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.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

        #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('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 JDL
            # Maybe, it worths to move this call into Submitter,
            # i.e. to create scheduler-specific file at submission time ?

            common.scheduler.createJDL(nj)

            # Create script

            script_writer.modifyTemplateScript(nj)
            os.chmod(common.job_list[nj].scriptFilename(), 0744)

            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:	1.10
Committed:	Wed Aug 10 16:52:51 2005 UTC (19 years, 8 months ago) by slacapra
Content type:	text/x-python
Branch:	MAIN
CVS Tags:	CRAB_1_0_0_pre3, CRAB_1_0_0_pre2
Changes since 1.9:	+1 -0 lines
Log Message:	more check on DB plus other
#	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	nsmirnov	1.2
25	nsmirnov	1.1	#
26
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
42			#TODO: deprecated code, not needed,
43			# will be eliminated when WorkSpace.saveConfiguration()
44			# will be improved.
45			#
46			# Set/Save Job Type name
47
48			jt_fname = common.work_space.shareDir() + 'jobtype'
49			if os.path.exists(jt_fname):
50			# Read stored job type name
51			jt_file = open(jt_fname, 'r')
52			jt = jt_file.read()
53			if self.job_type_name:
54			if ( jt != self.job_type_name+'\n' ):
55			msg = 'Job Type mismatch: requested <' + self.job_type_name
56			msg += '>, found <' + jt[:-1] + '>.'
57			raise CrabException(msg)
58			pass
59			else:
60			self.job_type_name = jt[:-1]
61			pass
62			jt_file.close()
63			pass
64			else:
65			# Save job type name
66			jt_file = open(jt_fname, 'w')
67			jt_file.write(self.job_type_name+'\n')
68			jt_file.close()
69			pass
70			#end of deprecated code
71
72	nsmirnov	1.3	common.logger.debug(5, "Creator constructor finished")
73	nsmirnov	1.1	return
74
75	nsmirnov	1.2	def defineTotalNumberOfJobs_(self):
76			"""
77			Calculates the total number of jobs to be created.
78			"""
79
80			try:
81	slacapra	1.8	self.first_event = int(self.cfg_params['USER.first_event'])
82	slacapra	1.7	except KeyError:
83			self.first_event = 0
84	slacapra	1.8	common.logger.debug(1,"First event ot be analyzed: "+str(self.first_event))
85	slacapra	1.7
86	slacapra	1.10	# TODO Could we find a better way to get this number?
87	slacapra	1.8	maxAvailableEvents = int(self.job_type.maxEvents)
88			common.logger.debug(1,"Available events: "+str(maxAvailableEvents))
89	slacapra	1.7
90	slacapra	1.8	# some sanity check
91			if self.first_event>=maxAvailableEvents:
92	slacapra	1.7	raise CrabException('First event is bigger than maximum number of available events!')
93
94	slacapra	1.8	# the total number of events to be analyzed
95	slacapra	1.7	try:
96	nsmirnov	1.2	n = self.cfg_params['USER.total_number_of_events']
97			if n == 'all': n = '-1'
98	slacapra	1.8	if n == '-1':
99			self.total_number_of_events = (maxAvailableEvents - self.first_event)
100			common.logger.debug(1,"Analysing all available events "+str(self.total_number_of_events))
101			else:
102			if maxAvailableEvents<(int(n)+self.first_event):
103			raise CrabException('(First event + total events)='+str(int(n)+self.first_event)+' is bigger than maximum number of available events '+str(maxAvailableEvents)+' !!')
104			self.total_number_of_events = int(n)
105	slacapra	1.7	except KeyError:
106			common.logger.message("total_number_of_events not defined, set it to maximum available")
107	slacapra	1.8	self.total_number_of_events = (maxAvailableEvents - self.first_event)
108	slacapra	1.7	pass
109	slacapra	1.8	common.logger.message("Total number of events to be analyzed: "+str(self.total_number_of_events))
110
111	slacapra	1.7
112	slacapra	1.8	# read user directives
113	slacapra	1.7	eventPerJob=0
114			try:
115			eventPerJob = self.cfg_params['USER.job_number_of_events']
116	nsmirnov	1.2	except KeyError:
117			pass
118
119	slacapra	1.7	jobsPerTask=0
120	nsmirnov	1.2	try:
121	slacapra	1.7	jobsPerTask = int(self.cfg_params['USER.total_number_of_jobs'])
122	nsmirnov	1.2	except KeyError:
123	slacapra	1.7	pass
124
125			# If both the above set, complain and use event per jobs
126			if eventPerJob>0 and jobsPerTask>0:
127	nsmirnov	1.2	msg = 'Warning. '
128	slacapra	1.7	msg += 'job_number_of_events and total_number_of_jobs are both defined '
129			msg += 'Using job_number_of_events.'
130	nsmirnov	1.2	common.logger.message(msg)
131	slacapra	1.7	jobsPerTask = 0
132			if eventPerJob==0 and jobsPerTask==0:
133			msg = 'Warning. '
134			msg += 'job_number_of_events and total_number_of_jobs are not defined '
135			msg += 'Creating just one job for all events.'
136			common.logger.message(msg)
137			jobsPerTask = 1
138
139			# first case: events per job defined
140			if eventPerJob>0:
141			n=eventPerJob
142			if n == 'all' or n == '-1' or (int(n)>self.total_number_of_events and self.total_number_of_events>0):
143			common.logger.message("Asking more events than available: set it to maximum available")
144			self.job_number_of_events = self.total_number_of_events
145	slacapra	1.8	self.total_njobs = 1
146	slacapra	1.7	else:
147			self.job_number_of_events = int(n)
148	slacapra	1.8	self.total_njobs = int((self.total_number_of_events-1)/self.job_number_of_events)+1
149	slacapra	1.7	# second case: jobs per task defined
150			elif jobsPerTask>0:
151	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+" JobPerTask "+str(jobsPerTask))
152			self.job_number_of_events = int(math.floor((self.total_number_of_events)/jobsPerTask))
153			self.total_njobs = jobsPerTask
154	slacapra	1.7	# should not happen...
155			else:
156			raise CrabException('Somthing wrong with splitting')
157
158	slacapra	1.8	common.logger.debug(2,"total number of events: "+str(self.total_number_of_events)+
159	slacapra	1.7	" events per job: "+str(self.job_number_of_events))
160	slacapra	1.8
161	slacapra	1.7	return
162
163			def writeJobsSpecsToDB(self):
164			"""
165			Write firstEvent and maxEvents in the DB for future use
166			"""
167
168			common.jobDB.load()
169			# case one: write first and max events
170			nJobs=self.nJobs()
171
172			firstEvent=self.first_event
173			# last jobs is different...
174			for job in range(nJobs-1):
175			common.jobDB.setFirstEvent(job, firstEvent)
176			common.jobDB.setMaxEvents(job, self.job_number_of_events)
177			firstEvent=firstEvent+self.job_number_of_events
178
179			# this is the last job
180			common.jobDB.setFirstEvent(nJobs-1, firstEvent)
181	slacapra	1.8	lastJobsNumberOfEvents= (self.total_number_of_events+self.first_event)-firstEvent
182			common.jobDB.setMaxEvents(nJobs-1, lastJobsNumberOfEvents)
183	slacapra	1.7
184	slacapra	1.8	common.logger.message('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')
185	slacapra	1.7
186			# case two (to be implemented) write eventCollections for each jobs
187	nsmirnov	1.2
188	slacapra	1.9	# save the DB
189			common.jobDB.save()
190	nsmirnov	1.2	return
191
192	nsmirnov	1.1	def nJobs(self):
193			return self.total_njobs
194
195			def createJobTypeObject(self):
196			file_name = 'cms_'+ string.lower(self.job_type_name)
197			klass_name = string.capitalize(self.job_type_name)
198
199			try:
200			klass = importName(file_name, klass_name)
201			except KeyError:
202			msg = 'No `class '+klass_name+'` found in file `'+file_name+'.py`'
203			raise CrabException(msg)
204			except ImportError, e:
205			msg = 'Cannot create job type '+self.job_type_name
206			msg += ' (file: '+file_name+', class '+klass_name+'):\n'
207			msg += str(e)
208			raise CrabException(msg)
209
210			self.job_type = klass(self.cfg_params)
211			return
212	nsmirnov	1.5
213			def jobType(self):
214			return self.job_type
215	nsmirnov	1.1
216			def run(self):
217	nsmirnov	1.4	"""
218			The main method of the class.
219			"""
220	nsmirnov	1.1
221	nsmirnov	1.3	common.logger.debug(5, "Creator::run() called")
222
223	nsmirnov	1.1	# Instantiate ScriptWriter
224
225			script_writer = ScriptWriter('crab_template.sh')
226
227			# Loop over jobs
228
229			njc = 0
230	nsmirnov	1.4	for nj in range(self.total_njobs):
231	nsmirnov	1.1	if njc == self.ncjobs : break
232	nsmirnov	1.4	st = common.jobDB.status(nj)
233	nsmirnov	1.1	if st != 'X': continue
234
235	nsmirnov	1.6	common.logger.message("Creating job # "+`(nj+1)`)
236	nsmirnov	1.3
237	nsmirnov	1.1	# Prepare configuration file
238
239	nsmirnov	1.4	self.job_type.modifySteeringCards(nj)
240	nsmirnov	1.1
241			# Create JDL
242	nsmirnov	1.4	# Maybe, it worths to move this call into Submitter,
243			# i.e. to create scheduler-specific file at submission time ?
244	nsmirnov	1.1
245	nsmirnov	1.4	common.scheduler.createJDL(nj)
246	nsmirnov	1.1
247			# Create script
248
249	nsmirnov	1.4	script_writer.modifyTemplateScript(nj)
250			os.chmod(common.job_list[nj].scriptFilename(), 0744)
251	nsmirnov	1.1
252	nsmirnov	1.4	common.jobDB.setStatus(nj, 'C')
253	slacapra	1.9	# common: write input and output sandbox
254			common.jobDB.setInputSandbox(nj, self.job_type.inputSandbox(nj))
255
256			outputSandbox=self.job_type.outputSandbox(nj)
257			stdout=common.job_list[nj].stdout()
258			stderr=common.job_list[nj].stderr()
259			outputSandbox.append(common.job_list[nj].stdout())
260			# check if out!=err
261			if stdout != stderr:
262			outputSandbox.append(common.job_list[nj].stderr())
263			common.jobDB.setOutputSandbox(nj, outputSandbox)
264
265	nsmirnov	1.1	njc = njc + 1
266			pass
267
268			####
269
270			common.jobDB.save()
271
272			msg = '\nTotal of %d jobs created'%njc
273			if njc != self.ncjobs: msg = msg + ' from %d requested'%self.ncjobs
274			msg = msg + '.\n'
275			common.logger.message(msg)
276			return