CRAB/python/Submitter.py

from Actor import *
from crab_util import *
import common
from ApmonIf import ApmonIf
import Statistic
from random import random
import time

class Submitter(Actor):
    def __init__(self, cfg_params, nj_list):
        self.cfg_params = cfg_params
        self.nj_list = nj_list
        
        try:
            self.ML = int(cfg_params['USER.activate_monalisa'])
        except KeyError:
            self.ML = 0
            pass

        return
    
    def run(self):
        """
        The main method of the class.
        """
        common.logger.debug(5, "Submitter::run() called")

        totalCreatedJobs= 0
        start = time.time()
        for nj in range(common.jobDB.nJobs()):
            if (common.jobDB.status(nj)=='C') or (common.jobDB.status(nj)=='RC'): totalCreatedJobs +=1
            pass

        if (totalCreatedJobs==0):
            common.logger.message("No jobs to be submitted: first create them")
            return
        
#         goodJob={}
#         for nj in self.nj_list:


#             if match:
#                 goodJob[nj]=1
#                 common.logger.message("Found "+str(match)+" compatible site(s) for job "+str(nj+1))
#             else:
#                 goodJob[nj]=0
#                 common.logger.message("No compatible site found, will not submit job "+str(nj+1))
# #                raise CrabException("No compatible site found!")
        #########
        # Loop over jobs
        njs = 0
        try:
            lastDest=''
            for nj in self.nj_list:
                # first check that status of the job is suitable for submission
                st = common.jobDB.status(nj)
                if st != 'C' and st != 'K' and st != 'A' and st != 'RC':
                    long_st = crabJobStatusToString(st)
                    msg = "Job # %d not submitted: status %s"%(nj+1, long_st)
                    common.logger.message(msg)
                    continue
                # SL perform listmatch only if destination has changed
                currDest=common.jobDB.destination(nj)
                match=0
                if (currDest!=lastDest):
                    match = common.scheduler.listMatch(nj)
                    lastDest = currDest
                if match:
                    common.logger.message("Found "+str(match)+" compatible site(s) for job "+str(nj+1)+" and those with same requirements")
                else:
                    common.logger.message("No compatible site found, will not submit job "+str(nj+1)+" and those with same requirements")
                    continue

                common.logger.message("Submitting job # "+`(nj+1)`)
                jid = common.scheduler.submit(nj)

                common.jobDB.setStatus(nj, 'S')
                common.jobDB.setJobId(nj, jid)
                common.jobDB.setTaskId(nj, self.cfg_params['taskId'])
                njs += 1

                ############################################   

                if st == 'C':
                    resFlag = 0
                elif st == 'RC':
                    resFlag = 2
                else:            
                    resFlag = 0
                    pass
                  
                try:
                    Statistic.Monitor('submit',resFlag,jid,'-----')
                except:
                    pass
                
                fl = open(common.work_space.shareDir() + '/' + self.cfg_params['apmon'].fName, 'r')
                self.cfg_params['sid'] = jid
                #### FF: per il momento commentiamo nevtJob che non c'e' piu' nel jobdb
                #nevtJob = common.jobDB.maxEvents(nj)

                # OLI: JobID treatment, special for Condor-G scheduler
                jobId = ''
                if common.scheduler.boss_scheduler_name == 'condor_g':
                    # create hash of cfg file
                    hash = makeCksum(common.work_space.cfgFileName())
                    jobId = str(nj + 1) + '_' + hash + '_' + self.cfg_params['sid']
                    common.logger.debug(5,'JobID for ML monitoring is created for CONDOR_G scheduler:'+jobId)
                else:
                    jobId = str(nj + 1) + '_' + self.cfg_params['sid']
                    common.logger.debug(5,'JobID for ML monitoring is created for EDG scheduler'+jobId)

                if ( jid.find(":") != -1 ) :
                    rb = jid.split(':')[1]
                    self.cfg_params['rb'] = rb.replace('//', '')
                else :
                    self.cfg_params['rb'] = 'OSG'

                #### FF: per il momento commentiamo nevtJob che non c'e' piu' nel jobdb
                #params = {'nevtJob': nevtJob, 'jobId': jobId, 'sid': self.cfg_params['sid'], \
                #          'broker': self.cfg_params['rb'], 'bossId': common.jobDB.bossId(nj)}
                params = {'jobId': jobId, 'sid': self.cfg_params['sid'], \
                          'broker': self.cfg_params['rb'], 'bossId': common.jobDB.bossId(nj)}
                for i in fl.readlines():
                    val = i.split(':')
                    params[val[0]] = string.strip(val[1])
                self.cfg_params['apmon'].sendToML(params)
        except:
            exctype, value = sys.exc_info()[:2]
            print "Type: %s Value: %s"%(exctype, value)
            common.logger.message("Submitter::run Exception raised: %s %s"%(exctype, value))
            common.jobDB.save()
        
        stop = time.time()
        common.logger.debug(1, "Submission Time: "+str(stop - start))
        #print "Submission Time: %d "%(stop - start)
        common.jobDB.save()
            
        msg = '\nTotal of %d jobs submitted'%njs
        if njs != len(self.nj_list) :
            msg += ' (from %d requested).'%(len(self.nj_list))
            pass
        else:
            msg += '.'
            pass
        common.logger.message(msg)
        return
Revision:	1.49
Committed:	Wed Aug 9 13:54:51 2006 UTC (18 years, 8 months ago) by slacapra
Content type:	text/x-python
Branch:	MAIN
Changes since 1.48:	+24 -17 lines
Log Message:	more clever code for listMatch: perform listmatch only if destination wrt previous job has changed
#	User	Rev	Content
1	nsmirnov	1.1	from Actor import *
2	nsmirnov	1.6	from crab_util import *
3	nsmirnov	1.2	import common
4	corvo	1.9	from ApmonIf import ApmonIf
5	spiga	1.10	import Statistic
6	corvo	1.27	from random import random
7	corvo	1.30	import time
8	nsmirnov	1.1
9			class Submitter(Actor):
10	corvo	1.40	def __init__(self, cfg_params, nj_list):
11	nsmirnov	1.1	self.cfg_params = cfg_params
12	nsmirnov	1.4	self.nj_list = nj_list
13	corvo	1.35
14	slacapra	1.25	try:
15			self.ML = int(cfg_params['USER.activate_monalisa'])
16			except KeyError:
17			self.ML = 0
18			pass
19	corvo	1.30
20	nsmirnov	1.1	return
21
22			def run(self):
23	nsmirnov	1.2	"""
24			The main method of the class.
25			"""
26			common.logger.debug(5, "Submitter::run() called")
27	slacapra	1.24
28	gutsche	1.42	totalCreatedJobs= 0
29	corvo	1.33	start = time.time()
30	slacapra	1.24	for nj in range(common.jobDB.nJobs()):
31	spiga	1.29	if (common.jobDB.status(nj)=='C') or (common.jobDB.status(nj)=='RC'): totalCreatedJobs +=1
32	slacapra	1.24	pass
33
34			if (totalCreatedJobs==0):
35			common.logger.message("No jobs to be submitted: first create them")
36	slacapra	1.28	return
37	slacapra	1.24
38	slacapra	1.49	# goodJob={}
39			# for nj in self.nj_list:
40
41
42			# if match:
43			# goodJob[nj]=1
44			# common.logger.message("Found "+str(match)+" compatible site(s) for job "+str(nj+1))
45			# else:
46			# goodJob[nj]=0
47			# common.logger.message("No compatible site found, will not submit job "+str(nj+1))
48			# # raise CrabException("No compatible site found!")
49	fanzago	1.16	#########
50	nsmirnov	1.2	# Loop over jobs
51	nsmirnov	1.6	njs = 0
52	corvo	1.15	try:
53	slacapra	1.49	lastDest=''
54	corvo	1.17	for nj in self.nj_list:
55	slacapra	1.49	# first check that status of the job is suitable for submission
56	corvo	1.17	st = common.jobDB.status(nj)
57			if st != 'C' and st != 'K' and st != 'A' and st != 'RC':
58			long_st = crabJobStatusToString(st)
59	slacapra	1.19	msg = "Job # %d not submitted: status %s"%(nj+1, long_st)
60	slacapra	1.18	common.logger.message(msg)
61	corvo	1.17	continue
62	slacapra	1.49	# SL perform listmatch only if destination has changed
63			currDest=common.jobDB.destination(nj)
64			match=0
65			if (currDest!=lastDest):
66			match = common.scheduler.listMatch(nj)
67			lastDest = currDest
68			if match:
69			common.logger.message("Found "+str(match)+" compatible site(s) for job "+str(nj+1)+" and those with same requirements")
70			else:
71			common.logger.message("No compatible site found, will not submit job "+str(nj+1)+" and those with same requirements")
72			continue
73	corvo	1.17
74			common.logger.message("Submitting job # "+`(nj+1)`)
75			jid = common.scheduler.submit(nj)
76
77			common.jobDB.setStatus(nj, 'S')
78			common.jobDB.setJobId(nj, jid)
79	corvo	1.33	common.jobDB.setTaskId(nj, self.cfg_params['taskId'])
80	corvo	1.17	njs += 1
81
82			############################################
83
84			if st == 'C':
85			resFlag = 0
86			elif st == 'RC':
87			resFlag = 2
88			else:
89			resFlag = 0
90			pass
91	spiga	1.26
92	spiga	1.38	try:
93			Statistic.Monitor('submit',resFlag,jid,'-----')
94			except:
95			pass
96	corvo	1.17
97	corvo	1.44	fl = open(common.work_space.shareDir() + '/' + self.cfg_params['apmon'].fName, 'r')
98			self.cfg_params['sid'] = jid
99	gutsche	1.47	#### FF: per il momento commentiamo nevtJob che non c'e' piu' nel jobdb
100			#nevtJob = common.jobDB.maxEvents(nj)
101	gutsche	1.46
102			# OLI: JobID treatment, special for Condor-G scheduler
103			jobId = ''
104			if common.scheduler.boss_scheduler_name == 'condor_g':
105			# create hash of cfg file
106			hash = makeCksum(common.work_space.cfgFileName())
107			jobId = str(nj + 1) + '_' + hash + '_' + self.cfg_params['sid']
108			common.logger.debug(5,'JobID for ML monitoring is created for CONDOR_G scheduler:'+jobId)
109			else:
110			jobId = str(nj + 1) + '_' + self.cfg_params['sid']
111			common.logger.debug(5,'JobID for ML monitoring is created for EDG scheduler'+jobId)
112
113	corvo	1.44	if ( jid.find(":") != -1 ) :
114			rb = jid.split(':')[1]
115			self.cfg_params['rb'] = rb.replace('//', '')
116			else :
117			self.cfg_params['rb'] = 'OSG'
118	gutsche	1.46
119	gutsche	1.47	#### FF: per il momento commentiamo nevtJob che non c'e' piu' nel jobdb
120			#params = {'nevtJob': nevtJob, 'jobId': jobId, 'sid': self.cfg_params['sid'], \
121			# 'broker': self.cfg_params['rb'], 'bossId': common.jobDB.bossId(nj)}
122			params = {'jobId': jobId, 'sid': self.cfg_params['sid'], \
123	corvo	1.44	'broker': self.cfg_params['rb'], 'bossId': common.jobDB.bossId(nj)}
124			for i in fl.readlines():
125			val = i.split(':')
126			params[val[0]] = string.strip(val[1])
127	corvo	1.45	self.cfg_params['apmon'].sendToML(params)
128	corvo	1.17	except:
129	corvo	1.27	exctype, value = sys.exc_info()[:2]
130	corvo	1.30	print "Type: %s Value: %s"%(exctype, value)
131	corvo	1.27	common.logger.message("Submitter::run Exception raised: %s %s"%(exctype, value))
132	corvo	1.17	common.jobDB.save()
133	corvo	1.30
134			stop = time.time()
135	gutsche	1.47	common.logger.debug(1, "Submission Time: "+str(stop - start))
136	fanzago	1.32	#print "Submission Time: %d "%(stop - start)
137	corvo	1.17	common.jobDB.save()
138	corvo	1.15
139	nsmirnov	1.6	msg = '\nTotal of %d jobs submitted'%njs
140			if njs != len(self.nj_list) :
141			msg += ' (from %d requested).'%(len(self.nj_list))
142			pass
143			else:
144			msg += '.'
145			pass
146	nsmirnov	1.2	common.logger.message(msg)
147	nsmirnov	1.1	return