#!/bin/bash
# ======================================================================
#
# Icinga/ Nagios Check
# CEPH STATUS / HEALTH
#
# ----------------------------------------------------------------------
#
# REQUIREMENTS:
#   - ceph
#
# SYNTAX:
#   - check_ceph_status
#     No parameter required
#
# RESULT:
#     OK        = all OSDs up
#     CRITITCAL = min. 1 OSD is down
#     UNKNOWN   = ceph osd tree is not executable
#
# ----------------------------------------------------------------------
# 2020-03-04  v1.0  <axel.hahn@iml.unibe.ch>
# 2020-03-05  v1.1  <axel.hahn@iml.unibe.ch>  added params -w -c 
# 2020-03-05  v1.2  <axel.hahn@iml.unibe.ch> switch to ph.* helper functions
# ======================================================================

. `dirname $0`/inc_pluginfunctions

tmpfile=/tmp/ceph_status_output_$$

# ----------------------------------------------------------------------
# FUNCTIONS
# ----------------------------------------------------------------------

# fetch lines of OSD entries only from output of ceph osd tree
function getOsd(){
    grep "^\ *[0-9]" $tmpfile
}

# ----------------------------------------------------------------------
# MAIN
# ----------------------------------------------------------------------
sudo /bin/ceph osd tree > $tmpfile 2>&1
if [ $? -ne 0 ]; then
    rm -f $tmpfile
    ph.abort "UNKNOWN: ceph is not available or no sudo permissions to execute ceph commands."
fi

# set default / override from command line params
typeset -i iWarnLimit=`     ph.getValueWithParam 1 w "$@"`
typeset -i iCriticalLimit=` ph.getValueWithParam 2 c "$@"`

typeset -i iOsdTotal=` getOsd | wc -l`
typeset -i iOsdDown=`  getOsd | awk '{ print $4 }' | grep "down" | wc -l`
typeset -i iOsdUp=`    getOsd | awk '{ print $4 }' | grep "up"   | wc -l`


if [ $iOsdDown -lt $iWarnLimit ]; then
    ph.setStatus "ok"
else
    if [ $iOsdDown -ge $iCriticalLimit ]; then
        ph.setStatus "critical"
    else
        ph.setStatus "warning"
    fi
fi

ph.status "Check of available OSDs - $iOsdTotal OSDs total .. $iOsdUp up .. $iOsdDown down (Limits: warn at $iWarnLimit; critical $iCriticalLimit)"
cat $tmpfile

rm -f $tmpfile

ph.perfadd "osd-total"    "${iOsdTotal}"  "" "" 0 ${iOsdTotal}
ph.perfadd "osd-up"       "${iOsdUp}"     "" "" 0 ${iOsdTotal}
ph.perfadd "osd-down"     "${iOsdDown}"   "" "" 0 ${iOsdTotal}

ph.exit

# ----------------------------------------------------------------------