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=$highPtTree
 155       echo ocdbStorage=${ocdbStorage}
 156       echo
 157
 158       #what if we have a zip archive?
 159       if [[ "$qaFile" =~ .*.zip$ ]]; then
 160         if unzip -l ${qaFile} | egrep "QAresults.root" &>/dev/null; then
 161           qaFile="${qaFile}#QAresults.root"
 162         else
 163           qaFile=""
 164         fi
 165       fi
 166       if [[ "$highPtTree" =~ .*.zip$ ]]; then
 167         if unzip -l ${highPtTree} | egrep "FilterEvents_Trees.root" &>/dev/null; then
 168           highPtTree="${highPtTree}#FilterEvents_Trees.root"
 169         else
 170           highPtTree=""
 171         fi
 172       fi
 173
 174       if [[ -n ${qaFile} && $(type -t runLevelQA) =~ "function" ]]; then
 175         echo running ${detector} runLevelQA for run ${runNumber} from ${qaFile}
 176         runLevelQA "${qaFile}" &> runLevelQA.log
 177         #perform some default actions:
 178         #if trending.root not created, create a default one
 179         if [[ ! -f trending.root ]]; then
 180           aliroot -b -q -l "$ALICE_ROOT/PWGPP/macros/simpleTrending.C(\"${qaFile}\",${runNumber},\"${detector}\",\"trending.root\",\"trending\",\"recreate\")" 2>&1 | tee -a runLevelQA.log
 181         fi
 182         if [[ -f trending.root ]]; then
 183           arrOfTouchedProductions[${tmpProductionDir}]=1
 184         else
 185           echo "trending.root not created"
 186         fi
 187       fi
 188       #expert QA based on high pt trees
 189       if [[ -n ${highPtTree} && $(type -t runLevelHighPtTreeQA) =~ "function" ]]; then
 190         echo running ${detector} runLevelHighPtTreeQA for run ${runNumber} from ${highPtTree}
 191         runLevelHighPtTreeQA "${highPtTree}" &> runLevelHighPtTreeQA.log
 192         arrOfTouchedProductions[${tmpProductionDir}]=1
 193       fi
 194
 195       cd ${tmpDetectorRunDir}
 196
 197     done < ${inputList}
 198
 199     #################################################################
 200     #cache which productions were (re)done
 201     echo "list of processed productions:"
 202     echo "    ${!arrOfTouchedProductions[@]}"
 203     echo
 204
 205     #################################################################
 206     #(re)do the merging/trending
 207     for tmpProductionDir in ${!arrOfTouchedProductions[@]}; do
 208       cd ${tmpProductionDir}
 209       echo
 210       echo "running period level stuff in ${tmpProductionDir}"
 211       echo $(date)
 212
 213       productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 214       echo productionDir=${outputDir}/${tmpProductionDir#${tmpPrefix}}
 215
 216       mkdir -p ${productionDir}
 217       if [[ ! -d ${productionDir} ]]; then
 218         echo "cannot make productionDir $productionDir" && continue
 219       fi
 220
 221       #move runs to final destination
 222       for dir in ${tmpProductionDir}/000*; do
 223         echo
 224         oldRunDir=${outputDir}/${dir#${tmpPrefix}}
 225         if ! guessRunData "${dir}/dummyName"; then
 226           echo "could not guess run data from ${dir}"
 227           continue
 228         fi
 229
 230         #before moving - VALIDATE!!!
 231         if ! validate ${dir}; then
 232           continue
 233         fi
 234
 235         #moving a dir is an atomic operation, no locking necessary
 236         if [[ -d ${oldRunDir} ]]; then
 237           echo "removing old ${oldRunDir}"
 238           rm -rf ${oldRunDir}
 239         fi
 240         echo "moving new ${runNumber} to ${productionDir}"
 241         mv -f ${dir} ${productionDir}
 242       done
 243
 244       #go to a temp dir to do the period level stuff in a completely clean dir
 245       tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
 246       echo
 247       echo tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
 248       if ! mkdir -p ${tmpPeriodLevelQAdir}; then continue; fi
 249       cd ${tmpPeriodLevelQAdir}
 250
 251       #link the final list of per-run dirs here, just the dirs
 252       #to have a clean working directory
 253       unset linkedStuff
 254       declare -a linkedStuff
 255       for x in ${productionDir}/000*; do [[ -d $x ]] && ln -s $x && linkedStuff+=(${x##*/}); done
 256
 257       #merge trending files if any
 258       if /bin/ls 000*/trending.root &>/dev/null; then
 259         hadd trending.root 000*/trending.root &> periodLevelQA.log
 260       fi
 261
 262       #run the period level trending/QA
 263       if [[ -f "trending.root" && $(type -t periodLevelQA) =~ "function" ]]; then
 264         echo running ${detector} periodLevelQA for production ${period}/${pass}
 265         periodLevelQA trending.root &>> periodLevelQA.log
 266       else
 267         echo "WARNING: not running ${detector} periodLevelQA for production ${period}/${pass}, no trending.root"
 268       fi
 269
 270       if ! validate ${PWD}; then continue; fi
 271
 272       #here we are validated so move the produced QA to the final place
 273       #clean up linked stuff first
 274       [[ -n ${linkedStuff[@]} ]] && rm ${linkedStuff[@]}
 275       periodLevelLock=${productionDir}/runQA.lock
 276       if [[ ! -f ${periodLevelLock} ]]; then
 277         #some of the output could be a directory, so handle that
 278         #TODO: maybe use rsync?
 279         #lock to avoid conflicts:
 280         echo "${HOSTNAME} ${dateString}" > ${periodLevelLock}
 281         for x in ${tmpPeriodLevelQAdir}/*; do
 282           if [[ -d ${x} ]]; then
 283             echo "removing ${productionDir}/${x##*/}"
 284             rm -rf ${productionDir}/${x##*/}
 285             echo "moving ${x} to ${productionDir}"
 286             mv ${x} ${productionDir}
 287           fi
 288           if [[ -f ${x} ]]; then
 289             echo "moving ${x} to ${productionDir}"
 290             mv -f ${x} ${productionDir}
 291           fi
 292         done
 293         rm -f ${periodLevelLock}
 294         #remove the temp dir
 295         rm -rf ${tmpPeriodLevelQAdir}
 296       else
 297         echo "locked! cannot move to destination"
 298         echo "check and maybe manually do:"
 299         echo " rm ${periodLevelLock}"
 300         echo " rsync -av ${tmpPeriodLevelQAdir}/ ${productionDir}/" >> ${logSummary}
 301         planB=1
 302       fi
 303
 304     done
 305
 306     cd ${workingDirectory}
 307
 308     if [[ -z ${planB} ]]; then
 309       echo
 310       echo removing ${tmpDetectorRunDir}
 311       rm -rf ${tmpDetectorRunDir}
 312     else
 313       executePlanB
 314     fi
 315   done #end of detector loop
 316
 317   #remove lock
 318   rm -f ${lockFile}
 319 }
 320
 321 executePlanB()
 322 {
 323   #in case of emergency
 324   if [[ -n ${MAILTO} ]]; then
 325     echo
 326     echo "trouble detected, sending email to ${MAILTO}"
 327
 328     grep BAD ${logSummary} | mail -s "qa in need of assistance" ${MAILTO}
 329   fi
 330 }
 331
 332 validate()
 333 {
 334   summarizeLogs ${1} >> ${logSummary}
 335   logStatus=$?
 336   if [[ ${logStatus} -ne 0 ]]; then
 337     echo "WARNING not validated: ${1}"
 338     planB=1
 339     return 1
 340   fi
 341   return 0
 342 }
 343
 344 summarizeLogs()
 345 {
 346   local dir=$1
 347   [[ ! -d ${dir} ]] && dir=${PWD}
 348
 349   #print a summary of logs
 350   logFiles=(
 351       "*.log"
 352       "stdout"
 353       "stderr"
 354   )
 355
 356   #check logs
 357   local logstatus=0
 358   for log in ${dir}/${logFiles[*]}; do
 359     finallog=${PWD%/}/${log}
 360     [[ ! -f ${log} ]] && continue
 361     errorSummary=$(validateLog ${log})
 362     validationStatus=$?
 363     [[ validationStatus -ne 0 ]] && logstatus=1
 364     if [[ ${validationStatus} -eq 0 ]]; then
 365       #in pretend mode randomly report an error in rec.log some cases
 366       if [[ -n ${pretend} && "${log}" == "rec.log" ]]; then
 367         [[ $(( ${RANDOM}%2 )) -ge 1 ]] && echo "${finallog} BAD random error" || echo "${finallog} OK"
 368       else
 369         echo "${finallog} OK"
 370       fi
 371     elif [[ ${validationStatus} -eq 1 ]]; then
 372       echo "${finallog} BAD ${errorSummary}"
 373     elif [[ ${validationStatus} -eq 2 ]]; then
 374       echo "${finallog} OK MWAH ${errorSummary}"
 375     fi
 376   done
 377
 378   #report core files
 379   while read x; do
 380     echo ${x}
 381     chmod 644 ${x}
 382     gdb --batch --quiet -ex "bt" -ex "quit" aliroot ${x} > stacktrace_${x//\//_}.log
 383   done < <(/bin/ls ${PWD}/*/core 2>/dev/null; /bin/ls ${PWD}/core 2>/dev/null)
 384
 385   return ${logstatus}
 386 }
 387
 388 validateLog()
 389 {
 390   log=${1}
 391   errorConditions=(
 392             'There was a crash'
 393             'floating'
 394             'error while loading shared libraries'
 395             'std::bad_alloc'
 396             's_err_syswatch_'
 397             'Thread [0-9]* (Thread'
 398             'AliFatal'
 399             'core dumped'
 400             '\.C.*error:.*\.h: No such file'
 401             'segmentation'
 402             'Interpreter error recovered'
 403   )
 404
 405   warningConditions=(
 406             'This is serious'
 407   )
 408
 409   local logstatus=0
 410   local errorSummary=""
 411   local warningSummary=""
 412
 413   for ((i=0; i<${#errorConditions[@]};i++)); do
 414     local tmp=$(grep -m1 -e "${errorConditions[${i}]}" ${log})
 415     [[ -n ${tmp} ]] && tmp+=" : "
 416     errorSummary+=${tmp}
 417   done
 418
 419   for ((i=0; i<${#warningConditions[@]};i++)); do
 420     local tmp=$(grep -m1 -e "${warningConditions[${i}]}" ${log})
 421     [[ -n ${tmp} ]] && tmp+=" : "
 422     warningSummary+=${tmp}
 423   done
 424
 425   if [[ -n ${errorSummary} ]]; then
 426     echo "${errorSummary}"
 427     return 1
 428   fi
 429
 430   if [[ -n ${warningSummary} ]]; then
 431     echo "${warningSummary}"
 432     return 2
 433   fi
 434
 435   return 0
 436 }
 437
 438 parseConfig()
 439 {
 440   args=("$@")
 441
 442   #config file
 443   configFile=""
 444   #where to search for qa files
 445   inputList=file.list
 446   #working directory
 447   workingDirectory="${PWD}"
 448   #where to place the final qa plots
 449   #outputDirectory="/afs/cern.ch/work/a/aliqa%det/www/"
 450   outputDirectory="${workingDirectory}/%DET"
 451   #filter out detector option
 452   excludeDetectors="EXAMPLE"
 453   #logs
 454   logDirectory=${workingDirectory}/logs
 455   #OCDB storage
 456   ocdbStorage="raw://"
 457   #email to
 458   #MAILTO="fbellini@cern.ch"
 459
 460   #first, check if the config file is configured
 461   #is yes - source it so that other options can override it
 462   #if any
 463   for opt in "${args[@]}"; do
 464     if [[ ${opt} =~ configFile=.* ]]; then
 465       eval "${opt}"
 466       [[ ! -f ${configFile} ]] && echo "configFile ${configFile} not found, exiting..." && return 1
 467       echo "using config file: ${configFile}"
 468       source "${configFile}"
 469       break
 470     fi
 471   done
 472
 473   #then, parse the options as they override the options from file
 474   for opt in "${args[@]}"; do
 475     if [[ ! "${opt}" =~ .*=.* ]]; then
 476       echo "badly formatted option ${var}, should be: option=value, stopping..."
 477       return 1
 478     fi
 479     local var="${opt%%=*}"
 480     local value="${opt#*=}"
 481     echo "${var}=${value}"
 482     export ${var}="${value}"
 483   done
 484 }
 485
 486 guessRunData()
 487 {
 488   #guess the period from the path, pick the rightmost one
 489   period=""
 490   runNumber=""
 491   year=""
 492   pass=""
 493   legoTrainRunNumber=""
 494   dataType=""
 495
 496   local shortRunNumber=""
 497   oldIFS=${IFS}
 498   local IFS="/"
 499   declare -a path=( $1 )
 500   IFS="${oldIFS}"
 501   local dirDepth=$(( ${#path[*]}-1 ))
 502   i=0
 503   for ((x=${dirDepth};x>=0;x--)); do
 504
 505     [[ $((x-1)) -ge 0 ]] && local fieldPrev=${path[$((x-1))]}
 506     local field=${path[${x}]}
 507     local fieldNext=${path[$((x+1))]}
 508
 509     [[ ${field} =~ ^[0-9]*$ && ${fieldNext} =~ (.*\.zip$|.*\.root$) ]] && legoTrainRunNumber=${field}
 510     [[ -n ${legoTrainRunNumber} && -z ${pass} ]] && pass=${fieldPrev}
 511     [[ ${field} =~ ^LHC[0-9][0-9][a-z].*$ ]] && period=${field%_*}
 512     [[ ${field} =~ ^000[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && runNumber=${field#000}
 513     [[ ${field} =~ ^[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && shortRunNumber=${field}
 514     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
 515     [[ ${field} =~ ^(^sim$|^data$) ]] && dataType=${field}
 516     (( i++ ))
 517   done
 518   [[ -z ${legoTrainRunNumber} ]] && pass=${path[$((dirDepth-1))]}
 519   [[ "${dataType}" =~ ^sim$ ]] && pass="passMC" && runNumber=${shortRunNumber}
 520
 521   #modify the OCDB: set the year
 522   ocdbStorage=$(setYear ${year} ${ocdbStorage})
 523
 524   #if [[ -z ${dataType} || -z ${year} || -z ${period} || -z ${runNumber}} || -z ${pass} ]];
 525   if [[ -z ${runNumber}} ]]
 526   then
 527     #error condition
 528     return 1
 529   else
 530     #ALL OK
 531     return 0
 532   fi
 533 }
 534
 535 substituteDetectorName()
 536 {
 537   local det=$1
 538   local dir=$2
 539   [[ ${dir} =~ \%det ]] && det=${det,,} && echo ${dir/\%det/${det}}
 540   [[ ${dir} =~ \%DET ]] && det=${det} && echo ${dir/\%DET/${det}}
 541 }
 542
 543 get_realpath()
 544 {
 545   if [[ -f "$1" ]]
 546   then
 547     # file *must* exist
 548     if cd "$(echo "${1%/*}")" &>/dev/null
 549     then
 550       # file *may* not be local
 551       # exception is ./file.ext
 552       # try 'cd .; cd -;' *works!*
 553       local tmppwd="$PWD"
 554       cd - &>/dev/null
 555     else
 556       # file *must* be local
 557       local tmppwd="$PWD"
 558     fi
 559   else
 560     # file *cannot* exist
 561     return 1 # failure
 562   fi
 563   # reassemble realpath
 564   echo "$tmppwd"/"${1##*/}"
 565   return 0 # success
 566 }
 567
 568 setYear()
 569 {
 570   #set the year
 571   #  ${1} - year to be set
 572   #  ${2} - where to set the year
 573   local year1=$(guessYear ${1})
 574   local year2=$(guessYear ${2})
 575   local path=${2}
 576   [[ ${year1} -ne ${year2} && -n ${year2} && -n ${year1} ]] && path=${2/\/${year2}\//\/${year1}\/}
 577   echo ${path}
 578   return 0
 579 }
 580
 581 guessYear()
 582 {
 583   #guess the year from the path, pick the rightmost one
 584   local IFS="/"
 585   declare -a pathArray=( ${1} )
 586   local field
 587   local year
 588   for field in ${pathArray[@]}; do
 589     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
 590   done
 591   echo ${year}
 592   return 0
 593 }
 594
 595 main "$@"