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     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} ${outputDirectory})
  89     tmpDetectorRunDir=${workingDirectory}/tmpQAtmpRunDir${detector}
  90     if ! mkdir -p ${tmpDetectorRunDir}; then
  91       echo "cannot create the temp dir $tmpDetectorRunDir"
  92       continue
  93     fi
  94     cd ${tmpDetectorRunDir}
  95
  96     tmpPrefix=${tmpDetectorRunDir}/${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     unset arrOfTouchedProductions
 110     declare -A arrOfTouchedProductions
 111     while read qaFile; do
 112       echo
 113
 114       if ! guessRunData ${qaFile}; then
 115         echo "could not guess run data from ${qaFile}"
 116         continue
 117       fi
 118
 119       tmpProductionDir=${tmpPrefix}/${dataType}/${year}/${period}/${pass}
 120       arrOfTouchedProductions[${tmpProductionDir}]=1
 121       tmpRunDir=${tmpProductionDir}/000${runNumber}
 122       mkdir -p ${tmpRunDir}
 123       cd ${tmpRunDir}
 124
 125       #handle the case of a zip archive
 126       [[ "$qaFile" =~ .*.zip$ ]] && qaFile="${qaFile}#QAresults.root"
 127
 128       echo running ${detector} runLevelQA for run ${runNumber} from ${qaFile}
 129       runLevelQA ${qaFile} &> runLevelQA.log
 130
 131       #perform some default actions:
 132       #if trending.root not created, create a default one
 133       if [[ ! -f trending.root ]]; then
 134         aliroot -b -q -l "$ALICE_ROOT/PWGPP/macros/simpleTrending.C(\"${qaFile}\",${runNumber},\"${detector}\",\"trending.root\",\"trending\",\"recreate\")" &>> runLevelQA.log
 135       fi
 136       cd ${tmpDetectorRunDir}
 137
 138     done < ${inputList}
 139
 140     #################################################################
 141     #cache which productions were (re)done
 142     echo "list of processed productions:"
 143     echo "    ${!arrOfTouchedProductions[@]}"
 144     echo
 145
 146     #################################################################
 147     #(re)do the merging/trending in the final destination
 148     for tmpProductionDir in ${!arrOfTouchedProductions[@]}; do
 149       echo
 150       echo "running period level stuff in ${tmpProductionDir}"
 151
 152       productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 153
 154       mkdir -p ${productionDir}
 155       if [[ ! -d ${productionDir} ]]; then
 156         echo "cannot make productionDir $productionDir" && continue
 157       fi
 158       cd ${productionDir}
 159
 160       #move to final destination
 161       for dir in ${tmpProductionDir}/*; do
 162         oldRunDir=${outputDir}/${dir#${tmpPrefix}}
 163         if ! guessRunData "${dir}/dummyName"; then
 164           echo "could not guess run data from ${dir}"
 165           continue
 166         fi
 167
 168         #before moving - VALIDATE!!!
 169         if ! validate ${dir}; then continue; fi
 170
 171         if [[ -d ${oldRunDir} ]]; then
 172           echo "removing old ${oldRunDir}"
 173           rm -rf ${oldRunDir}
 174         fi
 175         echo "moving new ${runNumber} to ${productionDir}"
 176         mv -f ${dir} ${productionDir}
 177       done
 178
 179       echo running ${detector} periodLevelQA for production ${period}/${pass}
 180       rm -f trending.root
 181
 182       #merge trending files if any
 183       if /bin/ls 000*/trending.root &>/dev/null; then
 184         hadd trending.root 000*/trending.root &> periodLevelQA.log
 185         periodLevelQA trending.root &>> periodLevelQA.log
 186       fi
 187
 188       if ! validate ${PWD}; then continue; fi
 189
 190       cd ${tmpDetectorRunDir}
 191
 192     done
 193
 194     cd ${workingDirectory}
 195
 196     if [[ -z ${planB} ]]; then
 197       echo
 198       echo removing ${tmpDetectorRunDir}
 199       rm -rf ${tmpDetectorRunDir}
 200     else
 201       executePlanB
 202     fi
 203   done
 204
 205   #remove lock
 206   rm -f ${lockFile}
 207 }
 208
 209 executePlanB()
 210 {
 211   #in case of emergency
 212   if [[ -n ${MAILTO} ]]; then
 213     echo
 214     echo "trouble detected, sending email to ${MAILTO}"
 215
 216     cat ${logSummary} | mail -s "qa in need of assistance" ${MAILTO}
 217   fi
 218 }
 219
 220 validate()
 221 {
 222   summarizeLogs ${1} >> ${logSummary}
 223   logStatus=$?
 224   if [[ ${logStatus} -ne 0 ]]; then
 225     echo "WARNING not validated: ${1}"
 226     planB=1
 227     return 1
 228   fi
 229   return 0
 230 }
 231
 232 summarizeLogs()
 233 {
 234   local dir=$1
 235   [[ ! -d ${dir} ]] && dir=${PWD}
 236
 237   #print a summary of logs
 238   logFiles=(
 239       "*.log"
 240       "stdout"
 241       "stderr"
 242   )
 243
 244   #check logs
 245   local logstatus=0
 246   for log in ${dir}/${logFiles[*]}; do
 247     finallog=${PWD%/}/${log}
 248     [[ ! -f ${log} ]] && continue
 249     errorSummary=$(validateLog ${log})
 250     validationStatus=$?
 251     [[ validationStatus -ne 0 ]] && logstatus=1
 252     if [[ ${validationStatus} -eq 0 ]]; then
 253       #in pretend mode randomly report an error in rec.log some cases
 254       if [[ -n ${pretend} && "${log}" == "rec.log" ]]; then
 255         [[ $(( ${RANDOM}%2 )) -ge 1 ]] && echo "${finallog} BAD random error" || echo "${finallog} OK"
 256       else
 257         echo "${finallog} OK"
 258       fi
 259     elif [[ ${validationStatus} -eq 1 ]]; then
 260       echo "${finallog} BAD ${errorSummary}"
 261     elif [[ ${validationStatus} -eq 2 ]]; then
 262       echo "${finallog} OK MWAH ${errorSummary}"
 263     fi
 264   done
 265
 266   #report core files
 267   while read x; do
 268     echo ${x}
 269     chmod 644 ${x}
 270     gdb --batch --quiet -ex "bt" -ex "quit" aliroot ${x} > stacktrace_${x//\//_}.log
 271   done < <(/bin/ls ${PWD}/*/core 2>/dev/null; /bin/ls ${PWD}/core 2>/dev/null)
 272
 273   return ${logstatus}
 274 }
 275
 276 validateLog()
 277 {
 278   log=${1}
 279   errorConditions=(
 280             'There was a crash'
 281             'floating'
 282             'error while loading shared libraries'
 283             'std::bad_alloc'
 284             's_err_syswatch_'
 285             'Thread [0-9]* (Thread'
 286             'AliFatal'
 287             'core dumped'
 288             '\.C.*error:.*\.h: No such file'
 289             'segmentation'
 290   )
 291
 292   warningConditions=(
 293             'This is serious'
 294   )
 295
 296   local logstatus=0
 297   local errorSummary=""
 298   local warningSummary=""
 299
 300   for ((i=0; i<${#errorConditions[@]};i++)); do
 301     local tmp=$(grep -m1 -e "${errorConditions[${i}]}" ${log})
 302     [[ -n ${tmp} ]] && tmp+=" : "
 303     errorSummary+=${tmp}
 304   done
 305
 306   for ((i=0; i<${#warningConditions[@]};i++)); do
 307     local tmp=$(grep -m1 -e "${warningConditions[${i}]}" ${log})
 308     [[ -n ${tmp} ]] && tmp+=" : "
 309     warningSummary+=${tmp}
 310   done
 311
 312   if [[ -n ${errorSummary} ]]; then
 313     echo "${errorSummary}"
 314     return 1
 315   fi
 316
 317   if [[ -n ${warningSummary} ]]; then
 318     echo "${warningSummary}"
 319     return 2
 320   fi
 321
 322   return 0
 323 }
 324
 325 parseConfig()
 326 {
 327   #config file
 328   configFile=""
 329   #where to search for qa files
 330   inputList=file.list
 331   #working directory
 332   workingDirectory="${PWD}"
 333   #where to place the final qa plots
 334   #outputDirectory="/afs/cern.ch/work/a/aliqa%det/www/"
 335   outputDirectory="${workingDirectory}/%DET"
 336   #filter out detector option
 337   excludeDetectors="EXAMPLE"
 338   #logs
 339   logDirectory=${workingDirectory}/logs
 340   #set aliroot
 341   #alirootEnv="/home/mkrzewic/alisoft/balice_master.sh"
 342   #OCDB storage
 343   #ocdbStorage="raw://"
 344   #email to
 345   #MAILTO="fbellini@cern.ch"
 346
 347   #first, check if the config file is configured
 348   #is yes - source it so that other options can override it
 349   #if any
 350   for opt in $@; do
 351     if [[ ${opt} =~ configFile=.* ]]; then
 352       eval "${opt}"
 353       configFile=$(readlink -f ${configFile})
 354       source "${configFile}"
 355       break
 356     fi
 357   done
 358
 359   #then, parse the options as they override the options from file
 360   while [[ -n ${1} ]]; do
 361     local var=${1#--}
 362     if [[ ${var} =~ .*=.* ]]; then
 363       eval "${var}"
 364     else
 365       echo "badly formatted option ${var}, should be: option=value, stopping..."
 366       return 1
 367     fi
 368     shift
 369   done
 370 }
 371
 372 guessRunData()
 373 {
 374   #guess the period from the path, pick the rightmost one
 375   period=""
 376   runNumber=""
 377   year=""
 378   pass=""
 379   legoTrainRunNumber=""
 380   dataType=""
 381
 382   local shortRunNumber=""
 383   local IFS="/"
 384   declare -a path=( $1 )
 385   local dirDepth=$(( ${#path[*]}-1 ))
 386   i=0
 387   for ((x=${dirDepth};x>=0;x--)); do
 388
 389     [[ $((x-1)) -ge 0 ]] && local fieldPrev=${path[$((x-1))]}
 390     local field=${path[${x}]}
 391     local fieldNext=${path[$((x+1))]}
 392
 393     [[ ${field} =~ ^[0-9]*$ && ${fieldNext} =~ (.*\.zip$|.*\.root$) ]] && legoTrainRunNumber=${field}
 394     [[ -n ${legoTrainRunNumber} && -z ${pass} ]] && pass=${fieldPrev}
 395     [[ ${field} =~ ^LHC[0-9][0-9][a-z].*$ ]] && period=${field%_*}
 396     [[ ${field} =~ ^000[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && runNumber=${field#000}
 397     [[ ${field} =~ ^[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && shortRunNumber=${field}
 398     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
 399     [[ ${field} =~ ^(^sim$|^data$) ]] && dataType=${field}
 400     (( i++ ))
 401   done
 402   [[ -z ${legoTrainRunNumber} ]] && pass=${path[$((dirDepth-1))]}
 403   [[ "${dataType}" =~ ^sim$ ]] && pass="passMC" && runNumber=${shortRunNumber}
 404
 405   #if [[ -z ${dataType} || -z ${year} || -z ${period} || -z ${runNumber}} || -z ${pass} ]];
 406   if [[ -z ${runNumber}} ]];
 407   then
 408     #error condition
 409     return 1
 410   else
 411     #ALL OK
 412     return 0
 413   fi
 414 }
 415
 416 substituteDetectorName()
 417 {
 418   local det=$1
 419   local dir=$2
 420   [[ ${dir} =~ \%det ]] && det=${det,,} && echo ${dir/\%det/${det}}
 421   [[ ${dir} =~ \%DET ]] && det=${det} && echo ${dir/\%DET/${det}}
 422 }
 423
 424 main $@