
#!/bin/ksh
# File  : cdot_rpt_qos_settings.sh
# By    : Maarten de Boer, 180417
# Subject       : Script to report (in Graphite) QoS
# set -x
#(0.2),180417	: Added -g & -m
#(0.3),200310	: Add filter for SVM (SFILTER). Add AGGR. MAX_QOS with PCT
#(0.4),200318	: Add; --noerror, for not sending error mail
#(0.5),200422	: Add; report on "wrong" aggr configured (AGGRCHK)
#               : Mod; -admin-state running,stopped
#(0.6),200429   : Mod; FILTER -> CFILER (-c|--cluster)
#(0.7),200429	: Mod; NO noerrmsg & No Single ErrMsg by mail
#(0.8),200429	: Mod; No ERROR at CC-clusters (EXCLUDES) 
PGM="`basename $0|cut -d\. -f1`"
VER="0.8"
TMP="/tmp/${PGM}.$$"
CSV="${PGM}.csv"
CLUSTERS="${HOME}/etc/clusters"
SSH="/usr/bin/ssh"
HOSTNAME="`hostname | cut -d\. -f1`"
LOG="${HOME}/log/${PGM}.log"
ASC="${PGM}.asc"
CFILTER="[?]*"
MAIL=""
MAILTO="maarten.deboer@atos.net"
NC="${HOME}/bin/nc"
GRAPHITESRV="10.192.109.224 2003"
GRAPHITE=""
CSV="/tmp/${PGM}.csv"
SFILTER="[?]*"
ERRFILE=${TMP}.errfile
EXCLUDES=${HOME}/etc/${PGM}.excl

AGGRCHK=""
AGGRDIAMOND="n[0-9][0-9]a[3][8][0-9][0-9]"
AGGRPLATINUM="n[0-9][0-9]a[3][8][0-9][0-9]"
AGGRGOLD="n[0-9][0-9]a[013][6928][0-9][0-9]"
AGGRSILVER="n[0-9][0-9]a20[0-9][0-9]"
AGGRBRONZE="n[0-9][0-9][ace][348]0[0-9][0-9]"


# PREFIX needed at 'vserver show'
PREFIX="nl"

# QoS -max-throughput
# Storage classes
# Diamond  |               : 5000 IOPS /TB (1-2ms)   | +200%
# Platinum | First (FC)    : 1500 IOPS /TB (3-5 ms)  | +200%
# Gold     | Business (BC) :  500 IOPS /TB (4-8 ms)  | +200%
# Silver   | Economy (EC)  :  250 IOPS /TB (6-12 ms) | +150%
# Bronze   | Archive (AC)  :   25 IOPS /TB (8-28 ms) | +100%
# MAX_PT = Max percentage as MAX QoS
STORAGECLASSES="diamond platinum gold silver bronze"
DIAMONT_TB=5000
DIAMONT_MAX_PT=120
PLATINUM_TB=1500
PLATINUM_MAX_PT=200
GOLD_TB=500
#GOLD_MAX_PT=150
GOLD_MAX_PT=200
SILVER_TB=250
SILVER_MAX_PT=150
BRONZE_TB=25
BRONZE_MAX_PT=200


SSHCMD()
# 1: Filername 2:Command-string
# When issue with connection to cluster, try the nodes (-01 & -02)
# "There are no entries matching your query." => EC=255
# "no connection" is also EC=255
{
  TMPERR="/tmp/${PGM}.$$.err"
  touch ${TMPERR}
  /usr/bin/ssh -n ${1} "${2}" 2> ${TMPERR}
  EC=${?}
  # Check if "ssh: connect to host 10.192.109.202 port 22: Connection refused" If so (EC2=0), the 2nd
  grep 'Connection refused' ${TMPERR}
  EC2=${?}
  if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
    sleep 1
    /usr/bin/ssh -n ${1}-01 "${2}" 2> ${TMPERR}
    EC=${?}
    grep 'Connection refused' ${TMPERR}
    EC2=${?}
    if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
      sleep 1
      /usr/bin/ssh -n ${1}-02 "${2}" 2> ${TMPERR}
      EC=${?}
      grep 'Connection refused' ${TMPERR}
      EC2=${?}
      if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
        sleep 1
        /usr/bin/ssh -n ${1}-03 "${2}" 2> ${TMPERR}
        EC=${?}
        grep 'Connection refused' ${TMPERR}
        EC2=${?}
        if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
          sleep 1
          /usr/bin/ssh -n ${1}-04 "${2}" 2> ${TMPERR}
          EC=${?}
          grep 'Connection refused' ${TMPERR}
          EC2=${?}
          if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
            sleep 1
            /usr/bin/ssh -n ${1}-05 "${2}" 2> ${TMPERR}
            EC=${?}
            grep 'Connection refused' ${TMPERR}
            EC2=${?}
            if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
              sleep 1
              /usr/bin/ssh -n ${1}-06 "${2}" 2> ${TMPERR}
              EC=${?}
              grep 'Connection refused' ${TMPERR}
              EC2=${?}
              if [ ${EC} -ne 0 ] && [ ${EC2} -eq 0 ]; then
                echo  "`date` ${PGM} ERROR with communication to ${1}. Connection to -01 - -06 failed too."|tee -a ${LOG} 
              fi  # EC=0 & EC2=0
            fi  # -06
          fi  # -05
        fi  # -04
      fi  # -03
    fi  # -02
  fi  # -01
  rm ${TMPERR}
}


USAGE()
{
  echo "Usage: ${PGM} [<options>]"
  echo "  Version: ${VER}"
  echo "  options         :"
  echo "    -a|--aggrchk  : Aggr check"
  echo "    -c|--cluster  : filter for clustername (${CFILTER})"
  echo "    -e|--etc      : Etc/clusters-file (${CLUSTERS})"
  echo "    -g|--graphite : send to Graphite"
  echo "    -h|--help     : this Help"
  echo "    -m|--mail     : do send Mail"
  echo "    -s|--svm      : filter for SVM (${SFILTER})"
  echo "    -V            : show Version"
  echo "    -x            : set -x"
  echo "    --mailto      : change MAILTO address & do send mail (${MAILTO})"
  echo "    --noerrmsg    : send NO Error-Msg (to ${MAILTO})"
}
# Check options
while [ $# -gt 0 ]
  do
  case $1 in
    -a | --aggrchk) AGGRCHK=1 ;;
    -e | --etc) CLUSTERS=$2; shift ;;
    -c | --cluster) CFILTER=$2; shift ;;
    -s | --svm) SFILTER=$2; shift ;;
    --mailto) MAILTO=$2; shift ;;
    -m | --mail) MAIL="1" ;;
    -g | --graphite) GRAPHITE="1" ;;
    -h | --help) USAGE; exit 1 ;;
    -V) echo "${PGM}: v${VER}"; exit 3 ;;
    -x)  set -x ;;
    *)  echo "Option ${1} not known. Exiting..."; echo; USAGE; exit 1 ;;
  esac
    shift
done  # case


echo "`date` ${PGM} v${VER} started"|tee -a ${LOG}
if [ ! -f ${NC} ]; then
  echo "  nc ($NC} NOT found. Exiting ..."|tee -a ${LOG}
  exit 4
fi
echo "AGGRCHK=${AGGRCHK}"
echo "CFILTER=${CFILTER}"
echo "CLUSTERS=${CLUSTERS}"
echo "ERRFILE=${ERRFILE}"
echo "EXCLUDES=${EXCLUDES}"
echo "GRAPHITE=${GRAPHITE}"
echo "GRAPHITESRV=${GRAPHITESRV}"
echo "MAIL=${MAIL}"
echo "MAILTO=${MAILTO}"
echo "SFILTER=${SFILTER}"

if [ -f ${EXCLUDES} ]; then
  echo "  EXCLUDES (${EXCLUDES}) found:"
  cat ${EXCLUDES}
  sleep 2
fi
sleep 1

touch ${TMP} ${ERRFILE}

cat "${CLUSTERS}"|grep -v \^#|awk -F\; '{print $1}'|sort|grep "${CFILTER}"|while read CLUSTER
do
  echo "  ${CLUSTER} ..."

# Getting the nodes 1st
# Also the DR (stopped) ones
  SSHCMD ${CLUSTER} "vserver show -admin-state running,stopped"|grep ${PREFIX}|grep ${SFILTER}|awk '{print $1}'|while read VSERVER
  do
    echo "  ${CLUSTER};${VSERVER} ..."

# Exclude root- & LS-volumes
    SSHCMD ${CLUSTER} "set -units GB -showseparator \";\"; vol show -vserver ${VSERVER} -field  aggregate,qos-policy-group,state,total,comment,create-time"|grep ${VSERVER}|egrep -v 'root|vol0ls'|while read LINE
    do
      SVM=`echo ${LINE}|awk -F\; '{print $1}'`
      VOL=`echo ${LINE}|awk -F\; '{print $2}'`
      AGGR=`echo ${LINE}|awk -F\; '{print $3}'`
      TOTAL_SIZE_GB=`echo ${LINE}|awk -F\; '{print $6}'|sed 's/GB//g'`
      QOS_POL_GR_NAME=`echo ${LINE}|awk -F\; '{print $8}'`
      STOR_CLASS=`echo ${LINE}|awk -F\; '{print $8}'|awk -F\_ '{print $2}'`
      touch ${TMP} 

# Check "right" AGGR
      case ${STOR_CLASS} in
        bronze)
          let MAX_QOS=${TOTAL_SIZE_GB}*${BRONZE_TB}*${BRONZE_MAX_PT}/100/1024
          if [ ${MAX_QOS} -lt ${BRONZE_TB} ]; then
            let MAX_QOS=${BRONZE_TB}*${BRONZE_MAX_PT}/100
          fi
          VOLISBRONZE=`echo "${AGGR}"|grep "${AGGRBRONZE}"`
          if [ "${VOLISBRONZE}" = "" ]; then
            if [ -f ${EXCLUDES} ]; then
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE}
            else
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|tee -a ${ERRFILE}
            fi  # EXCLUDES
          fi
          ;;
        silver) 
          let MAX_QOS=${TOTAL_SIZE_GB}*${SILVER_TB}*${SILVER_MAX_PT}/100/1024
          if [ ${MAX_QOS} -lt ${SILVER_TB} ]; then
            let MAX_QOS=${SILVER_TB}*${SILVER_MAX_PT}/100
          fi
          VOLISSILVER=`echo "${AGGR}"|grep "${AGGRSILVER}"`
          if [ "${VOLISSILVER}" = "" ]; then
            if [ -f ${EXCLUDES} ]; then
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE}
            else
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|tee -a ${ERRFILE}
            fi
          fi
          ;;
        gold) 
          let MAX_QOS=${TOTAL_SIZE_GB}*${GOLD_TB}*${GOLD_MAX_PT}/100/1024
          if [ ${MAX_QOS} -lt ${GOLD_TB} ]; then
            let MAX_QOS=${GOLD_TB}*${GOLD_MAX_PT}/100
          fi
          VOLISGOLD=`echo ${AGGR}|grep "${AGGRGOLD}"`
          if [ "${VOLISGOLD}" = "" ]; then
            if [ -f ${EXCLUDES} ]; then
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE}
            else
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|tee -a ${ERRFILE}
            fi
          fi
          ;;
        platinum) 
          let MAX_QOS=${TOTAL_SIZE_GB}*${PLATINUM_TB}*${PLATINUM_MAX_PT}/100/1024 
          if [ ${MAX_QOS} -lt ${PLATINUM_TB} ]; then
            let MAX_QOS=${PLATINUM_TB}*${PLATINUM_MAX_PT}
          fi
          VOLISPLATINUM=`echo "${AGGR}"|grep "${AGGRPLATINUM}"`
          if [ "${VOLISPLATINUM}" = "" ]; then
            if [ -f ${EXCLUDES} ]; then
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE}
            else
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|tee -a ${ERRFILE}
            fi
          fi
          ;;
        diamond) 
          let MAX_QOS=${TOTAL_SIZE_GB}*${DIAMONT_TB}*${DIAMONT_MAX_PT}/100/1024 
          if [ ${MAX_QOS} -lt ${DIAMONT_TB} ]; then
            let MAX_QOS=${DIAMONT_TB}*${DIAMONT_MAX_PT}
          fi
          VOLISDIAMOND=`echo "${AGGR}"|grep "${AGGRDIAMOND}"`
          if [ "${VOLISDIAMOND}" = "" ]; then
            if [ -f ${EXCLUDES} ]; then
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE}
            else
              echo "  ERROR: Wrong storage-class? Based on QoS-policy:${STOR_CLASS} and Aggr:${AGGR} of ${CLUSTER};${SVM};${AGGR};${VOL}"|tee -a ${ERRFILE}
            fi
          fi
          ;;
        *) MAX_QOS=0 ;;
      esac

      echo "${CLUSTER};${SVM};${AGGR};${VOL};${TOTAL_SIZE_GB};${QOS_POL_GR_NAME};${STOR_CLASS};${MAX_QOS};${DATE_S}"|tee -a ${TMP}

# Check QoS-policy
# And send to Graphite
      if [ "${QOS_POL_GR_NAME}" != "-" ]; then
        if [ ${GRAPHITE} ]; then
          DATE_S="`date +%s`"
        else
          DATE_S=""
        fi
        if [ ${GRAPHITE} ]; then
          echo "atos.qos.nl.${CLUSTER}.svm.${SVM}.vol.${VOL}.size.total ${TOTAL_SIZE_GB} ${DATE_S}"|${NC} ${GRAPHITESRV}
          echo "atos.qos.nl.${CLUSTER}.svm.${SVM}.vol.${VOL}.qos.max ${MAX_QOS} ${DATE_S}"|${NC} ${GRAPHITESRV}
        fi  # GRAPHITE
      else
        if [ -f ${EXCLUDES} ]; then
          echo "  ! ${CLUSTER};${SVM};${AGGR};${VOL} has NO QoS-policy-group defined (${QOS_POL_GR_NAME})"|grep -v -f ${EXCLUDES}|tee -a ${ERRFILE} ${LOG}
        else
          echo "  ! ${CLUSTER};${SVM};${AGGR};${VOL} has NO QoS-policy-group defined (${QOS_POL_GR_NAME})"|tee -a ${ERRFILE} ${LOG}
        fi  # EXCLUDES

        sleep 1
      fi  # ! -

    done  # read LINE

  done  # "vserver show
# If ERRFILE is > 0, then add HR (--------------------)
  if [ -s ${ERRFILE} ]; then # If size > 0
    echo "--------------------------------------------------" >> ${ERRFILE}
  fi
done  # CLUSTER


if [ ${MAIL} ]; then
  echo "# CLUSTER;SVM;AGGR;VOL;TOTAL_SIZE_GB;QOS_POL_GR_NAME;STOR_CLASS;MAX_QOS;DATE_S (if empty not send to Graphite)" > ${CSV}
  cat ${TMP} >> ${CSV}
# If ERRFILE, the add this to the mail. Otherwise only add the CSV to the mail
  if [ -s ${ERRFILE} ]; then # If size > 0
    echo "`date` CFILTER=${CFILTER} SFILTER=${SFILTER} GRAPHITE=${GRAPHITE} (to ${GRAPHITESRV}) `cat ${ERRFILE}`" | mailx -a ${CSV} -s ":${HOSTNAME}: ERROR & QoS report [${PGM} v${VER}]" ${MAILTO}
    echo "  Mailed ERROR (${ERRFILE}) & report (${CSV}) to ${MAILTO}"|tee -a ${LOG}
  else
    echo "`date` CFILTER=${CFILTER} SFILTER=${SFILTER} GRAPHITE=${GRAPHITE} (to ${GRAPHITESRV})" | mailx -a ${CSV} -s ":${HOSTNAME}: QoS report [${PGM} v${VER}]" ${MAILTO}
    echo "  Mailed report (${CSV}) to ${MAILTO}"|tee -a ${LOG}
  fi  # ERRFILE
  rm ${CSV}
fi  # MAIL

rm ${TMP} ${ERRFILE}
echo "`date` ${PGM} v${VER} finished."|tee -a ${LOG}
exit 0

