nagios: adding check_btrfs_quota
This commit is contained in:
parent
6363625979
commit
737f61e844
1 changed files with 244 additions and 0 deletions
244
nagios/check_btrfs_quota.sh
Executable file
244
nagios/check_btrfs_quota.sh
Executable file
|
|
@ -0,0 +1,244 @@
|
|||
#!/bin/sh
|
||||
|
||||
# Little check for quota usage on Btrfs subvolumes
|
||||
# GPL v3+ (copyright chl-dev@bugness.org)
|
||||
|
||||
PROGPATH=$( echo $0 | sed -e 's,[\\/][^\\/][^\\/]*$,,' )
|
||||
REVISION="0.1"
|
||||
|
||||
# Stop at the first non-catched error
|
||||
set -e
|
||||
|
||||
# Include check_range()
|
||||
# Not needed at the moment
|
||||
#. $PROGPATH/utils.sh
|
||||
STATE_OK=0
|
||||
STATE_WARNING=1
|
||||
STATE_CRITICAL=2
|
||||
STATE_UNKNOWN=3
|
||||
STATE_DEPENDENT=4
|
||||
|
||||
# Default values
|
||||
RANGE_WARNING="20%"
|
||||
RANGE_CRITICAL="10%"
|
||||
|
||||
# Output
|
||||
OUTPUT_EXIT_STATUS=$STATE_OK
|
||||
OUTPUT_DETAIL_OK=""
|
||||
OUTPUT_DETAIL_WARNING=""
|
||||
OUTPUT_DETAIL_CRITICAL=""
|
||||
OUTPUT_PERFDATA=""
|
||||
|
||||
#
|
||||
# Help function
|
||||
#
|
||||
usage() {
|
||||
cat <<EOF
|
||||
Usage :
|
||||
$0 [-w warning-free-space%] [-c critical-free-space%] -V volume -V...
|
||||
|
||||
Example :
|
||||
./check_btrfs_quota -w 20% -c 10% -V /home
|
||||
|
||||
We adopt the same behaviour than check_disks so thresholds are matched
|
||||
against "free" space. The example above will trigger a warning if less
|
||||
than 20% of maxrfer or maxecl is available.
|
||||
|
||||
Note: Since the volumes are checked against the latest thresholds specified, order
|
||||
of the arguments is important. Ex:
|
||||
./check_btrfs_quota -w 20% -V /home -w 50% -V /tmp
|
||||
|
||||
Default values:
|
||||
warning-free-space: $RANGE_WARNING
|
||||
critical-free-space: $RANGE_CRITICAL
|
||||
EOF
|
||||
}
|
||||
|
||||
check_range_syntax() {
|
||||
return 0
|
||||
}
|
||||
# check_range 0 "$1" >/dev/null 2>&1
|
||||
# if [ "$?" -eq "2" ]; then
|
||||
# return 1
|
||||
# fi
|
||||
# return 0
|
||||
#}
|
||||
|
||||
#
|
||||
# Remove the '%' suffix (or error if no '%')
|
||||
#
|
||||
get_percent() {
|
||||
if [ "$( echo -n "$1" | tail -c 1 )" = "%" ]; then
|
||||
echo -n "$1" | head -c -1
|
||||
return 0
|
||||
fi
|
||||
return 1
|
||||
}
|
||||
|
||||
#
|
||||
# Returns 0 if freespace is above threshold, 1 if not
|
||||
#
|
||||
# args:
|
||||
# - "free" space value
|
||||
# - quota (aka. maximum space possible)
|
||||
# - threshold (in percent with a '%' suffix, or plain unit)
|
||||
#
|
||||
check_freespace_is_above_threshold() {
|
||||
local VALUE="$1"
|
||||
local MAXVALUE="$2"
|
||||
local THRESHOLD="$3"
|
||||
|
||||
if get_percent "$THRESHOLD" >/dev/null; then
|
||||
VALUE="$(( $VALUE * 100 / $MAXVALUE ))"
|
||||
THRESHOLD="$( get_percent "$THRESHOLD" )"
|
||||
fi
|
||||
|
||||
if [ "$VALUE" -le "$THRESHOLD" ]; then
|
||||
return 1
|
||||
fi
|
||||
return 0
|
||||
}
|
||||
|
||||
#
|
||||
# Factorize output in the form of:
|
||||
# /home:0/260 (refr:10M(50%) excl:5M(20%)
|
||||
# args:
|
||||
# 1: VOLUME
|
||||
# 2: QGROUPID
|
||||
# 3: REFR
|
||||
# 4: EXCL
|
||||
# 5: MAXRFER
|
||||
# 6: MAXEXCL
|
||||
format_output_item() {
|
||||
local LOCAL_OUTPUT=""
|
||||
if [ "$5" != "none" ]; then
|
||||
LOCAL_OUTPUT="$( printf "rfer:%s(%d%%)" \
|
||||
"$( numfmt --to=iec $(( $5 - $3 )) )" \
|
||||
"$(( 100 * ($5 - $3) / $5 ))" \
|
||||
)"
|
||||
fi
|
||||
if [ "$6" != "none" ]; then
|
||||
[ -n "$LOCAL_OUTPUT" ] && LOCAL_OUTPUT="$LOCAL_OUTPUT "
|
||||
LOCAL_OUTPUT="$( printf "excl:%s(%d%%)" \
|
||||
"$( numfmt --to=iec $(( $6 - $4 )) )" \
|
||||
"$(( 100 * ($6 - $4) / $6 ))" \
|
||||
)"
|
||||
fi
|
||||
if [ -n "$LOCAL_OUTPUT" ]; then
|
||||
printf "%s:%s (%s)" "$1" "$2" "$LOCAL_OUTPUT"
|
||||
fi
|
||||
}
|
||||
|
||||
# Some early checks
|
||||
if ! which btrfs 2>&1 >/dev/null; then
|
||||
echo "UNKNOWN 'btrfs' command not found."
|
||||
exit $STATE_UNKNOWN
|
||||
fi
|
||||
|
||||
#
|
||||
# Loop on parameters + tests
|
||||
#
|
||||
while getopts hw:c:V: f; do
|
||||
case "$f" in
|
||||
'h')
|
||||
usage
|
||||
exit
|
||||
;;
|
||||
|
||||
'w')
|
||||
if check_range_syntax "$OPTARG" >/dev/null; then
|
||||
RANGE_WARNING="$OPTARG"
|
||||
else
|
||||
echo "UNKNOWN: invalid range."
|
||||
exit 3
|
||||
fi
|
||||
;;
|
||||
|
||||
'c')
|
||||
if check_range_syntax "$OPTARG" >/dev/null; then
|
||||
RANGE_CRITICAL="$OPTARG"
|
||||
else
|
||||
echo "UNKNOWN: invalid range."
|
||||
exit 3
|
||||
fi
|
||||
;;
|
||||
|
||||
'V')
|
||||
VOLUME="$OPTARG"
|
||||
|
||||
# Quickly check if we have enough permission to launch btrfs commands
|
||||
if ! btrfs qgroup show "$VOLUME" >/dev/null 2>&1 ; then
|
||||
echo "UNKNOWN: unable to launch 'btrfs qgroup show $VOLUME' command."
|
||||
exit $STATE_UNKNOWN
|
||||
fi
|
||||
|
||||
# Loop on each qgroup
|
||||
# (the kinda weird <<EOF at the end is to avoid entering a sub-shell, so we can keep
|
||||
# our precious vars even with prosix-strict dash)
|
||||
while read QGROUPID RFER EXCL MAXRFER MAXEXCL; do
|
||||
LOOP_OUTPUT_STATUS=$STATE_OK
|
||||
if [ "$MAXRFER" != "none" ]; then
|
||||
FREERFER="$(( $MAXRFER - $RFER ))"
|
||||
if ! check_freespace_is_above_threshold "$FREERFER" "$MAXRFER" "$RANGE_CRITICAL"; then
|
||||
OUTPUT_EXIT_STATUS=$STATE_CRITICAL
|
||||
LOOP_OUTPUT_STATUS=$STATE_CRITICAL
|
||||
elif ! check_freespace_is_above_threshold "$FREERFER" "$MAXRFER" "$RANGE_WARNING"; then
|
||||
[ "$OUTPUT_EXIT_STATUS" = "$STATE_OK" ] && OUTPUT_EXIT_STATUS=$STATE_WARNING
|
||||
LOOP_OUTPUT_STATUS=$STATE_WARNING
|
||||
fi
|
||||
fi
|
||||
if [ "$MAXEXCL" != "none" ]; then
|
||||
FREEEXCL="$(( $MAXEXCL - $EXCL ))"
|
||||
if ! check_freespace_is_above_threshold "$FREEEXCL" "$MAXEXCL" "$RANGE_CRITICAL"; then
|
||||
OUTPUT_EXIT_STATUS=$STATE_CRITICAL
|
||||
LOOP_OUTPUT_STATUS=$STATE_CRITICAL
|
||||
elif ! check_freespace_is_above_threshold "$FREEEXCL" "$MAXEXCL" "$RANGE_WARNING"; then
|
||||
[ "$OUTPUT_EXIT_STATUS" = "$STATE_OK" ] && OUTPUT_EXIT_STATUS=$STATE_WARNING
|
||||
[ "$LOOP_OUTPUT_STATUS" = "$STATE_OK" ] && LOOP_OUTPUT_STATUS=$STATE_WARNING
|
||||
fi
|
||||
fi
|
||||
|
||||
# Depending on the alerts raised during this loop, we store
|
||||
# the output in ok/warning/critical. This way, in case of critical
|
||||
# alert, we don't list all the ok data.
|
||||
TMP="$( format_output_item "$VOLUME" "$QGROUPID" "$RFER" "$EXCL" "$MAXRFER" "$MAXEXCL" )"
|
||||
case "$LOOP_OUTPUT_STATUS" in
|
||||
"$STATE_OK")
|
||||
OUTPUT_DETAIL_OK="$OUTPUT_DETAIL_OK $TMP" ;;
|
||||
"$STATE_WARNING")
|
||||
OUTPUT_DETAIL_WARNING="$OUTPUT_DETAIL_WARNING $TMP" ;;
|
||||
"$STATE_CRITICAL")
|
||||
OUTPUT_DETAIL_CRITICAL="$OUTPUT_DETAIL_CRITICAL $TMP" ;;
|
||||
esac
|
||||
|
||||
OUTPUT_PERFDATA="$( printf "%s\n%s\n%s" "$OUTPUT_PERFDATA" "$VOLUME:$QGROUPID:rfer=$RFER" "$VOLUME:$QGROUPID:excl=$EXCL" )"
|
||||
done <<EOF
|
||||
$( btrfs qgroup show -re --raw "$VOLUME" | tail -n +3 )
|
||||
EOF
|
||||
;;
|
||||
|
||||
\?)
|
||||
usage
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
done
|
||||
|
||||
case "$OUTPUT_EXIT_STATUS" in
|
||||
'0')
|
||||
printf "OK %s" "$OUTPUT_DETAIL_OK"
|
||||
;;
|
||||
'1')
|
||||
printf "WARNING %s" "$OUTPUT_DETAIL_WARNING"
|
||||
;;
|
||||
'2')
|
||||
printf "CRITICAL %s" "$OUTPUT_DETAIL_CRITICAL"
|
||||
;;
|
||||
*)
|
||||
printf "UNKNOWN"
|
||||
;;
|
||||
esac
|
||||
|
||||
# We sort the perfdata because of a weird bug in some icinga/pnp4nagios interactions
|
||||
printf "|%s\n" "$( printf "%s" "$OUTPUT_PERFDATA" | sort | grep -v "^$" | tr "\n" " " )"
|
||||
exit $OUTPUT_EXIT_STATUS
|
||||
Loading…
Reference in a new issue