CRAB/python/Status.py

from Actor import *
import common, crab_util
import string, os
import Statistic

class Status(Actor):
    def __init__(self, cfg_params, nj_list=[]):
        if nj_list==[]:
            self.nj_list = range(len(common.job_list))
        else:
            self.nj_list = nj_list
        self.countDone = 0
        self.countReady = 0
        self.countSched = 0
        self.countRun = 0
        self.countAbort = 0
        self.countCancel = 0
        self.countCleared = 0
        self.countToTjob = 0
        self.cfg_params = cfg_params
        

        #Status = crab_util.importName('edg_wl_userinterface_common_LbWrapper', 'Status')
        # Bypass edg-job-status interfacing directly to C++ API
        # Job attribute vector to retrieve status without edg-job-status
        self.level = 0
        # Instance of the Status class provided by LB API
        #self.jobStat = Status()

        self.states = [ "Acl", "cancelReason", "cancelling","ce_node","children", \
          "children_hist","children_num","children_states","condorId","condor_jdl", \
          "cpuTime","destination", "done_code","exit_code","expectFrom", \
          "expectUpdate","globusId","jdl","jobId","jobtype", \
          "lastUpdateTime","localId","location", "matched_jdl","network_server", \
          "owner","parent_job", "reason","resubmitted","rsl","seed",\
          "stateEnterTime","stateEnterTimes","subjob_failed", \
          "user tags" , "status" , "status_code","hierarchy"]
        self.hstates = {}
        for key in self.states:
            self.hstates[key]=''

        return

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

        self.compute()
        self.PrintReport_()
        pass

    def status(self) :
        """ Return #jobs for each status as a tuple"""
        return (self.countToTjob,self.countReady,self.countSched,self.countRun,self.countCleared,self.countAbort,self.countCancel,self.countDone)

    def compute(self):
        """
        Update the status to DB
        """

        common.jobDB.load()
        for nj in self.nj_list:
            st = common.jobDB.status(nj)
            self.countToTjob = self.countToTjob + 1
            jid = common.jobDB.jobId(nj)
            if st == 'S' or st == 'A' or st == 'D' or st == 'K':
                result = common.scheduler.queryStatus(jid)
                self.processResult_(nj, result, jid)
                exit = common.jobDB.exitStatus(nj)
                print 'Job %03d:'%(nj+1),jid,result,exit
                dest = common.scheduler.queryDest(jid)
                if ( dest.find(":") != -1 ) :
                    dest = destination.split(":")[0]
                if int(self.cfg_params['USER.activate_monalisa']) == 1:
                    self.cfg_params['apmon'].fillDict({'taskId': 'JobStatus', 'jobId': jid, \
                                                   'StatusValueReason': common.scheduler.getStatusAttribute_(jid, 'reason'), \
                                                   'StatusValue': st, 'StatusEnterTime': common.scheduler.getStatusAttribute_(jid, 'stateEnterTime'), 'StatusDestination': dest})
                    self.cfg_params['apmon'].sendToML()
                pass
            else:
                exit = common.jobDB.exitStatus(nj)
                #print 'Job %03d:'%(nj+1),jid,crab_util.crabJobStatusToString(st),exit
                pass

        common.jobDB.save()
        pass

    def processResult_(self, nj, result,jid):

        destination = common.scheduler.queryDest(jid)
        if ( destination.find(":") != -1 ) :
            destination = destination.split(":")[0]
            
        if ( jid.find(":") != -1 ) :
            ID3 =  jid.split("/")[3]
            broker = jid.split("/")[2].split(":")[0]
        else :
            ID3 = jid
            broker = 'OSG'
            
        resFlag = 0
        ### TODO: set relevant status also to DB

        try: 
            if result == 'Done': 
                self.countDone = self.countDone + 1
                exCode = common.scheduler.getExitStatus(jid)
                common.jobDB.setStatus(nj, 'D')
                jid = common.jobDB.jobId(nj)
                exit = common.scheduler.getExitStatus(jid)
                common.jobDB.setExitStatus(nj, exit)
                Statistic.Monitor('checkstatus',resFlag,jid,exCode)
            elif result == 'Ready':
                self.countReady = self.countReady + 1
                Statistic.Monitor('checkstatus',resFlag,jid,'-----')
            elif result == 'Scheduled':
                self.countSched = self.countSched + 1
                Statistic.Monitor('checkstatus',resFlag,jid,'-----')
            elif result == 'Running':
                self.countRun = self.countRun + 1
                Statistic.Monitor('checkstatus',resFlag,jid,'-----')
            elif result == 'Aborted':
                common.jobDB.setStatus(nj, 'A')
                self.countAbort = self.countAbort + 1
                Statistic.Monitor('checkstatus',resFlag,jid,'abort')
                pass
            elif result == 'Cancelled':
                common.jobDB.setStatus(nj, 'K')
                self.countCancel = self.countCancel + 1
                Statistic.Monitor('checkstatus',resFlag,jid,'cancel')
                pass
            elif result == 'Cleared':
                exCode = common.scheduler.getExitStatus(jid) 
                Statistic.Monitor('checkstatus',resFlag,jid,exCode) 
                self.countCleared = self.countCleared + 1
        except UnboundLocalError:
            common.logger.message('ERROR: UnboundLocalError with ')

    def PrintReport_(self) :

        """ Report #jobs for each status  """  

        #job_stat = common.job_list.loadStatus()

        print ''
        print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)

        if (self.countReady != 0):
            print ''
            print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
        if (self.countSched != 0):
            print ''
            print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
        if (self.countRun != 0):
            print ''
            print ">>>>>>>>> %i Jobs Running" % (self.countRun)
        if (self.countCleared != 0):
            print ''
            print ">>>>>>>>> %i Jobs Retrieved (=Cleared)" % (self.countCleared)
            print "          You can resubmit them specifying JOB numbers: crab.py -resubmit JOB_number (or range of JOB) -continue" 
            print "          (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"       
        # if job_stat[6] or job_stat[7]:
        #     print ''
        #     print ">>>>>>>>> %i Jobs aborted or killed(=cancelled by user)" % (job_stat[6] + job_stat[7])
        #     print "          Resubmit them with: crab.py -resubmit -continue to resubmit all" 
        #     print "          or specifying JOB numbers (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"       
        #     print "           "       
        if (self.countDone != 0):
            print ">>>>>>>>> %i Jobs Done" % (self.countDone)
            print "          Retrieve them with: crab.py -getoutput -continue to retrieve all" 
            print "          or specifying JOB numbers (i.e -getoutput 1-3 => 1 and 2 and 3 or -getoutput 1,3 => 1 and 3)"
            print('\n')  
        pass

Revision:	1.18
Committed:	Mon Mar 27 17:36:35 2006 UTC (19 years, 1 month ago) by gutsche
Content type:	text/x-python
Branch:	MAIN
Changes since 1.17:	+17 -7 lines
Log Message:	commented out call to edg_wl_userinterface_common_LbWrapper because it would break condor_g scheduler execution on OSG-client only machines and secure parsing of jid and destination, in case of condor, no https grid id is available
#	User	Rev	Content
1	slacapra	1.1	from Actor import *
2			import common, crab_util
3			import string, os
4	spiga	1.9	import Statistic
5	slacapra	1.1
6			class Status(Actor):
7	corvo	1.16	def __init__(self, cfg_params, nj_list=[]):
8	slacapra	1.15	if nj_list==[]:
9			self.nj_list = range(len(common.job_list))
10			else:
11			self.nj_list = nj_list
12	slacapra	1.1	self.countDone = 0
13			self.countReady = 0
14			self.countSched = 0
15			self.countRun = 0
16	slacapra	1.15	self.countAbort = 0
17			self.countCancel = 0
18	slacapra	1.1	self.countCleared = 0
19			self.countToTjob = 0
20	corvo	1.16	self.cfg_params = cfg_params
21	slacapra	1.1
22	spiga	1.9
23	gutsche	1.18	#Status = crab_util.importName('edg_wl_userinterface_common_LbWrapper', 'Status')
24	slacapra	1.1	# Bypass edg-job-status interfacing directly to C++ API
25			# Job attribute vector to retrieve status without edg-job-status
26			self.level = 0
27			# Instance of the Status class provided by LB API
28	gutsche	1.18	#self.jobStat = Status()
29	slacapra	1.1
30			self.states = [ "Acl", "cancelReason", "cancelling","ce_node","children", \
31			"children_hist","children_num","children_states","condorId","condor_jdl", \
32			"cpuTime","destination", "done_code","exit_code","expectFrom", \
33			"expectUpdate","globusId","jdl","jobId","jobtype", \
34			"lastUpdateTime","localId","location", "matched_jdl","network_server", \
35	slacapra	1.5	"owner","parent_job", "reason","resubmitted","rsl","seed",\
36			"stateEnterTime","stateEnterTimes","subjob_failed", \
37	slacapra	1.1	"user tags" , "status" , "status_code","hierarchy"]
38			self.hstates = {}
39	slacapra	1.6	for key in self.states:
40			self.hstates[key]=''
41	slacapra	1.1
42			return
43
44			def run(self):
45			"""
46			The main method of the class.
47			"""
48			common.logger.debug(5, "Status::run() called")
49
50	slacapra	1.15	self.compute()
51			self.PrintReport_()
52			pass
53
54			def status(self) :
55			""" Return #jobs for each status as a tuple"""
56			return (self.countToTjob,self.countReady,self.countSched,self.countRun,self.countCleared,self.countAbort,self.countCancel,self.countDone)
57
58			def compute(self):
59			"""
60			Update the status to DB
61			"""
62
63	slacapra	1.1	common.jobDB.load()
64			for nj in self.nj_list:
65			st = common.jobDB.status(nj)
66			self.countToTjob = self.countToTjob + 1
67			jid = common.jobDB.jobId(nj)
68	gutsche	1.18	if st == 'S' or st == 'A' or st == 'D' or st == 'K':
69	slacapra	1.5	result = common.scheduler.queryStatus(jid)
70	slacapra	1.11	self.processResult_(nj, result, jid)
71	slacapra	1.10	exit = common.jobDB.exitStatus(nj)
72	slacapra	1.4	print 'Job %03d:'%(nj+1),jid,result,exit
73	gutsche	1.18	dest = common.scheduler.queryDest(jid)
74			if ( dest.find(":") != -1 ) :
75			dest = destination.split(":")[0]
76	corvo	1.16	if int(self.cfg_params['USER.activate_monalisa']) == 1:
77			self.cfg_params['apmon'].fillDict({'taskId': 'JobStatus', 'jobId': jid, \
78			'StatusValueReason': common.scheduler.getStatusAttribute_(jid, 'reason'), \
79			'StatusValue': st, 'StatusEnterTime': common.scheduler.getStatusAttribute_(jid, 'stateEnterTime'), 'StatusDestination': dest})
80			self.cfg_params['apmon'].sendToML()
81	slacapra	1.1	pass
82			else:
83	slacapra	1.10	exit = common.jobDB.exitStatus(nj)
84	fanzago	1.13	#print 'Job %03d:'%(nj+1),jid,crab_util.crabJobStatusToString(st),exit
85	slacapra	1.1	pass
86	corvo	1.16
87	slacapra	1.2	common.jobDB.save()
88	slacapra	1.1	pass
89
90	spiga	1.9	def processResult_(self, nj, result,jid):
91
92	gutsche	1.18	destination = common.scheduler.queryDest(jid)
93			if ( destination.find(":") != -1 ) :
94			destination = destination.split(":")[0]
95
96			if ( jid.find(":") != -1 ) :
97			ID3 = jid.split("/")[3]
98			broker = jid.split("/")[2].split(":")[0]
99			else :
100			ID3 = jid
101			broker = 'OSG'
102
103	spiga	1.9	resFlag = 0
104	slacapra	1.3	### TODO: set relevant status also to DB
105
106	slacapra	1.1	try:
107			if result == 'Done':
108			self.countDone = self.countDone + 1
109	spiga	1.9	exCode = common.scheduler.getExitStatus(jid)
110	slacapra	1.3	common.jobDB.setStatus(nj, 'D')
111	slacapra	1.10	jid = common.jobDB.jobId(nj)
112			exit = common.scheduler.getExitStatus(jid)
113			common.jobDB.setExitStatus(nj, exit)
114	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,exCode)
115	slacapra	1.1	elif result == 'Ready':
116			self.countReady = self.countReady + 1
117	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
118	slacapra	1.1	elif result == 'Scheduled':
119			self.countSched = self.countSched + 1
120	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
121	slacapra	1.1	elif result == 'Running':
122			self.countRun = self.countRun + 1
123	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
124	slacapra	1.1	elif result == 'Aborted':
125	slacapra	1.2	common.jobDB.setStatus(nj, 'A')
126	slacapra	1.15	self.countAbort = self.countAbort + 1
127	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'abort')
128	slacapra	1.1	pass
129			elif result == 'Cancelled':
130	slacapra	1.3	common.jobDB.setStatus(nj, 'K')
131	slacapra	1.15	self.countCancel = self.countCancel + 1
132	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'cancel')
133	slacapra	1.1	pass
134			elif result == 'Cleared':
135	spiga	1.9	exCode = common.scheduler.getExitStatus(jid)
136	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,exCode)
137	slacapra	1.1	self.countCleared = self.countCleared + 1
138			except UnboundLocalError:
139			common.logger.message('ERROR: UnboundLocalError with ')
140
141	slacapra	1.15	def PrintReport_(self) :
142	slacapra	1.1
143			""" Report #jobs for each status """
144
145			#job_stat = common.job_list.loadStatus()
146
147			print ''
148			print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
149
150			if (self.countReady != 0):
151			print ''
152			print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
153			if (self.countSched != 0):
154			print ''
155			print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
156			if (self.countRun != 0):
157			print ''
158			print ">>>>>>>>> %i Jobs Running" % (self.countRun)
159			if (self.countCleared != 0):
160			print ''
161			print ">>>>>>>>> %i Jobs Retrieved (=Cleared)" % (self.countCleared)
162			print " You can resubmit them specifying JOB numbers: crab.py -resubmit JOB_number (or range of JOB) -continue"
163			print " (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
164			# if job_stat[6] or job_stat[7]:
165			# print ''
166			# print ">>>>>>>>> %i Jobs aborted or killed(=cancelled by user)" % (job_stat[6] + job_stat[7])
167			# print " Resubmit them with: crab.py -resubmit -continue to resubmit all"
168			# print " or specifying JOB numbers (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
169			# print " "
170			if (self.countDone != 0):
171			print ">>>>>>>>> %i Jobs Done" % (self.countDone)
172			print " Retrieve them with: crab.py -getoutput -continue to retrieve all"
173			print " or specifying JOB numbers (i.e -getoutput 1-3 => 1 and 2 and 3 or -getoutput 1,3 => 1 and 3)"
174			print('\n')
175			pass
176