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.ncjobs = 0                  # nb of jobs to be created
        self.total_number_of_events = 0
        self.job_number_of_events = 0
        self.first_event = 0
        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.logDir()+"/.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)+' !!')
                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)
    
        common.logger.message('Will be created '+str(self.total_njobs-1)+' jobs for '+str(self.job_number_of_events)+' each plus 1 for '+str(lastJobsNumberOfEvents)+' 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)

            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)

            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:	1.12
Committed:	Wed Aug 31 16:13:57 2005 UTC (19 years, 8 months ago) by fanzago
Content type:	text/x-python
Branch:	MAIN
CVS Tags:	CRAB_1_0_0_pre1_boss
Changes since 1.11:	+6 -20 lines
Log Message:	changed flow...now jdl creation is before sh creation
#	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.ncjobs = 0 # nb of jobs to be created
21			self.total_number_of_events = 0
22			self.job_number_of_events = 0
23	slacapra	1.7	self.first_event = 0
24	spiga	1.11	self.owner = cfg_params['USER.owner']
25			self.dataset = cfg_params['USER.dataset']
26	nsmirnov	1.1
27			self.createJobTypeObject()
28	nsmirnov	1.3	common.logger.debug(5, __name__+": JobType "+self.job_type.name()+" created")
29	nsmirnov	1.1
30			self.job_type.prepareSteeringCards()
31	nsmirnov	1.3	common.logger.debug(5, __name__+": Steering cards prepared")
32	nsmirnov	1.1
33	nsmirnov	1.2	self.defineTotalNumberOfJobs_()
34	nsmirnov	1.3	common.logger.debug(5, __name__+": total # of jobs = "+`self.total_njobs`)
35	nsmirnov	1.1
36			# Set number of jobs to be created
37
38			self.ncjobs = ncjobs
39			if ncjobs == 'all' : self.ncjobs = self.total_njobs
40			if ncjobs > self.total_njobs : self.ncjobs = self.total_njobs
41	spiga	1.11
42			fileCODE1 = open(common.work_space.logDir()+"/.code","a")
43			fileCODE1.write('::'+str(self.ncjobs)+'::'+str(self.dataset)+'::'+str(self.owner))
44			fileCODE1.close()
45
46	nsmirnov	1.1
47			#TODO: deprecated code, not needed,
48			# will be eliminated when WorkSpace.saveConfiguration()
49			# will be improved.
50			#
51			# Set/Save Job Type name
52
53			jt_fname = common.work_space.shareDir() + 'jobtype'
54			if os.path.exists(jt_fname):
55			# Read stored job type name
56			jt_file = open(jt_fname, 'r')
57			jt = jt_file.read()
58			if self.job_type_name:
59			if ( jt != self.job_type_name+'\n' ):
60			msg = 'Job Type mismatch: requested <' + self.job_type_name
61			msg += '>, found <' + jt[:-1] + '>.'
62			raise CrabException(msg)
63			pass
64			else:
65			self.job_type_name = jt[:-1]
66			pass
67			jt_file.close()
68			pass
69			else:
70			# Save job type name
71			jt_file = open(jt_fname, 'w')
72			jt_file.write(self.job_type_name+'\n')
73			jt_file.close()
74			pass
75			#end of deprecated code
76
77	nsmirnov	1.3	common.logger.debug(5, "Creator constructor finished")
78	nsmirnov	1.1	return
79
80	nsmirnov	1.2	def defineTotalNumberOfJobs_(self):
81			"""
82			Calculates the total number of jobs to be created.
83			"""
84
85			try:
86	slacapra	1.8	self.first_event = int(self.cfg_params['USER.first_event'])
87	slacapra	1.7	except KeyError:
88			self.first_event = 0
89	slacapra	1.8	common.logger.debug(1,"First event ot be analyzed: "+str(self.first_event))
90	slacapra	1.7
91	slacapra	1.10	# TODO Could we find a better way to get this number?
92	slacapra	1.8	maxAvailableEvents = int(self.job_type.maxEvents)
93			common.logger.debug(1,"Available events: "+str(maxAvailableEvents))
94	slacapra	1.7
95	slacapra	1.8	# some sanity check
96			if self.first_event>=maxAvailableEvents:
97	slacapra	1.7	raise CrabException('First event is bigger than maximum number of available events!')
98
99	slacapra	1.8	# the total number of events to be analyzed
100	slacapra	1.7	try:
101	nsmirnov	1.2	n = self.cfg_params['USER.total_number_of_events']
102			if n == 'all': n = '-1'
103	slacapra	1.8	if n == '-1':
104			self.total_number_of_events = (maxAvailableEvents - self.first_event)
105			common.logger.debug(1,"Analysing all available events "+str(self.total_number_of_events))
106			else:
107			if maxAvailableEvents<(int(n)+self.first_event):
108			raise CrabException('(First event + total events)='+str(int(n)+self.first_event)+' is bigger than maximum number of available events '+str(maxAvailableEvents)+' !!')
109			self.total_number_of_events = int(n)
110	slacapra	1.7	except KeyError:
111			common.logger.message("total_number_of_events not defined, set it to maximum available")
112	slacapra	1.8	self.total_number_of_events = (maxAvailableEvents - self.first_event)
113	slacapra	1.7	pass
114	slacapra	1.8	common.logger.message("Total number of events to be analyzed: "+str(self.total_number_of_events))
115
116	slacapra	1.7
117	slacapra	1.8	# read user directives
118	slacapra	1.7	eventPerJob=0
119			try:
120			eventPerJob = self.cfg_params['USER.job_number_of_events']
121	nsmirnov	1.2	except KeyError:
122			pass
123
124	slacapra	1.7	jobsPerTask=0
125	nsmirnov	1.2	try:
126	slacapra	1.7	jobsPerTask = int(self.cfg_params['USER.total_number_of_jobs'])
127	nsmirnov	1.2	except KeyError:
128	slacapra	1.7	pass
129
130			# If both the above set, complain and use event per jobs
131			if eventPerJob>0 and jobsPerTask>0:
132	nsmirnov	1.2	msg = 'Warning. '
133	slacapra	1.7	msg += 'job_number_of_events and total_number_of_jobs are both defined '
134			msg += 'Using job_number_of_events.'
135	nsmirnov	1.2	common.logger.message(msg)
136	slacapra	1.7	jobsPerTask = 0
137			if eventPerJob==0 and jobsPerTask==0:
138			msg = 'Warning. '
139			msg += 'job_number_of_events and total_number_of_jobs are not defined '
140			msg += 'Creating just one job for all events.'
141			common.logger.message(msg)
142			jobsPerTask = 1
143
144			# first case: events per job defined
145			if eventPerJob>0:
146			n=eventPerJob
147			if n == 'all' or n == '-1' or (int(n)>self.total_number_of_events and self.total_number_of_events>0):
148			common.logger.message("Asking more events than available: set it to maximum available")
149			self.job_number_of_events = self.total_number_of_events
150	slacapra	1.8	self.total_njobs = 1
151	slacapra	1.7	else:
152			self.job_number_of_events = int(n)
153	slacapra	1.8	self.total_njobs = int((self.total_number_of_events-1)/self.job_number_of_events)+1
154	slacapra	1.7	# second case: jobs per task defined
155			elif jobsPerTask>0:
156	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+" JobPerTask "+str(jobsPerTask))
157			self.job_number_of_events = int(math.floor((self.total_number_of_events)/jobsPerTask))
158			self.total_njobs = jobsPerTask
159	slacapra	1.7	# should not happen...
160			else:
161			raise CrabException('Somthing wrong with splitting')
162
163	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+
164	slacapra	1.7	" events per job: "+str(self.job_number_of_events))
165	slacapra	1.8
166	slacapra	1.7	return
167
168			def writeJobsSpecsToDB(self):
169			"""
170			Write firstEvent and maxEvents in the DB for future use
171			"""
172
173			common.jobDB.load()
174			# case one: write first and max events
175			nJobs=self.nJobs()
176
177			firstEvent=self.first_event
178			# last jobs is different...
179			for job in range(nJobs-1):
180			common.jobDB.setFirstEvent(job, firstEvent)
181			common.jobDB.setMaxEvents(job, self.job_number_of_events)
182			firstEvent=firstEvent+self.job_number_of_events
183
184			# this is the last job
185			common.jobDB.setFirstEvent(nJobs-1, firstEvent)
186	slacapra	1.8	lastJobsNumberOfEvents= (self.total_number_of_events+self.first_event)-firstEvent
187			common.jobDB.setMaxEvents(nJobs-1, lastJobsNumberOfEvents)
188	slacapra	1.7
189	fanzago	1.12	common.logger.message('Will be created '+str(self.total_njobs-1)+' jobs for '+str(self.job_number_of_events)+' each plus 1 for '+str(lastJobsNumberOfEvents)+' for a total of '+str(self.job_number_of_events*(self.total_njobs-1)+lastJobsNumberOfEvents)+' events')
190	slacapra	1.7
191			# case two (to be implemented) write eventCollections for each jobs
192	nsmirnov	1.2
193	slacapra	1.9	# save the DB
194			common.jobDB.save()
195	nsmirnov	1.2	return
196
197	nsmirnov	1.1	def nJobs(self):
198			return self.total_njobs
199
200			def createJobTypeObject(self):
201			file_name = 'cms_'+ string.lower(self.job_type_name)
202			klass_name = string.capitalize(self.job_type_name)
203
204			try:
205			klass = importName(file_name, klass_name)
206			except KeyError:
207			msg = 'No `class '+klass_name+'` found in file `'+file_name+'.py`'
208			raise CrabException(msg)
209			except ImportError, e:
210			msg = 'Cannot create job type '+self.job_type_name
211			msg += ' (file: '+file_name+', class '+klass_name+'):\n'
212			msg += str(e)
213			raise CrabException(msg)
214
215			self.job_type = klass(self.cfg_params)
216			return
217	nsmirnov	1.5
218			def jobType(self):
219			return self.job_type
220	nsmirnov	1.1
221			def run(self):
222	nsmirnov	1.4	"""
223			The main method of the class.
224			"""
225	nsmirnov	1.1
226	nsmirnov	1.3	common.logger.debug(5, "Creator::run() called")
227
228	nsmirnov	1.1	# Instantiate ScriptWriter
229
230			script_writer = ScriptWriter('crab_template.sh')
231
232			# Loop over jobs
233
234			njc = 0
235	nsmirnov	1.4	for nj in range(self.total_njobs):
236	nsmirnov	1.1	if njc == self.ncjobs : break
237	nsmirnov	1.4	st = common.jobDB.status(nj)
238	nsmirnov	1.1	if st != 'X': continue
239
240	nsmirnov	1.6	common.logger.message("Creating job # "+`(nj+1)`)
241	nsmirnov	1.3
242	nsmirnov	1.1	# Prepare configuration file
243
244	nsmirnov	1.4	self.job_type.modifySteeringCards(nj)
245	fanzago	1.12	# Create script (sh)
246	nsmirnov	1.1
247	nsmirnov	1.4	script_writer.modifyTemplateScript(nj)
248			os.chmod(common.job_list[nj].scriptFilename(), 0744)
249	nsmirnov	1.1
250	fanzago	1.12	# Create scheduler scripts (jdl)
251			common.scheduler.createSchScript(nj)
252
253	nsmirnov	1.4	common.jobDB.setStatus(nj, 'C')
254	slacapra	1.9	# common: write input and output sandbox
255			common.jobDB.setInputSandbox(nj, self.job_type.inputSandbox(nj))
256
257			outputSandbox=self.job_type.outputSandbox(nj)
258			stdout=common.job_list[nj].stdout()
259			stderr=common.job_list[nj].stderr()
260			outputSandbox.append(common.job_list[nj].stdout())
261			# check if out!=err
262			if stdout != stderr:
263			outputSandbox.append(common.job_list[nj].stderr())
264			common.jobDB.setOutputSandbox(nj, outputSandbox)
265
266	nsmirnov	1.1	njc = njc + 1
267			pass
268
269			####
270	fanzago	1.12
271	nsmirnov	1.1	common.jobDB.save()
272
273			msg = '\nTotal of %d jobs created'%njc
274			if njc != self.ncjobs: msg = msg + ' from %d requested'%self.ncjobs
275			msg = msg + '.\n'
276			common.logger.message(msg)
277			return