PWGPP/QA/scripts/runQA.sh

   1 #!/bin/bash
   2 main()
   3 {
   4   if [[ -z $1 ]]; then
   5     echo "Usage: "
   6     echo "  ${0##*/} option=value [option=value]"
   7     echo "  at least one option, either inputFile or configFile should be specified,"
   8     echo "  options override config file (if any), e.g.:"
   9     echo "  ${0##*/} configFile=runQA.config inputList=file.list outputDirectory=%det"
  10     return 1
  11   fi
  12
  13   if ! parseConfig $@; then
  14     ${0}
  15     return 1
  16   fi
  17
  18   [[ -z $ALICE_ROOT ]] && source ${alirootEnv}
  19   [[ -z $ALICE_ROOT ]] && echo "ALICE_ROOT not defined" && return 1
  20
  21   ocdbregex='raw://'
  22   if [[ ${ocdbStorage} =~ ${ocdbregex} ]]; then
  23     alien-token-init
  24   fi
  25
  26   updateQA $@
  27 }
  28
  29 updateQA()
  30 {
  31   umask 0002
  32   parseConfig $@
  33
  34   #be paranoid and make some full paths
  35   inputList=$(readlink -f ${inputList})
  36   [[ ! -f ${inputList} ]] && echo "no input list: ${inputList}" && return 1
  37   workingDirectory=$(readlink -f ${workingDirectory})
  38   mkdir -p ${workingDirectory}
  39   if [[ ! -d ${workingDirectory} ]]; then
  40     echo "working dir $workingDirectory does not exist and cannot be created"
  41     return 1
  42   fi
  43   cd ${workingDirectory}
  44
  45   echo JOB config:
  46   echo inputList=$inputList
  47   echo outputDirectory=$outputDirectory
  48   echo
  49
  50   dateString=$(date +%Y-%m-%d-%H-%M)
  51   echo "Start time QA process: $dateString"
  52
  53   #logging
  54   mkdir -p $logDirectory
  55   [[ ! -d $logDirectory ]] && echo "no log dir $logDirectory" && return 1
  56   logFile="$logDirectory/${0##*/}.${dateString}.log"
  57   touch ${logFile}
  58   [[ ! -f ${logFile} ]] && echo "cannot write logfile $logfile" && return 1
  59   echo "logFile = $logFile"
  60   exec &>${logFile}
  61
  62   #check lock
  63   lockFile=${logDirectory}/runQA.lock
  64   [[ -f ${lockFile} ]] && echo "lock ${lockFile} exists!" && return 1
  65   touch ${lockFile}
  66   [[ ! -f ${lockFile} ]] && echo "cannot lock $lockFile" && return 1
  67
  68   ################################################################
  69   #ze detector loop
  70   for detectorScript in $ALICE_ROOT/PWGPP/QA/detectorQAscripts/*; do
  71
  72     [[ ! ${detectorScript} =~ .*\.sh ]] && continue
  73     detector=${detectorScript%.sh}
  74     detector=${detector##*/}
  75
  76     #skip if excluded
  77     skipDetector=0
  78     for excluded in ${excludeDetectors}; do
  79       if [[ ${detector} =~ ${excluded} ]]; then
  80         echo "${detector} is excluded in config, skipping..."
  81         skipDetector=1
  82         break
  83       fi
  84     done
  85     [[ ${skipDetector} -eq 1 ]] && continue
  86
  87     logSummary=${logDirectory}/summary-${detector}-${dateString}.log
  88     outputDir=$(substituteDetectorName ${detector} ${outputDirectory})
  89     tmpRunDir=${workingDirectory}/tmpQArunDir${detector}
  90     if ! mkdir -p ${tmpRunDir}; then
  91       echo "cannot create the temp dir $tmpRunDir"
  92       continue
  93     fi
  94     cd ${tmpRunDir}
  95
  96     tmpPrefix=${tmpRunDir}/${outputDir}
  97     echo
  98     echo "##############################################"
  99     echo "running QA for ${detector}"
 100     echo "  outputDir=$outputDir"
 101     echo "  tmpPrefix=$tmpPrefix"
 102
 103     unset -f runLevelQA
 104     unset -f periodLevelQA
 105     source ${detectorScript}
 106
 107     #################################################################
 108     #produce the QA and trending tree for each file (run)
 109     while read qaFile; do
 110       echo
 111
 112       if ! guessRunData ${qaFile}; then
 113         echo "could not guess run data from ${qaFile}"
 114         continue
 115       fi
 116
 117       runDir=${tmpPrefix}/${dataType}/${year}/${period}/${pass}/000${runNumber}
 118       mkdir -p ${runDir}
 119       cd ${runDir}
 120
 121       #handle the case of a zip archive
 122       [[ "$qaFile" =~ .*.zip$ ]] && qaFile="${qaFile}#QAresults.root"
 123
 124       echo running ${detector} runLevelQA for run ${runNumber} from ${qaFile}
 125       runLevelQA ${qaFile} &> runLevelQA.log
 126
 127       cd ${tmpRunDir}
 128
 129     done < ${inputList}
 130
 131     #################################################################
 132     #cache which productions were (re)done
 133     arrOfTouchedProductions=( $(/bin/ls -d ${tmpPrefix}/*/*/*/*) )
 134     echo "list of processed productions:"
 135     echo "    ${arrOfTouchedProductions[@]}"
 136     echo
 137     #################################################################
 138     #(re)do the merging/trending in the final destination
 139     for tmpProductionDir in ${arrOfTouchedProductions[@]}; do
 140       echo
 141       echo "running period level stuff in ${tmpProductionDir}"
 142
 143       productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 144
 145       mkdir -p ${productionDir}
 146       if [[ ! -d ${productionDir} ]]; then
 147         echo "cannot make productionDir $productionDir" && continue
 148       fi
 149       cd ${productionDir}
 150
 151       #move to final destination
 152       for dir in ${tmpProductionDir}/*; do
 153         oldRunDir=${outputDir}/${dir#${tmpPrefix}}
 154         if ! guessRunData "${dir}/dummyName"; then
 155           echo "could not guess run data from ${dir}"
 156           continue
 157         fi
 158
 159         #before moving - VALIDATE!!!
 160         if ! validate ${dir}; then continue; fi
 161
 162         #summarizeLogs ${dir} >> ${logSummary}
 163         #logStatus=$?
 164         #if [[ ${logStatus} -ne 0 ]]; then
 165         #  echo "WARNING: run not validated: ${dir}"
 166         #  planB=1
 167         #  continue
 168         #fi
 169
 170         if [[ -d ${oldRunDir} ]]; then
 171           echo "removing old ${period}/${pass}/${runNumber}"
 172           rm -rf ${oldRunDir}
 173         fi
 174         echo "moving new ${runNumber} to ${productionDir}"
 175         mv -f ${dir} ${productionDir}
 176       done
 177
 178       echo running ${detector} periodLevelQA for production ${period}/${pass}
 179       rm -f trending.root
 180       hadd trending.root 000*/trending.root &> periodLevelQA.log
 181       periodLevelQA trending.root &>> periodLevelQA.log
 182
 183       if ! validate ${PWD}; then continue; fi
 184       #summarizeLogs ${PWD} >> ${logSummary}
 185       #logStatus=$?
 186       #if [[ ${logStatus} -ne 0 ]]; then
 187       #  echo "WARNING: period ${period} not validated: ${dir}"
 188       #  planB=1
 189       #  continue
 190       #fi
 191
 192       cd ${tmpRunDir}
 193
 194     done
 195
 196     cd ${workingDirectory}
 197
 198     if [[ -z ${planB} ]]; then
 199       echo
 200       echo removing ${tmpRunDir}
 201       rm -rf ${tmpRunDir}
 202     else
 203       executePlanB
 204     fi
 205   done
 206
 207   #remove lock
 208   rm -f ${lockFile}
 209 }
 210
 211 executePlanB()
 212 {
 213   #in case of emergency
 214   if [[ -n ${MAILTO} ]]; then
 215     echo
 216     echo "trouble detected, sending email to ${MAILTO}"
 217
 218     cat ${logSummary} | mail -s "qa in need of assistance" ${MAILTO}
 219   fi
 220 }
 221
 222 validate()
 223 {
 224   summarizeLogs ${1} >> ${logSummary}
 225   logStatus=$?
 226   if [[ ${logStatus} -ne 0 ]]; then
 227     echo "WARNING not validated: ${1}"
 228     planB=1
 229     return 1
 230   fi
 231   return 0
 232 }
 233
 234 summarizeLogs()
 235 {
 236   local dir=$1
 237   [[ -z ${dir} ]] && dir=${PWD}
 238
 239   #print a summary of logs
 240   logFiles=(
 241       "*.log"
 242       "stdout"
 243       "stderr"
 244   )
 245
 246   #check logs
 247   local logstatus=0
 248   for log in ${dir}/${logFiles[*]}; do
 249     finallog=${PWD%/}/${log}
 250     [[ ! -f ${log} ]] && continue
 251     errorSummary=$(validateLog ${log})
 252     validationStatus=$?
 253     [[ validationStatus -ne 0 ]] && logstatus=1
 254     if [[ ${validationStatus} -eq 0 ]]; then
 255       #in pretend mode randomly report an error in rec.log some cases
 256       if [[ -n ${pretend} && "${log}" == "rec.log" ]]; then
 257         [[ $(( ${RANDOM}%2 )) -ge 1 ]] && echo "${finallog} BAD random error" || echo "${finallog} OK"
 258       else
 259         echo "${finallog} OK"
 260       fi
 261     elif [[ ${validationStatus} -eq 1 ]]; then
 262       echo "${finallog} BAD ${errorSummary}"
 263     elif [[ ${validationStatus} -eq 2 ]]; then
 264       echo "${finallog} OK MWAH ${errorSummary}"
 265     fi
 266   done
 267
 268   #report core files
 269   while read x; do
 270     echo ${x}
 271     chmod 644 ${x}
 272     gdb --batch --quiet -ex "bt" -ex "quit" aliroot ${x} > stacktrace_${x//\//_}.log
 273   done < <(/bin/ls ${PWD}/*/core 2>/dev/null; /bin/ls ${PWD}/core 2>/dev/null)
 274
 275   return ${logstatus}
 276 }
 277
 278 validateLog()
 279 {
 280   log=${1}
 281   errorConditions=(
 282             'There was a crash'
 283             'floating'
 284             'error while loading shared libraries'
 285             'std::bad_alloc'
 286             's_err_syswatch_'
 287             'Thread [0-9]* (Thread'
 288             'AliFatal'
 289             'core dumped'
 290             '\.C.*error:.*\.h: No such file'
 291             'segmentation'
 292   )
 293
 294   warningConditions=(
 295             'This is serious'
 296   )
 297
 298   local logstatus=0
 299   local errorSummary=""
 300   local warningSummary=""
 301
 302   for ((i=0; i<${#errorConditions[@]};i++)); do
 303     local tmp=$(grep -m1 -e "${errorConditions[${i}]}" ${log})
 304     [[ -n ${tmp} ]] && tmp+=" : "
 305     errorSummary+=${tmp}
 306   done
 307
 308   for ((i=0; i<${#warningConditions[@]};i++)); do
 309     local tmp=$(grep -m1 -e "${warningConditions[${i}]}" ${log})
 310     [[ -n ${tmp} ]] && tmp+=" : "
 311     warningSummary+=${tmp}
 312   done
 313
 314   if [[ -n ${errorSummary} ]]; then
 315     echo "${errorSummary}"
 316     return 1
 317   fi
 318
 319   if [[ -n ${warningSummary} ]]; then
 320     echo "${warningSummary}"
 321     return 2
 322   fi
 323
 324   return 0
 325 }
 326
 327 parseConfig()
 328 {
 329   #config file
 330   configFile=""
 331   #where to search for qa files
 332   inputList=file.list
 333   #working directory
 334   workingDirectory="${PWD}"
 335   #where to place the final qa plots
 336   #outputDirectory="/afs/cern.ch/work/a/aliqa%det/www/"
 337   outputDirectory="${workingDirectory}/%DET"
 338   #filter out detector option
 339   excludeDetectors="EXAMPLE"
 340   #logs
 341   logDirectory=${workingDirectory}/logs
 342   #set aliroot
 343   #alirootEnv="/home/mkrzewic/alisoft/balice_master.sh"
 344   #OCDB storage
 345   #ocdbStorage="raw://"
 346   #email to
 347   #MAILTO="fbellini@cern.ch"
 348
 349   #first, check if the config file is configured
 350   #is yes - source it so that other options can override it
 351   #if any
 352   for opt in $@; do
 353     if [[ ${opt} =~ configFile=.* ]]; then
 354       eval "${opt}"
 355       configFile=$(readlink -f ${configFile})
 356       source "${configFile}"
 357       break
 358     fi
 359   done
 360
 361   #then, parse the options as they override the options from file
 362   while [[ -n ${1} ]]; do
 363     local var=${1#--}
 364     if [[ ${var} =~ .*=.* ]]; then
 365       eval "${var}"
 366     else
 367       echo "badly formatted option ${var}, should be: option=value, stopping..."
 368       return 1
 369     fi
 370     shift
 371   done
 372 }
 373
 374 guessRunData()
 375 {
 376   #guess the period from the path, pick the rightmost one
 377   period=""
 378   runNumber=""
 379   year=""
 380   pass=""
 381   legoTrainRunNumber=""
 382   dataType=""
 383
 384   local shortRunNumber=""
 385   local IFS="/"
 386   declare -a path=( $1 )
 387   local dirDepth=$(( ${#path[*]}-1 ))
 388   i=0
 389   for ((x=${dirDepth};x>=0;x--)); do
 390
 391     [[ $((x-1)) -ge 0 ]] && local fieldPrev=${path[$((x-1))]}
 392     local field=${path[${x}]}
 393     local fieldNext=${path[$((x+1))]}
 394
 395     [[ ${field} =~ ^[0-9]*$ && ${fieldNext} =~ (.*\.zip$|.*\.root$) ]] && legoTrainRunNumber=${field}
 396     [[ -n ${legoTrainRunNumber} && -z ${pass} ]] && pass=${fieldPrev}
 397     [[ ${field} =~ ^LHC[0-9][0-9][a-z].*$ ]] && period=${field%_*}
 398     [[ ${field} =~ ^000[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && runNumber=${field#000}
 399     [[ ${field} =~ ^[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && shortRunNumber=${field}
 400     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
 401     [[ ${field} =~ ^(^sim$|^data$) ]] && dataType=${field}
 402     (( i++ ))
 403   done
 404   [[ -z ${legoTrainRunNumber} ]] && pass=${path[$((dirDepth-1))]}
 405   [[ "${dataType}" =~ ^sim$ ]] && pass="passMC" && runNumber=${shortRunNumber}
 406
 407   if [[ -z ${dataType} || -z ${year} || -z ${period} || -z ${runNumber}} || -z ${pass} ]];
 408   then
 409     #error condition
 410     return 1
 411   else
 412     #ALL OK
 413     return 0
 414   fi
 415 }
 416
 417 substituteDetectorName()
 418 {
 419   local det=$1
 420   local dir=$2
 421   [[ ${dir} =~ \%det ]] && det=${det,,} && echo ${dir/\%det/${det}}
 422   [[ ${dir} =~ \%DET ]] && det=${det} && echo ${dir/\%DET/${det}}
 423 }
 424
 425 main $@