1 |
|
from Actor import * |
2 |
|
from crab_util import * |
3 |
< |
import EdgLoggingInfo |
4 |
< |
import CondorGLoggingInfo |
3 |
> |
#import EdgLoggingInfo |
4 |
> |
#import CondorGLoggingInfo |
5 |
|
import common |
6 |
|
import string, os |
7 |
|
|
9 |
|
def __init__(self, cfg_params, nj_list): |
10 |
|
self.cfg_params = cfg_params |
11 |
|
self.nj_list = nj_list |
12 |
+ |
self.all_jobs=common._db.nJobs('list') |
13 |
+ |
|
14 |
+ |
self.fname_base = common.work_space.jobDir() + self.cfg_params['CRAB.jobtype'].upper() + '_' |
15 |
|
|
13 |
– |
if common.scheduler.boss_scheduler_name == 'condor_g': |
14 |
– |
# create hash of cfg file |
15 |
– |
self.hash = makeCksum(common.work_space.cfgFileName()) |
16 |
– |
else: |
17 |
– |
self.hash = '' |
18 |
– |
|
16 |
|
return |
17 |
|
|
18 |
|
def run(self): |
21 |
|
""" |
22 |
|
common.logger.debug(5, "PostMortem::run() called") |
23 |
|
|
24 |
< |
for c, v in self.nj_list.iteritems(): |
25 |
< |
id = int(c) |
29 |
< |
out = common.scheduler.loggingInfo(v) |
30 |
< |
job = common.job_list[id - 1] |
31 |
< |
jobnum_str = '%06d' % (id) |
32 |
< |
fname = common.work_space.jobDir() + '/' + self.cfg_params['CRAB.jobtype'].upper() + '_' + jobnum_str + '.loggingInfo' |
33 |
< |
if os.path.exists(fname): |
34 |
< |
common.logger.message('Logging info for job ' + str(id) + ' already present in '+fname+'\nRemove it for update') |
35 |
< |
continue |
36 |
< |
jdl = open(fname, 'w') |
37 |
< |
for line in out: jdl.write(line) |
38 |
< |
jdl.close() |
39 |
< |
|
40 |
< |
reason = '' |
41 |
< |
## SL this if-elif is the negation of OO! Mus disappear ASAP |
42 |
< |
if common.scheduler.boss_scheduler_name == "edg" or common.scheduler.boss_scheduler_name == "glite" or common.scheduler.boss_scheduler_name == "glitecoll": |
43 |
< |
loggingInfo = EdgLoggingInfo.EdgLoggingInfo() |
44 |
< |
reason = loggingInfo.decodeReason(out) |
45 |
< |
elif common.scheduler.boss_scheduler_name == "condor_g" : |
46 |
< |
loggingInfo = CondorGLoggingInfo.CondorGLoggingInfo() |
47 |
< |
reason = loggingInfo.decodeReason(out) |
48 |
< |
else : |
49 |
< |
reason = out |
50 |
< |
|
51 |
< |
common.logger.message('Logging info for job '+ str(id) +': '+reason+'\n written to '+fname) |
52 |
< |
|
53 |
< |
# ML reporting |
54 |
< |
jobId = '' |
55 |
< |
if common.scheduler.boss_scheduler_name == 'condor_g': |
56 |
< |
jobId = str(id) + '_' + self.hash + '_' + v |
57 |
< |
else: |
58 |
< |
jobId = str(id) + '_' + v |
59 |
< |
|
60 |
< |
params = {'taskId': self.cfg_params['taskId'], 'jobId': jobId, \ |
61 |
< |
'sid': v, |
62 |
< |
'PostMortemCategory': loggingInfo.getCategory(), \ |
63 |
< |
'PostMortemReason': loggingInfo.getReason()} |
64 |
< |
self.cfg_params['apmon'].sendToML(params) |
65 |
< |
pass |
24 |
> |
self.collectLogging() |
25 |
> |
|
26 |
|
|
27 |
+ |
def collectLogging(self): |
28 |
+ |
for id in self.nj_list: |
29 |
+ |
if id not in self.all_jobs: |
30 |
+ |
common.logger.message('Warning: job # ' + str(id) + ' does not exist! Not possible to ask for postMortem ') |
31 |
+ |
continue |
32 |
+ |
else: |
33 |
+ |
fname = self.fname_base + str(id) + '.LoggingInfo' |
34 |
+ |
if os.path.exists(fname): |
35 |
+ |
common.logger.message('Logging info for job ' + str(id) + ' already present in '+fname+'\nRemove it for update') |
36 |
+ |
continue |
37 |
+ |
common.scheduler.loggingInfo(id,self.fname_base+str(id)) |
38 |
+ |
fl = open(fname, 'r') |
39 |
+ |
out = "".join(fl.readlines()) |
40 |
+ |
fl.close() |
41 |
+ |
reason = self.decodeLogging(out) |
42 |
+ |
common.logger.message('Logging info for job '+ str(id) +': '+str(reason)+'\n written to '+str(fname) ) |
43 |
|
return |
44 |
+ |
|
45 |
+ |
def decodeLogging(self, out): |
46 |
+ |
""" |
47 |
+ |
""" |
48 |
+ |
return common.scheduler.decodeLogInfo(out) |
49 |
|
|