CRAB/python/StatusServer.py

from Actor import *
from crab_util import *
import common
from ApmonIf import ApmonIf
import Statistic
import time
from ProgressBar import ProgressBar
from TerminalController import TerminalController

import xml.dom.minidom
import xml.dom.ext
import TaskDB

class StatusServer(Actor):
 
    def __init__(self, cfg_params,):
        self.cfg_params = cfg_params

        self.countNotSubmit = 0
        self.countSubmit = 0
        self.countSubmitting = 0
        self.countWait = 0
        self.countDone = 0
        self.countReady = 0
        self.countSched = 0
        self.countRun = 0
        self.countAbort = 0
        self.countCancel = 0
        self.countRet = 0
        self.countKilled = 0
        self.countCleared = 0
        self.countToTjob = 0

        try:  
            self.server_name = self.cfg_params['CRAB.server_name'] # gsiftp://pcpg01.cern.ch/data/SEDir/
        except KeyError:
            msg = 'No server selected ...' 
            msg = msg + 'Please specify a server in the crab cfg file' 
            raise CrabException(msg) 

        return

    def translateStatus(self, status):
        """
        simmetric as server
        """

        stateConverting = {'Running': 'R', 'Aborted': 'A', 'Done': 'D', 'Done (Failed)': 'D',\
                           'Cleared': 'D', 'Cancelled': 'K', 'Killed': 'K', 'NotSubmitted': 'C',\
                           'Retrieving by the server': 'R' }

        if status in stateConverting:
            return stateConverting[status]
        return None


    def run(self):
        """
        The main method of the class: check the status of the task
        """
        common.logger.debug(5, "status server::run() called")
        start = time.time()

        totalCreatedJobs = 0
        flagSubmit = 1
        for nj in range(common.jobDB.nJobs()):
            if (common.jobDB.status(nj)!='S'):
                totalCreatedJobs +=1
        #        flagSubmit = 0

        if not flagSubmit:
            common.logger.message("Not all jobs are submitted: before checking the status submit all the jobs.")
            return

        common.scheduler.checkProxy()

        common.taskDB.load()
        WorkDirName =os.path.basename(os.path.split(common.work_space.topDir())[0])
        projectUniqName = 'crab_'+str(WorkDirName)+'_'+common.taskDB.dict('TasKUUID')     
        try: 
            common.logger.message ("Checking the status...\n")
            cmd = 'lcg-cp --vo cms  gsiftp://' + str(self.server_name) + str(projectUniqName)+'/res/xmlReportFile.xml file://'+common.work_space.resDir()+'xmlReportFile.xml'
            common.logger.debug(6, cmd)
            os.system(cmd +' >& /dev/null')  

        except: 
            #msg = ("task status not yet available")
            msg = "The server is managing your task."
            msg += "\n      A detailed report will be ready soon.\n"
            raise CrabException(msg)

        try:     
            file = open(common.work_space.resDir()+"xmlReportFile.xml", "r")
            doc = xml.dom.minidom.parse(common.work_space.resDir()+ "xmlReportFile.xml" )
        
        except: 
            #msg = ("problems reading report file")
            msg = "The server is managing your task."
            msg += "\n      A detailed report will be ready soon.\n"
            raise CrabException(msg)

        ###  <Job status='Submitted' job_exit='NULL' id='1' exe_exit='NULL'/>

        task     = doc.childNodes[0].childNodes[1].getAttribute("taskName")
        self.countToTjob = int(doc.childNodes[0].childNodes[1].getAttribute("totJob") )
       
        addTree = 3

        common.jobDB.load()

        if doc.childNodes[0].childNodes[3].getAttribute("id") == "all":
            if doc.childNodes[0].childNodes[3].getAttribute("status") == "Submitted":
                self.countSubmitting = common.jobDB.nJobs()
                for nj in range(common.jobDB.nJobs()):
                    common.jobDB.setStatus(nj, 'S')
            elif doc.childNodes[0].childNodes[3].getAttribute("status") == "Killed":
                self.countKilled = common.jobDB.nJobs()
                for nj in range(common.jobDB.nJobs()):
                    common.jobDB.setStatus(nj, 'K')
                    self.countKilled = common.jobDB.nJobs()
            elif doc.childNodes[0].childNodes[3].getAttribute("status") == "NotSubmitted":
                self.countNotSubmit = common.jobDB.nJobs()
                for nj in range(common.jobDB.nJobs()):
                    common.jobDB.setStatus(nj, 'C')
            self.countToTjob = common.jobDB.nJobs()
        else:
            printline = ''
            printline+= "%-10s %-20s %-20s %-20s %-18s %-20s" % ('JOBID','STATUS','SITE','JOB_EXIT_STATUS','EXE_EXIT_CODE','RESUBMIT')
            print printline
            print '-------------------------------------------------------------------------------------------------------'

            for job in range( self.countToTjob ):
                idJob = doc.childNodes[0].childNodes[job+addTree].getAttribute("id")
                stato = doc.childNodes[0].childNodes[job+addTree].getAttribute("status")
                exe_exit_code = doc.childNodes[0].childNodes[job+addTree].getAttribute("job_exit")
                job_exit_status = doc.childNodes[0].childNodes[job+addTree].getAttribute("exe_exit")
                cleared = doc.childNodes[0].childNodes[job+addTree].getAttribute("cleared")

                try:
                    site = doc.childNodes[0].childNodes[job+addTree].getAttribute("site")
                    resub = doc.childNodes[0].childNodes[job+addTree].getAttribute("resubmit")
                    if site == "none" or site == "NULL" or site=="None":
                        site=''
                    if resub == "none" or resub =="None" or resub == "0":
                        resub=''
                    if stato == "Killed":
                        resub=''
                except Excpetion, ex:
                    common.logger.message ("Problem reading report file: are you using the latest crab version?")
                    common.logger.debug( 1 , str(ex) )

                jobDbStatus = self.translateStatus(stato)
                if jobDbStatus != None:
                    common.logger.debug(5, '*** Updating jobdb for job %s ***' %idJob)
                    if common.jobDB.status( str(int(idJob)-1) ) != "Y":
                        if jobDbStatus == 'D' and int(cleared) != 1:#exe_exit_code =='' and job_exit_status=='':
                            ## 'Done' but not yet cleared (server side) still showing 'Running'
                            ##stato = 'Running'
                            stato = 'Retrieving by the server'  ## changed - is this user friendly? 
                            jobDbStatus = 'R'
                        common.jobDB.setStatus( str(int(idJob)-1), self.translateStatus(stato) )
                    else:
                        stato = "Cleared"
                    common.jobDB.setExitStatus(  str(int(idJob)-1), job_exit_status )
                if stato != "Done" and stato != "Cleared" and stato != "Aborted" and stato != "Done (Failed)":
                    print "%-10s %-20s %-20s %-20s %-18s %-20s" % (idJob,stato,site,'','',resub)
                else:
                    print "%-10s %-20s %-20s %-20s %-18s %-20s" % (idJob,stato,site,exe_exit_code,job_exit_status,resub)

                if stato == 'Running':
                    self.countRun += 1
                elif stato == 'Aborted':
                    self.countAbort += 1
                elif stato == 'Done' or stato == 'Done (Failed)':
                    self.countDone += 1
                elif stato == 'Cancelled':
                    self.countCancel += 1
                elif stato == 'Submitted':
                    self.countSubmit += 1
                elif stato == 'Submitting':
                    self.countSubmitting += 1
                elif stato == 'Ready':
                    self.countReady += 1
                elif stato == 'Scheduled':
                    self.countSched += 1
                elif stato == 'Cleared':
                    self.countCleared += 1
                elif stato == 'NotSubmitted':
                    self.countSubmitting += 1
                elif stato == 'Waiting':
                    self.countWait += 1
                elif stato == 'Retrieving by the server':
                    self.countRet += 1
                elif stato == 'Killed':
                    self.countKilled += 1

                addTree += 1
        common.jobDB.save()

        self.PrintReport_()


    def PrintReport_(self) :

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


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

        if (self.countSubmitting != 0) :
            print ">>>>>>>>> %i Jobs Submitting by the server" % (self.countSubmitting)
        if (self.countNotSubmit != 0):
            print ">>>>>>>>> %i Jobs Not Submitted to the grid" % (self.countNotSubmit)
        if (self.countSubmit != 0):
            print ">>>>>>>>> %i Jobs Submitted" % (self.countSubmit)
        if (self.countWait != 0):
            print ">>>>>>>>> %i Jobs Waiting" % (self.countWait)
        if (self.countReady != 0):
            print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
        if (self.countSched != 0):
            print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
        if (self.countRun != 0):
            print ">>>>>>>>> %i Jobs Running" % (self.countRun)
        if (self.countRet != 0):
            print ">>>>>>>>> %i Jobs Retrieving by the server" % (self.countRet)
        if (self.countDone != 0):
            print ">>>>>>>>> %i Jobs Done" % (self.countDone)
            print "          Retrieve them with: crab -getoutput -continue"
        if (self.countKilled != 0):
            print ">>>>>>>>> %i Jobs Killed" % (self.countKilled)
#            print "          Retrieve more information with: crab -postMortem -continue"
        if (self.countAbort != 0):
            print ">>>>>>>>> %i Jobs Aborted" % (self.countAbort)
        if (self.countCleared != 0):
            print ">>>>>>>>> %i Jobs Cleared" % (self.countCleared)

        countUnderMngmt = self.countToTjob - (self.countSubmitting+ self.countNotSubmit + self.countSubmit)
        countUnderMngmt -= (self.countReady + self.countSched + self.countRun + self.countDone + self.countRet) 
        countUnderMngmt -= (self.countKilled + self.countAbort + self.countCleared + self.countWait)
        if (countUnderMngmt != 0):
            print ">>>>>>>>> %i Jobs Waiting or Under Server Management" % (countUnderMngmt)

        print ''
        pass

Revision:	1.15
Committed:	Mon Jul 16 17:51:55 2007 UTC (17 years, 9 months ago) by mcinquil
Content type:	text/x-python
Branch:	MAIN
Changes since 1.14:	+18 -5 lines
Log Message:	Added site and n.resubmit to the status output
#	User	Rev	Content
1	spiga	1.1	from Actor import *
2			from crab_util import *
3			import common
4			from ApmonIf import ApmonIf
5			import Statistic
6			import time
7			from ProgressBar import ProgressBar
8			from TerminalController import TerminalController
9
10			import xml.dom.minidom
11			import xml.dom.ext
12			import TaskDB
13
14			class StatusServer(Actor):
15
16			def __init__(self, cfg_params,):
17			self.cfg_params = cfg_params
18	mcinquil	1.2
19	mcinquil	1.8	self.countNotSubmit = 0
20	mcinquil	1.2	self.countSubmit = 0
21			self.countSubmitting = 0
22	mcinquil	1.11	self.countWait = 0
23	mcinquil	1.2	self.countDone = 0
24			self.countReady = 0
25			self.countSched = 0
26			self.countRun = 0
27			self.countAbort = 0
28			self.countCancel = 0
29	mcinquil	1.12	self.countRet = 0
30	mcinquil	1.6	self.countKilled = 0
31	mcinquil	1.2	self.countCleared = 0
32			self.countToTjob = 0
33
34	spiga	1.3	try:
35			self.server_name = self.cfg_params['CRAB.server_name'] # gsiftp://pcpg01.cern.ch/data/SEDir/
36			except KeyError:
37			msg = 'No server selected ...'
38			msg = msg + 'Please specify a server in the crab cfg file'
39			raise CrabException(msg)
40
41	spiga	1.1	return
42	mcinquil	1.2
43			def translateStatus(self, status):
44			"""
45			simmetric as server
46			"""
47
48	mcinquil	1.4	stateConverting = {'Running': 'R', 'Aborted': 'A', 'Done': 'D', 'Done (Failed)': 'D',\
49	mcinquil	1.12	'Cleared': 'D', 'Cancelled': 'K', 'Killed': 'K', 'NotSubmitted': 'C',\
50	spiga	1.13	'Retrieving by the server': 'R' }
51	mcinquil	1.2
52			if status in stateConverting:
53			return stateConverting[status]
54			return None
55
56
57	spiga	1.1	def run(self):
58			"""
59			The main method of the class: check the status of the task
60			"""
61			common.logger.debug(5, "status server::run() called")
62			start = time.time()
63
64			totalCreatedJobs = 0
65			flagSubmit = 1
66			for nj in range(common.jobDB.nJobs()):
67			if (common.jobDB.status(nj)!='S'):
68			totalCreatedJobs +=1
69	mcinquil	1.2	# flagSubmit = 0
70	spiga	1.1
71			if not flagSubmit:
72			common.logger.message("Not all jobs are submitted: before checking the status submit all the jobs.")
73			return
74
75			common.scheduler.checkProxy()
76
77			common.taskDB.load()
78			WorkDirName =os.path.basename(os.path.split(common.work_space.topDir())[0])
79			projectUniqName = 'crab_'+str(WorkDirName)+'_'+common.taskDB.dict('TasKUUID')
80			try:
81			common.logger.message ("Checking the status...\n")
82	spiga	1.3	cmd = 'lcg-cp --vo cms gsiftp://' + str(self.server_name) + str(projectUniqName)+'/res/xmlReportFile.xml file://'+common.work_space.resDir()+'xmlReportFile.xml'
83	mcinquil	1.2	common.logger.debug(6, cmd)
84	spiga	1.1	os.system(cmd +' >& /dev/null')
85
86			except:
87	mcinquil	1.10	#msg = ("task status not yet available")
88			msg = "The server is managing your task."
89			msg += "\n A detailed report will be ready soon.\n"
90	spiga	1.1	raise CrabException(msg)
91
92			try:
93			file = open(common.work_space.resDir()+"xmlReportFile.xml", "r")
94			doc = xml.dom.minidom.parse(common.work_space.resDir()+ "xmlReportFile.xml" )
95
96			except:
97	mcinquil	1.10	#msg = ("problems reading report file")
98			msg = "The server is managing your task."
99			msg += "\n A detailed report will be ready soon.\n"
100	spiga	1.1	raise CrabException(msg)
101
102	mcinquil	1.2	### <Job status='Submitted' job_exit='NULL' id='1' exe_exit='NULL'/>
103
104	spiga	1.1	task = doc.childNodes[0].childNodes[1].getAttribute("taskName")
105	mcinquil	1.2	self.countToTjob = int(doc.childNodes[0].childNodes[1].getAttribute("totJob") )
106
107			addTree = 3
108
109			common.jobDB.load()
110
111			if doc.childNodes[0].childNodes[3].getAttribute("id") == "all":
112			if doc.childNodes[0].childNodes[3].getAttribute("status") == "Submitted":
113			self.countSubmitting = common.jobDB.nJobs()
114			for nj in range(common.jobDB.nJobs()):
115			common.jobDB.setStatus(nj, 'S')
116	mcinquil	1.6	elif doc.childNodes[0].childNodes[3].getAttribute("status") == "Killed":
117			self.countKilled = common.jobDB.nJobs()
118			for nj in range(common.jobDB.nJobs()):
119			common.jobDB.setStatus(nj, 'K')
120	mcinquil	1.14	self.countKilled = common.jobDB.nJobs()
121	mcinquil	1.8	elif doc.childNodes[0].childNodes[3].getAttribute("status") == "NotSubmitted":
122			self.countNotSubmit = common.jobDB.nJobs()
123	mcinquil	1.2	for nj in range(common.jobDB.nJobs()):
124	mcinquil	1.8	common.jobDB.setStatus(nj, 'C')
125	mcinquil	1.2	self.countToTjob = common.jobDB.nJobs()
126	spiga	1.1	else:
127	mcinquil	1.2	printline = ''
128	mcinquil	1.15	printline+= "%-10s %-20s %-20s %-20s %-18s %-20s" % ('JOBID','STATUS','SITE','JOB_EXIT_STATUS','EXE_EXIT_CODE','RESUBMIT')
129	mcinquil	1.2	print printline
130	mcinquil	1.15	print '-------------------------------------------------------------------------------------------------------'
131	mcinquil	1.2
132			for job in range( self.countToTjob ):
133			idJob = doc.childNodes[0].childNodes[job+addTree].getAttribute("id")
134			stato = doc.childNodes[0].childNodes[job+addTree].getAttribute("status")
135			exe_exit_code = doc.childNodes[0].childNodes[job+addTree].getAttribute("job_exit")
136			job_exit_status = doc.childNodes[0].childNodes[job+addTree].getAttribute("exe_exit")
137	mcinquil	1.5	cleared = doc.childNodes[0].childNodes[job+addTree].getAttribute("cleared")
138	mcinquil	1.15
139			try:
140			site = doc.childNodes[0].childNodes[job+addTree].getAttribute("site")
141			resub = doc.childNodes[0].childNodes[job+addTree].getAttribute("resubmit")
142			if site == "none" or site == "NULL" or site=="None":
143			site=''
144			if resub == "none" or resub =="None" or resub == "0":
145			resub=''
146			if stato == "Killed":
147			resub=''
148			except Excpetion, ex:
149			common.logger.message ("Problem reading report file: are you using the latest crab version?")
150			common.logger.debug( 1 , str(ex) )
151
152	mcinquil	1.2	jobDbStatus = self.translateStatus(stato)
153			if jobDbStatus != None:
154			common.logger.debug(5, '* Updating jobdb for job %s *' %idJob)
155			if common.jobDB.status( str(int(idJob)-1) ) != "Y":
156	mcinquil	1.5	if jobDbStatus == 'D' and int(cleared) != 1:#exe_exit_code =='' and job_exit_status=='':
157			## 'Done' but not yet cleared (server side) still showing 'Running'
158	mcinquil	1.12	##stato = 'Running'
159	spiga	1.13	stato = 'Retrieving by the server' ## changed - is this user friendly?
160	mcinquil	1.5	jobDbStatus = 'R'
161	mcinquil	1.2	common.jobDB.setStatus( str(int(idJob)-1), self.translateStatus(stato) )
162			else:
163			stato = "Cleared"
164			common.jobDB.setExitStatus( str(int(idJob)-1), job_exit_status )
165	mcinquil	1.4	if stato != "Done" and stato != "Cleared" and stato != "Aborted" and stato != "Done (Failed)":
166	mcinquil	1.15	print "%-10s %-20s %-20s %-20s %-18s %-20s" % (idJob,stato,site,'','',resub)
167	mcinquil	1.2	else:
168	mcinquil	1.15	print "%-10s %-20s %-20s %-20s %-18s %-20s" % (idJob,stato,site,exe_exit_code,job_exit_status,resub)
169	mcinquil	1.2
170			if stato == 'Running':
171			self.countRun += 1
172			elif stato == 'Aborted':
173			self.countAbort += 1
174	mcinquil	1.11	elif stato == 'Done' or stato == 'Done (Failed)':
175	mcinquil	1.2	self.countDone += 1
176			elif stato == 'Cancelled':
177			self.countCancel += 1
178			elif stato == 'Submitted':
179			self.countSubmit += 1
180			elif stato == 'Submitting':
181			self.countSubmitting += 1
182			elif stato == 'Ready':
183			self.countReady += 1
184			elif stato == 'Scheduled':
185			self.countSched += 1
186			elif stato == 'Cleared':
187			self.countCleared += 1
188	mcinquil	1.8	elif stato == 'NotSubmitted':
189			self.countSubmitting += 1
190	mcinquil	1.11	elif stato == 'Waiting':
191			self.countWait += 1
192	spiga	1.13	elif stato == 'Retrieving by the server':
193	mcinquil	1.12	self.countRet += 1
194	mcinquil	1.14	elif stato == 'Killed':
195			self.countKilled += 1
196	mcinquil	1.2
197			addTree += 1
198			common.jobDB.save()
199
200			self.PrintReport_()
201
202
203			def PrintReport_(self) :
204
205			""" Report #jobs for each status """
206
207
208			print ''
209			print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
210			print ''
211
212			if (self.countSubmitting != 0) :
213			print ">>>>>>>>> %i Jobs Submitting by the server" % (self.countSubmitting)
214	mcinquil	1.8	if (self.countNotSubmit != 0):
215			print ">>>>>>>>> %i Jobs Not Submitted to the grid" % (self.countNotSubmit)
216	mcinquil	1.2	if (self.countSubmit != 0):
217			print ">>>>>>>>> %i Jobs Submitted" % (self.countSubmit)
218	mcinquil	1.11	if (self.countWait != 0):
219			print ">>>>>>>>> %i Jobs Waiting" % (self.countWait)
220	mcinquil	1.6	if (self.countReady != 0):
221			print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
222	mcinquil	1.2	if (self.countSched != 0):
223			print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
224	mcinquil	1.14	if (self.countRun != 0):
225			print ">>>>>>>>> %i Jobs Running" % (self.countRun)
226	mcinquil	1.12	if (self.countRet != 0):
227	spiga	1.13	print ">>>>>>>>> %i Jobs Retrieving by the server" % (self.countRet)
228	mcinquil	1.6	if (self.countDone != 0):
229			print ">>>>>>>>> %i Jobs Done" % (self.countDone)
230			print " Retrieve them with: crab -getoutput -continue"
231			if (self.countKilled != 0):
232			print ">>>>>>>>> %i Jobs Killed" % (self.countKilled)
233	mcinquil	1.14	# print " Retrieve more information with: crab -postMortem -continue"
234	mcinquil	1.6	if (self.countAbort != 0):
235			print ">>>>>>>>> %i Jobs Aborted" % (self.countAbort)
236	mcinquil	1.2	if (self.countCleared != 0):
237	farinafa	1.9	print ">>>>>>>>> %i Jobs Cleared" % (self.countCleared)
238
239			countUnderMngmt = self.countToTjob - (self.countSubmitting+ self.countNotSubmit + self.countSubmit)
240	mcinquil	1.14	countUnderMngmt -= (self.countReady + self.countSched + self.countRun + self.countDone + self.countRet)
241	mcinquil	1.11	countUnderMngmt -= (self.countKilled + self.countAbort + self.countCleared + self.countWait)
242	farinafa	1.9	if (countUnderMngmt != 0):
243			print ">>>>>>>>> %i Jobs Waiting or Under Server Management" % (countUnderMngmt)
244
245	mcinquil	1.2	print ''
246			pass
247	spiga	1.1