ViewVC Help
View File | Revision Log | Show Annotations | Root Listing
root/cvsroot/COMP/CRAB/python/Submitter.py
Revision: 1.74
Committed: Mon Dec 4 11:13:20 2006 UTC (18 years, 5 months ago) by corvo
Content type: text/x-python
Branch: MAIN
CVS Tags: CRAB_1_5_0_pre6, CRAB_1_5_0_pre5, CRAB_1_5_0_pre4, CRAB_1_5_0_pre3, CRAB_1_5_0_pre2
Changes since 1.73: +6 -6 lines
Log Message:
pbar recovered

File Contents

# 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 slacapra 1.60 #from random import random
7 corvo 1.30 import time
8 slacapra 1.60 from ProgressBar import ProgressBar
9     from TerminalController import TerminalController
10 nsmirnov 1.1
11     class Submitter(Actor):
12 corvo 1.40 def __init__(self, cfg_params, nj_list):
13 nsmirnov 1.1 self.cfg_params = cfg_params
14 nsmirnov 1.4 self.nj_list = nj_list
15 gutsche 1.58
16     if common.scheduler.boss_scheduler_name == 'condor_g':
17     # create hash of cfg file
18     self.hash = makeCksum(common.work_space.cfgFileName())
19     else:
20     self.hash = ''
21 corvo 1.30
22 nsmirnov 1.1 return
23    
24     def run(self):
25 nsmirnov 1.2 """
26 slacapra 1.53 The main method of the class: submit jobs in range self.nj_list
27 nsmirnov 1.2 """
28     common.logger.debug(5, "Submitter::run() called")
29 slacapra 1.24
30 gutsche 1.42 totalCreatedJobs= 0
31 corvo 1.33 start = time.time()
32 slacapra 1.24 for nj in range(common.jobDB.nJobs()):
33 spiga 1.29 if (common.jobDB.status(nj)=='C') or (common.jobDB.status(nj)=='RC'): totalCreatedJobs +=1
34 slacapra 1.24 pass
35    
36     if (totalCreatedJobs==0):
37     common.logger.message("No jobs to be submitted: first create them")
38 slacapra 1.28 return
39 slacapra 1.60
40 gutsche 1.70 # submit pre DashBoard information
41     params = {'jobId':'TaskMeta'}
42    
43     fl = open(common.work_space.shareDir() + '/' + self.cfg_params['apmon'].fName, 'r')
44     for i in fl.readlines():
45     val = i.split(':')
46     params[val[0]] = string.strip(val[1])
47     fl.close()
48    
49     common.logger.debug(5,'Submission DashBoard Pre-Submission report: '+str(params))
50    
51     self.cfg_params['apmon'].sendToML(params)
52    
53 fanzago 1.16 #########
54 nsmirnov 1.2 # Loop over jobs
55 nsmirnov 1.6 njs = 0
56 corvo 1.15 try:
57 slacapra 1.69 list=[]
58 spiga 1.66 list_of_list = []
59 slacapra 1.57 lastBlock=-1
60 spiga 1.66 count = 0
61 corvo 1.17 for nj in self.nj_list:
62 slacapra 1.53 same=0
63 slacapra 1.49 # first check that status of the job is suitable for submission
64 corvo 1.17 st = common.jobDB.status(nj)
65 slacapra 1.69 if st != 'C' and st != 'K' and st != 'A' and st != 'RC':
66 corvo 1.17 long_st = crabJobStatusToString(st)
67 slacapra 1.19 msg = "Job # %d not submitted: status %s"%(nj+1, long_st)
68 slacapra 1.18 common.logger.message(msg)
69 corvo 1.17 continue
70 corvo 1.72
71 slacapra 1.57 currBlock = common.jobDB.block(nj)
72     # SL perform listmatch only if block has changed
73     if (currBlock!=lastBlock):
74 gutsche 1.54 if common.scheduler.boss_scheduler_name != "condor_g" :
75 slacapra 1.57 match = common.scheduler.listMatch(nj, currBlock)
76 gutsche 1.54 else :
77     match = "1"
78 slacapra 1.57 lastBlock = currBlock
79 mkirn 1.50 else:
80 slacapra 1.53 common.logger.debug(1,"Sites for job "+str(nj+1)+" the same as previous job")
81     same=1
82 corvo 1.72
83 slacapra 1.49 if match:
84 slacapra 1.53 if not same:
85     common.logger.message("Found "+str(match)+" compatible site(s) for job "+str(nj+1))
86     else:
87     common.logger.debug(1,"Found "+str(match)+" compatible site(s) for job "+str(nj+1))
88 slacapra 1.69 list.append(common.jobDB.bossId(nj))
89 corvo 1.72
90 slacapra 1.69 if nj == self.nj_list[-1]: # check that is not the last job in the list
91     list_of_list.append([currBlock,list])
92     else: # check if next job has same group
93     nextBlock = common.jobDB.block(nj+1)
94     if currBlock != nextBlock : # if not, close this group and reset
95 spiga 1.66 list_of_list.append([currBlock,list])
96 slacapra 1.69 list=[]
97 slacapra 1.49 else:
98 mkirn 1.50 common.logger.message("No compatible site found, will not submit job "+str(nj+1))
99 slacapra 1.49 continue
100 spiga 1.66 count += 1
101 gutsche 1.65 ### Progress Bar indicator, deactivate for debug
102     if not common.logger.debugLevel() :
103     term = TerminalController()
104 corvo 1.72
105 spiga 1.66 for ii in range(len(list_of_list)): # Add loop DS
106 slacapra 1.68 common.logger.debug(1,'Submitting jobs '+str(list_of_list[ii][1]))
107 corvo 1.74 if not common.logger.debugLevel() :
108     try: pbar = ProgressBar(term, 'Submitting '+str(len(list_of_list[ii][1]))+' jobs')
109     except: pbar = None
110 corvo 1.72
111 corvo 1.67 jidLista, bjidLista = common.scheduler.submit(list_of_list[ii])
112 slacapra 1.69 bjidLista = map(int, bjidLista) # cast all bjidLista to int
113 corvo 1.72
114 corvo 1.74 if not common.logger.debugLevel():
115     if pbar :
116     pbar.update(float(ii+1)/float(len(list_of_list)),'please wait')
117 corvo 1.72
118 corvo 1.67 for jj in bjidLista: # Add loop over SID returned from group submission DS
119 slacapra 1.69 tmpNj = jj - 1
120 corvo 1.67 jid=jidLista[bjidLista.index(jj)]
121 corvo 1.72 common.logger.debug(5,"Submitted job # "+ `(jj)`)
122 slacapra 1.69 common.jobDB.setStatus(tmpNj, 'S')
123     common.jobDB.setJobId(tmpNj, jid)
124     common.jobDB.setTaskId(tmpNj, self.cfg_params['taskId'])
125 spiga 1.56 njs += 1
126    
127 slacapra 1.69 ##### DashBoard report #####################
128     try:
129 spiga 1.56 resFlag = 0
130 slacapra 1.69 if st == 'RC': resFlag = 2
131 spiga 1.71 Statistic.Monitor('submit',resFlag,jid,'-----','dest')
132 spiga 1.56 except:
133     pass
134    
135     # OLI: JobID treatment, special for Condor-G scheduler
136     jobId = ''
137     if common.scheduler.boss_scheduler_name == 'condor_g':
138 corvo 1.72 jobId = str(jj) + '_' + self.hash + '_' + jid
139 spiga 1.56 common.logger.debug(5,'JobID for ML monitoring is created for CONDOR_G scheduler:'+jobId)
140     else:
141 corvo 1.72 jobId = str(jj) + '_' + jid
142 spiga 1.56 common.logger.debug(5,'JobID for ML monitoring is created for EDG scheduler'+jobId)
143    
144     if ( jid.find(":") != -1 ) :
145     rb = jid.split(':')[1]
146 slacapra 1.69 rb = rb.replace('//', '')
147 spiga 1.56 else :
148 slacapra 1.69 rb = 'OSG'
149 spiga 1.56
150     params = {'jobId': jobId, \
151 slacapra 1.69 'sid': jid, \
152     'broker': rb, \
153 corvo 1.72 'bossId': jj, \
154     'TargetSE': string.join((common.jobDB.destination(tmpNj)),",")}
155 spiga 1.56
156 slacapra 1.69 fl = open(common.work_space.shareDir() + '/' + self.cfg_params['apmon'].fName, 'r')
157 spiga 1.56 for i in fl.readlines():
158     val = i.split(':')
159     params[val[0]] = string.strip(val[1])
160 slacapra 1.69 fl.close()
161 corvo 1.72
162 gutsche 1.65 common.logger.debug(5,'Submission DashBoard report: '+str(params))
163    
164 spiga 1.56 self.cfg_params['apmon'].sendToML(params)
165 slacapra 1.69 pass
166     pass
167 corvo 1.17
168     except:
169 corvo 1.27 exctype, value = sys.exc_info()[:2]
170 corvo 1.30 print "Type: %s Value: %s"%(exctype, value)
171 corvo 1.27 common.logger.message("Submitter::run Exception raised: %s %s"%(exctype, value))
172 corvo 1.17 common.jobDB.save()
173 corvo 1.30
174     stop = time.time()
175 gutsche 1.47 common.logger.debug(1, "Submission Time: "+str(stop - start))
176 slacapra 1.60 common.logger.write("Submission time :"+str(stop - start))
177 corvo 1.17 common.jobDB.save()
178 corvo 1.15
179 nsmirnov 1.6 msg = '\nTotal of %d jobs submitted'%njs
180     if njs != len(self.nj_list) :
181     msg += ' (from %d requested).'%(len(self.nj_list))
182     pass
183     else:
184     msg += '.'
185     pass
186 nsmirnov 1.2 common.logger.message(msg)
187 slacapra 1.73 ## add some more verbose message in case submission is not complete
188     if (njs < len(self.nj_list)):
189     msg = 'Submission performed using the Requirements: \n'
190     msg += string.upper(common.taskDB.dict("JobType"))+' version: '+string.upper(common.taskDB.dict("codeVersion"))+'\n'
191     try: msg += 'SE White List: '+self.cfg_params['EDG.se_white_list']+'\n'
192     except KeyError: pass
193     try: msg += 'SE Black List: '+self.cfg_params['EDG.se_black_list']+'\n'
194     except KeyError: pass
195     try: msg += 'CE White List: '+self.cfg_params['EDG.ce_white_list']+'\n'
196     except KeyError: pass
197     try: msg += 'CE Black List: '+self.cfg_params['EDG.ce_black_list']+'\n'
198     except KeyError: pass
199     msg += '(Hint: please check if '+string.upper(common.taskDB.dict("JobType"))+' is available at the Sites)\n'
200    
201     common.logger.message(msg)
202    
203 nsmirnov 1.1 return