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 inputList should be specified, or configFile containing it:"
   8     echo "  ${0##*/} inputList=file.list"
   9     echo "  options override config file (if any), e.g.:"
  10     echo "  ${0##*/} configFile=runQA.config inputList=file.list outputDirectory=%det"
  11     return 1
  12   fi
  13
  14   if ! parseConfig $@; then
  15     ${0}
  16     return 1
  17   fi
  18
  19   [[ -z $ALICE_ROOT ]] && echo "ALICE_ROOT not defined" && return 1
  20
  21   ocdbregex='raw://'
  22   if [[ ${ocdbStorage} =~ ${ocdbregex} ]]; then
  23     alien-token-init ${alienUserName}
  24     #this is a hack! alien-token init seems not enough
  25     #but the gclient_env script messes up the LD_LIBRARY_PATH
  26     while read x; do
  27       eval ${x};
  28     done < <(grep -v "LD_LIBRARY_PATH" /tmp/gclient_env_${UID})
  29   fi
  30
  31   updateQA $@
  32 }
  33
  34 updateQA()
  35 {
  36   umask 0002
  37   parseConfig $@
  38
  39   #be paranoid and make some full paths
  40   [[ ! -f ${inputList} ]] && echo "no input list: ${inputList}" && return 1
  41   inputList=$(get_realpath ${inputList})
  42   mkdir -p ${workingDirectory}
  43   workingDirectory=$(workingDirectory=${workingDirectory%/}; cd ${workingDirectory%/*}; echo "${PWD}/${workingDirectory##*/}")
  44   if [[ ! -d ${workingDirectory} ]]; then
  45     echo "working dir $workingDirectory does not exist and cannot be created"
  46     return 1
  47   fi
  48   cd ${workingDirectory}
  49
  50   echo JOB config:
  51   echo inputList=$inputList
  52   echo outputDirectory=$outputDirectory
  53   echo
  54
  55   dateString=$(date +%Y-%m-%d-%H-%M)
  56   echo "Start time QA process: $dateString"
  57
  58   #logging
  59   mkdir -p $logDirectory
  60   [[ ! -d $logDirectory ]] && echo "no log dir $logDirectory" && return 1
  61   logFile="$logDirectory/${0##*/}.${dateString}.log"
  62   touch ${logFile}
  63   [[ ! -f ${logFile} ]] && echo "cannot write logfile $logfile" && return 1
  64   echo "logFile = $logFile"
  65
  66   #check lock
  67   lockFile=${logDirectory}/runQA.lock
  68   [[ -f ${lockFile} ]] && echo "lock ${lockFile} exists!" | tee ${logFile} && return 1
  69   touch ${lockFile}
  70   [[ ! -f ${lockFile} ]] && echo "cannot lock $lockFile" | tee ${logFile} && return 1
  71
  72   exec &>${logFile}
  73
  74   ################################################################
  75   #ze detector loop
  76   for detectorScript in $ALICE_ROOT/PWGPP/QA/detectorQAscripts/*; do
  77     echo
  78     echo "##############################################"
  79     echo $(date)
  80     unset planB
  81     [[ ! ${detectorScript} =~ .*\.sh$ ]] && continue
  82     detector=${detectorScript%.sh}
  83     detector=${detector##*/}
  84
  85     #skip if excluded
  86     if [[ "${excludeDetectors}" =~ ${detector} ]]; then
  87       echo "${detector} is excluded in config, skipping..."
  88       continue
  89     fi
  90
  91     #if includeDetectors set, only process thoe detectors specified there
  92     if [[ -n ${includeDetectors} && ! "${includeDetectors}" =~ ${detector} ]]; then
  93       echo "${detector} not included in includeDetectors, skipping..."
  94       continue
  95     fi
  96
  97     logSummary=${logDirectory}/summary-${detector}-${dateString}.log
  98     outputDir=$(substituteDetectorName ${detector} ${outputDirectory})
  99     tmpDetectorRunDir=${workingDirectory}/tmpQAtmpRunDir${detector}-${dateString}
 100     if ! mkdir -p ${tmpDetectorRunDir}; then
 101       echo "cannot create the temp dir $tmpDetectorRunDir"
 102       continue
 103     fi
 104     cd ${tmpDetectorRunDir}
 105
 106     tmpPrefix=${tmpDetectorRunDir}/${outputDir}
 107     echo "running QA for ${detector}"
 108     echo "  outputDir=$outputDir"
 109     echo "  tmpPrefix=$tmpPrefix"
 110
 111     #unset the detector functions from previous iterations (detectors)
 112     unset -f runLevelQA
 113     unset -f periodLevelQA
 114     unset -f runLevelHighPtTreeQA
 115     unset -f periodLevelHighPtTreeQA
 116     source ${detectorScript}
 117
 118     #################################################################
 119     #produce the QA and trending tree for each file (run)
 120     unset arrOfTouchedProductions
 121     declare -A arrOfTouchedProductions
 122     while read qaFile; do
 123       echo
 124       echo $(date)
 125
 126       #first check if input file exists
 127       [[ ! -f ${qaFile%\#*} ]] && echo "file ${qaFile%\#*} not accessible" && continue
 128
 129       if ! guessRunData ${qaFile}; then
 130         echo "could not guess run data from ${qaFile}"
 131         continue
 132       fi
 133
 134       tmpProductionDir=${tmpPrefix}/${dataType}/${year}/${period}/${pass}
 135       tmpRunDir=${tmpProductionDir}/000${runNumber}
 136       mkdir -p ${tmpRunDir}
 137       cd ${tmpRunDir}
 138
 139       #by default we expect to have everything in the same archive
 140       highPtTree=${qaFile}
 141
 142       #maybe the input is not an archive, but a file
 143       [[ "${qaFile}" =~ QAresults.root$ ]] && highPtTree=""
 144       [[ "${qaFile}" =~ FilterEvents_Trees.root$ ]] && qaFile=""
 145
 146       #it is possible we get the highPt trees from somewhere else
 147       #search the list of high pt trees for the proper run number
 148       if [[ -n ${inputListHighPtTrees} ]]; then
 149         highPtTree=$(egrep -m1 ${runNumber} ${inputListHighPtTrees})
 150         echo "loaded the highPtTree ${highPtTree} from external file ${inputListHighPtTrees}"
 151       fi
 152
 153       echo qaFile=$qaFile
 154       echo highPtTree=$highPtTree
 155
 156       #what if we have a zip archive?
 157       if [[ "$qaFile" =~ .*.zip$ ]]; then
 158         if unzip -l ${qaFile} | egrep "QAresults.root" &>/dev/null; then
 159           qaFile="${qaFile}#QAresults.root"
 160         else
 161           qaFile=""
 162         fi
 163       fi
 164       if [[ "$highPtTree" =~ .*.zip$ ]]; then
 165         if unzip -l ${highPtTree} | egrep "FilterEvents_Trees.root" &>/dev/null; then
 166           highPtTree="${highPtTree}#FilterEvents_Trees.root"
 167         else
 168           highPtTree=""
 169         fi
 170       fi
 171
 172       if [[ -n ${qaFile} && $(type -t runLevelQA) =~ "function" ]]; then
 173         echo running ${detector} runLevelQA for run ${runNumber} from ${qaFile}
 174         runLevelQA "${qaFile}" &> runLevelQA.log
 175         #perform some default actions:
 176         #if trending.root not created, create a default one
 177         if [[ ! -f trending.root ]]; then
 178           aliroot -b -q -l "$ALICE_ROOT/PWGPP/macros/simpleTrending.C(\"${qaFile}\",${runNumber},\"${detector}\",\"trending.root\",\"trending\",\"recreate\")" 2>&1 | tee -a runLevelQA.log
 179         fi
 180         if [[ -f trending.root ]]; then
 181           arrOfTouchedProductions[${tmpProductionDir}]=1
 182         else
 183           echo "trending.root not created"
 184         fi
 185       fi
 186       #expert QA based on high pt trees
 187       if [[ -n ${highPtTree} && $(type -t runLevelHighPtTreeQA) =~ "function" ]]; then
 188         echo running ${detector} runLevelHighPtTreeQA for run ${runNumber} from ${highPtTree}
 189         runLevelHighPtTreeQA "${highPtTree}" &> runLevelHighPtTreeQA.log
 190         arrOfTouchedProductions[${tmpProductionDir}]=1
 191       fi
 192
 193       cd ${tmpDetectorRunDir}
 194
 195     done < ${inputList}
 196
 197     #################################################################
 198     #cache which productions were (re)done
 199     echo "list of processed productions:"
 200     echo "    ${!arrOfTouchedProductions[@]}"
 201     echo
 202
 203     #################################################################
 204     #(re)do the merging/trending
 205     for tmpProductionDir in ${!arrOfTouchedProductions[@]}; do
 206       cd ${tmpProductionDir}
 207       echo
 208       echo "running period level stuff in ${tmpProductionDir}"
 209       echo $(date)
 210
 211       productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 212       echo productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 213
 214       mkdir -p ${productionDir}
 215       if [[ ! -d ${productionDir} ]]; then
 216         echo "cannot make productionDir $productionDir" && continue
 217       fi
 218
 219       #move runs to final destination
 220       for dir in ${tmpProductionDir}/000*; do
 221         echo
 222         oldRunDir=${outputDir}/${dir#${tmpPrefix}}
 223         if ! guessRunData "${dir}/dummyName"; then
 224           echo "could not guess run data from ${dir}"
 225           continue
 226         fi
 227
 228         #before moving - VALIDATE!!!
 229         if ! validate ${dir}; then
 230           continue
 231         fi
 232
 233         #moving a dir is an atomic operation, no locking necessary
 234         if [[ -d ${oldRunDir} ]]; then
 235           echo "removing old ${oldRunDir}"
 236           rm -rf ${oldRunDir}
 237         fi
 238         echo "moving new ${runNumber} to ${productionDir}"
 239         mv -f ${dir} ${productionDir}
 240       done
 241
 242       #go to a temp dir to do the period level stuff in a completely clean dir
 243       tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
 244       echo
 245       echo tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
 246       if ! mkdir -p ${tmpPeriodLevelQAdir}; then continue; fi
 247       cd ${tmpPeriodLevelQAdir}
 248
 249       #link the final list of per-run dirs here, just the dirs
 250       #to have a clean working directory
 251       unset linkedStuff
 252       declare -a linkedStuff
 253       for x in ${productionDir}/000*; do [[ -d $x ]] && ln -s $x && linkedStuff+=(${x##*/}); done
 254
 255       #merge trending files if any
 256       if /bin/ls 000*/trending.root &>/dev/null; then
 257         hadd trending.root 000*/trending.root &> periodLevelQA.log
 258       fi
 259
 260       #run the period level trending/QA
 261       if [[ -f "trending.root" && $(type -t periodLevelQA) =~ "function" ]]; then
 262         echo running ${detector} periodLevelQA for production ${period}/${pass}
 263         periodLevelQA trending.root &>> periodLevelQA.log
 264       else
 265         echo "WARNING: not running ${detector} periodLevelQA for production ${period}/${pass}, no trending.root"
 266       fi
 267
 268       if ! validate ${PWD}; then continue; fi
 269
 270       #here we are validated so move the produced QA to the final place
 271       #clean up linked stuff first
 272       [[ -n ${linkedStuff[@]} ]] && rm ${linkedStuff[@]}
 273       periodLevelLock=${productionDir}/runQA.lock
 274       if [[ ! -f ${periodLevelLock} ]]; then
 275         #some of the output could be a directory, so handle that
 276         #TODO: maybe use rsync?
 277         #lock to avoid conflicts:
 278         echo "${HOSTNAME} ${dateString}" > ${periodLevelLock}
 279         for x in ${tmpPeriodLevelQAdir}/*; do
 280           if [[ -d ${x} ]]; then
 281             echo "removing ${productionDir}/${x##*/}"
 282             rm -rf ${productionDir}/${x##*/}
 283             echo "moving ${x} to ${productionDir}"
 284             mv ${x} ${productionDir}
 285           fi
 286           if [[ -f ${x} ]]; then
 287             echo "moving ${x} to ${productionDir}"
 288             mv -f ${x} ${productionDir}
 289           fi
 290         done
 291         rm -f ${periodLevelLock}
 292         #remove the temp dir
 293         rm -rf ${tmpPeriodLevelQAdir}
 294       else
 295         echo "locked! cannot move to destination"
 296         echo "check and maybe manually do:"
 297         echo " rm ${periodLevelLock}"
 298         echo " rsync -av ${tmpPeriodLevelQAdir}/ ${productionDir}/"
 299       fi
 300
 301     done
 302
 303     cd ${workingDirectory}
 304
 305     if [[ -z ${planB} ]]; then
 306       echo
 307       echo removing ${tmpDetectorRunDir}
 308       rm -rf ${tmpDetectorRunDir}
 309     else
 310       executePlanB
 311     fi
 312   done #end of detector loop
 313
 314   #remove lock
 315   rm -f ${lockFile}
 316 }
 317
 318 executePlanB()
 319 {
 320   #in case of emergency
 321   if [[ -n ${MAILTO} ]]; then
 322     echo
 323     echo "trouble detected, sending email to ${MAILTO}"
 324
 325     grep BAD ${logSummary} | mail -s "qa in need of assistance" ${MAILTO}
 326   fi
 327 }
 328
 329 validate()
 330 {
 331   summarizeLogs ${1} >> ${logSummary}
 332   logStatus=$?
 333   if [[ ${logStatus} -ne 0 ]]; then
 334     echo "WARNING not validated: ${1}"
 335     planB=1
 336     return 1
 337   fi
 338   return 0
 339 }
 340
 341 summarizeLogs()
 342 {
 343   local dir=$1
 344   [[ ! -d ${dir} ]] && dir=${PWD}
 345
 346   #print a summary of logs
 347   logFiles=(
 348       "*.log"
 349       "stdout"
 350       "stderr"
 351   )
 352
 353   #check logs
 354   local logstatus=0
 355   for log in ${dir}/${logFiles[*]}; do
 356     finallog=${PWD%/}/${log}
 357     [[ ! -f ${log} ]] && continue
 358     errorSummary=$(validateLog ${log})
 359     validationStatus=$?
 360     [[ validationStatus -ne 0 ]] && logstatus=1
 361     if [[ ${validationStatus} -eq 0 ]]; then
 362       #in pretend mode randomly report an error in rec.log some cases
 363       if [[ -n ${pretend} && "${log}" == "rec.log" ]]; then
 364         [[ $(( ${RANDOM}%2 )) -ge 1 ]] && echo "${finallog} BAD random error" || echo "${finallog} OK"
 365       else
 366         echo "${finallog} OK"
 367       fi
 368     elif [[ ${validationStatus} -eq 1 ]]; then
 369       echo "${finallog} BAD ${errorSummary}"
 370     elif [[ ${validationStatus} -eq 2 ]]; then
 371       echo "${finallog} OK MWAH ${errorSummary}"
 372     fi
 373   done
 374
 375   #report core files
 376   while read x; do
 377     echo ${x}
 378     chmod 644 ${x}
 379     gdb --batch --quiet -ex "bt" -ex "quit" aliroot ${x} > stacktrace_${x//\//_}.log
 380   done < <(/bin/ls ${PWD}/*/core 2>/dev/null; /bin/ls ${PWD}/core 2>/dev/null)
 381
 382   return ${logstatus}
 383 }
 384
 385 validateLog()
 386 {
 387   log=${1}
 388   errorConditions=(
 389             'There was a crash'
 390             'floating'
 391             'error while loading shared libraries'
 392             'std::bad_alloc'
 393             's_err_syswatch_'
 394             'Thread [0-9]* (Thread'
 395             'AliFatal'
 396             'core dumped'
 397             '\.C.*error:.*\.h: No such file'
 398             'segmentation'
 399             'Interpreter error recovered'
 400   )
 401
 402   warningConditions=(
 403             'This is serious'
 404   )
 405
 406   local logstatus=0
 407   local errorSummary=""
 408   local warningSummary=""
 409
 410   for ((i=0; i<${#errorConditions[@]};i++)); do
 411     local tmp=$(grep -m1 -e "${errorConditions[${i}]}" ${log})
 412     [[ -n ${tmp} ]] && tmp+=" : "
 413     errorSummary+=${tmp}
 414   done
 415
 416   for ((i=0; i<${#warningConditions[@]};i++)); do
 417     local tmp=$(grep -m1 -e "${warningConditions[${i}]}" ${log})
 418     [[ -n ${tmp} ]] && tmp+=" : "
 419     warningSummary+=${tmp}
 420   done
 421
 422   if [[ -n ${errorSummary} ]]; then
 423     echo "${errorSummary}"
 424     return 1
 425   fi
 426
 427   if [[ -n ${warningSummary} ]]; then
 428     echo "${warningSummary}"
 429     return 2
 430   fi
 431
 432   return 0
 433 }
 434
 435 parseConfig()
 436 {
 437   args=("$@")
 438
 439   #config file
 440   configFile=""
 441   #where to search for qa files
 442   inputList=file.list
 443   #working directory
 444   workingDirectory="${PWD}"
 445   #where to place the final qa plots
 446   #outputDirectory="/afs/cern.ch/work/a/aliqa%det/www/"
 447   outputDirectory="${workingDirectory}/%DET"
 448   #filter out detector option
 449   excludeDetectors="EXAMPLE"
 450   #logs
 451   logDirectory=${workingDirectory}/logs
 452   #OCDB storage
 453   ocdbStorage="raw://"
 454   #email to
 455   #MAILTO="fbellini@cern.ch"
 456
 457   #first, check if the config file is configured
 458   #is yes - source it so that other options can override it
 459   #if any
 460   for opt in "${args[@]}"; do
 461     if [[ ${opt} =~ configFile=.* ]]; then
 462       eval "${opt}"
 463       [[ ! -f ${configFile} ]] && echo "configFile ${configFile} not found, exiting..." && return 1
 464       echo "using config file: ${configFile}"
 465       source "${configFile}"
 466       break
 467     fi
 468   done
 469
 470   #then, parse the options as they override the options from file
 471   for opt in "${args[@]}"; do
 472     if [[ ! "${opt}" =~ .*=.* ]]; then
 473       echo "badly formatted option ${var}, should be: option=value, stopping..."
 474       return 1
 475     fi
 476     local var="${opt%%=*}"
 477     local value="${opt#*=}"
 478     echo "${var} = ${value}"
 479     export ${var}="${value}"
 480   done
 481 }
 482
 483 guessRunData()
 484 {
 485   #guess the period from the path, pick the rightmost one
 486   period=""
 487   runNumber=""
 488   year=""
 489   pass=""
 490   legoTrainRunNumber=""
 491   dataType=""
 492
 493   local shortRunNumber=""
 494   local IFS="/"
 495   declare -a path=( $1 )
 496   local dirDepth=$(( ${#path[*]}-1 ))
 497   i=0
 498   for ((x=${dirDepth};x>=0;x--)); do
 499
 500     [[ $((x-1)) -ge 0 ]] && local fieldPrev=${path[$((x-1))]}
 501     local field=${path[${x}]}
 502     local fieldNext=${path[$((x+1))]}
 503
 504     [[ ${field} =~ ^[0-9]*$ && ${fieldNext} =~ (.*\.zip$|.*\.root$) ]] && legoTrainRunNumber=${field}
 505     [[ -n ${legoTrainRunNumber} && -z ${pass} ]] && pass=${fieldPrev}
 506     [[ ${field} =~ ^LHC[0-9][0-9][a-z].*$ ]] && period=${field%_*}
 507     [[ ${field} =~ ^000[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && runNumber=${field#000}
 508     [[ ${field} =~ ^[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && shortRunNumber=${field}
 509     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
 510     [[ ${field} =~ ^(^sim$|^data$) ]] && dataType=${field}
 511     (( i++ ))
 512   done
 513   [[ -z ${legoTrainRunNumber} ]] && pass=${path[$((dirDepth-1))]}
 514   [[ "${dataType}" =~ ^sim$ ]] && pass="passMC" && runNumber=${shortRunNumber}
 515
 516   #if [[ -z ${dataType} || -z ${year} || -z ${period} || -z ${runNumber}} || -z ${pass} ]];
 517   if [[ -z ${runNumber}} ]];
 518   then
 519     #error condition
 520     return 1
 521   else
 522     #ALL OK
 523     return 0
 524   fi
 525 }
 526
 527 substituteDetectorName()
 528 {
 529   local det=$1
 530   local dir=$2
 531   [[ ${dir} =~ \%det ]] && det=${det,,} && echo ${dir/\%det/${det}}
 532   [[ ${dir} =~ \%DET ]] && det=${det} && echo ${dir/\%DET/${det}}
 533 }
 534
 535 get_realpath()
 536 {
 537   if [[ -f "$1" ]]
 538   then
 539     # file *must* exist
 540     if cd "$(echo "${1%/*}")" &>/dev/null
 541     then
 542       # file *may* not be local
 543       # exception is ./file.ext
 544       # try 'cd .; cd -;' *works!*
 545       local tmppwd="$PWD"
 546       cd - &>/dev/null
 547     else
 548       # file *must* be local
 549       local tmppwd="$PWD"
 550     fi
 551   else
 552     # file *cannot* exist
 553     return 1 # failure
 554   fi
 555   # reassemble realpath
 556   echo "$tmppwd"/"${1##*/}"
 557   return 0 # success
 558 }
 559
 560 main $@