#!/bin/bash

PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

# script can be executed as user

# although we could switch to ksh93 to be able to use floats, and calculate more precise percentage ourselves
# it's maybe preferable to keep using bash for compatibility

# todo - check what happens on busybox systems
# todo - skip/continue resource when unknown - or is it fine for all resources already since var is empty?
# todo - port to bsd systems?  no posix output format there

[[ ! -x `which df` ]] && echo install coreutils first && exit 1
[[ ! -x `which free` ]] && echo install procps first && exit 1
[[ ! -x `which iostat` ]] && echo install sysstat first && exit 1
[[ ! -x `which uptime` ]] && echo install procps first && exit 1

#
# cpu & disk i/o usage
#

output=`iostat -x`

avg_cpu=`echo "$output" | grep -A1 ^avg-cpu:`

[[ -z $avg_cpu ]] && echo [UNKNOWN] could not define cpu usage && (( fail = 1 ))

# avoid floats and the need to switch to ksh93
idle=`echo "$avg_cpu" | tail -1 | awk '{print $NF}' | cut -f1 -d.`

if (( idle <= 5 )); then
	echo [WARNING] average cpu usage reached 95%:
	echo "$avg_cpu"
	(( warn = 1 ))
else
	echo [OK] average cpu usage is fine
fi

util_disks=`echo "$output" | sed -n '/^Device/,$p' | sed '1d; /^$/d' | grep -v ^loop`

[[ -z $util_disks ]] && echo [UNKNOWN] could not define disk i/o usage && (( fail = 1 ))

[[ -n $util_disks ]] && while read line; do
	disk_name=`echo $line | awk '{print $1}'`

	# avoid floats again
	util_percent=`echo $line | awk '{print $NF}' | cut -f1 -d.`

	if (( util_percent >= 100 )); then
		echo [WARNING] $disk_name disk reached 100% i/o usage \(fine for ssd drives\):
		echo "$util_disks"
		(( warn = 1 ))
	else
		echo [OK] $disk_name disk i/o usage is fine
	fi

	unset util_percent disk_name
done <<< "$util_disks"

unset output avg_cpu idle util_disks

#
# cpu load
#

# avoid floats 
load15m=`uptime | awk '{print $NF}' | cut -f1 -d.`
nproc=`grep ^processor /proc/cpuinfo | wc -l`

[[ -z $load15m ]] && echo [UNKNOWN] could not define 15 minutes cpu load && (( fail = 1 ))
[[ -z $nproc ]] && echo [UNKNOWN] could not define amount of cores && (( fail = 1 ))

if (( load15m / nproc >= 100 )); then
	echo [WARNING] average 15 minutes cpu load reached 100% per core:
	echo "$load15m"
	echo \(for $nproc cores\)
	(( warn = 1 ))
else
	echo [OK] average 15 minutes cpu load is fine
fi

unset load15m nproc

#
# ram
#

ram_mebi=`free -m`
total_mebi=`echo "$ram_mebi" | grep ^Mem: | awk '{print $2}'`
available_mebi=`echo "$ram_mebi" | grep ^Mem: | awk '{print $NF}'`

[[ -z $total_mebi ]] && echo [UNKNOWN] could not define total ram && (( fail = 1 ))
[[ -z $available_mebi ]] && echo [UNKNOWN] could not define available ram && (( fail = 1 ))

if (( 100 * $available_mebi / $total_mebi <= 10 )); then
	echo [WARNING] ram usage reached 90%:
	echo "$ram_mebi"
	(( warn = 1 ))
else
	echo [OK] ram usage is fine
fi

unset available_mebi total_mebi ram_mebi

#
# network tx/rx
#

sar_output=`sar -n DEV`

nics=`sar -n DEV | grep ^Average:`

if [[ -z $nics ]]; then
	echo [UNKNOWN] enable sysstat sa1/collect for network interface statistics
	echo "$sar_output"
	(( fail = 1 ))
fi

[[ -n $nics ]] && while read line; do
	nic_name=`echo $line | awk '{print $2}'`

	# avoid floats altogether
	nic_util=`echo $line | awk '{print $NF}' | cut -f1 -d.`

	if (( nic_util >= 50 )); then
		echo [WARNING] $nic_name network interface reached 50% usage:
		echo "$sar_output"
		(( warn = 1 ))
	else
		echo [OK] $nic_name network interface usage is fine
	fi

	unset nic_name nic_util
done <<< "$nics"

unset nics sar_output

# todo
#sar_errors_output=`sar -n EDEV`

#
# filesystem usage
#

posix_output=`df -P -BG`

filesystems=`echo "$posix_output" | sed 1d | grep -vE '^overlay|^tmpfs|^devtmpfs|^udev'`

[[ -z $filesystems ]] && echo [UNKNOWN] could not find filesystem usage statistics && (( fail = 1 ))

[[ -n $filesystems ]] && while read line; do
        filesystem_name=`echo $line | awk '{print $NF}'`
       filesystem_total=`echo $line | awk '{print $2}' | sed 's/G$//'`
        filesystem_util=`echo $line | awk '{print $5}' | sed 's/%$//'`

	# conf file needs to be readable as user
        [[ -f /etc/check_hw_resources.conf ]] && source /etc/check_hw_resources.conf

        if [[ $filesystem_special_name = $filesystem_name ]]; then
                percent_trigger_float=$filesystem_special_percent_trigger
                percent_trigger=$filesystem_special_percent_trigger
		special="special "
        else
		if (( filesystem_total <= 25 )); then
			(( percent_trigger_float = 75 ))
			(( percent_trigger = 75 ))
                else
                        # raise trigger according to disk size
                        # https://en.wikipedia.org/wiki/Sigmoid_function
                        # https://en.wikipedia.org/wiki/Logistic_function
                        percent_trigger_float=`echo $filesystem_total | awk '{print ( 25 / \
                          ( 1 + exp \
                          (-6 * (1 / 5000 * $1 - 0.5) ) \
                          ) ) - 1.18 + 75 }'`
			percent_trigger=${percent_trigger_float%%\.*}
		fi
        fi

        if (( percent_trigger < 75 || percent_trigger > 99 )); then
                echo [UNKNOWN] could not define filesystem percent_trigger
                echo debug filesystem_total is $filesystem_total G
                echo debug percent_trigger is $percent_trigger %
                (( fail = 1 ))
        fi

        if (( filesystem_util >= percent_trigger )); then
                echo [WARNING] $filesystem_name filesystem total size ${filesystem_total}G reached $special$filesystem_util/$percent_trigger_float%:
                echo "$posix_output" | grep -vE '^overlay|^tmpfs|^devtmpfs|^udev'
                (( warn = 1 ))
        else
                echo [OK] $filesystem_name filesystem total size ${filesystem_total}G usage $special$filesystem_util/$percent_trigger_float% is fine
        fi

	unset filesystem_util filesystem_total filesystem_name
	unset percent_trigger_float percent_trigger
        unset filesystem_special_name filesystem_special_percent_trigger special
done <<< "$filesystems"

unset filesystems posix_output

#
# nagios-style conclusion
# https://nagios-plugins.org/doc/guidelines.html#AEN78
#

(( crit == 1 )) && exit 2 || true
(( fail == 1 )) && exit 3 || true
(( warn == 1 )) && exit 1 || true