mirror of
https://github.com/munin-monitoring/contrib.git
synced 2018-11-08 00:59:34 +01:00
165 lines
4.4 KiB
Plaintext
165 lines
4.4 KiB
Plaintext
|
#!/bin/bash
|
||
|
#
|
||
|
# Wildcard-plugin to monitor Grid Engine queue state.
|
||
|
# To monitor a queue, link sge_queue_xml_<queue> to this file. E.g.
|
||
|
#
|
||
|
# ln -s /usr/share/munin/plugins/sge_queue_xml_ /etc/munin/plugins/sge_queue_xml_all.q
|
||
|
#
|
||
|
# inspired by the sge_queue_ plugin of steveschnepp
|
||
|
#
|
||
|
# Runs 'qstat -g c -xml' for the specified queue and gets the number of
|
||
|
# used, reserved, unavailable, available and total queue slots.
|
||
|
# Unavailable slots have queue states Disabled/Suspended/Error/Ambiguous config/Unknown
|
||
|
#
|
||
|
# Parameters understood:
|
||
|
#
|
||
|
# config (required)
|
||
|
# autoconf (optional - used by munin-config)
|
||
|
# suggest (optional - used by munin-config)
|
||
|
#
|
||
|
# Configurable variables:
|
||
|
#
|
||
|
# env.sge_settings - Path to SGE settings.sh script, defaults to /opt/sge/default/common/settings.sh
|
||
|
# env.title - Graph title, overrides "SGE Queue state".
|
||
|
# env.options - Additional command line options to qstat.
|
||
|
# env.queues - list of queues to summarize
|
||
|
#
|
||
|
# Revisions:
|
||
|
# v1.0 2013-10-14
|
||
|
#
|
||
|
# Magic markers
|
||
|
#%# family=contrib
|
||
|
#%# capabilities=autoconf suggest
|
||
|
|
||
|
# env.sge_settings -- alternatively you can just modify the path below
|
||
|
SGE_SETTINGS=${sge_settings:-/opt/sge/default/common/settings.sh}
|
||
|
|
||
|
|
||
|
# queues to monitor
|
||
|
# priority 1: queue name in symlink
|
||
|
QUEUE=${0##*_}
|
||
|
# priority 2: queue names from environment
|
||
|
QUEUE=${QUEUE:-$queues}
|
||
|
# priority 3: summary of all GridEngine Queues
|
||
|
QUEUE=${QUEUE:-"Summary"}
|
||
|
|
||
|
# env.title
|
||
|
GRAPHTITLE=${title:-"GridEngine Queue state (${QUEUE})"}
|
||
|
|
||
|
# env.options
|
||
|
OPTIONS="-g c -xml $options"
|
||
|
|
||
|
if [ "$1" = "config" ]; then
|
||
|
echo "graph_title "$GRAPHTITLE""
|
||
|
echo "graph_order total used reserved unavailable available"
|
||
|
echo "graph_args --lower-limit 0 "
|
||
|
echo "graph_vlabel Queue slots"
|
||
|
echo "graph_scale no"
|
||
|
echo "graph_info Shows global Grid Engine queue state."
|
||
|
echo "graph_category SGE"
|
||
|
echo "graph_period minute"
|
||
|
echo "used.label Used"
|
||
|
echo "used.draw AREA"
|
||
|
echo "used.type GAUGE"
|
||
|
echo "used.info Currently used slots."
|
||
|
echo "unavailable.label Unavailable"
|
||
|
echo "unavailable.draw LINE1"
|
||
|
echo "unavailable.type GAUGE"
|
||
|
echo "unavailable.info Queue state Disabled/Suspended/Error/Ambiguous config/Unknown."
|
||
|
echo "reserved.label Reserved"
|
||
|
echo "reserved.draw LINE1"
|
||
|
echo "reserved.type GAUGE"
|
||
|
echo "reserved.info Reserved slots."
|
||
|
echo "available.label Available"
|
||
|
echo "available.draw LINE1"
|
||
|
echo "available.type GAUGE"
|
||
|
echo "available.info Available slots."
|
||
|
echo "total.label Total"
|
||
|
echo "total.draw AREA"
|
||
|
echo "total.type GAUGE"
|
||
|
echo "total.info Total slots."
|
||
|
exit 0
|
||
|
fi
|
||
|
|
||
|
# source settings script, needed for qstat
|
||
|
if [ -f "$SGE_SETTINGS" ]; then
|
||
|
source "$SGE_SETTINGS"
|
||
|
fi
|
||
|
QSTAT=$( which qstat )
|
||
|
XMLSTARLET=$( which xmlstarlet )
|
||
|
|
||
|
if [ "$1" = "autoconf" ]; then
|
||
|
if [ -n "$QSTAT" -a -n "$XMLSTARLET" ]; then
|
||
|
echo "yes"
|
||
|
exit 0
|
||
|
else
|
||
|
echo "no"
|
||
|
exit 1
|
||
|
fi
|
||
|
fi
|
||
|
|
||
|
# check requirements
|
||
|
[ -z "$QSTAT" ] && { echo "qstat not found" 1>&2 ; exit 1; }
|
||
|
[ -z "$XMLSTARLET" ] && { echo "xmlstarlet not found" 1>&2 ; exit 1; }
|
||
|
|
||
|
ALL_QUEUES=$( $QSTAT -g c -xml | $XMLSTARLET sel -T -t -m "//cluster_queue_summary/name" -v "node()" -o "," )
|
||
|
[ "$QUEUE" == "Summary" ] && QUEUE="$ALL_QUEUES"
|
||
|
|
||
|
if [ "$1" = "suggest" ]; then
|
||
|
/bin/echo -en "${ALL_QUEUES//,/\n}"
|
||
|
exit 0
|
||
|
fi
|
||
|
|
||
|
xmldemangle() {
|
||
|
# _SGE_XML hash array to store values
|
||
|
# _SGE_OUT xml output of qstat
|
||
|
# _SGE_ENTITY entity to read values from (optional)
|
||
|
|
||
|
[ -z "$_SGE_OUT" ] && return 1
|
||
|
|
||
|
local ENTITY=${_SGE_ENTITY:-"cluster_queue_summary"}
|
||
|
|
||
|
local IFSBAK=$IFS;IFS=$
|
||
|
for node in $( echo "$_SGE_OUT" | $XMLSTARLET sel -T -t -m "//${ENTITY}/*" -o "_SGE_XML[" -v "name()" -o ']=' -v "node()" -n )
|
||
|
do
|
||
|
eval $node
|
||
|
done
|
||
|
IFS=$IFSBAK
|
||
|
}
|
||
|
|
||
|
printvalues() {
|
||
|
local IFSBAK=$IFS; unset IFS
|
||
|
for i in ${!_SGE_QUEUE_KEYS[@]}
|
||
|
do
|
||
|
echo "${_SGE_QUEUE_KEYS[$i]}.value ${_SGE_QUEUE_VALUES[$i]}"
|
||
|
done
|
||
|
IFS=$IFSBAK
|
||
|
}
|
||
|
|
||
|
declare -A _SGE_XML
|
||
|
declare -A _SGE_QUEUE_VALUES
|
||
|
declare -A _SGE_QUEUE_KEYS
|
||
|
_SGE_QUEUE_KEYS=(
|
||
|
[used]="used"
|
||
|
[resv]="reserved"
|
||
|
[manual_intervention]="unavailable"
|
||
|
[available]="available"
|
||
|
[total]="total"
|
||
|
)
|
||
|
|
||
|
for qu in $( /bin/echo -e "${QUEUE//,/\n}" )
|
||
|
do
|
||
|
_SGE_OUT=$( "$QSTAT" $OPTIONS -q $qu )
|
||
|
xmldemangle || echo "Error on QUEUE: $qu" 1>&2
|
||
|
|
||
|
for i in ${!_SGE_QUEUE_KEYS[@]}
|
||
|
do
|
||
|
_SGE_QUEUE_VALUES[$i]=${_SGE_QUEUE_VALUES[$i]:-0}
|
||
|
let "_SGE_QUEUE_VALUES[$i] += ${_SGE_XML[$i]:-0}"
|
||
|
done
|
||
|
done
|
||
|
|
||
|
printvalues
|
||
|
|
||
|
exit 0
|