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