source: trunk/libIGCM/AA_RunChecker @ 758

Last change on this file since 758 was 757, checked in by labetoulle, 12 years ago

RunChecker? :

  • Added the number of IGCM_sys_Put[Buffer]_Out to the output for rebuild and pack_* jobs to get a more precise idea of how things went ;
  • Added the possibility to choose the path from a list when more than one is found ;
  • Eclude IGCM_OUT/ (where part of the atlas and monitoring files are stored on Curie) from search path to save some time.
  • Property svn:executable set to *
  • Property svn:keywords set to Revision Author Date
File size: 15.3 KB
Line 
1#!/bin/ksh
2
3#**************************************************************
4# Author: Sonia Labetoulle
5# Contact: sonia.labetoulle__at__ipsl.jussieu.fr
6# $Revision::                                          $ Revision of last commit
7# $Author::                                            $ Author of last commit
8# $Date::                                              $ Date of last commit
9# IPSL (2012)
10#  This software is governed by the CeCILL licence see libIGCM/libIGCM_CeCILL.LIC
11#
12#**************************************************************
13
14#======================================================================#
15# Display a report on how a simulation is faring.
16
17
18function ChangeUsr {
19
20  echo $1 | sed -e "s/${CurrentUsr}/${TargetUsr}/" \
21                -e "s/${CurrentGrp}/${TargetGrp}/"
22
23}
24
25
26function SearchCatalog {
27
28  typeset num
29  unset SUBMIT_DIR
30
31  fg_new=false
32
33  NbOcc=$( awk -v JobName=${JobName} \
34           'BEGIN {x=0}  $1 ~ JobName {++x} END {print x}' ${SimuCatalog} )
35
36  if ( [ ${NbOcc} -eq 0 ] && ( $fg_path ) ) ; then
37    set -A FileList $( ls ${ConfigPath}/Job_* )
38    if [ X$FileList == X ] ; then
39      NbOcc=0
40    else
41      NbOcc=${#FileList[@]}
42      fg_new=true
43    fi
44  fi
45
46  if ( [ ${NbOcc} -eq 0 ] && ( ${fg_search} ) ) ; then
47    SEARCH_DIR=${WORKDIR}
48    if [ ${TargetUsr} != ${CurrentUsr} ] ; then
49      SEARCH_DIR=$( ChangeUsr ${SEARCH_DIR})
50    fi
51    echo "${JobName} not in Catalog, we'll try to find it in ${SEARCH_DIR}"
52
53    set -A FileList $( find ${SEARCH_DIR}/ \
54                            -path ${SEARCH_DIR}/IGCM_OUT -prune -o \
55                            -name Job_${JobName} -print )
56    if [ X$FileList == X ] ; then
57      NbOcc=0
58    else
59      NbOcc=${#FileList[@]}
60      fg_new=true
61    fi
62  fi
63
64  if [ ${NbOcc} -gt 1 ] ; then
65    echo "More than one job"
66    ind=0
67    while [ ${ind} -lt ${NbOcc} ] ; do
68      printf '%2i) %-30s\n' ${ind} ${FileList[${ind}]}
69      (( ind = ind + 1 ))
70    done
71    echo "Give your choice number or 'q' to quit : "
72    read Choice
73    if [ X${Choice} == Xq ] ; then
74      exit
75    else
76      fg_new=true
77      FileList=${FileList[${Choice}]}
78      NbOcc=1
79    fi
80  fi
81
82  case ${NbOcc} in
83    0)
84      echo "${JobName} not found."
85      echo "You can try : *) '-s' option to automatically search your \$WORKDIR, "
86      echo "              *) '-p' option to provide the config.card path, "
87      echo "              *)  manually editing your ${SimuCatalog}"
88      exit ;;
89    1)
90      if ( ${fg_new} ) ; then
91        JobName=${JobName:=$( basename ${FileList} | awk -F"_" '{ print $2 }' )}
92        SUBMIT_DIR=$( dirname ${FileList} )
93        echo "${JobName} ${TargetUsr} ${HostName} ${SUBMIT_DIR}"
94        echo "${JobName} ${TargetUsr} ${HostName} ${SUBMIT_DIR}" >> ${SimuCatalog}
95      elif ( [ ${TargetUsr} == $( awk -v JobName=${JobName} \
96                                      '$1 ~ JobName {print $2}' \
97                                      ${SimuCatalog} ) ] \
98          && [ ${HostName}  == $( awk -v JobName=${JobName} \
99                                      '$1 ~ JobName {print $3}' \
100                                      ${SimuCatalog} ) ] ) ; then
101        JobName=$( awk -v JobName=${JobName} '$1 ~ JobName {print $1}' ${SimuCatalog} )
102        SUBMIT_DIR=$( awk -v JobName=${JobName} '$1 ~ JobName {print $4}' ${SimuCatalog} )
103      else
104        echo "${JobName} not in Catalog."
105        exit
106      fi
107      break ;;
108    *)
109      break ;;
110  esac
111
112  return
113
114}
115
116
117function AffichResult {
118
119  fg_first=false
120  fg_last=false
121  fg_job=false
122
123  while [ $# -ne 0 ] ; do
124    case ${1} in
125      -f|--first)
126        fg_first=true
127        shift 1 ;;
128      -l|--last)
129        fg_last=true
130        shift 1 ;;
131      -j|--job)
132        fg_job=true
133        shift 1 ;;
134      -*)
135        echo "usage: ${0}"
136        echo "       options = -f; -l"
137        exit ;;
138      *)
139        break ;;
140    esac
141  done
142
143
144  # Define colors
145  # =============
146  ColEsc="\033["
147  ColNon="${ColEsc}0m"       # Return to normal
148  ColExp="${ColEsc}1m"       # Blanc - gras
149  ColFat="${ColEsc}1;31m"    # Fatal
150  ColCpl="${ColEsc}1;32m"    # Completed
151  ColAtt="${ColEsc}1;30m"    # Waiting
152  ColDef="${ColEsc}1;34m"    # Default
153  ColRbl="${ColEsc}31m"      # Rebuild
154
155
156 
157  # Print header
158  # ============
159  if ( ${fg_first} ) ; then
160    Dum=""
161    (( len = 67 - ${#JobName} ))
162    echo "|===========================================================================================================|"
163    printf "| JobName = ${ColExp}%-${#JobName}s${ColNon}" ${JobName}
164    printf "%-${len}s" ${Dum} 
165    printf "run.card : ${ColExp}%-17s${ColNon}|\n" "${LastWrite} "
166    echo "|-------------------------|-------------|-------------------------|-------------|-----:----------:----------|"
167    echo "|                         |             |                         |             |     Pending Rebuilds      |"
168    echo "| Date Begin - DateEnd    | PeriodState | Current Period          | CumulPeriod | Nb  : from     : to       |"
169    echo "|-------------------------|-------------|-------------------------|-------------|-----:----------:----------|"
170
171    printf "| %-10s - %-10s | " \
172           $DateBegin $DateEnd 
173
174    case $PeriodState in
175      Fatal)
176        Color=${ColFat}
177        break ;;
178      Completed)
179        Color=${ColCpl}
180        break ;;
181      Waiting|OnQueue)
182        Color=${ColAtt}
183        break ;;
184      *)
185        Color=${ColDef}
186        break ;;
187    esac
188    printf "${Color}%-11s${ColNon} | " $PeriodState
189
190    printf "%-10s - %-10s | %11s | " \
191           $PeriodDateBegin $PeriodDateEnd $CumulPeriod
192
193    if ( [ X${NbRebuild} != X. ] && [ X${NbRebuild} != X0 ] ) ; then
194      printf "${ColRbl}%3s : %-8s : %-8s${ColNon} |\n" \
195             $NbRebuild $FirstRebuild $LastRebuild
196    else
197      printf "%3s : %-8s : %-8s |\n" \
198             $NbRebuild $FirstRebuild $LastRebuild
199    fi
200
201    printf "|-----------------------------------------------------------------------------------------------------------|\n"
202    printf "|                                                      Last                                                 |\n"
203    printf "|     Rebuild      |   Pack_Output    |   Pack_Restart   |    Pack_Debug    |  Monitoring  |     Atlas      |\n"
204    printf "|------------------|------------------|------------------|------------------|--------------|----------------|\n"
205
206    return
207  fi
208
209  # Print Post-processing job status
210  # ================================
211  if ( ${fg_job} ) ; then
212    printf "|"
213
214    # Print rebuild and pack jobs
215    # ---------------------------
216    for JobType in ${JobType_list[*]} ; do
217      eval Date=\${${JobType}_Date[${ind}]}
218      eval Status=\${${JobType}_Status[${ind}]}
219      eval Nb=\${${JobType}_Nb[${ind}]}
220
221      if [ X${Status} == XOK  ] ; then
222        Color=${ColCpl}
223      else
224        Color=${ColFat}
225      fi
226      printf "  ${Color}%-8s${ColNon} : %3i  |" ${Date} ${Nb}
227    done
228
229    Color=${ColExp}
230
231    # Print monitoring jobs
232    # ---------------------
233    JobType=monitoring
234    if [ $ind -eq 0 ] ; then
235      eval Date=\${${JobType}_Date}
236    else
237      Date=""
238    fi
239    printf "     ${Color}%-4s${ColNon}     |" ${Date}
240
241    # Print atlas jobs
242    # ----------------
243    JobType=atlas
244    eval Date=\${${JobType}_Date[${ind}]}
245    printf "  ${Color}%-12s${ColNon}  |" ${Date}
246
247    printf "\n"
248
249    return
250  fi
251
252  # Print footer
253  # ============
254  if ( ${fg_last} ) ; then
255    printf "|===========================================================================================================|\n"
256    date +"${DateFormat}"
257    return
258  fi
259
260}
261
262#======================================================================#
263
264#D- Task type (computing or post-processing)
265TaskType=post-processing
266typeset -i Verbosity=0
267
268CurrentUsr=$( whoami )
269CurrentGrp=$( groups $CurrentUsr | awk '{print $3}' )
270
271if ( [ $# -eq 0 ] ) ; then
272  $0 -h
273  exit
274fi
275
276# Arguments
277# =========
278# Default argument values
279# -----------------------
280TargetUsr=${CurrentUsr}
281HostName=$( hostname | sed -e "s/[0-9].*//" )
282
283fg_color=true
284fg_search=false
285fg_quiet=false
286fg_path=false
287NbHisto=10
288
289# Get arguments from command line
290# -------------------------------
291while [ $# -ne 0 ] ; do
292  case $1 in
293    -h|--help|-help)
294      echo "usage: $0 [-u user] [-n] [-q] [-j n] job_name"
295      echo "       $0 [-u user] [-n] [-q] [-j n] -p config.card_path"
296      echo ""
297      echo "options :"
298      echo "  -h : print this help and exit"
299      echo "  -u : owner of the job"
300      echo "  -q : quiet"
301      echo "  -j : print n post-processing jobs (default is 10)"
302      echo "  -s : search for a new job in \$WORKDIR and fill in "
303      echo "       the catalog before printing information"
304      echo "  -p : give the config.card path to a new simulation, "
305      echo "       instead of the job name."
306      exit ;;
307#    -b|-nocolor)
308#      fg_color=false
309#      shift 1 ;;
310    -s|-search)
311      fg_search=true
312      shift 1 ;;
313    -q|-quiet)
314      fg_quiet=true
315      shift 1 ;;
316    -p|-config-path)
317      ConfigPath="$2"
318      fg_path=true
319      shift 2 ;;
320    -j|-job-number)
321      NbHisto="$2"
322      shift 2 ;;
323    -u|-user)
324      TargetUsr="$2"
325      shift 2 ;;
326    -*)
327      $0 -h
328      exit ;;
329    *)
330      break ;;
331  esac
332done
333
334
335if ( ( ! $fg_path ) && [ $# -lt 1 ] ) ; then
336  $0 -h
337  exit
338fi
339
340if ( ( ${fg_path} ) && ( ${fg_search} ) ) ; then
341  echo "You cannot use -s and -p at the same time"
342  exit
343fi
344
345
346# Load libIGCM library
347# ====================
348libIGCM=${libIGCM:=::modipsl::/libIGCM}
349
350. ${libIGCM}/libIGCM_debug/libIGCM_debug.ksh
351     ( ${DEBUG_debug} ) && IGCM_debug_Check
352. ${libIGCM}/libIGCM_card/libIGCM_card.ksh
353     ( ${DEBUG_debug} ) && IGCM_card_Check
354. ${libIGCM}/libIGCM_date/libIGCM_date.ksh
355     ( ${DEBUG_debug} ) && IGCM_date_Check
356#-------
357. ${libIGCM}/libIGCM_sys/libIGCM_sys.ksh
358. ${libIGCM}/libIGCM_config/libIGCM_config.ksh
359
360
361JobName=$1
362
363if ( ${fg_quiet} ) ; then
364  NbHisto=1
365fi
366
367echo "TargetUsr = ${TargetUsr}"
368echo "HostName = ${HostName}"
369
370TargetGrp=$( groups $TargetUsr | awk '{print $3}' )
371
372# Define the catalog in which the known simulations are stored
373SimuCatalog="$( ccc_home )/.simucatalog.dat"
374if [ ! -s ${SimuCatalog} ] ; then
375  touch ${SimuCatalog}
376fi
377
378# Date format
379DateFormat="%d/%m/%y %R:%S"
380
381# Find SUBMIT_DIR in catalog
382# ==========================
383SearchCatalog
384
385if [ ! X${SUBMIT_DIR} == X ] ; then
386
387  echo "Submit:  >${SUBMIT_DIR}<"
388  cd $SUBMIT_DIR
389
390
391  # Extract usefull information from run.card and config.card
392  # =========================================================
393
394  RunFile="${SUBMIT_DIR}/run.card"
395  ConfFile="${SUBMIT_DIR}/config.card"
396
397  IGCM_config_CommonConfiguration ${SUBMIT_DIR}/config.card
398
399
400  if [ -s ${RunFile} ] ; then
401    IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodState
402    PeriodState=${run_Configuration_PeriodState}
403  else
404    PeriodState="Waiting"
405  fi
406
407  if ( [ X${PeriodState} == XRunning ] || [ X${PeriodState} == XOnQueue ] ) ; then
408    NbRun=$( ccc_mstat -f | grep -c ${JobName} )
409
410    if [ ${NbRun} -eq 0 ] ; then
411      PeriodState="Fatal"
412    fi
413  fi
414
415  DateBegin=${config_UserChoices_DateBegin}
416  DateEnd=${config_UserChoices_DateEnd}
417  TagName=${config_UserChoices_TagName}
418  ExperimentName=${config_UserChoices_ExperimentName}
419  SpaceName=${config_UserChoices_SpaceName}
420
421  IGCM_card_DefineVariableFromOption ${RunFile} Configuration CumulPeriod
422  CumulPeriod=${run_Configuration_CumulPeriod}
423
424  if ( [ ! X${PeriodState} == XWaiting ] && [ ! X${PeriodState} == XCompleted ] ) ; then
425    IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodDateBegin
426    IGCM_card_DefineVariableFromOption ${RunFile} Configuration PeriodDateEnd
427    PeriodDateBegin=${run_Configuration_PeriodDateBegin}
428    PeriodDateEnd=${run_Configuration_PeriodDateEnd}
429  else
430    PeriodDateBegin="."
431    PeriodDateEnd="."
432  fi
433
434  DATA_DIR=${R_SAVE}
435  POST_DIR=${R_BUFR}/Out
436  CWORK_DIR=${R_FIGR}
437  if [ X${config_Post_RebuildFromArchive} = Xtrue ] ; then
438    REBUILD_DIR=${R_SAVE}/TMP
439    RebuildJob="rebuild_fromArchive"
440  else
441    REBUILD_DIR=${BIG_DIR}/${config_UserChoices_TagName}/${config_UserChoices_JobName}
442    RebuildJob="rebuild_fromWorkdir"
443  fi
444
445  if [ ${TargetUsr} != ${CurrentUsr} ] ; then
446    DATA_DIR=$( ChangeUsr ${DATA_DIR})
447    POST_DIR=$( ChangeUsr ${POST_DIR} )
448    CWORK_DIR=$( ChangeUsr ${CWORK_DIR} )
449    REBUILD_DIR=$( ChangeUsr ${REBUILD_DIR} )
450  fi
451
452  echo "Data:    >${DATA_DIR}<"
453  echo "Rebuild: >${REBUILD_DIR}<"
454  echo "Post:    >${POST_DIR}<"
455  echo "Work:    >${CWORK_DIR}<"
456
457  if [ $PeriodState != "Waiting" ] ; then
458
459    # Check pending rebuilds
460    # ======================
461
462    set -A RebuildList $( find ${REBUILD_DIR}/ -name "REBUILD_*" | sort )
463    if [ ${#RebuildList[*]} -gt 0 ] ; then
464      NbRebuild=$( IGCM_sys_CountFileArchive ${REBUILD_DIR} )
465
466      FirstRebuild=$( basename ${RebuildList[0]} | cut -f2 -d\_ )
467      LastRebuild=$( basename ${RebuildList[ (( NbRebuild=${NbRebuild}-1 )) ]} | cut -f2 -d\_ )
468    else
469      NbRebuild="."
470
471      FirstRebuild="."
472      LastRebuild="."
473    fi
474  else
475    NbRebuild="."
476
477    FirstRebuild="."
478    LastRebuild="."
479  fi
480
481
482  # Check last REBUILD and PACK* jobs
483  # =================================
484  # Define input parameters
485  # -----------------------
486  set -A JobType_list "${RebuildJob}" "pack_output" "pack_restart" "pack_debug"
487
488  for JobType in ${JobType_list[*]} ; do
489    typeset    name1="${JobType}_String"
490    typeset    name2="${JobType}_Field"
491    if [ X${JobType} == X${RebuildJob} ] ; then
492      eval ${name1}=IGCM_sys_PutBuffer_Out
493      eval ${name2}=4
494    else
495      eval ${name1}=IGCM_sys_Put_Out
496      eval ${name2}=3
497    fi
498  done
499
500  # Check jobs
501  # ----------
502  NbLines=0
503  for JobType in ${JobType_list[*]} ; do
504    eval String=\${${JobType}_String}
505    eval Field=\${${JobType}_Field}
506
507    set -A FileList $( ls ${POST_DIR}/${JobType}.*.out | tail -n ${NbHisto} )
508
509    if [ ${#FileList[*]} -gt ${NbLines} ] ; then
510      NbLines=${#FileList[*]}
511    fi
512
513    (( ind = 0 ))
514    for FileName in ${FileList[*]} ; do
515      LastDate=$( basename ${FileName} | awk -F"." '{ print $(NF-1) }' )
516
517      dum=$( basename $( grep ^${String} ${POST_DIR}/${JobType}.${LastDate}.out | \
518               tail -1 | awk -v Field=${Field} '{ print $( Field ) }' ) | \
519               awk -F"_" '{ print $3 }' )
520
521      if [[ X${dum} == X${LastDate} ]] ; then
522        Status=OK
523      else
524        Status=KO
525      fi
526
527      eval ${JobType}_Date[$ind]=${LastDate}
528      eval ${JobType}_Status[$ind]=${Status}
529      eval ${JobType}_Nb[$ind]=$( grep -c ^${String} ${POST_DIR}/${JobType}.${LastDate}.out )
530     
531      (( ind = ind + 1 ))
532    done
533  done
534
535  # Check last MONITORING jobs
536  # ==========================
537  JobType=monitoring
538  if [ -d ${CWORK_DIR}/MONITORING ] ; then
539    LastDate=$( cdo showyear ${CWORK_DIR}/MONITORING/files/ATM_bils_global_ave.nc 2> /dev/null | \
540                    awk '{ print $NF }' )
541    eval ${JobType}_Date=${LastDate}
542  fi
543
544  # Check last ATLAS jobs
545  # =====================
546  JobType=atlas
547  if [ -d ${CWORK_DIR}/ATLAS ] ; then
548    set -A FileList $( ls ${CWORK_DIR}/ATLAS | tail -n ${NbHisto} )
549
550    if [ ${#FileList[*]} -gt ${NbLines} ] ; then
551      NbLines=${#FileList[*]}
552    fi
553
554    (( ind = 0 ))
555    for FileName in ${FileList[*]} ; do
556      eval ${JobType}_Date[$ind]=${FileName}
557      (( ind = ind + 1 ))
558    done
559  fi
560
561
562  # Time of last write on run.card
563  # ==============================
564  LastWrite=$( ls -l --time-style=+"${DateFormat}" ${SUBMIT_DIR}/run.card | awk '{print $6 " " $7}' )
565   
566
567  # Print results
568  # =============
569  AffichResult -f
570  ind=0
571  while [ $ind -lt $NbLines ] ; do
572    AffichResult -j
573    (( ind = ind + 1 ))
574  done
575  AffichResult -l
576
577fi
578
Note: See TracBrowser for help on using the repository browser.