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