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.countSubmit = 0
        self.countSubmitting = 0
        self.countDone = 0
        self.countReady = 0
        self.countSched = 0
        self.countRun = 0
        self.countAbort = 0
        self.countCancel = 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'}

        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")
            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")
            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')
            else:
                self.countAbort = common.jobDB.nJobs()
                for nj in range(common.jobDB.nJobs()):
                    common.jobDB.setStatus(nj, 'A')
            self.countToTjob = common.jobDB.nJobs()
        else:
            printline = ''
            printline+= "%-10s %-20s %-20s %-25s" % ('JOBID','STATUS','EXE_EXIT_CODE','JOB_EXIT_STATUS')
            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")
                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'
                            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" % (idJob,stato)
                else:
                    print "%-10s %-20s %-20s %-25s" % (idJob,stato,exe_exit_code,job_exit_status)

                if stato == 'Running':
                    self.countRun += 1
                elif stato == 'Aborted':
                    self.countAbort += 1
                elif stato == 'Done':
                    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

                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.countSubmit != 0):
            print ">>>>>>>>> %i Jobs Submitted" % (self.countSubmit)
        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.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.countRun)
        print ''
        pass

Revision:	1.6
Committed:	Thu May 31 15:54:47 2007 UTC (17 years, 11 months ago) by mcinquil
Content type:	text/x-python
Branch:	MAIN
Changes since 1.5:	+15 -7 lines
Log Message:	Support of kill functionalities (printout & jobDB)
#	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			self.countSubmit = 0
20			self.countSubmitting = 0
21			self.countDone = 0
22			self.countReady = 0
23			self.countSched = 0
24			self.countRun = 0
25			self.countAbort = 0
26			self.countCancel = 0
27	mcinquil	1.6	self.countKilled = 0
28	mcinquil	1.2	self.countCleared = 0
29			self.countToTjob = 0
30
31	spiga	1.3	try:
32			self.server_name = self.cfg_params['CRAB.server_name'] # gsiftp://pcpg01.cern.ch/data/SEDir/
33			except KeyError:
34			msg = 'No server selected ...'
35			msg = msg + 'Please specify a server in the crab cfg file'
36			raise CrabException(msg)
37
38	spiga	1.1	return
39	mcinquil	1.2
40			def translateStatus(self, status):
41			"""
42			simmetric as server
43			"""
44
45	mcinquil	1.4	stateConverting = {'Running': 'R', 'Aborted': 'A', 'Done': 'D', 'Done (Failed)': 'D',\
46	mcinquil	1.2	'Cleared': 'D', 'Cancelled': 'K', 'Killed': 'K'}
47
48			if status in stateConverting:
49			return stateConverting[status]
50			return None
51
52
53	spiga	1.1	def run(self):
54			"""
55			The main method of the class: check the status of the task
56			"""
57			common.logger.debug(5, "status server::run() called")
58			start = time.time()
59
60			totalCreatedJobs = 0
61			flagSubmit = 1
62			for nj in range(common.jobDB.nJobs()):
63			if (common.jobDB.status(nj)!='S'):
64			totalCreatedJobs +=1
65	mcinquil	1.2	# flagSubmit = 0
66	spiga	1.1
67			if not flagSubmit:
68			common.logger.message("Not all jobs are submitted: before checking the status submit all the jobs.")
69			return
70
71			common.scheduler.checkProxy()
72
73			common.taskDB.load()
74			WorkDirName =os.path.basename(os.path.split(common.work_space.topDir())[0])
75			projectUniqName = 'crab_'+str(WorkDirName)+'_'+common.taskDB.dict('TasKUUID')
76			try:
77			common.logger.message ("Checking the status...\n")
78	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'
79	mcinquil	1.2	common.logger.debug(6, cmd)
80	spiga	1.1	os.system(cmd +' >& /dev/null')
81
82			except:
83			msg = ("task status not yet available")
84			raise CrabException(msg)
85
86			try:
87			file = open(common.work_space.resDir()+"xmlReportFile.xml", "r")
88			doc = xml.dom.minidom.parse(common.work_space.resDir()+ "xmlReportFile.xml" )
89
90			except:
91			msg = ("problems reading report file")
92			raise CrabException(msg)
93
94	mcinquil	1.2	### <Job status='Submitted' job_exit='NULL' id='1' exe_exit='NULL'/>
95
96	spiga	1.1	task = doc.childNodes[0].childNodes[1].getAttribute("taskName")
97	mcinquil	1.2	self.countToTjob = int(doc.childNodes[0].childNodes[1].getAttribute("totJob") )
98
99			addTree = 3
100
101			common.jobDB.load()
102
103			if doc.childNodes[0].childNodes[3].getAttribute("id") == "all":
104			if doc.childNodes[0].childNodes[3].getAttribute("status") == "Submitted":
105			self.countSubmitting = common.jobDB.nJobs()
106			for nj in range(common.jobDB.nJobs()):
107			common.jobDB.setStatus(nj, 'S')
108	mcinquil	1.6	elif doc.childNodes[0].childNodes[3].getAttribute("status") == "Killed":
109			self.countKilled = common.jobDB.nJobs()
110			for nj in range(common.jobDB.nJobs()):
111			common.jobDB.setStatus(nj, 'K')
112	mcinquil	1.2	else:
113			self.countAbort = common.jobDB.nJobs()
114			for nj in range(common.jobDB.nJobs()):
115			common.jobDB.setStatus(nj, 'A')
116			self.countToTjob = common.jobDB.nJobs()
117	spiga	1.1	else:
118	mcinquil	1.2	printline = ''
119			printline+= "%-10s %-20s %-20s %-25s" % ('JOBID','STATUS','EXE_EXIT_CODE','JOB_EXIT_STATUS')
120			print printline
121			print '-------------------------------------------------------------------------------------'
122
123			for job in range( self.countToTjob ):
124			idJob = doc.childNodes[0].childNodes[job+addTree].getAttribute("id")
125			stato = doc.childNodes[0].childNodes[job+addTree].getAttribute("status")
126			exe_exit_code = doc.childNodes[0].childNodes[job+addTree].getAttribute("job_exit")
127			job_exit_status = doc.childNodes[0].childNodes[job+addTree].getAttribute("exe_exit")
128	mcinquil	1.5	cleared = doc.childNodes[0].childNodes[job+addTree].getAttribute("cleared")
129	mcinquil	1.2	jobDbStatus = self.translateStatus(stato)
130	spiga	1.1
131	mcinquil	1.2	if jobDbStatus != None:
132			common.logger.debug(5, '* Updating jobdb for job %s *' %idJob)
133			if common.jobDB.status( str(int(idJob)-1) ) != "Y":
134	mcinquil	1.5	if jobDbStatus == 'D' and int(cleared) != 1:#exe_exit_code =='' and job_exit_status=='':
135			## 'Done' but not yet cleared (server side) still showing 'Running'
136			stato = 'Running'
137			jobDbStatus = 'R'
138	mcinquil	1.2	common.jobDB.setStatus( str(int(idJob)-1), self.translateStatus(stato) )
139			else:
140			stato = "Cleared"
141			common.jobDB.setExitStatus( str(int(idJob)-1), job_exit_status )
142	mcinquil	1.4	if stato != "Done" and stato != "Cleared" and stato != "Aborted" and stato != "Done (Failed)":
143	mcinquil	1.2	print "%-10s %-20s" % (idJob,stato)
144			else:
145			print "%-10s %-20s %-20s %-25s" % (idJob,stato,exe_exit_code,job_exit_status)
146
147			if stato == 'Running':
148			self.countRun += 1
149			elif stato == 'Aborted':
150			self.countAbort += 1
151			elif stato == 'Done':
152			self.countDone += 1
153			elif stato == 'Cancelled':
154			self.countCancel += 1
155			elif stato == 'Submitted':
156			self.countSubmit += 1
157			elif stato == 'Submitting':
158			self.countSubmitting += 1
159			elif stato == 'Ready':
160			self.countReady += 1
161			elif stato == 'Scheduled':
162			self.countSched += 1
163			elif stato == 'Cleared':
164			self.countCleared += 1
165
166			addTree += 1
167			common.jobDB.save()
168
169			self.PrintReport_()
170
171
172			def PrintReport_(self) :
173
174			""" Report #jobs for each status """
175
176
177			print ''
178			print ">>>>>>>>> %i Total Jobs " % (self.countToTjob)
179			print ''
180
181			if (self.countSubmitting != 0) :
182			print ">>>>>>>>> %i Jobs Submitting by the server" % (self.countSubmitting)
183			if (self.countSubmit != 0):
184			print ">>>>>>>>> %i Jobs Submitted" % (self.countSubmit)
185	mcinquil	1.6	if (self.countReady != 0):
186			print ">>>>>>>>> %i Jobs Ready" % (self.countReady)
187	mcinquil	1.2	if (self.countSched != 0):
188			print ">>>>>>>>> %i Jobs Scheduled" % (self.countSched)
189			if (self.countRun != 0):
190			print ">>>>>>>>> %i Jobs Running" % (self.countRun)
191	mcinquil	1.6	if (self.countDone != 0):
192			print ">>>>>>>>> %i Jobs Done" % (self.countDone)
193			print " Retrieve them with: crab -getoutput -continue"
194			if (self.countKilled != 0):
195			print ">>>>>>>>> %i Jobs Killed" % (self.countKilled)
196			print " Retrieve more information with: crab -postMortem -continue"
197			if (self.countAbort != 0):
198			print ">>>>>>>>> %i Jobs Aborted" % (self.countAbort)
199	mcinquil	1.2	if (self.countCleared != 0):
200			print ">>>>>>>>> %i Jobs Cleared" % (self.countRun)
201			print ''
202			pass
203	spiga	1.1