| 1 |
|
from Actor import * |
| 2 |
< |
import EdgLoggingInfo |
| 2 |
> |
from crab_util import * |
| 3 |
|
import common |
| 4 |
|
import string, os |
| 5 |
|
|
| 6 |
|
class PostMortem(Actor): |
| 7 |
< |
def __init__(self, cfg_params, nj_list, use_boss): |
| 7 |
> |
def __init__(self, cfg_params, nj_list): |
| 8 |
|
self.cfg_params = cfg_params |
| 9 |
|
self.nj_list = nj_list |
| 10 |
< |
self.flag_useboss = use_boss |
| 10 |
> |
self.all_jobs=common._db.nJobs('list') |
| 11 |
> |
|
| 12 |
> |
self.fname_base = common.work_space.jobDir() + self.cfg_params['CRAB.jobtype'].upper() + '_' |
| 13 |
> |
|
| 14 |
|
return |
| 15 |
|
|
| 16 |
|
def run(self): |
| 17 |
|
""" |
| 18 |
|
The main method of the class. |
| 19 |
|
""" |
| 20 |
< |
common.logger.debug(5, "PostMortem::run() called") |
| 20 |
> |
common.logger.debug( "PostMortem::run() called") |
| 21 |
|
|
| 22 |
< |
if len(self.nj_list)==0: |
| 20 |
< |
common.logger.debug(5, "No jobs to check") |
| 21 |
< |
return |
| 22 |
> |
self.collectLogging() |
| 23 |
|
|
| 24 |
< |
# run a list-match on first job |
| 25 |
< |
for nj in self.nj_list: |
| 26 |
< |
#nj parte da 1 --> nj = internal_id di boss |
| 27 |
< |
if self.flag_useboss == 1 : |
| 28 |
< |
id = common.scheduler.boss_SID(nj) |
| 29 |
< |
#print "id = ", id |
| 30 |
< |
#else: |
| 31 |
< |
# id = common.jobDB.jobId(nj) |
| 32 |
< |
out = common.scheduler.loggingInfo(id) |
| 33 |
< |
# job_list inizia a contare da zero |
| 34 |
< |
job = common.job_list[nj-1] |
| 35 |
< |
#print "job.jdlFilename()", job.jdlFilename() |
| 36 |
< |
jdl_fname = string.replace(job.jdlFilename(),'jdl','loggingInfo') |
| 37 |
< |
#print "jdl_fname = ", jdl_fname |
| 38 |
< |
if os.path.exists(jdl_fname): |
| 39 |
< |
common.logger.message('Logging info for job '+str(nj)+' already present in '+jdl_fname+' Remove it for update') |
| 40 |
< |
continue |
| 41 |
< |
jdl = open(jdl_fname, 'w') |
| 42 |
< |
for line in out: jdl.write(line) |
| 43 |
< |
jdl.close() |
| 44 |
< |
|
| 44 |
< |
loggingInfo = EdgLoggingInfo.EdgLoggingInfo() |
| 45 |
< |
|
| 46 |
< |
reason = loggingInfo.decodeReason(out) |
| 47 |
< |
|
| 48 |
< |
common.logger.message('Logging info for job '+str(nj)+': '+reason+'\n written to '+jdl_fname) |
| 49 |
< |
|
| 50 |
< |
# ML reporting |
| 51 |
< |
jobId = '' |
| 52 |
< |
if common.scheduler.boss_scheduler_name == 'condor_g': |
| 53 |
< |
# create hash of cfg file |
| 54 |
< |
hash = makeCksum(common.work_space.cfgFileName()) |
| 55 |
< |
jobId = str(nj) + '_' + hash + '_' + id |
| 56 |
< |
else: |
| 57 |
< |
jobId = str(nj) + '_' + id |
| 58 |
< |
|
| 59 |
< |
params = {'taskId': self.cfg_params['taskId'], 'jobId': jobId, \ |
| 60 |
< |
'sid': id, |
| 61 |
< |
'PostMortemCategory': loggingInfo.getCategory(), \ |
| 62 |
< |
'PostMortemReason': loggingInfo.getReason()} |
| 63 |
< |
self.cfg_params['apmon'].sendToML(params) |
| 64 |
< |
pass |
| 24 |
> |
def collectOneLogging(self, id): |
| 25 |
> |
job=self.up_task.getJob(id) |
| 26 |
> |
if not job: #id not in self.all_jobs: |
| 27 |
> |
common.logger.info('Warning: job # ' + str(id) + ' does not exist! Not possible to ask for postMortem ') |
| 28 |
> |
return |
| 29 |
> |
elif job.runningJob['state'] == 'Created': |
| 30 |
> |
common.logger.info('Warning: job # ' + str(id) + ' just Created ! Not possible to ask for postMortem ') |
| 31 |
> |
else: |
| 32 |
> |
fname = self.fname_base + str(id) + '.LoggingInfo' |
| 33 |
> |
if os.path.exists(fname): |
| 34 |
> |
common.logger.info('Logging info for job ' + str(id) + ' already present in '+fname+'\nRemove it for update') |
| 35 |
> |
return |
| 36 |
> |
common.scheduler.loggingInfo(id,fname) |
| 37 |
> |
fl = open(fname, 'r') |
| 38 |
> |
out = "".join(fl.readlines()) |
| 39 |
> |
fl.close() |
| 40 |
> |
reason = self.decodeLogging(out) |
| 41 |
> |
common.logger.info('Logging info for job '+ str(id) +' written to '+str(fname)) |
| 42 |
> |
common.logger.info('Reason for job status is:\n\n'+str(reason)+'\n') |
| 43 |
> |
return |
| 44 |
> |
|
| 45 |
|
|
| 46 |
+ |
def collectLogging(self): |
| 47 |
+ |
self.up_task = common._db.getTask( self.nj_list ) |
| 48 |
+ |
for id in self.nj_list: |
| 49 |
+ |
self.collectOneLogging(id) |
| 50 |
|
return |
| 51 |
+ |
|
| 52 |
+ |
def decodeLogging(self, out): |
| 53 |
+ |
""" |
| 54 |
+ |
""" |
| 55 |
+ |
return common.scheduler.decodeLogInfo(out) |
| 56 |
|
|
