.user_warning
  env.user_critical

  env.nice_warning
  env.nice_critical

  env.idle_warning
  env.idle_critical

For some kernels there is also the following settings:

  env.iowait_warning
  env.iowait_critical
  env.irq_warning
  env.irq_critical
  env.softirq_warning
  env.softirq_critical
  env.steal_warning
  env.steal_critical
  env.guest_warning
  env.guest_critical

=head1 INTERPRETATION

The plugin shows cpu usage in percent. In case of more than one core
it displays 100% for each core.

If a core is 100% busy there will be no "iowait" showing, that only
shows if the CPU has nothing else to do while it waits on IO.
Therefore a 100% busy core can hide a lot of iowait.  Please refer to
the IO latency and other disk related graphs for further information
about IO performance.

=head1 MAGIC MARKERS

  #%# family=auto
  #%# capabilities=autoconf


=head1 BUGS

Some combinations of hardware and Linux (probably only 2.4 kernels)
use 1000 units/second in /proc/stat corresponding to the systems
HZ. (see /usr/src/linux/include/asm/param.h). But Almost all systems
use 100 units/second and this is our default. Even if
Documentation/proc.txt in the kernel source says otherwise. - Finding
and fix by dz@426.ch

Otherwise none known

=head1 AUTHOR

Unknown

=head1 LICENSE

GPLv2

=cut


. "$MUNIN_LIBDIR/plugins/plugin.sh"

if [ "$1" = "autoconf" ]; then
	if [ -r /proc/stat ]; then
		echo yes
		exit 0
	else
		echo no
		exit 0
	fi
fi

HZ=${HZ:-100}
scaleto100=${scaleto100:-no}
ignore_fields=${ignore_fields:-}


maybe_hide_field() {
    local fieldname="$1"
    if echo "$ignore_fields" | grep -wq "$fieldname"; then
        echo "$fieldname.graph no"
    fi
}


extinfo=""

if grep -qE '^cpu +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+' /proc/stat; then
	extinfo="iowait irq softirq"
	if grep -qE '^cpu +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+' /proc/stat; then
		extextinfo="steal"
	fi
	if grep -qE '^cpu +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+ +[0-9]+' /proc/stat; then
		extextextinfo="guest"
	fi

fi

if [ "$1" = "config" ]; then

	NCPU=$(grep -E '^cpu[0-9]+ ' /proc/stat | wc -l)
	if [ "$scaleto100" = "yes" ]; then
		graphlimit=100
	else
		graphlimit=$((NCPU * 100))
	fi
	echo 'graph_title CPU usage'
	echo "graph_order system user nice idle $extinfo"
	echo "graph_args --base 1000 -r --lower-limit 0 --upper-limit $graphlimit"
	echo 'graph_vlabel %'
	echo 'graph_scale no'
	echo 'graph_info This graph shows how CPU time is spent.'
	echo 'graph_category system'
	echo 'graph_period second'
	echo 'system.label system'
	echo 'system.draw AREA'
	echo 'system.min 0'
	echo 'system.type DERIVE'
	echo "system.info CPU time spent by the kernel in system activities"
	echo 'user.label user'
	echo 'user.draw STACK'
	echo 'user.min 0'
	echo 'user.type DERIVE'
	echo 'user.info CPU time spent by normal programs and daemons'
	echo 'nice.label nice'
	echo 'nice.draw STACK'
	echo 'nice.min 0'
	echo 'nice.type DERIVE'
	echo 'nice.info CPU time spent by nice(1)d programs'
	echo 'idle.label idle'
	echo 'idle.draw STACK'
	echo 'idle.min 0'
	echo 'idle.type DERIVE'
	echo 'idle.info Idle CPU time'

	for field in system user nice idle; do
		print_adjusted_thresholds "$field" "$graphlimit"
		maybe_hide_field "$field"
	done

	if [ "$scaleto100" = "yes" ]; then
		echo "system.cdef system,$NCPU,/"
		echo "user.cdef user,$NCPU,/"
		echo "nice.cdef nice,$NCPU,/"
		echo "idle.cdef idle,$NCPU,/"
	fi
	if [ -n "$extinfo" ]
	then
		echo 'iowait.label iowait'
		echo 'iowait.draw STACK'
		echo 'iowait.min 0'
		echo 'iowait.type DERIVE'
		echo 'iowait.info CPU time spent waiting for I/O operations to finish when there is nothing else to do.'
		echo 'irq.label irq'
		echo 'irq.draw STACK'
		echo 'irq.min 0'
		echo 'irq.type DERIVE'
		echo 'irq.info CPU time spent handling interrupts'
		echo 'softirq.label softirq'
		echo 'softirq.draw STACK'
		echo 'softirq.min 0'
		echo 'softirq.type DERIVE'
		echo 'softirq.info CPU time spent handling "batched" interrupts'
		if [ "$scaleto100" = "yes" ]; then
			echo "iowait.cdef iowait,$NCPU,/"
			echo "irq.cdef irq,$NCPU,/"
			echo "softirq.cdef softirq,$NCPU,/"
		fi
		for field in iowait irq softirq; do
			print_adjusted_thresholds "$field" "$graphlimit"
			maybe_hide_field "$field"
		done
	fi

	if [ -n "$extextinfo" ]
	then
		echo 'steal.label steal'
		echo 'steal.draw STACK'
		echo 'steal.min 0'
		echo 'steal.type DERIVE'
		echo 'steal.info The time that a virtual CPU had runnable tasks, but the virtual CPU itself was not running'
		maybe_hide_field "steal"
		if [ "$scaleto100" = "yes" ]; then
			echo "steal.cdef steal,$NCPU,/"
		fi
		print_adjusted_thresholds "steal" "$graphlimit"
	fi

	if [ -n "$extextextinfo" ]
	then
		echo 'guest.label guest'
		echo 'guest.draw STACK'
		echo 'guest.min 0'
		echo 'guest.type DERIVE'
		echo 'guest.info The time spent running a virtual CPU for guest operating systems under the control of the Linux kernel.'
		maybe_hide_field "guest"
		if [ "$scaleto100" = "yes" ]; then
			echo "guest.cdef guest,$NCPU,/"
		fi
		print_adjusted_thresholds "guest" "$graphlimit"
	fi

	exit 0
fi

# Note: Counters/derive need to report integer values.  Also we need
# to avoid 10e+09 and the like %.0f should do this.

if [ -n "$extextextinfo" ]; then
	awk -v "hz=$HZ" '/^cpu / { printf "user.value %.0f\nnice.value %.0f\nsystem.value %.0f\nidle.value %.0f\niowait.value %.0f\nirq.value %.0f\nsoftirq.value %.0f\nsteal.value %.0f\nguest.value %.0f\n", $2*100/hz, $3*100/hz, $4*100/hz, $5*100/hz, $6*100/hz, $7*100/hz, $8*100/hz, $9*100/hz, $10*100/hz }' < /proc/stat
elif [ -n "$extextinfo" ]; then
	awk -v "hz=$HZ" '/^cpu / { printf "user.value %.0f\nnice.value %.0f\nsystem.value %.0f\nidle.value %.0f\niowait.value %.0f\nirq.value %.0f\nsoftirq.value %.0f\nsteal.value %.0f\n", $2*100/hz, $3*100/hz, $4*100/hz, $5*100/hz, $6*100/hz, $7*100/hz, $8*100/hz, $9*100/hz }' < /proc/stat
elif [ -n "$extinfo" ]; then
	awk -v "hz=$HZ" '/^cpu / { printf "user.value %.0f\nnice.value %.0f\nsystem.value %.0f\nidle.value %.0f\niowait.value %.0f\nirq.value %.0f\nsoftirq.value %.0f\n", $2*100/hz, $3*100/hz, $4*100/hz, $5*100/hz, $6*100/hz, $7*100/hz, $8*100/hz }' < /proc/stat
else
	awk -v "hz=$HZ" '/^cpu / { printf "user.value %.0f\nnice.value %.0f\nsystem.value %.0f\nidle.value %.0f\n", $2*100/hz, $3*100/hz, $4*100/hz, $5*100/hz }' < /proc/stat
fi