CRAB/python/Creator.py

from Actor import Actor
from WorkSpace import WorkSpace
from JobList import JobList
from JobDB import JobDB
from ScriptWriter import ScriptWriter
from Scheduler import Scheduler
from crab_logger import Logger
from crab_exceptions import *
from crab_util import *
import common

import os, string, math

class Creator(Actor):
    def __init__(self, job_type_name, cfg_params, ncjobs):
        self.job_type_name = job_type_name
        self.job_type = None
        self.cfg_params = cfg_params
        self.total_njobs = 0
        self.total_number_of_events = 0
        self.job_number_of_events = 0
        self.first_event = 0

        ## SL No. These must not be here!            
        self.owner = cfg_params['USER.owner']
        self.dataset = cfg_params['USER.dataset'] 

        self.createJobTypeObject()
        common.logger.debug(5, __name__+": JobType "+self.job_type.name()+" created")

        self.job_type.prepareSteeringCards()
        common.logger.debug(5, __name__+": Steering cards prepared")

        self.defineTotalNumberOfJobs_()
        common.logger.debug(5, __name__+": total # of jobs = "+`self.total_njobs`)

        # Set number of jobs to be created

        self.ncjobs = ncjobs
        if ncjobs == 'all' : self.ncjobs = self.total_njobs
        if ncjobs > self.total_njobs : self.ncjobs = self.total_njobs
        
        fileCODE1 = open(common.work_space.shareDir()+"/code","a")
        fileCODE1.write('::'+str(self.ncjobs)+'::'+str(self.dataset)+'::'+str(self.owner))
        fileCODE1.close()


        #TODO: deprecated code, not needed,
        # will be eliminated when WorkSpace.saveConfiguration()
        # will be improved.
        #
        # Set/Save Job Type name

        jt_fname = common.work_space.shareDir() + 'jobtype'
        if os.path.exists(jt_fname):
            # Read stored job type name
            jt_file = open(jt_fname, 'r')
            jt = jt_file.read()
            if self.job_type_name:
                if ( jt != self.job_type_name+'\n' ):
                    msg = 'Job Type mismatch: requested <' + self.job_type_name
                    msg += '>, found <' + jt[:-1] + '>.'
                    raise CrabException(msg)
                pass
            else:
                self.job_type_name = jt[:-1]
                pass
            jt_file.close()
            pass
        else:
            # Save job type name
            jt_file = open(jt_fname, 'w')
            jt_file.write(self.job_type_name+'\n')
            jt_file.close()
            pass
        #end of deprecated code

        common.logger.debug(5, "Creator constructor finished")
        return

    def defineTotalNumberOfJobs_(self):
        """
        Calculates the total number of jobs to be created.
        """

        try:
            self.first_event = int(self.cfg_params['USER.first_event'])
        except KeyError:
            self.first_event = 0
        common.logger.debug(1,"First event ot be analyzed: "+str(self.first_event))

        # TODO Could we find a better way to get this number?
        maxAvailableEvents = int(self.job_type.maxEvents)
        common.logger.debug(1,"Available events: "+str(maxAvailableEvents))

        # some sanity check
        if self.first_event>=maxAvailableEvents:
            raise CrabException('First event is bigger than maximum number of available events!')

        # the total number of events to be analyzed
        try:
            n = self.cfg_params['USER.total_number_of_events']
            if n == 'all': n = '-1'
            if n == '-1':
                self.total_number_of_events = (maxAvailableEvents - self.first_event)
                common.logger.debug(1,"Analysing all available events "+str(self.total_number_of_events))
            else:
                if maxAvailableEvents<(int(n)+self.first_event):
                    raise CrabException('(First event + total events)='+str(int(n)+self.first_event)+' is bigger than maximum number of available events '+str(maxAvailableEvents)+' !! Use "total_number_of_events=-1" to analyze to whole dataset')
                self.total_number_of_events = int(n)
        except KeyError:
            common.logger.message("total_number_of_events not defined, set it to maximum available")
            self.total_number_of_events = (maxAvailableEvents - self.first_event)
            pass
        common.logger.message("Total number of events to be analyzed: "+str(self.total_number_of_events))


        # read user directives
        eventPerJob=0
        try:
            eventPerJob = self.cfg_params['USER.job_number_of_events']
        except KeyError:
            pass
        
        jobsPerTask=0
        try:
            jobsPerTask = int(self.cfg_params['USER.total_number_of_jobs'])
        except KeyError:
            pass

        # If both the above set, complain and use event per jobs
        if eventPerJob>0 and jobsPerTask>0:
            msg = 'Warning. '
            msg += 'job_number_of_events and total_number_of_jobs are both defined '
            msg += 'Using job_number_of_events.'
            common.logger.message(msg)
            jobsPerTask = 0 
        if eventPerJob==0 and jobsPerTask==0:
            msg = 'Warning. '
            msg += 'job_number_of_events and total_number_of_jobs are not defined '
            msg += 'Creating just one job for all events.'
            common.logger.message(msg)
            jobsPerTask = 1 

        # first case: events per job defined
        if eventPerJob>0:
            n=eventPerJob
            if n == 'all' or n == '-1' or (int(n)>self.total_number_of_events and self.total_number_of_events>0):
                common.logger.message("Asking more events than available: set it to maximum available")
                self.job_number_of_events = self.total_number_of_events
                self.total_njobs = 1
            else:
                self.job_number_of_events = int(n)
                self.total_njobs = int((self.total_number_of_events-1)/self.job_number_of_events)+1
        # second case: jobs per task defined
        elif jobsPerTask>0:
            common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+" JobPerTask "+str(jobsPerTask))
            self.job_number_of_events = int(math.floor((self.total_number_of_events)/jobsPerTask))
            self.total_njobs = jobsPerTask
        # should not happen...
        else:
            raise CrabException('Somthing wrong with splitting')

        common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+
            " events per job: "+str(self.job_number_of_events))

        return

    def writeJobsSpecsToDB(self):
        """
        Write firstEvent and maxEvents in the DB for future use
        """

        common.jobDB.load()
        # case one: write first and max events
        nJobs=self.nJobs()

        firstEvent=self.first_event
        # last jobs is different...
        for job in range(nJobs-1):
            common.jobDB.setFirstEvent(job, firstEvent)
            common.jobDB.setMaxEvents(job, self.job_number_of_events)
            firstEvent=firstEvent+self.job_number_of_events

        # this is the last job
        common.jobDB.setFirstEvent(nJobs-1, firstEvent)
        lastJobsNumberOfEvents= (self.total_number_of_events+self.first_event)-firstEvent
        common.jobDB.setMaxEvents(nJobs-1, lastJobsNumberOfEvents)
    
        if (lastJobsNumberOfEvents!=self.job_number_of_events):
            common.logger.message(str(self.total_njobs-1)+' jobs will be created for '+str(self.job_number_of_events)+' events each plus 1 for '+str(lastJobsNumberOfEvents)+' events for a total of '+str(self.job_number_of_events*(self.total_njobs-1)+lastJobsNumberOfEvents)+' events')
        else:
            common.logger.message(str(self.total_njobs)+' jobs will be created for '+str(self.job_number_of_events)+' events each for a total of '+str(self.job_number_of_events*(self.total_njobs-1)+lastJobsNumberOfEvents)+' events')

        # case two (to be implemented) write eventCollections for each jobs

        # save the DB
        common.jobDB.save()
        return

    def nJobs(self):
        return self.total_njobs
    
    def createJobTypeObject(self):
        file_name = 'cms_'+ string.lower(self.job_type_name)
        klass_name = string.capitalize(self.job_type_name)

        try:
            klass = importName(file_name, klass_name)
        except KeyError:
            msg = 'No `class '+klass_name+'` found in file `'+file_name+'.py`'
            raise CrabException(msg)
        except ImportError, e:
            msg = 'Cannot create job type '+self.job_type_name
            msg += ' (file: '+file_name+', class '+klass_name+'):\n'
            msg += str(e)
            raise CrabException(msg)

        self.job_type = klass(self.cfg_params)
        return

    def jobType(self):
        return self.job_type

    def run(self):
        """
        The main method of the class.
        """

        common.logger.debug(5, "Creator::run() called")

        # Instantiate ScriptWriter

        script_writer = ScriptWriter('crab_template.sh')

        # Loop over jobs

        njc = 0
        for nj in range(self.total_njobs):
            if njc == self.ncjobs : break
            st = common.jobDB.status(nj)
            if st != 'X': continue

            common.logger.message("Creating job # "+`(nj+1)`)

            # Prepare configuration file

            self.job_type.modifySteeringCards(nj)
            # Create script (sh)
            #print "nel for prima del modifyTemplateScript, nj vale", nj
            script_writer.modifyTemplateScript(nj)
            os.chmod(common.job_list[nj].scriptFilename(), 0744)

            # Create scheduler scripts (jdl)
            common.scheduler.createSchScript(nj)

            common.jobDB.setStatus(nj, 'C')
            # common: write input and output sandbox
            common.jobDB.setInputSandbox(nj, self.job_type.inputSandbox(nj))

            outputSandbox=self.job_type.outputSandbox(nj)
            stdout=common.job_list[nj].stdout()
            stderr=common.job_list[nj].stderr()
            outputSandbox.append(common.job_list[nj].stdout())
            # check if out!=err
            if stdout != stderr:
                outputSandbox.append(common.job_list[nj].stderr())
            common.jobDB.setOutputSandbox(nj, outputSandbox)
            common.jobDB.setTaskId(nj, self.cfg_params['taskId'])

            njc = njc + 1
            pass

        ####

        common.jobDB.save()

        msg = '\nTotal of %d jobs created'%njc
        if njc != self.ncjobs: msg = msg + ' from %d requested'%self.ncjobs
        msg = msg + '.\n'
        common.logger.message(msg)
        return
Revision:	1.18
Committed:	Fri Jan 20 16:58:47 2006 UTC (19 years, 3 months ago) by corvo
Content type:	text/x-python
Branch:	MAIN
Changes since 1.17:	+1 -0 lines
Log Message:	Moved here the task ID insertion to db/jobs
#	User	Rev	Content
1	nsmirnov	1.1	from Actor import Actor
2			from WorkSpace import WorkSpace
3			from JobList import JobList
4			from JobDB import JobDB
5			from ScriptWriter import ScriptWriter
6			from Scheduler import Scheduler
7			from crab_logger import Logger
8			from crab_exceptions import *
9			from crab_util import *
10			import common
11
12	slacapra	1.8	import os, string, math
13	nsmirnov	1.1
14			class Creator(Actor):
15			def __init__(self, job_type_name, cfg_params, ncjobs):
16			self.job_type_name = job_type_name
17			self.job_type = None
18			self.cfg_params = cfg_params
19	nsmirnov	1.2	self.total_njobs = 0
20			self.total_number_of_events = 0
21			self.job_number_of_events = 0
22	slacapra	1.7	self.first_event = 0
23	slacapra	1.14
24			## SL No. These must not be here!
25	spiga	1.11	self.owner = cfg_params['USER.owner']
26			self.dataset = cfg_params['USER.dataset']
27	nsmirnov	1.1
28			self.createJobTypeObject()
29	nsmirnov	1.3	common.logger.debug(5, __name__+": JobType "+self.job_type.name()+" created")
30	nsmirnov	1.1
31			self.job_type.prepareSteeringCards()
32	nsmirnov	1.3	common.logger.debug(5, __name__+": Steering cards prepared")
33	nsmirnov	1.1
34	nsmirnov	1.2	self.defineTotalNumberOfJobs_()
35	nsmirnov	1.3	common.logger.debug(5, __name__+": total # of jobs = "+`self.total_njobs`)
36	nsmirnov	1.1
37			# Set number of jobs to be created
38
39			self.ncjobs = ncjobs
40			if ncjobs == 'all' : self.ncjobs = self.total_njobs
41			if ncjobs > self.total_njobs : self.ncjobs = self.total_njobs
42	spiga	1.11
43	slacapra	1.16	fileCODE1 = open(common.work_space.shareDir()+"/code","a")
44	spiga	1.11	fileCODE1.write('::'+str(self.ncjobs)+'::'+str(self.dataset)+'::'+str(self.owner))
45			fileCODE1.close()
46
47	nsmirnov	1.1
48			#TODO: deprecated code, not needed,
49			# will be eliminated when WorkSpace.saveConfiguration()
50			# will be improved.
51			#
52			# Set/Save Job Type name
53
54			jt_fname = common.work_space.shareDir() + 'jobtype'
55			if os.path.exists(jt_fname):
56			# Read stored job type name
57			jt_file = open(jt_fname, 'r')
58			jt = jt_file.read()
59			if self.job_type_name:
60			if ( jt != self.job_type_name+'\n' ):
61			msg = 'Job Type mismatch: requested <' + self.job_type_name
62			msg += '>, found <' + jt[:-1] + '>.'
63			raise CrabException(msg)
64			pass
65			else:
66			self.job_type_name = jt[:-1]
67			pass
68			jt_file.close()
69			pass
70			else:
71			# Save job type name
72			jt_file = open(jt_fname, 'w')
73			jt_file.write(self.job_type_name+'\n')
74			jt_file.close()
75			pass
76			#end of deprecated code
77
78	nsmirnov	1.3	common.logger.debug(5, "Creator constructor finished")
79	nsmirnov	1.1	return
80
81	nsmirnov	1.2	def defineTotalNumberOfJobs_(self):
82			"""
83			Calculates the total number of jobs to be created.
84			"""
85
86			try:
87	slacapra	1.8	self.first_event = int(self.cfg_params['USER.first_event'])
88	slacapra	1.7	except KeyError:
89			self.first_event = 0
90	slacapra	1.8	common.logger.debug(1,"First event ot be analyzed: "+str(self.first_event))
91	slacapra	1.7
92	slacapra	1.10	# TODO Could we find a better way to get this number?
93	slacapra	1.8	maxAvailableEvents = int(self.job_type.maxEvents)
94			common.logger.debug(1,"Available events: "+str(maxAvailableEvents))
95	slacapra	1.7
96	slacapra	1.8	# some sanity check
97			if self.first_event>=maxAvailableEvents:
98	slacapra	1.7	raise CrabException('First event is bigger than maximum number of available events!')
99
100	slacapra	1.8	# the total number of events to be analyzed
101	slacapra	1.7	try:
102	nsmirnov	1.2	n = self.cfg_params['USER.total_number_of_events']
103			if n == 'all': n = '-1'
104	slacapra	1.8	if n == '-1':
105			self.total_number_of_events = (maxAvailableEvents - self.first_event)
106			common.logger.debug(1,"Analysing all available events "+str(self.total_number_of_events))
107			else:
108			if maxAvailableEvents<(int(n)+self.first_event):
109	slacapra	1.17	raise CrabException('(First event + total events)='+str(int(n)+self.first_event)+' is bigger than maximum number of available events '+str(maxAvailableEvents)+' !! Use "total_number_of_events=-1" to analyze to whole dataset')
110	slacapra	1.8	self.total_number_of_events = int(n)
111	slacapra	1.7	except KeyError:
112			common.logger.message("total_number_of_events not defined, set it to maximum available")
113	slacapra	1.8	self.total_number_of_events = (maxAvailableEvents - self.first_event)
114	slacapra	1.7	pass
115	slacapra	1.8	common.logger.message("Total number of events to be analyzed: "+str(self.total_number_of_events))
116
117	slacapra	1.7
118	slacapra	1.8	# read user directives
119	slacapra	1.7	eventPerJob=0
120			try:
121			eventPerJob = self.cfg_params['USER.job_number_of_events']
122	nsmirnov	1.2	except KeyError:
123			pass
124
125	slacapra	1.7	jobsPerTask=0
126	nsmirnov	1.2	try:
127	slacapra	1.7	jobsPerTask = int(self.cfg_params['USER.total_number_of_jobs'])
128	nsmirnov	1.2	except KeyError:
129	slacapra	1.7	pass
130
131			# If both the above set, complain and use event per jobs
132			if eventPerJob>0 and jobsPerTask>0:
133	nsmirnov	1.2	msg = 'Warning. '
134	slacapra	1.7	msg += 'job_number_of_events and total_number_of_jobs are both defined '
135			msg += 'Using job_number_of_events.'
136	nsmirnov	1.2	common.logger.message(msg)
137	slacapra	1.7	jobsPerTask = 0
138			if eventPerJob==0 and jobsPerTask==0:
139			msg = 'Warning. '
140			msg += 'job_number_of_events and total_number_of_jobs are not defined '
141			msg += 'Creating just one job for all events.'
142			common.logger.message(msg)
143			jobsPerTask = 1
144
145			# first case: events per job defined
146			if eventPerJob>0:
147			n=eventPerJob
148			if n == 'all' or n == '-1' or (int(n)>self.total_number_of_events and self.total_number_of_events>0):
149			common.logger.message("Asking more events than available: set it to maximum available")
150			self.job_number_of_events = self.total_number_of_events
151	slacapra	1.8	self.total_njobs = 1
152	slacapra	1.7	else:
153			self.job_number_of_events = int(n)
154	slacapra	1.8	self.total_njobs = int((self.total_number_of_events-1)/self.job_number_of_events)+1
155	slacapra	1.7	# second case: jobs per task defined
156			elif jobsPerTask>0:
157	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+" JobPerTask "+str(jobsPerTask))
158			self.job_number_of_events = int(math.floor((self.total_number_of_events)/jobsPerTask))
159			self.total_njobs = jobsPerTask
160	slacapra	1.7	# should not happen...
161			else:
162			raise CrabException('Somthing wrong with splitting')
163
164	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+
165	slacapra	1.7	" events per job: "+str(self.job_number_of_events))
166	slacapra	1.8
167	slacapra	1.7	return
168
169			def writeJobsSpecsToDB(self):
170			"""
171			Write firstEvent and maxEvents in the DB for future use
172			"""
173
174			common.jobDB.load()
175			# case one: write first and max events
176			nJobs=self.nJobs()
177
178			firstEvent=self.first_event
179			# last jobs is different...
180			for job in range(nJobs-1):
181			common.jobDB.setFirstEvent(job, firstEvent)
182			common.jobDB.setMaxEvents(job, self.job_number_of_events)
183			firstEvent=firstEvent+self.job_number_of_events
184
185			# this is the last job
186			common.jobDB.setFirstEvent(nJobs-1, firstEvent)
187	slacapra	1.8	lastJobsNumberOfEvents= (self.total_number_of_events+self.first_event)-firstEvent
188			common.jobDB.setMaxEvents(nJobs-1, lastJobsNumberOfEvents)
189	slacapra	1.7
190	slacapra	1.15	if (lastJobsNumberOfEvents!=self.job_number_of_events):
191			common.logger.message(str(self.total_njobs-1)+' jobs will be created for '+str(self.job_number_of_events)+' events each plus 1 for '+str(lastJobsNumberOfEvents)+' events for a total of '+str(self.job_number_of_events*(self.total_njobs-1)+lastJobsNumberOfEvents)+' events')
192			else:
193			common.logger.message(str(self.total_njobs)+' jobs will be created for '+str(self.job_number_of_events)+' events each for a total of '+str(self.job_number_of_events*(self.total_njobs-1)+lastJobsNumberOfEvents)+' events')
194	slacapra	1.7
195			# case two (to be implemented) write eventCollections for each jobs
196	nsmirnov	1.2
197	slacapra	1.9	# save the DB
198			common.jobDB.save()
199	nsmirnov	1.2	return
200
201	nsmirnov	1.1	def nJobs(self):
202			return self.total_njobs
203
204			def createJobTypeObject(self):
205			file_name = 'cms_'+ string.lower(self.job_type_name)
206			klass_name = string.capitalize(self.job_type_name)
207
208			try:
209			klass = importName(file_name, klass_name)
210			except KeyError:
211			msg = 'No `class '+klass_name+'` found in file `'+file_name+'.py`'
212			raise CrabException(msg)
213			except ImportError, e:
214			msg = 'Cannot create job type '+self.job_type_name
215			msg += ' (file: '+file_name+', class '+klass_name+'):\n'
216			msg += str(e)
217			raise CrabException(msg)
218
219			self.job_type = klass(self.cfg_params)
220			return
221	nsmirnov	1.5
222			def jobType(self):
223			return self.job_type
224	nsmirnov	1.1
225			def run(self):
226	nsmirnov	1.4	"""
227			The main method of the class.
228			"""
229	nsmirnov	1.1
230	nsmirnov	1.3	common.logger.debug(5, "Creator::run() called")
231
232	nsmirnov	1.1	# Instantiate ScriptWriter
233
234			script_writer = ScriptWriter('crab_template.sh')
235
236			# Loop over jobs
237
238			njc = 0
239	nsmirnov	1.4	for nj in range(self.total_njobs):
240	nsmirnov	1.1	if njc == self.ncjobs : break
241	nsmirnov	1.4	st = common.jobDB.status(nj)
242	nsmirnov	1.1	if st != 'X': continue
243
244	nsmirnov	1.6	common.logger.message("Creating job # "+`(nj+1)`)
245	nsmirnov	1.3
246	nsmirnov	1.1	# Prepare configuration file
247
248	nsmirnov	1.4	self.job_type.modifySteeringCards(nj)
249	fanzago	1.12	# Create script (sh)
250	fanzago	1.13	#print "nel for prima del modifyTemplateScript, nj vale", nj
251	nsmirnov	1.4	script_writer.modifyTemplateScript(nj)
252			os.chmod(common.job_list[nj].scriptFilename(), 0744)
253	nsmirnov	1.1
254	fanzago	1.12	# Create scheduler scripts (jdl)
255			common.scheduler.createSchScript(nj)
256
257	nsmirnov	1.4	common.jobDB.setStatus(nj, 'C')
258	slacapra	1.9	# common: write input and output sandbox
259			common.jobDB.setInputSandbox(nj, self.job_type.inputSandbox(nj))
260
261			outputSandbox=self.job_type.outputSandbox(nj)
262			stdout=common.job_list[nj].stdout()
263			stderr=common.job_list[nj].stderr()
264			outputSandbox.append(common.job_list[nj].stdout())
265			# check if out!=err
266			if stdout != stderr:
267			outputSandbox.append(common.job_list[nj].stderr())
268			common.jobDB.setOutputSandbox(nj, outputSandbox)
269	corvo	1.18	common.jobDB.setTaskId(nj, self.cfg_params['taskId'])
270	slacapra	1.9
271	nsmirnov	1.1	njc = njc + 1
272			pass
273
274			####
275	fanzago	1.12
276	nsmirnov	1.1	common.jobDB.save()
277
278			msg = '\nTotal of %d jobs created'%njc
279			if njc != self.ncjobs: msg = msg + ' from %d requested'%self.ncjobs
280			msg = msg + '.\n'
281			common.logger.message(msg)
282			return