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':
                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.getStatusAttribute_(jid, '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

        if int(self.cfg_params['USER.activate_monalisa']) == 1:
            self.cfg_params['apmon'].free()
            pass

        common.jobDB.save()
        pass

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

        destination = common.scheduler.queryDest(jid).split(":")[0]
        ID3 =  jid.split("/")[3]
        broker = jid.split("/")[2].split(":")[0]
        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:	1.16
Committed:	Sat Dec 17 17:12:29 2005 UTC (19 years, 4 months ago) by corvo
Content type:	text/x-python
Branch:	MAIN
CVS Tags:	CRAB_1_0_6, CRAB_1_0_5, CRAB_1_0_4
Changes since 1.15:	+11 -1 lines
Log Message:	Fixed ML bug
#	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	slacapra	1.1	Status = crab_util.importName('edg_wl_userinterface_common_LbWrapper', 'Status')
24			# 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			self.jobStat = Status()
29
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			if st == 'S':
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	corvo	1.16	dest = common.scheduler.getStatusAttribute_(jid, 'destination').split(":")[0]
74			if int(self.cfg_params['USER.activate_monalisa']) == 1:
75			self.cfg_params['apmon'].fillDict({'taskId': 'JobStatus', 'jobId': jid, \
76			'StatusValueReason': common.scheduler.getStatusAttribute_(jid, 'reason'), \
77			'StatusValue': st, 'StatusEnterTime': common.scheduler.getStatusAttribute_(jid, 'stateEnterTime'), 'StatusDestination': dest})
78			self.cfg_params['apmon'].sendToML()
79	slacapra	1.1	pass
80			else:
81	slacapra	1.10	exit = common.jobDB.exitStatus(nj)
82	fanzago	1.13	#print 'Job %03d:'%(nj+1),jid,crab_util.crabJobStatusToString(st),exit
83	slacapra	1.1	pass
84	corvo	1.16
85			if int(self.cfg_params['USER.activate_monalisa']) == 1:
86			self.cfg_params['apmon'].free()
87	slacapra	1.1	pass
88
89	slacapra	1.2	common.jobDB.save()
90	slacapra	1.1	pass
91
92	spiga	1.9	def processResult_(self, nj, result,jid):
93
94			destination = common.scheduler.queryDest(jid).split(":")[0]
95			ID3 = jid.split("/")[3]
96			broker = jid.split("/")[2].split(":")[0]
97			resFlag = 0
98	slacapra	1.3	### TODO: set relevant status also to DB
99
100	slacapra	1.1	try:
101			if result == 'Done':
102			self.countDone = self.countDone + 1
103	spiga	1.9	exCode = common.scheduler.getExitStatus(jid)
104	slacapra	1.3	common.jobDB.setStatus(nj, 'D')
105	slacapra	1.10	jid = common.jobDB.jobId(nj)
106			exit = common.scheduler.getExitStatus(jid)
107			common.jobDB.setExitStatus(nj, exit)
108	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,exCode)
109	slacapra	1.1	elif result == 'Ready':
110			self.countReady = self.countReady + 1
111	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
112	slacapra	1.1	elif result == 'Scheduled':
113			self.countSched = self.countSched + 1
114	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
115	slacapra	1.1	elif result == 'Running':
116			self.countRun = self.countRun + 1
117	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'-----')
118	slacapra	1.1	elif result == 'Aborted':
119	slacapra	1.2	common.jobDB.setStatus(nj, 'A')
120	slacapra	1.15	self.countAbort = self.countAbort + 1
121	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'abort')
122	slacapra	1.1	pass
123			elif result == 'Cancelled':
124	slacapra	1.3	common.jobDB.setStatus(nj, 'K')
125	slacapra	1.15	self.countCancel = self.countCancel + 1
126	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,'cancel')
127	slacapra	1.1	pass
128			elif result == 'Cleared':
129	spiga	1.9	exCode = common.scheduler.getExitStatus(jid)
130	spiga	1.14	Statistic.Monitor('checkstatus',resFlag,jid,exCode)
131	slacapra	1.1	self.countCleared = self.countCleared + 1
132			except UnboundLocalError:
133			common.logger.message('ERROR: UnboundLocalError with ')
134
135	slacapra	1.15	def PrintReport_(self) :
136	slacapra	1.1
137			""" Report #jobs for each status """
138
139			#job_stat = common.job_list.loadStatus()
140
141			print ''
142			print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
143
144			if (self.countReady != 0):
145			print ''
146			print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
147			if (self.countSched != 0):
148			print ''
149			print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
150			if (self.countRun != 0):
151			print ''
152			print ">>>>>>>>> %i Jobs Running" % (self.countRun)
153			if (self.countCleared != 0):
154			print ''
155			print ">>>>>>>>> %i Jobs Retrieved (=Cleared)" % (self.countCleared)
156			print " You can resubmit them specifying JOB numbers: crab.py -resubmit JOB_number (or range of JOB) -continue"
157			print " (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
158			# if job_stat[6] or job_stat[7]:
159			# print ''
160			# print ">>>>>>>>> %i Jobs aborted or killed(=cancelled by user)" % (job_stat[6] + job_stat[7])
161			# print " Resubmit them with: crab.py -resubmit -continue to resubmit all"
162			# print " or specifying JOB numbers (i.e -resubmit 1-3 => 1 and 2 and 3 or -resubmit 1,3 => 1 and 3)"
163			# print " "
164			if (self.countDone != 0):
165			print ">>>>>>>>> %i Jobs Done" % (self.countDone)
166			print " Retrieve them with: crab.py -getoutput -continue to retrieve all"
167			print " or specifying JOB numbers (i.e -getoutput 1-3 => 1 and 2 and 3 or -getoutput 1,3 => 1 and 3)"
168			print('\n')
169			pass
170