handle bash args properly using arrays
[u/mrichter/AliRoot.git] / 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         if [[ -d ${oldRunDir} ]]; then
234           echo "removing old ${oldRunDir}"
235           rm -rf ${oldRunDir}
236         fi
237         echo "moving new ${runNumber} to ${productionDir}"
238         mv -f ${dir} ${productionDir}
239       done
240    
241       #go to a temp dir to do the period level stuff in a completely clean dir
242       tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
243       echo
244       echo tmpPeriodLevelQAdir="${tmpProductionDir}/periodLevelQA"
245       if ! mkdir -p ${tmpPeriodLevelQAdir}; then continue; fi
246       cd ${tmpPeriodLevelQAdir}
247
248       #link the final list of per-run dirs here, just the dirs
249       #to have a clean working directory
250       unset linkedStuff
251       declare -a linkedStuff
252       for x in ${productionDir}/000*; do [[ -d $x ]] && ln -s $x && linkedStuff+=(${x##*/}); done
253
254       #merge trending files if any
255       if /bin/ls 000*/trending.root &>/dev/null; then
256         hadd trending.root 000*/trending.root &> periodLevelQA.log
257       fi
258       
259       #run the period level trending/QA
260       if [[ -f "trending.root" && $(type -t periodLevelQA) =~ "function" ]]; then
261         echo running ${detector} periodLevelQA for production ${period}/${pass}
262         periodLevelQA trending.root &>> periodLevelQA.log
263       else 
264         echo "WARNING: not running ${detector} periodLevelQA for production ${period}/${pass}, no trending.root"
265       fi
266
267       if ! validate ${PWD}; then continue; fi
268
269       #here we are validated so move the produced QA to the final place
270       #clean up linked stuff first
271       [[ -n ${linkedStuff[@]} ]] && rm ${linkedStuff[@]}
272       #some of the output could be a directory, so handle that
273       #TODO: maybe use rsync?
274       for x in ${tmpPeriodLevelQAdir}/*; do  
275         if [[ -d ${x} ]]; then
276           echo "removing ${productionDir}/${x##*/}"
277           rm -rf ${productionDir}/${x##*/}
278           echo "moving ${x} to ${productionDir}"
279           mv ${x} ${productionDir}
280         fi
281         if [[ -f ${x} ]]; then
282           echo "moving ${x} to ${productionDir}"
283           mv -f ${x} ${productionDir} 
284         fi
285       done
286
287       #remove the temp dir
288       rm -rf ${tmpPeriodLevelQAdir}
289     
290     done
291
292     cd ${workingDirectory}
293
294     if [[ -z ${planB} ]]; then
295       echo
296       echo removing ${tmpDetectorRunDir}
297       rm -rf ${tmpDetectorRunDir}
298     else
299       executePlanB
300     fi
301   done #end of detector loop
302
303   #remove lock
304   rm -f ${lockFile}
305 }
306
307 executePlanB()
308 {
309   #in case of emergency
310   if [[ -n ${MAILTO} ]]; then 
311     echo
312     echo "trouble detected, sending email to ${MAILTO}"
313
314     grep BAD ${logSummary} | mail -s "qa in need of assistance" ${MAILTO}
315   fi
316 }
317
318 validate()
319 {
320   summarizeLogs ${1} >> ${logSummary}
321   logStatus=$?
322   if [[ ${logStatus} -ne 0 ]]; then 
323     echo "WARNING not validated: ${1}"
324     planB=1
325     return 1
326   fi
327   return 0
328 }
329
330 summarizeLogs()
331 {
332   local dir=$1
333   [[ ! -d ${dir} ]] && dir=${PWD}
334
335   #print a summary of logs
336   logFiles=(
337       "*.log"
338       "stdout"
339       "stderr"
340   )
341
342   #check logs
343   local logstatus=0
344   for log in ${dir}/${logFiles[*]}; do
345     finallog=${PWD%/}/${log}
346     [[ ! -f ${log} ]] && continue
347     errorSummary=$(validateLog ${log})
348     validationStatus=$?
349     [[ validationStatus -ne 0 ]] && logstatus=1
350     if [[ ${validationStatus} -eq 0 ]]; then 
351       #in pretend mode randomly report an error in rec.log some cases
352       if [[ -n ${pretend} && "${log}" == "rec.log" ]]; then
353         [[ $(( ${RANDOM}%2 )) -ge 1 ]] && echo "${finallog} BAD random error" || echo "${finallog} OK"
354       else
355         echo "${finallog} OK"
356       fi
357     elif [[ ${validationStatus} -eq 1 ]]; then
358       echo "${finallog} BAD ${errorSummary}"
359     elif [[ ${validationStatus} -eq 2 ]]; then
360       echo "${finallog} OK MWAH ${errorSummary}"
361     fi
362   done
363
364   #report core files
365   while read x; do
366     echo ${x}
367     chmod 644 ${x}
368     gdb --batch --quiet -ex "bt" -ex "quit" aliroot ${x} > stacktrace_${x//\//_}.log
369   done < <(/bin/ls ${PWD}/*/core 2>/dev/null; /bin/ls ${PWD}/core 2>/dev/null)
370
371   return ${logstatus}
372 }
373
374 validateLog()
375 {
376   log=${1}
377   errorConditions=(
378             'There was a crash'
379             'floating'
380             'error while loading shared libraries'
381             'std::bad_alloc'
382             's_err_syswatch_'
383             'Thread [0-9]* (Thread'
384             'AliFatal'
385             'core dumped'
386             '\.C.*error:.*\.h: No such file'
387             'segmentation'
388             'Interpreter error recovered'
389   )
390
391   warningConditions=(
392             'This is serious'
393   )
394
395   local logstatus=0
396   local errorSummary=""
397   local warningSummary=""
398
399   for ((i=0; i<${#errorConditions[@]};i++)); do
400     local tmp=$(grep -m1 -e "${errorConditions[${i}]}" ${log})
401     [[ -n ${tmp} ]] && tmp+=" : "
402     errorSummary+=${tmp}
403   done
404
405   for ((i=0; i<${#warningConditions[@]};i++)); do
406     local tmp=$(grep -m1 -e "${warningConditions[${i}]}" ${log})
407     [[ -n ${tmp} ]] && tmp+=" : "
408     warningSummary+=${tmp}
409   done
410
411   if [[ -n ${errorSummary} ]]; then 
412     echo "${errorSummary}"
413     return 1
414   fi
415
416   if [[ -n ${warningSummary} ]]; then
417     echo "${warningSummary}"
418     return 2
419   fi
420
421   return 0
422 }
423
424 parseConfig()
425 {
426   args=("$@")
427
428   #config file
429   configFile=""
430   #where to search for qa files
431   inputList=file.list
432   #working directory
433   workingDirectory="${PWD}"
434   #where to place the final qa plots
435   #outputDirectory="/afs/cern.ch/work/a/aliqa%det/www/"
436   outputDirectory="${workingDirectory}/%DET"
437   #filter out detector option
438   excludeDetectors="EXAMPLE"
439   #logs
440   logDirectory=${workingDirectory}/logs
441   #OCDB storage
442   ocdbStorage="raw://"
443   #email to
444   #MAILTO="fbellini@cern.ch"
445
446   #first, check if the config file is configured
447   #is yes - source it so that other options can override it
448   #if any
449   for opt in "${args[@]}"; do
450     if [[ ${opt} =~ configFile=.* ]]; then
451       eval "${opt}"
452       [[ ! -f ${configFile} ]] && echo "configFile ${configFile} not found, exiting..." && return 1
453       echo "using config file: ${configFile}"
454       source "${configFile}"
455       break
456     fi
457   done
458
459   #then, parse the options as they override the options from file
460   for opt in "${args[@]}"; do
461     if [[ ! "${opt}" =~ .*=.* ]]; then
462       echo "badly formatted option ${var}, should be: option=value, stopping..."
463       return 1
464     fi
465     local var="${opt%%=*}"
466     local value="${opt#*=}"
467     echo "${var} = ${value}"
468     export ${var}="${value}"
469   done
470 }
471
472 guessRunData()
473 {
474   #guess the period from the path, pick the rightmost one
475   period=""
476   runNumber=""
477   year=""
478   pass=""
479   legoTrainRunNumber=""
480   dataType=""
481
482   local shortRunNumber=""
483   local IFS="/"
484   declare -a path=( $1 )
485   local dirDepth=$(( ${#path[*]}-1 ))
486   i=0
487   for ((x=${dirDepth};x>=0;x--)); do
488
489     [[ $((x-1)) -ge 0 ]] && local fieldPrev=${path[$((x-1))]}
490     local field=${path[${x}]}
491     local fieldNext=${path[$((x+1))]}
492
493     [[ ${field} =~ ^[0-9]*$ && ${fieldNext} =~ (.*\.zip$|.*\.root$) ]] && legoTrainRunNumber=${field}
494     [[ -n ${legoTrainRunNumber} && -z ${pass} ]] && pass=${fieldPrev}
495     [[ ${field} =~ ^LHC[0-9][0-9][a-z].*$ ]] && period=${field%_*}
496     [[ ${field} =~ ^000[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && runNumber=${field#000}
497     [[ ${field} =~ ^[0-9][0-9][0-9][0-9][0-9][0-9]$ ]] && shortRunNumber=${field}
498     [[ ${field} =~ ^20[0-9][0-9]$ ]] && year=${field}
499     [[ ${field} =~ ^(^sim$|^data$) ]] && dataType=${field}
500     (( i++ ))
501   done
502   [[ -z ${legoTrainRunNumber} ]] && pass=${path[$((dirDepth-1))]}
503   [[ "${dataType}" =~ ^sim$ ]] && pass="passMC" && runNumber=${shortRunNumber}
504   
505   #if [[ -z ${dataType} || -z ${year} || -z ${period} || -z ${runNumber}} || -z ${pass} ]];
506   if [[ -z ${runNumber}} ]];
507   then
508     #error condition
509     return 1
510   else
511     #ALL OK
512     return 0
513   fi
514 }
515
516 substituteDetectorName()
517 {
518   local det=$1
519   local dir=$2
520   [[ ${dir} =~ \%det ]] && det=${det,,} && echo ${dir/\%det/${det}}
521   [[ ${dir} =~ \%DET ]] && det=${det} && echo ${dir/\%DET/${det}}
522 }
523
524 get_realpath() 
525 {
526   if [[ -f "$1" ]]
527   then
528     # file *must* exist
529     if cd "$(echo "${1%/*}")" &>/dev/null
530     then
531       # file *may* not be local
532       # exception is ./file.ext
533       # try 'cd .; cd -;' *works!*
534       local tmppwd="$PWD"
535       cd - &>/dev/null
536     else
537       # file *must* be local
538       local tmppwd="$PWD"
539     fi
540   else
541     # file *cannot* exist
542     return 1 # failure
543   fi
544   # reassemble realpath
545   echo "$tmppwd"/"${1##*/}"
546   return 0 # success
547 }
548
549 main $@