def setResourceThresholds(self, site, **options):
"""
_setResourceThresholds_
Utility to set resource thresholds
"""
if not options:
options = {'state' : 'Normal',
'runningSlots' : 10,
'pendingSlots' : 5,
'tasks' : ['Processing', 'Merge'],
'Processing' : {'pendingSlots' : 5,
'runningSlots' : 10},
'Merge' : {'pendingSlots' : 2,
'runningSlots' : 5}}
resourceControl = ResourceControl()
resourceControl.insertSite(siteName = site, seName = 'se.%s' % (site),
ceName = site, plugin = "MockPlugin", pendingSlots = options['pendingSlots'],
runningSlots = options['runningSlots'], cmsName = site)
for task in options['tasks']:
resourceControl.insertThreshold(siteName = site, taskType = task,
maxSlots = options[task]['runningSlots'],
pendingSlots = options[task]['pendingSlots'])
if options.get('state'):
resourceControl.changeSiteState(site, options.get('state'))
return
def setUp(self):
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
#self.tearDown()
self.testInit.setSchema(customModules = ["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault = False)
self.daoFactory = DAOFactory(package = "WMCore.BossAir",
logger = myThread.logger,
dbinterface = myThread.dbi)
resourceControl = ResourceControl()
resourceControl.insertSite(siteName = 'Xanadu', seName = 'se.Xanadu',
ceName = 'Xanadu', plugin = "TestPlugin")
resourceControl.insertThreshold(siteName = 'Xanadu', taskType = 'Processing', \
maxSlots = 10000, pendingSlots = 10000)
# Create user
wmbsFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
newuser = wmbsFactory(classname = "Users.New")
newuser.execute(dn = "mnorman", group_name = "phgroup", role_name = "cmsrole")
def setResourceThresholds(self, site, **options):
"""
_setResourceThresholds_
Utility to set resource thresholds
"""
if not options:
options = {'state': 'Normal',
'runningSlots': 10,
'pendingSlots': 5,
'tasks': ['Processing', 'Merge'],
'Processing': {'pendingSlots': 5,
'runningSlots': 10},
'Merge': {'pendingSlots': 2,
'runningSlots': 5}}
resourceControl = ResourceControl()
resourceControl.insertSite(siteName=site, pnn='se.%s' % (site),
ceName=site, plugin="MockPlugin", pendingSlots=options['pendingSlots'],
runningSlots=options['runningSlots'], cmsName=site)
for task in options['tasks']:
resourceControl.insertThreshold(siteName=site, taskType=task,
maxSlots=options[task]['runningSlots'],
pendingSlots=options[task]['pendingSlots'])
if options.get('state'):
resourceControl.changeSiteState(site, options.get('state'))
return
def setUp(self):
"""
_setUp_
Setup the database and logging connection. Try to create all of the
WMBS tables. Also, create some dummy locations.
"""
myThread = threading.currentThread()
self.sites = ['T2_US_Florida', 'T2_US_UCSD', 'T2_TW_Taiwan', 'T1_CH_CERN']
self.testInit = TestInit(__file__)
self.testInit.setLogging(logLevel = logging.DEBUG)
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules = ['WMCore.WMBS',
'WMCore.ResourceControl',
'WMCore.Agent.Database'], useDefault = False)
self.testInit.setupCouch("dashboardreporter_t/jobs", "JobDump")
self.testInit.setupCouch("dashboardreporter_t/fwjrs", "FWJRDump")
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName = site, seName = site, ceName = site)
resourceControl.insertThreshold(siteName = site, taskType = 'Processing', \
maxSlots = 10000)
self.testDir = self.testInit.generateWorkDir()
self.alertsReceiver = None
return
def setUp(self):
"""
_setUp_
Set everything up.
"""
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules = ["WMCore.WMBS", "WMCore.BossAir",
"WMCore.ResourceControl"],
useDefault = False)
self.testInit.setupCouch("jobsubmittercaching_t/jobs", "JobDump")
self.testInit.setupCouch("jobsubmittercaching_t/fwjrs", "FWJRDump")
resourceControl = ResourceControl()
for siteName in ["T1_US_FNAL", "T1_UK_RAL"]:
resourceControl.insertSite(siteName = siteName, pnn = "se.%s" % (siteName),
ceName = siteName, plugin = "CondorPlugin", cmsName = siteName)
resourceControl.insertThreshold(siteName = siteName, taskType = "Processing",
maxSlots = 10000, pendingSlots = 10000)
self.testDir = self.testInit.generateWorkDir()
self.configFile = EmulatorSetup.setupWMAgentConfig()
return
def testThresholdsForSite(self):
"""
_testThresholdsForSite_
Check that we can get the thresholds in intelligible form
for each site
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, 40, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3)
result = myResourceControl.thresholdBySite(siteName="testSite1")
procInfo = {}
mergInfo = {}
for res in result:
if res['task_type'] == 'Processing':
procInfo = res
elif res['task_type'] == 'Merge':
mergInfo = res
self.assertEqual(procInfo.get('pending_slots', None), 20)
self.assertEqual(procInfo.get('running_slots', None), 40)
self.assertEqual(procInfo.get('max_slots', None), 10)
self.assertEqual(procInfo.get('task_pending_slots', None), 8)
self.assertEqual(mergInfo.get('pending_slots', None), 20)
self.assertEqual(mergInfo.get('running_slots', None), 40)
self.assertEqual(mergInfo.get('max_slots', None), 5)
self.assertEqual(mergInfo.get('task_pending_slots', None), 3)
return
def testThresholdsForSite(self):
"""
_testThresholdsForSite_
Check that we can get the thresholds in intelligible form
for each site
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10)
myResourceControl.insertThreshold("testSite1", "Merge", 5)
result = myResourceControl.thresholdBySite(siteName="testSite1")
procInfo = {}
mergInfo = {}
for res in result:
if res["task_type"] == "Processing":
procInfo = res
elif res["task_type"] == "Merge":
mergInfo = res
self.assertEqual(procInfo.get("job_slots", None), 20)
self.assertEqual(procInfo.get("max_slots", None), 10)
self.assertEqual(mergInfo.get("job_slots", None), 20)
self.assertEqual(mergInfo.get("max_slots", None), 5)
return
def testThresholdPriority(self):
"""
_testThresholdPriority_
Test that we get things back in priority order
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, 40, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3)
# test default task priorities
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds']['Merge']['priority'], 4)
self.assertEqual(result['testSite1']['thresholds']['Processing']['priority'], 0)
myResourceControl.changeTaskPriority("Merge", 3)
myResourceControl.changeTaskPriority("Processing", 1)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds']['Merge']['priority'], 3)
self.assertEqual(result['testSite1']['thresholds']['Processing']['priority'], 1)
myResourceControl.changeTaskPriority("Merge", 1)
myResourceControl.changeTaskPriority("Processing", 3)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds']['Merge']['priority'], 1)
self.assertEqual(result['testSite1']['thresholds']['Processing']['priority'], 3)
return
def testF_OverloadTest(self):
"""
_OverloadTest_
Test and see what happens if you put in more jobs
Then the sites can handle
"""
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertThreshold(siteName=site, taskType="Silly", maxSlots=1)
nRunning = getCondorRunningJobs(self.user)
self.assertEqual(nRunning, 0, "User currently has %i running jobs. Test will not continue" % (nRunning))
workloadName = "basicWorkload"
myThread = threading.currentThread()
workload = self.createTestWorkload()
config = self.getConfig()
changeState = ChangeState(config)
nSubs = 2
nJobs = 10
cacheDir = os.path.join(self.testDir, "CacheDir")
jobGroupList = self.createJobGroups(
nSubs=nSubs,
nJobs=nJobs,
task=workload.getTask("ReReco"),
workloadSpec=os.path.join(self.testDir, "workloadTest", workloadName),
type="Silly",
)
for group in jobGroupList:
changeState.propagate(group.jobs, "created", "new")
jobSubmitter = JobSubmitterPoller(config=config)
# Actually run it
jobSubmitter.algorithm()
# Should be one job for each site
nSites = len(self.sites)
nRunning = getCondorRunningJobs(self.user)
self.assertEqual(nRunning, nSites)
getJobsAction = self.daoFactory(classname="Jobs.GetAllJobs")
result = getJobsAction.execute(state="Executing", jobType="Silly")
self.assertEqual(len(result), nSites)
result = getJobsAction.execute(state="Created", jobType="Silly")
self.assertEqual(len(result), nJobs * nSubs - nSites)
# Now clean-up
command = ["condor_rm", self.user]
pipe = Popen(command, stdout=PIPE, stderr=PIPE, shell=False)
pipe.communicate()
del jobSubmitter
return
def setUp(self):
"""
_setUp_
Setup the database and logging connection. Try to create all of the
WMBS tables. Also, create some dummy locations.
"""
super(JobCreatorTest, self).setUp()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules=[
'WMCore.WMBS', 'WMCore.ResourceControl', 'WMCore.Agent.Database'
],
useDefault=False)
self.couchdbname = "jobcreator_t"
self.testInit.setupCouch("%s/jobs" % self.couchdbname, "JobDump")
self.testInit.setupCouch("%s/fwjrs" % self.couchdbname, "FWJRDump")
self.configFile = EmulatorSetup.setupWMAgentConfig()
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
locationAction = self.daoFactory(classname="Locations.New")
for site in self.sites:
locationAction.execute(siteName=site, pnn=site)
# Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName=site, pnn=site, ceName=site)
resourceControl.insertThreshold(siteName=site,
taskType='Processing',
maxSlots=10000,
pendingSlots=10000)
self.resourceControl = resourceControl
self._setup = True
self._teardown = False
self.testDir = self.testInit.generateWorkDir()
self.cwd = os.getcwd()
# Set heartbeat
self.componentName = 'JobCreator'
self.heartbeatAPI = HeartbeatAPI(self.componentName)
self.heartbeatAPI.registerComponent()
if PY3:
self.assertItemsEqual = self.assertCountEqual
return
def setUp(self):
"""
_setUp_
Setup the database and logging connection. Try to create all of the
WMBS tables. Also, create some dummy locations.
"""
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
#self.tearDown()
self.testInit.setSchema(customModules = ['WMCore.WMBS',
'WMCore.ResourceControl',
'WMCore.Agent.Database'], useDefault = False)
self.couchdbname = "jobcreator_t"
self.testInit.setupCouch("%s/jobs" % self.couchdbname, "JobDump")
self.testInit.setupCouch("%s/fwjrs" % self.couchdbname, "FWJRDump")
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
locationAction = self.daoFactory(classname = "Locations.New")
for site in self.sites:
locationAction.execute(siteName = site, seName = site)
#Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName = site, seName = site, ceName = site)
resourceControl.insertThreshold(siteName = site, taskType = 'Processing', \
maxSlots = 10000, pendingSlots = 10000)
self.resourceControl = resourceControl
self._setup = True
self._teardown = False
self.testDir = self.testInit.generateWorkDir()
self.cwd = os.getcwd()
# Set heartbeat
self.componentName = 'JobCreator'
self.heartbeatAPI = HeartbeatAPI(self.componentName)
self.heartbeatAPI.registerComponent()
return
def setUp(self):
"""
setup for test.
"""
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection(destroyAllDatabase=True)
self.tearDown()
self.testInit.setSchema(
customModules=["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault=False)
self.testInit.setupCouch("bossair_t/jobs", "JobDump")
self.testInit.setupCouch("bossair_t/fwjrs", "FWJRDump")
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
# Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName=site, pnn='%s_PNN' % site, cmsName=site,
ceName=site, plugin="SimpleCondorPlugin", pendingSlots=1000,
runningSlots=2000)
resourceControl.insertThreshold(siteName=site, taskType='Processing',
maxSlots=1000, pendingSlots=1000)
site = 'T3_US_Xanadu'
resourceControl.insertSite(siteName=site, pnn='%s_PNN' % site, cmsName=site,
ceName=site, plugin="TestPlugin")
resourceControl.insertThreshold(siteName=site, taskType='Processing',
maxSlots=10000, pendingSlots=10000)
# Create user
newuser = self.daoFactory(classname="Users.New")
newuser.execute(dn="tapas", group_name="phgroup", role_name="cmsrole")
# We actually need the user name
self.user = getpass.getuser()
# Change this to the working dir to keep track of error and log files from condor
self.testDir = self.testInit.generateWorkDir()
# Set heartbeat
componentName = 'test'
self.heartbeatAPI = HeartbeatAPI(componentName)
self.heartbeatAPI.registerComponent()
componentName = 'JobTracker'
self.heartbeatAPI2 = HeartbeatAPI(componentName)
self.heartbeatAPI2.registerComponent()
return
def setUp(self):
"""
_setUp_
Set up vital components
"""
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules = ["WMCore.WMBS",'WMCore.MsgService',
'WMCore.ResourceControl', 'WMCore.ThreadPool',
'WMCore.Agent.Database'],
useDefault = False)
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
locationAction = self.daoFactory(classname = "Locations.New")
pendingSlots = self.daoFactory(classname = "Locations.SetPendingSlots")
for site in self.sites:
locationAction.execute(siteName = site, seName = 'se.%s' % (site), ceName = site)
pendingSlots.execute(siteName = site, pendingSlots = 1000)
#Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName = site, seName = 'se.%s' % (site), ceName = site)
resourceControl.insertThreshold(siteName = site, taskType = 'Processing', \
maxSlots = 10000, pendingSlots = 10000)
self.testDir = self.testInit.generateWorkDir()
# Set heartbeat
for component in self.components:
heartbeatAPI = HeartbeatAPI(component)
heartbeatAPI.registerComponent()
return
def testAbortedState(self):
"""
_testAbortedState_
Check that we can kill jobs when a site is set to aborted
### We no longer need this test as we are not killing jobs that are running
"""
self.tempDir = self.testInit.generateWorkDir()
config = self.createConfig()
myResourceControl = ResourceControl(config)
myResourceControl.insertSite("testSite1", 10, 20, "testSE1", "testCE1", "T1_US_FNAL", "MockPlugin")
myResourceControl.insertSite("testSite2", 20, 40, "testSE2", "testCE2", "T1_IT_CNAF", "MockPlugin")
myResourceControl.insertThreshold("testSite1", "Processing", 20, 10)
myResourceControl.insertThreshold("testSite1", "Merge", 200, 100)
myResourceControl.insertThreshold("testSite2", "Processing", 50, 25)
myResourceControl.insertThreshold("testSite2", "Merge", 135, 65)
self.createJobs()
myResourceControl.changeSiteState("testSite1", "Aborted")
## Now check the tempDir for a FWJR for the killed job
reportPath = os.path.join(self.tempDir, "Report.0.pkl")
report = Report()
report.load(reportPath)
self.assertEqual(report.getExitCode(), 71301)
return
def testDrain(self):
"""Drain a site"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, priority=1)
myResourceControl.drainSite("testSite1")
result = myResourceControl.listThresholdsForCreate()
self.assertTrue(result["testSite1"]["drain"])
# re-renable
myResourceControl.drainSite("testSite1", drain=False)
result = myResourceControl.listThresholdsForCreate()
self.assertFalse(result["testSite1"]["drain"])
def setUp(self):
"""
Standard setup: Now with 100% more couch
"""
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection(destroyAllDatabase=True)
self.testInit.setSchema(
customModules=["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault=False,
)
self.testInit.setupCouch("jobsubmitter_t/jobs", "JobDump")
self.testInit.setupCouch("jobsubmitter_t/fwjrs", "FWJRDump")
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi)
locationAction = self.daoFactory(classname="Locations.New")
locationSlots = self.daoFactory(classname="Locations.SetJobSlots")
# We actually need the user name
self.user = getpass.getuser()
self.ceName = "127.0.0.1"
# Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(
siteName=site,
seName="se.%s" % (site),
ceName=site,
plugin="CondorPlugin",
pendingSlots=10000,
runningSlots=20000,
cmsName=site,
)
resourceControl.insertThreshold(siteName=site, taskType="Processing", maxSlots=10000)
self.testDir = self.testInit.generateWorkDir()
# Set heartbeat
self.componentName = "JobSubmitter"
self.heartbeatAPI = HeartbeatAPI(self.componentName)
self.heartbeatAPI.registerComponent()
return
def testChangeState(self):
"""
_testChangeState_
Check that we can change the state between different values and
retrieve it through the threshold methods
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, 40, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, 5)
result = myResourceControl.listThresholdsForCreate()
self.assertEqual(result['testSite1']['state'], 'Normal', 'Error: Wrong site state')
myResourceControl.changeSiteState("testSite1", "Down")
result = myResourceControl.listThresholdsForCreate()
self.assertEqual(result['testSite1']['state'], 'Down', 'Error: Wrong site state')
def setUp(self):
"""
setup for test.
"""
super(JobTrackerTest, self).setUp()
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
# self.testInit.clearDatabase(modules = ["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl"])
self.testInit.setSchema(customModules=["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl"],
useDefault=False)
self.testInit.setupCouch("jobtracker_t/jobs", "JobDump")
self.testInit.setupCouch("jobtracker_t/fwjrs", "FWJRDump")
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
# Create sites in resourceControl
resourceControl = ResourceControl()
resourceControl.insertSite(siteName='malpaquet', pnn='se.malpaquet',
ceName='malpaquet', plugin="CondorPlugin")
resourceControl.insertThreshold(siteName='malpaquet', taskType='Processing', \
maxSlots=10000, pendingSlots=10000)
locationAction = self.daoFactory(classname="Locations.New")
locationAction.execute(siteName="malpaquet", pnn="malpaquet",
ceName="malpaquet", plugin="CondorPlugin")
# Create user
newuser = self.daoFactory(classname="Users.New")
newuser.execute(dn="jchurchill")
# We actually need the user name
self.user = getpass.getuser()
self.testDir = self.testInit.generateWorkDir()
self.configFile = EmulatorSetup.setupWMAgentConfig()
def setUp(self):
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
#self.tearDown()
self.testInit.setSchema(customModules = ["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault = False)
self.daoFactory = DAOFactory(package = "WMCore.BossAir",
logger = myThread.logger,
dbinterface = myThread.dbi)
resourceControl = ResourceControl()
resourceControl.insertSite(siteName = 'Xanadu', seName = 'se.Xanadu',
ceName = 'Xanadu', plugin = "TestPlugin")
resourceControl.insertThreshold(siteName = 'Xanadu', taskType = 'Processing', \
maxSlots = 10000)
def setResourceThresholds(self, site, **options):
"""
_setResourceThresholds_
Utility to set resource thresholds
"""
if not options:
options = {
"state": "Normal",
"runningSlots": 10,
"pendingSlots": 5,
"tasks": ["Processing", "Merge"],
"Processing": {"pendingSlots": 5, "runningSlots": 10},
"Merge": {"pendingSlots": 2, "runningSlots": 5},
}
resourceControl = ResourceControl()
resourceControl.insertSite(
siteName=site,
pnn="se.%s" % (site),
ceName=site,
plugin="MockPlugin",
pendingSlots=options["pendingSlots"],
runningSlots=options["runningSlots"],
cmsName=site,
)
for task in options["tasks"]:
resourceControl.insertThreshold(
siteName=site,
taskType=task,
maxSlots=options[task]["runningSlots"],
pendingSlots=options[task]["pendingSlots"],
)
if options.get("state"):
resourceControl.changeSiteState(site, options.get("state"))
return
class ResourceControlUpdater(BaseWorkerThread):
"""
Update site status and thresholds from SSB
"""
def __init__(self, config):
"""
Initialize
"""
BaseWorkerThread.__init__(self)
# set the workqueue service for REST call
self.config = config
self.setVariables(self.config)
def setVariables(self, config):
"""
load all the variables from the config file
"""
# get dashboard url, set metric columns from config
self.dashboard = config.AgentStatusWatcher.dashboard
self.siteStatusMetric = config.AgentStatusWatcher.siteStatusMetric
self.cpuBoundMetric = config.AgentStatusWatcher.cpuBoundMetric
self.ioBoundMetric = config.AgentStatusWatcher.ioBoundMetric
# set pending percentages from config
self.pendingSlotsSitePercent = config.AgentStatusWatcher.pendingSlotsSitePercent
self.pendingSlotsTaskPercent = config.AgentStatusWatcher.pendingSlotsTaskPercent
self.runningExpressPercent = config.AgentStatusWatcher.runningExpressPercent
self.runningRepackPercent = config.AgentStatusWatcher.runningRepackPercent
# forced site list
self.forcedSiteList = config.AgentStatusWatcher.forcedSiteList
# agent teams (for dynamic threshold) and queueParams (drain mode)
self.teamNames = config.Agent.teamName
self.queueParams = config.WorkQueueManager.queueParams
self.agentsNumByTeam = getattr(config.AgentStatusWatcher, 'defaultAgentsNumByTeam', 5)
# only SSB sites
self.onlySSB = config.AgentStatusWatcher.onlySSB
# tier mode
self.tier0Mode = hasattr(config, "Tier0Feeder")
self.t1SitesCores = config.AgentStatusWatcher.t1SitesCores
# switch this component on/off
self.enabled = getattr(config.AgentStatusWatcher, 'enabled', True)
def setup(self, parameters):
"""
Set db connection and prepare resource control
"""
# Interface to WMBS/BossAir db
myThread = threading.currentThread()
# set resource control
self.resourceControl = ResourceControl(config = self.config)
# wmstats connection
self.centralCouchDBReader = WMStatsReader(self.config.AgentStatusWatcher.centralWMStatsURL)
# init variables
self.agentsByTeam = {}
def algorithm(self, parameters):
"""
_algorithm_
Update site info about state and thresholds
1. Get information from SSB
2. Get information about teams and agents from WMStats
3. Set site status and set therholds for each valid site
Sites from SSB are validated with PhEDEx node names
"""
# set variables every polling cycle
self.setVariables(self.config)
if not self.enabled:
logging.info("This component is not enabled in the configuration. Doing nothing.")
return
try:
# Get sites in Resource Control
currentSites = self.resourceControl.listCurrentSites()
logging.debug("Starting algorithm, getting site info from SSB")
stateBySite, slotsCPU, slotsIO = self.getInfoFromSSB()
if not stateBySite or not slotsCPU or not slotsIO:
logging.error("One or more of the SSB metrics is down. Please contact the Dashboard team.")
return
logging.debug("Setting status and thresholds for all sites, site pending: %s%%, task pending: %s%%" %
(str(self.pendingSlotsSitePercent), str(self.pendingSlotsTaskPercent)))
# get number of agents working in the same team (not in DrainMode)
agentsByTeam = self.getAgentsByTeam()
if not agentsByTeam:
logging.debug("agentInfo couch view is not available, use previous agent count %s" % self.agentsNumByTeam)
else:
self.agentsByTeam = agentsByTeam
teams = self.teamNames.split(',')
agentsCount = []
for team in teams:
if team not in self.agentsByTeam:
agentsCount.append(1)
else:
agentsCount.append(self.agentsByTeam[team])
self.agentsNumByTeam = min(agentsCount) # If agent is in several teams, we choose the team with less agents
logging.debug("Number of agents not in DrainMode running in the same team: %s" % str(self.agentsNumByTeam))
# set site status and thresholds
listSites = stateBySite.keys()
if self.forcedSiteList:
if set(self.forcedSiteList).issubset(set(listSites)):
listSites = self.forcedSiteList
logging.info("Forcing site list: %s" % (', '.join(self.forcedSiteList)))
else:
listSites = self.forcedSiteList
logging.warn("Forcing site list: %s. Some site(s) are not in SSB" % (', '.join(self.forcedSiteList)))
for site in listSites:
if site in currentSites:
sitestate = stateBySite.get(site,'Normal')
if site not in slotsCPU or site not in slotsIO:
logging.warn("%s not available in SSB. Changing only site status to %s." % (site,sitestate))
pluginResponse = self.updateSiteInfo(site, sitestate, None, None, self.agentsNumByTeam)
continue
pluginResponse = self.updateSiteInfo(site, sitestate, slotsCPU[site], slotsIO[site], self.agentsNumByTeam)
if not pluginResponse:
continue
logging.info('Setting site %s to %s, CPUBound: %s, IOBound: %s' %
(site, sitestate, slotsCPU[site], slotsIO[site]))
else:
logging.debug("Site '%s' has not been added to Resource Control" % site)
# if onlySSB sites or forcedSiteList, force to down all the sites not in SSB/forcedSiteList
if self.onlySSB or self.forcedSiteList:
for site in set(currentSites).difference(set(listSites)):
pluginResponse = self.updateSiteInfo(site, 'Down', 0, 0, self.agentsNumByTeam)
if not pluginResponse:
continue
logging.info('Only SSBsites/forcedSiteList, forcing site %s to Down, CPUBound: 0, IOBound: 0' % site)
logging.info("Resource update is completed, waiting for the next cycle.\n")
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s" % traceback.format_exc())
def getAgentsByTeam(self):
"""
_getAgentsByTeam_
Get the WMStats view about agents and teams
"""
agentsByTeam = []
try:
agentsByTeam = self.centralCouchDBReader.agentsByTeam()
return agentsByTeam
except Exception as ex:
logging.error("WMStats is not available or is unresponsive. Don't divide thresholds by team")
return agentsByTeam
def getInfoFromSSB(self):
"""
_getInfoFromSSB_
Get site status, CPU bound and IO bound from dashboard (SSB)
"""
# urls from site status board
url_site_state = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.siteStatusMetric)
url_cpu_bound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.cpuBoundMetric)
url_io_bound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.ioBoundMetric)
# get info from dashboard
sites = urllib2.urlopen(url_site_state).read()
cpu_bound = urllib2.urlopen(url_cpu_bound).read()
io_bound = urllib2.urlopen(url_io_bound).read()
# parse from json format to dictionary, get only 'csvdata'
site_state = json.loads(sites)['csvdata']
cpu_slots = json.loads(cpu_bound)['csvdata']
io_slots = json.loads(io_bound)['csvdata']
# dictionaries with status/thresholds info by VOName
stateBySite = self.siteStatusByVOName(site_state)
slotsCPU = self.thresholdsByVOName(cpu_slots)
slotsIO = self.thresholdsByVOName(io_slots)
return stateBySite, slotsCPU, slotsIO
def thresholdsByVOName(self, sites):
"""
_thresholdsByVOName_
Creates a dictionary with keys->VOName and values->threshold:
"""
thresholdbyVOName = {}
for site in sites:
voname = site['VOName']
value = site['Value']
if voname not in thresholdbyVOName:
if value is None:
logging.warn('Site %s does not have threholds in SSB, assuming 0' % voname)
thresholdbyVOName[voname] = 0
else:
thresholdbyVOName[voname] = int(value)
else:
logging.error('I have a duplicated threshold entry in SSB for %s' % voname)
return thresholdbyVOName
def siteStatusByVOName(self, sites):
"""
_siteStatusByVOName_
Creates a dictionary with keys->VOName and values->status:
"""
statusBySite = {}
for site in sites:
voname = site['VOName']
status = site['Status']
if voname not in statusBySite:
if not status:
logging.error('Site %s does not have status in SSB' % voname)
continue
new_status = self.getState(str(status))
if not new_status:
logging.error("Unkwown status '%s' for site %s, please check SSB" % (str(status), voname))
continue
statusBySite[voname] = new_status
else:
logging.error('I have a duplicated status entry in SSB for %s' % voname)
return statusBySite
def getState(self, stateFromSSB):
"""
_getState_
Translate SSB states into resource control state
"""
if stateFromSSB == "on":
return "Normal"
elif stateFromSSB == "drain":
return "Draining"
elif stateFromSSB == "tier0":
logging.debug('There is a site in tier0 status (Tier0Mode is %s)' % self.tier0Mode )
if self.tier0Mode:
return "Normal"
else:
return "Draining"
elif stateFromSSB == "down":
return "Down"
elif stateFromSSB == "skip":
return "Down"
else:
return None
def updateSiteInfo(self, siteName, state, CPUBound, IOBound, agentsNum):
"""
_updateSiteInfo_
Update information about a site in the database. Also set thresholds for a given site
pending_jobs policy:
sitePending is CPUBound*(pendingSlotsSitePercent/100)
taskPending is (CPUBound or IOBound)*(pendingSlotsTaskPercent/100) depending on the task type
This allows to maintain the right pressure in the queue, and keep the agent safe.
The site threshold is higger than each task threshold. This allow to have different task type jobs in the queue.
When there is several agents in the same team, we divide the pending threshold between the number of agents running.
"""
if self.resourceControl.listSiteInfo(siteName) is None:
logging.warn("Site %s has not been added to the resource control. Please check if the site was added by the condor plugin" % siteName)
return False
# set site state:
self.resourceControl.changeSiteState(siteName, state)
if CPUBound == None or IOBound == None:
return True
# tier0 T1 cores utilization
if self.tier0Mode and 'T1_' in siteName:
CPUBound = CPUBound*self.t1SitesCores/100
IOBound = IOBound*self.t1SitesCores/100
# Thresholds:
sitePending = int(CPUBound/agentsNum*self.pendingSlotsSitePercent/100)
taskCPUPending = int(CPUBound/agentsNum*self.pendingSlotsTaskPercent/100)
taskIOPending = int(IOBound/agentsNum*self.pendingSlotsTaskPercent/100)
# min pending values for thresholds
if taskCPUPending < 10 and taskCPUPending > 0:
taskCPUPending = 10
if taskIOPending < 10 and taskIOPending > 0:
taskIOPending = 10
# Set site main thresholds
self.resourceControl.setJobSlotsForSite(siteName = siteName,
pendingJobSlots = sitePending,
runningJobSlots = CPUBound)
# Set thresholds for CPU bound task types
cpuTasks = ['Processing', 'Production', 'Analysis']
for task in cpuTasks:
self.resourceControl.insertThreshold(siteName = siteName, taskType = task,
maxSlots = CPUBound, pendingSlots = taskCPUPending)
# Set thresholds for IO bound task types
ioTasks = ['Merge', 'Cleanup', 'Harvesting', 'LogCollect', 'Skim']
for task in ioTasks:
self.resourceControl.insertThreshold(siteName = siteName, taskType = task,
maxSlots = IOBound, pendingSlots = taskIOPending)
if self.tier0Mode:
# Set thresholds for tier0 task types
expressSlots = int(CPUBound*self.runningExpressPercent/100)
pendingExpress = int(expressSlots*self.pendingSlotsTaskPercent/100)
self.resourceControl.insertThreshold(siteName = siteName, taskType = 'Express',
maxSlots = expressSlots, pendingSlots = pendingExpress)
repackSlots = int(CPUBound*self.runningRepackPercent/100)
pendingRepack = int(repackSlots*self.pendingSlotsTaskPercent/100)
self.resourceControl.insertThreshold(siteName = siteName, taskType = 'Repack',
maxSlots = repackSlots, pendingSlots = pendingRepack)
return True
class ResourceControlUpdater(BaseWorkerThread):
"""
Update site status and thresholds from SSB
"""
def __init__(self, config):
"""
Initialize
"""
BaseWorkerThread.__init__(self)
self.config = config
self.tasksCPU = ['Processing', 'Production']
self.tasksIO = ['Merge', 'Cleanup', 'Harvesting', 'LogCollect', 'Skim']
self.minCPUSlots = 50
self.minIOSlots = 25
# get dashboard url, set metric columns from config
self.dashboard = config.AgentStatusWatcher.dashboard
self.siteStatusMetric = config.AgentStatusWatcher.siteStatusMetric
self.cpuBoundMetric = config.AgentStatusWatcher.cpuBoundMetric
self.ioBoundMetric = config.AgentStatusWatcher.ioBoundMetric
self.ssb = Dashboard(self.dashboard)
# set pending percentages from config
self.pendingSlotsSitePercent = config.AgentStatusWatcher.pendingSlotsSitePercent
self.pendingSlotsTaskPercent = config.AgentStatusWatcher.pendingSlotsTaskPercent
self.runningExpressPercent = config.AgentStatusWatcher.runningExpressPercent
self.runningRepackPercent = config.AgentStatusWatcher.runningRepackPercent
# sites forced to down
self.forceSiteDown = getattr(config.AgentStatusWatcher, 'forceSiteDown', [])
# agent team (for dynamic threshold) and queueParams (drain mode)
self.teamName = config.Agent.teamName
self.agentsNumByTeam = getattr(config.AgentStatusWatcher, 'defaultAgentsNumByTeam', 5)
# only SSB sites
self.onlySSB = config.AgentStatusWatcher.onlySSB
# tier mode
self.tier0Mode = hasattr(config, "Tier0Feeder")
self.t1SitesCores = config.AgentStatusWatcher.t1SitesCores
# switch this component on/off
self.enabled = getattr(config.AgentStatusWatcher, 'enabled', True)
# set resource control
self.resourceControl = ResourceControl(config=self.config)
# wmstats connection
self.centralCouchDBReader = WMStatsReader(self.config.AgentStatusWatcher.centralWMStatsURL)
@timeFunction
def algorithm(self, parameters):
"""
_algorithm_
Update site state and thresholds, based on differences between resource
control database and info available in SSB.
1. Get info from Resource Control database
2. Get info from SSB
3. Get information about teams and number of agents from WMStats
4. Change site state when needed (this triggers a condor clasAd fetch)
5. Change site thresholds when needed (and task thresholds)
Sites from SSB are validated with PhEDEx node names
"""
if not self.enabled:
logging.info("This component is not enabled in the configuration. Doing nothing.")
return
try:
sitesRC = self.resourceControl.listSitesSlots()
logging.debug("Info from resource control: %s", sitesRC)
# first, update site status
ssbSiteStatus = self.getSiteStatus()
self.checkStatusChanges(sitesRC, ssbSiteStatus)
# now fetch site slots thresholds
sitesSSB = self.getInfoFromSSB()
if not sitesSSB:
logging.error("One or more of the SSB metrics is down. Please contact the Dashboard team.")
return
logging.debug("Info from SSB: %s", sitesSSB)
# get number of agents working in the same team (not in DrainMode)
self.getAgentsByTeam()
# Check which site slots need to be updated in the database
self.checkSlotsChanges(sitesRC, sitesSSB)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s", traceback.format_exc())
logging.info("Resource control cycle finished updating site state and thresholds.")
def getAgentsByTeam(self):
"""
_getAgentsByTeam_
Get the WMStats view for agents and teams
"""
if isDrainMode(self.config):
# maximize pending thresholds to get this agent drained ASAP
self.agentsNumByTeam = 1
return
agentsByTeam = {}
try:
agentsByTeam = self.centralCouchDBReader.agentsByTeam(filterDrain=True)
except Exception:
logging.error("WMStats is not available or is unresponsive.")
if not agentsByTeam:
logging.warning("agentInfo couch view is not available, use default value %s", self.agentsNumByTeam)
else:
self.agentsNumByTeam = agentsByTeam.get(self.teamName, self.agentsNumByTeam)
logging.debug("Agents connected to the same team (not in DrainMode): %d", self.agentsNumByTeam)
return
def getInfoFromSSB(self):
"""
_getInfoFromSSB_
Get site status, CPU bound and IO bound from dashboard (SSB).
Returns a dict of dicts where the first key is the site name.
"""
ssbCpuSlots = self.ssb.getMetric(self.cpuBoundMetric)
ssbIoSlots = self.ssb.getMetric(self.ioBoundMetric)
ssbSiteSlots = self.thresholdsByVOName(ssbCpuSlots, ssbIoSlots)
return ssbSiteSlots
def checkStatusChanges(self, infoRC, infoSSB):
"""
_checkStatusChanges_
Checks which sites need to have their site state updated in
resource control, based on:
1. settings defined for the component (config.py)
2. site state changes between SSB and RC
"""
# First sets list of forced sites to down (HLT @FNAL is an example)
for site in self.forceSiteDown:
if site in infoRC and infoRC[site]['state'] != 'Down':
logging.info("Forcing site %s to Down", site)
self.updateSiteState(site, 'Down')
infoSSB.pop(site, None)
# if onlySSB sites, force all the sites not in SSB to down
if self.onlySSB:
for site in set(infoRC).difference(set(infoSSB)):
if infoRC[site]['state'] != 'Down':
logging.info('Only SSBsites, forcing site %s to Down', site)
self.updateSiteState(site, 'Down')
# normally set all the others
for site in set(infoRC).intersection(set(infoSSB)):
if infoRC[site]['state'] != infoSSB[site]['state']:
logging.info('Changing %s state from %s to %s', site, infoRC[site]['state'], infoSSB[site]['state'])
self.updateSiteState(site, infoSSB[site]['state'])
return
def checkSlotsChanges(self, infoRC, infoSSB):
"""
_checkSlotsChanges_
Checks which sites need to have their running and/or pending
slots updated in resource control database, based on:
1. number of agents connected to the same team
2. and slots provided by the Dashboard team (SSB)
If site slots are updated, then updates the task level too.
"""
logging.debug("Settings for site and task pending slots: %s%% and %s%%", self.pendingSlotsSitePercent,
self.pendingSlotsTaskPercent)
for site in set(infoRC).intersection(set(infoSSB)):
if self.tier0Mode and site.startswith('T1_'):
# T1 cores utilization for Tier0
infoSSB[site]['slotsCPU'] *= self.t1SitesCores / 100
infoSSB[site]['slotsIO'] *= self.t1SitesCores / 100
else:
# round very small sites to the bare minimum
infoSSB[site]['slotsCPU'] = max(infoSSB[site]['slotsCPU'], self.minCPUSlots)
infoSSB[site]['slotsIO'] = max(infoSSB[site]['slotsIO'], self.minIOSlots)
CPUBound = infoSSB[site]['slotsCPU']
IOBound = infoSSB[site]['slotsIO']
sitePending = max(int(CPUBound / self.agentsNumByTeam * self.pendingSlotsSitePercent / 100),
self.minCPUSlots)
# update site slots, if needed
if infoRC[site]['running_slots'] != CPUBound or infoRC[site]['pending_slots'] != sitePending:
# Update site running and pending slots
logging.info("Updating %s site thresholds for pend/runn: %d/%d", site, sitePending, CPUBound)
self.resourceControl.setJobSlotsForSite(site, pendingJobSlots=sitePending,
runningJobSlots=CPUBound)
# now handle the task level thresholds
self.checkTaskSlotsChanges(site, CPUBound, IOBound)
def thresholdsByVOName(self, infoCpu, infoIo):
"""
_thresholdsByVOName_
Creates a dictionary with CPU and IO slots keyed by the site name.
If any of the thresholds is missing or has an invalid value, the whole
site thresholds is skipped.
"""
ssbSiteSlots = {}
for entry in infoCpu:
if entry['Value'] is None:
logging.warn('Site %s has invalid thresholds in SSB. Taking no action', entry['VOName'])
continue
ssbSiteSlots[entry['VOName']] = {'slotsCPU': int(entry['Value'])}
# then iterate over the IO slots
for entry in infoIo:
if entry['VOName'] not in ssbSiteSlots:
logging.warn('Site %s does not have CPU thresholds in SSB. Taking no action', entry['VOName'])
ssbSiteSlots.pop(entry['VOName'], None)
continue
if entry['Value'] is None:
logging.warn('Site %s has invalid thresholds in SSB. Taking no action', entry['VOName'])
ssbSiteSlots.pop(entry['VOName'], None)
continue
ssbSiteSlots[entry['VOName']]['slotsIO'] = int(entry['Value'])
return ssbSiteSlots
def getSiteStatus(self):
"""
_getSiteStatus_
Fetch site state from SSB and map it to agent state
"""
ssbState = self.ssb.getMetric(self.siteStatusMetric)
ssbSiteState = {}
for site in ssbState:
voname = site['VOName']
status = site['Status']
if voname not in ssbSiteState:
statusAgent = self.getState(str(status))
if not statusAgent:
logging.error("Unknown status '%s' for site %s, please check SSB", status, voname)
else:
ssbSiteState[voname] = {'state': statusAgent}
else:
logging.warning('I have a duplicated status entry in SSB for %s', voname)
return ssbSiteState
def getState(self, stateSSB):
"""
_getState_
Translates SSB states into resource control state
"""
ssb2agent = {'enabled': 'Normal',
'drain': 'Draining',
'disabled': 'Down',
'test': 'Draining'}
# 'test' state behaviour varies between production and tier0 agents
ssb2agent['test'] = 'Normal' if self.tier0Mode else "Draining"
return ssb2agent.get(stateSSB)
def updateSiteState(self, siteName, state):
"""
_updateSiteState_
Update only the site state in the resource control database.
"""
try:
self.resourceControl.changeSiteState(siteName, state)
except Exception as ex:
logging.error("Failed to update %s state to %s:", siteName, state)
logging.error(str(ex))
logging.error("Traceback: \n%s", traceback.format_exc())
return
def checkTaskSlotsChanges(self, siteName, CPUBound, IOBound):
"""
_checkTaskSlotsChanges_
Update the CPU and IOBound slots for a given site.
"""
siteTaskSlots = self.resourceControl.thresholdBySite(siteName)
taskCPUPending = max(int(CPUBound / self.agentsNumByTeam * self.pendingSlotsTaskPercent / 100),
self.minCPUSlots)
taskIOPending = max(int(IOBound / self.agentsNumByTeam * self.pendingSlotsTaskPercent / 100), self.minIOSlots)
updateTasks = False
if siteTaskSlots[0]['task_type'] in self.tasksCPU and siteTaskSlots[0]['task_pending_slots'] != taskCPUPending:
updateTasks = True
elif siteTaskSlots[0]['task_type'] in self.tasksIO and siteTaskSlots[0]['task_pending_slots'] != taskIOPending:
updateTasks = True
if updateTasks:
logging.info("Updating %s CPU tasks thresholds for pend/runn: %d/%d", siteName,
taskCPUPending, CPUBound)
self.resourceControl.insertThreshold(siteName, taskType=self.tasksCPU, maxSlots=CPUBound,
pendingSlots=taskCPUPending)
logging.info("Updating %s IO tasks thresholds for pend/runn: %d/%d", siteName,
taskIOPending, IOBound)
self.resourceControl.insertThreshold(siteName, taskType=self.tasksIO, maxSlots=IOBound,
pendingSlots=taskIOPending)
if self.tier0Mode:
# Set task thresholds for Tier0
logging.debug("Updating %s Express and Repack task thresholds.", siteName)
expressSlots = int(CPUBound * self.runningExpressPercent / 100)
pendingExpress = int(expressSlots * self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(siteName, 'Express', expressSlots, pendingExpress)
repackSlots = int(CPUBound * self.runningRepackPercent / 100)
pendingRepack = int(repackSlots * self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(siteName, 'Repack', repackSlots, pendingRepack)
def setUp(self):
"""
setup for test.
"""
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.tearDown()
self.testInit.setSchema(
customModules=["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault=False,
)
self.testInit.setupCouch("bossair_t/jobs", "JobDump")
self.testInit.setupCouch("bossair_t/fwjrs", "FWJRDump")
self.daoFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi)
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
# Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(
siteName=site,
pnn="se.%s" % (site),
cmsName=site,
ceName=site,
plugin="CondorPlugin",
pendingSlots=1000,
runningSlots=2000,
)
resourceControl.insertThreshold(siteName=site, taskType="Processing", maxSlots=1000, pendingSlots=1000)
resourceControl.insertSite(
siteName="Xanadu", pnn="se.Xanadu", cmsName=site, ceName="Xanadu", plugin="TestPlugin"
)
resourceControl.insertThreshold(siteName="Xanadu", taskType="Processing", maxSlots=10000, pendingSlots=10000)
resourceControl.insertSite(
siteName="jade-cms.hip.fi",
pnn="madhatter.csc.fi",
cmsName=site,
ceName="jade-cms.hip.fi",
plugin="ARCPlugin",
)
resourceControl.insertThreshold(
siteName="jade-cms.hip.fi", taskType="Processing", maxSlots=100, pendingSlots=100
)
# using this for glite submissions
resourceControl.insertSite(
siteName="grid-ce-01.ba.infn.it",
pnn="storm-se-01.ba.infn.it",
cmsName=site,
ceName="grid-ce-01.ba.infn.it",
plugin="gLitePlugin",
)
resourceControl.insertThreshold(
siteName="grid-ce-01.ba.infn.it", taskType="Processing", maxSlots=50, pendingSlots=50
)
# Create user
newuser = self.daoFactory(classname="Users.New")
newuser.execute(dn="tapas", group_name="phgroup", role_name="cmsrole")
# We actually need the user name
self.user = getpass.getuser()
# Change this to the working dir to keep track of error and log files from condor
self.testDir = self.testInit.generateWorkDir()
# Set heartbeat
componentName = "test"
self.heartbeatAPI = HeartbeatAPI(componentName)
self.heartbeatAPI.registerComponent()
componentName = "JobTracker"
self.heartbeatAPI2 = HeartbeatAPI(componentName)
self.heartbeatAPI2.registerComponent()
return
def createTestJob(self):
"""
_createTestJobs_
Create several jobs
"""
#Create sites in resourceControl
resourceControl = ResourceControl()
resourceControl.insertSite(siteName = 'malpaquet', seName = 'se.malpaquet',
ceName = 'malpaquet', plugin = "CondorPlugin")
resourceControl.insertThreshold(siteName = 'malpaquet', taskType = 'Processing', \
maxSlots = 10000)
locationAction = self.daoFactory(classname = "Locations.New")
locationAction.execute(siteName = "malpaquet", seName = "malpaquet",
ceName = "malpaquet", plugin = "CondorPlugin")
jobAction = self.daoFactory(classname = "Jobs.New")
jobAction.execute()
# Create user
newuser = self.daoFactory(classname = "Users.New")
newuser.execute(dn = "jchurchill")
testWorkflow = Workflow(spec = "spec.xml", owner = "Simon",
name = "meloam_ASYNCTEST1_120810_170823_8981", task="Test")
testWorkflow.create()
testWMBSFileset = Fileset(name = "TestFileset")
testWMBSFileset.create()
testSubscription = Subscription(fileset = testWMBSFileset,
workflow = testWorkflow,
type = "Processing",
split_algo = "FileBased")
testSubscription.create()
testJobGroup = JobGroup(subscription = testSubscription)
testJobGroup.create()
# Create a file
testFileA = File(lfn = "/this/is/a/lfnA", size = 1024, events = 10)
testFileA.addRun(Run(10, *[12312]))
testFileA.setLocation('malpaquet')
testFileA.create()
baseName = makeUUID()
# make a copy of the FWJR since it will be modified
shutil.copyfile(os.path.join(WMCore.WMBase.getTestBase(),
'WMComponent_t',
'AsyncStageoutTracker_t',
'oneaso.pkl'),
os.path.join(self.testDir, "oneaso.pkl") )
# Now create a job
testJob = Job(name = '%s-%i' % (baseName, 1))
testJob.addFile(testFileA)
testJob['location'] = 'malpaquet'
testJob['retry_count'] = 1
testJob['retry_max'] = 10
testJob['fwjr_path'] = os.path.join(self.testDir, "oneaso.pkl")
testJob.create(testJobGroup)
testJob.save()
testJobGroup.add(testJob)
testJobGroup.commit()
stateAction = self.daoFactory(classname = "Jobs.ChangeState")
stateAction.execute( [{'id' : testJob['id'],
'state' : 'asopending',
'couch_record' : testJob['couch_record']}])
return testJob
return testJobGroup
def testList(self):
"""
_testList_
Test the functions that list thresholds for creating jobs and submitting
jobs.
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 10, 20, "testSE1", "testCE1", "T1_US_FNAL", "LsfPlugin")
myResourceControl.insertSite("testSite2", 20, 40, "testSE2", "testCE2", "T3_US_FNAL", "LsfPlugin")
myResourceControl.insertThreshold("testSite1", "Processing", 20, 10)
myResourceControl.insertThreshold("testSite1", "Merge", 200, 100)
myResourceControl.insertThreshold("testSite2", "Processing", 50, 25)
myResourceControl.insertThreshold("testSite2", "Merge", 135, 65)
self.createJobs()
createThresholds = myResourceControl.listThresholdsForCreate()
submitThresholds = myResourceControl.listThresholdsForSubmit()
self.assertEqual(len(createThresholds.keys()), 2,
"Error: Wrong number of sites in create thresholds")
self.assertEqual(createThresholds["testSite1"]["total_slots"], 10,
"Error: Wrong number of slots for site 1")
self.assertEqual(createThresholds["testSite2"]["total_slots"], 20,
"Error: Wrong number of slots for site 2")
# We should have two running jobs with locations at site one,
# two running jobs without locations at site two, and one running
# job without a location at site one and two.
self.assertEqual(createThresholds["testSite1"]["pending_jobs"], {0: 4},
"Error: Wrong number of pending jobs for site 1")
# We should have one running job with a location at site 2 and
# another running job without a location.
self.assertEqual(createThresholds["testSite2"]["pending_jobs"], {0: 2},
"Error: Wrong number of pending jobs for site 2")
# We should also have a phedex_name
self.assertEqual(createThresholds["testSite1"]["cms_name"], "T1_US_FNAL")
self.assertEqual(createThresholds["testSite2"]["cms_name"], "T3_US_FNAL")
mergeThreshold1 = None
mergeThreshold2 = None
procThreshold1 = None
procThreshold2 = None
self.assertEqual(set(submitThresholds.keys()), set(["testSite1", "testSite2"]))
for taskType, threshold in submitThresholds["testSite1"]["thresholds"].items():
if taskType == "Merge":
mergeThreshold1 = threshold
elif taskType == "Processing":
procThreshold1 = threshold
for taskType, threshold in submitThresholds["testSite2"]["thresholds"].items():
if taskType == "Merge":
mergeThreshold2 = threshold
elif taskType == "Processing":
procThreshold2 = threshold
self.assertEqual(submitThresholds["testSite1"]["total_running_jobs"], 0,
"Error: Wrong number of running jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite2"]["total_running_jobs"], 1,
"Error: Wrong number of running jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite1"]["total_pending_jobs"], 1,
"Error: Wrong number of pending jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite2"]["total_pending_jobs"], 0,
"Error: Wrong number of pending jobs for submit thresholds.")
self.assertEqual(mergeThreshold1["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold1["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold1["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold1["task_pending_jobs"], 1,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold2["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold2["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold2["task_running_jobs"], 1,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold2["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
return
def testInsert(self):
"""
_testInsert_
Verify that inserting sites and thresholds works correctly, even if the
site or threshold already exists.
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 10, 20, "testSE1", "testCE1")
myResourceControl.insertSite("testSite1", 10, 20, "testSE1", "testCE1")
myResourceControl.insertSite("testSite2", 100, 200, "testSE2", "testCE2")
myResourceControl.insertThreshold("testSite1", "Processing", 20, 10)
myResourceControl.insertThreshold("testSite1", "Merge", 200, 100)
myResourceControl.insertThreshold("testSite1", "Merge", 250, 150)
myResourceControl.insertThreshold("testSite2", "Processing", 50, 30)
myResourceControl.insertThreshold("testSite2", "Merge", 135, 100)
createThresholds = myResourceControl.listThresholdsForCreate()
self.assertEqual(len(createThresholds.keys()), 2,
"Error: Wrong number of site in Resource Control DB")
self.assertTrue("testSite1" in createThresholds.keys(),
"Error: Test Site 1 missing from thresholds.")
self.assertTrue("testSite2" in createThresholds.keys(),
"Error: Test Site 2 missing from thresholds.")
self.assertEqual(createThresholds["testSite1"]["total_slots"], 10,
"Error: Wrong number of total slots.")
self.assertEqual(createThresholds["testSite1"]["pending_jobs"], {0: 0},
"Error: Wrong number of running jobs: %s" %
createThresholds["testSite1"]["pending_jobs"])
self.assertEqual(createThresholds["testSite2"]["total_slots"], 100,
"Error: Wrong number of total slots.")
self.assertEqual(createThresholds["testSite2"]["pending_jobs"], {0: 0},
"Error: Wrong number of running jobs.")
thresholds = myResourceControl.listThresholdsForSubmit()
self.assertEqual(len(thresholds.keys()), 2,
"Error: Wrong number of sites in Resource Control DB")
self.assertTrue("testSite1" in thresholds.keys(),
"Error: testSite1 missing from thresholds.")
self.assertTrue("testSite2" in thresholds.keys(),
"Error: testSite2 missing from thresholds.")
site1Info = thresholds["testSite1"]
site2Info = thresholds["testSite2"]
site1Thresholds = site1Info["thresholds"]
site2Thresholds = site2Info["thresholds"]
procThreshold1 = None
procThreshold2 = None
mergeThreshold1 = None
mergeThreshold2 = None
for taskType, threshold in site1Thresholds.items():
if taskType == "Merge":
mergeThreshold1 = threshold
elif taskType == "Processing":
procThreshold1 = threshold
for taskType, threshold in site2Thresholds.items():
if taskType == "Merge":
mergeThreshold2 = threshold
elif taskType == "Processing":
procThreshold2 = threshold
self.assertEqual(len(site1Thresholds), 2,
"Error: Wrong number of task types.")
self.assertEqual(len(site2Thresholds), 2,
"Error: Wrong number of task types.")
self.assertNotEqual(procThreshold1, None)
self.assertNotEqual(procThreshold2, None)
self.assertNotEqual(mergeThreshold1, None)
self.assertNotEqual(mergeThreshold2, None)
self.assertEqual(site1Info["total_pending_slots"], 10,
"Error: Site thresholds wrong")
self.assertEqual(site1Info["total_running_slots"], 20,
"Error: Site thresholds wrong")
self.assertEqual(site1Info["total_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(site1Info["total_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold1["task_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold1["task_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold1["max_slots"], 20,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold1["pending_slots"], 10,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold1["task_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold1["task_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold1["max_slots"], 250,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold1["pending_slots"], 150,
"Error: Site thresholds wrong")
self.assertEqual(site2Info["total_pending_slots"], 100,
"Error: Site thresholds wrong")
self.assertEqual(site2Info["total_running_slots"], 200,
"Error: Site thresholds wrong")
self.assertEqual(site2Info["total_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(site2Info["total_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold2["task_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold2["task_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold2["max_slots"], 50,
"Error: Site thresholds wrong")
self.assertEqual(procThreshold2["pending_slots"], 30,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold2["task_running_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold2["task_pending_jobs"], 0,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold2["max_slots"], 135,
"Error: Site thresholds wrong")
self.assertEqual(mergeThreshold2["pending_slots"], 100,
"Error: Site thresholds wrong")
class ResourceControlUpdater(BaseWorkerThread):
"""
Update site status and thresholds from SSB
"""
def __init__(self, config):
"""
Initialize
"""
BaseWorkerThread.__init__(self)
# set the workqueue service for REST call
self.config = config
self.setVariables(self.config)
def setVariables(self, config):
"""
load all the variables from the config file
"""
# get dashboard url, set metric columns from config
self.dashboard = config.AgentStatusWatcher.dashboard
self.siteStatusMetric = config.AgentStatusWatcher.siteStatusMetric
self.cpuBoundMetric = config.AgentStatusWatcher.cpuBoundMetric
self.ioBoundMetric = config.AgentStatusWatcher.ioBoundMetric
# set pending percentages from config
self.pendingSlotsSitePercent = config.AgentStatusWatcher.pendingSlotsSitePercent
self.pendingSlotsTaskPercent = config.AgentStatusWatcher.pendingSlotsTaskPercent
self.runningExpressPercent = config.AgentStatusWatcher.runningExpressPercent
self.runningRepackPercent = config.AgentStatusWatcher.runningRepackPercent
# sites forced to down
self.forceSiteDown = getattr(config.AgentStatusWatcher, 'forceSiteDown', [])
# agent teams (for dynamic threshold) and queueParams (drain mode)
self.teamNames = config.Agent.teamName
self.agentsNumByTeam = getattr(config.AgentStatusWatcher, 'defaultAgentsNumByTeam', 5)
# only SSB sites
self.onlySSB = config.AgentStatusWatcher.onlySSB
# tier mode
self.tier0Mode = hasattr(config, "Tier0Feeder")
self.t1SitesCores = config.AgentStatusWatcher.t1SitesCores
# switch this component on/off
self.enabled = getattr(config.AgentStatusWatcher, 'enabled', True)
def setup(self, parameters):
"""
Set db connection and prepare resource control
"""
# Interface to WMBS/BossAir db
myThread = threading.currentThread()
# set resource control
self.resourceControl = ResourceControl(config = self.config)
# wmstats connection
self.centralCouchDBReader = WMStatsReader(self.config.AgentStatusWatcher.centralWMStatsURL)
def algorithm(self, parameters):
"""
_algorithm_
Update site state and thresholds, based on differences between resource
control database and info available in SSB.
1. Get info from Resource Control database
2. Get info from SSB
3. Get information about teams and number of agents from WMStats
4. Change site state when needed (this triggers a condor clasAd fetch)
5. Change site thresholds when needed (and task thresholds)
Sites from SSB are validated with PhEDEx node names
"""
# set variables every polling cycle
self.setVariables(self.config)
if not self.enabled:
logging.info("This component is not enabled in the configuration. Doing nothing.")
return
try:
sitesRC = self.resourceControl.listSitesSlots()
logging.debug("Info from resource control: %s" % sitesRC)
sitesSSB = self.getInfoFromSSB()
if not sitesSSB:
return
logging.debug("Info from SSB: %s" % sitesSSB)
# Check which site states need to be updated in the database
sitesRC = self.checkStatusChanges(sitesRC, sitesSSB)
# get number of agents working in the same team (not in DrainMode)
self.getAgentsByTeam()
# Check which site slots need to be updated in the database
self.checkSlotsChanges(sitesRC, sitesSSB, self.agentsNumByTeam)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s" % traceback.format_exc())
logging.info("Resource control cycle finished updating site state and thresholds.")
def getAgentsByTeam(self):
"""
_getAgentsByTeam_
Get the WMStats view about agents and teams
"""
agentsByTeam = {}
try:
agentsByTeam = self.centralCouchDBReader.agentsByTeam()
except Exception as ex:
logging.error("WMStats is not available or is unresponsive.")
if not agentsByTeam:
logging.debug("agentInfo couch view is not available, use default value %s" % self.agentsNumByTeam)
else:
self.agentsByTeam = agentsByTeam
agentsCount = []
for team in self.teamNames.split(','):
if team not in self.agentsByTeam:
agentsCount.append(1)
else:
agentsCount.append(self.agentsByTeam[team])
# If agent is in several teams, we choose the team with less agents
self.agentsNumByTeam = min(agentsCount, self.agentsNumByTeam)
logging.debug("Agents connected to the same team (not in DrainMode): %d" % self.agentsNumByTeam)
return
def getInfoFromSSB(self):
"""
_getInfoFromSSB_
Get site status, CPU bound and IO bound from dashboard (SSB).
Returns a dict of dicts where the first key is the site name.
"""
# urls from site status board
url_site_state = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.siteStatusMetric)
url_cpu_bound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.cpuBoundMetric)
url_io_bound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(self.ioBoundMetric)
# get info from dashboard
sites = urllib2.urlopen(url_site_state).read()
cpu_bound = urllib2.urlopen(url_cpu_bound).read()
io_bound = urllib2.urlopen(url_io_bound).read()
# parse from json format to dictionary, get only 'csvdata'
site_state = json.loads(sites)['csvdata']
cpu_slots = json.loads(cpu_bound)['csvdata']
io_slots = json.loads(io_bound)['csvdata']
# dictionaries with status/thresholds info by VOName
stateBySite = self.siteStatusByVOName(site_state)
slotsCPU = self.thresholdsByVOName(cpu_slots)
slotsIO = self.thresholdsByVOName(io_slots)
sitesSSB = {}
if not stateBySite or not slotsCPU or not slotsIO:
logging.error("One or more of the SSB metrics is down. Please contact the Dashboard team.")
return sitesSSB
for k,v in stateBySite.iteritems():
sitesSSB[k] = {'state': v}
sitesSSB[k]['slotsCPU'] = slotsCPU[k] if k in slotsCPU else None
sitesSSB[k]['slotsIO'] = slotsIO[k] if k in slotsIO else None
return sitesSSB
def checkStatusChanges(self, infoRC, infoSSB):
"""
_checkStatusChanges_
Checks which sites need to have their site state updated in
resource control, based on:
1. settings defined for the component (config.py)
2. site state changes between SSB and RC
Returns the new infoRC dict (where a few key/value pairs were
deleted - no need to update slots information)
"""
# First sets list of forced sites to down (HLT @FNAL is an example)
for site in self.forceSiteDown:
if site in infoRC and infoRC[site]['state'] != 'Down':
logging.info("Forcing site %s to Down" % site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# if onlySSB sites, force all the sites not in SSB to down
if self.onlySSB:
for site in set(infoRC).difference(set(infoSSB)):
if infoRC[site]['state'] != 'Down':
logging.info('Only SSBsites, forcing site %s to Down' % site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# this time don't update infoRC since we still want to update slots info
for site in set(infoRC).intersection(set(infoSSB)):
if infoRC[site]['state'] != infoSSB[site]['state']:
logging.info('Changing %s state from %s to %s' % (site, infoRC[site]['state'],
infoSSB[site]['state']))
self.updateSiteState(site, infoSSB[site]['state'])
return infoRC
def checkSlotsChanges(self, infoRC, infoSSB, agentsCount):
"""
_checkSlotsChanges_
Checks which sites need to have their running and/or pending
slots updated in resource control database, based on:
1. number of agents connected to the same team
2. and slots provided by the Dashboard team (SSB)
If site slots are updated, then also updates its tasks.
"""
tasksCPU = ['Processing', 'Production']
tasksIO = ['Merge', 'Cleanup', 'Harvesting', 'LogCollect', 'Skim']
minCPUSlots, minIOSlots = 50, 25
logging.debug("Settings for site and task pending slots: %s%% and %s%%" %
(self.pendingSlotsSitePercent, self.pendingSlotsTaskPercent))
for site in set(infoRC).intersection(set(infoSSB)):
if self.tier0Mode and 'T1_' in site:
# T1 cores utilization for Tier0
infoSSB[site]['slotsCPU'] = infoSSB[site]['slotsCPU'] * self.t1SitesCores/100
infoSSB[site]['slotsIO'] = infoSSB[site]['slotsIO'] * self.t1SitesCores/100
# round very small sites to the bare minimum
if infoSSB[site]['slotsCPU'] < minCPUSlots:
infoSSB[site]['slotsCPU'] = minCPUSlots
if infoSSB[site]['slotsIO'] < minIOSlots:
infoSSB[site]['slotsIO'] = minIOSlots
CPUBound = infoSSB[site]['slotsCPU']
IOBound = infoSSB[site]['slotsIO']
sitePending = max(int(CPUBound/agentsCount * self.pendingSlotsSitePercent/100), minCPUSlots)
taskCPUPending = max(int(CPUBound/agentsCount * self.pendingSlotsTaskPercent/100), minCPUSlots)
taskIOPending = max(int(IOBound/agentsCount * self.pendingSlotsTaskPercent/100), minIOSlots)
if infoRC[site]['running_slots'] != CPUBound or infoRC[site]['pending_slots'] != sitePending:
# Update site running and pending slots
logging.debug("Updating %s site thresholds for pend/runn: %d/%d" % (site, sitePending, CPUBound))
self.resourceControl.setJobSlotsForSite(site, pendingJobSlots = sitePending,
runningJobSlots = CPUBound)
# Update site CPU tasks running and pending slots (large running slots)
logging.debug("Updating %s tasksCPU thresholds for pend/runn: %d/%d" % (site, taskCPUPending,
CPUBound))
for task in tasksCPU:
self.resourceControl.insertThreshold(site, taskType = task, maxSlots = CPUBound,
pendingSlots = taskCPUPending)
# Update site IO tasks running and pending slots
logging.debug("Updating %s tasksIO thresholds for pend/runn: %d/%d" % (site, taskIOPending,
IOBound))
for task in tasksIO:
self.resourceControl.insertThreshold(site, taskType = task, maxSlots = IOBound,
pendingSlots = taskIOPending)
if self.tier0Mode:
# Set task thresholds for Tier0
logging.debug("Updating %s Express and Repack task thresholds." % site)
expressSlots = int(CPUBound * self.runningExpressPercent/100)
pendingExpress = int(expressSlots * self.pendingSlotsTaskPercent/100)
self.resourceControl.insertThreshold(site, 'Express', expressSlots, pendingExpress)
repackSlots = int(CPUBound * self.runningRepackPercent/100)
pendingRepack = int(repackSlots * self.pendingSlotsTaskPercent/100)
self.resourceControl.insertThreshold(site, 'Repack', repackSlots, pendingRepack)
def thresholdsByVOName(self, sites):
"""
_thresholdsByVOName_
Creates a dictionary with keys->VOName and values->threshold:
"""
thresholdbyVOName = {}
for site in sites:
voname = site['VOName']
value = site['Value']
if voname not in thresholdbyVOName:
if value is None:
logging.warn('Site %s does not have thresholds in SSB, assuming 0' % voname)
thresholdbyVOName[voname] = 0
else:
thresholdbyVOName[voname] = int(value)
else:
logging.error('I have a duplicated threshold entry in SSB for %s' % voname)
return thresholdbyVOName
def siteStatusByVOName(self, sites):
"""
_siteStatusByVOName_
Creates a dictionary with keys->VOName and values->status:
"""
statusBySite = {}
for site in sites:
voname = site['VOName']
status = site['Status']
if not status:
logging.error('Site %s does not have status in SSB' % voname)
continue
if voname not in statusBySite:
statusAgent = self.getState(str(status))
if not statusAgent:
logging.error("Unkwown status '%s' for site %s, please check SSB" % (status, voname))
continue
statusBySite[voname] = statusAgent
else:
logging.error('I have a duplicated status entry in SSB for %s' % voname)
return statusBySite
def getState(self, stateSSB):
"""
_getState_
Translates SSB states into resource control state
"""
ssb2agent = {'on': 'Normal',
'drain': 'Draining',
'down': 'Down',
'skip': 'Down'}
if stateSSB in ssb2agent:
return ssb2agent[stateSSB]
elif stateSSB == "tier0":
logging.debug('There is a site in tier0 status (Tier0Mode is %s)' % self.tier0Mode )
if self.tier0Mode:
return "Normal"
else:
return "Draining"
else:
return None
def updateSiteState(self, siteName, state):
"""
_updateSiteState_
Update only the site state in the resource control database.
"""
try:
self.resourceControl.changeSiteState(siteName, state)
except Exception as ex:
logging.error("Failed to update %s state to %s:" % (siteName, state))
logging.error(str(ex))
logging.error("Traceback: \n%s" % traceback.format_exc())
return
class ResourceControlUpdater(BaseWorkerThread):
"""
Update site status and thresholds from SSB
"""
def __init__(self, config):
"""
Initialize
"""
BaseWorkerThread.__init__(self)
self.config = config
# get dashboard url, set metric columns from config
self.dashboard = config.AgentStatusWatcher.dashboard
self.siteStatusMetric = config.AgentStatusWatcher.siteStatusMetric
self.cpuBoundMetric = config.AgentStatusWatcher.cpuBoundMetric
self.ioBoundMetric = config.AgentStatusWatcher.ioBoundMetric
# set pending percentages from config
self.pendingSlotsSitePercent = config.AgentStatusWatcher.pendingSlotsSitePercent
self.pendingSlotsTaskPercent = config.AgentStatusWatcher.pendingSlotsTaskPercent
self.runningExpressPercent = config.AgentStatusWatcher.runningExpressPercent
self.runningRepackPercent = config.AgentStatusWatcher.runningRepackPercent
# sites forced to down
self.forceSiteDown = getattr(config.AgentStatusWatcher,
'forceSiteDown', [])
# agent team (for dynamic threshold) and queueParams (drain mode)
self.teamName = config.Agent.teamName
self.agentsNumByTeam = getattr(config.AgentStatusWatcher,
'defaultAgentsNumByTeam', 5)
# only SSB sites
self.onlySSB = config.AgentStatusWatcher.onlySSB
# tier mode
self.tier0Mode = hasattr(config, "Tier0Feeder")
self.t1SitesCores = config.AgentStatusWatcher.t1SitesCores
# switch this component on/off
self.enabled = getattr(config.AgentStatusWatcher, 'enabled', True)
# set resource control
self.resourceControl = ResourceControl(config=self.config)
# wmstats connection
self.centralCouchDBReader = WMStatsReader(
self.config.AgentStatusWatcher.centralWMStatsURL)
def algorithm(self, parameters):
"""
_algorithm_
Update site state and thresholds, based on differences between resource
control database and info available in SSB.
1. Get info from Resource Control database
2. Get info from SSB
3. Get information about teams and number of agents from WMStats
4. Change site state when needed (this triggers a condor clasAd fetch)
5. Change site thresholds when needed (and task thresholds)
Sites from SSB are validated with PhEDEx node names
"""
if not self.enabled:
logging.info(
"This component is not enabled in the configuration. Doing nothing."
)
return
try:
sitesRC = self.resourceControl.listSitesSlots()
logging.debug("Info from resource control: %s", sitesRC)
sitesSSB = self.getInfoFromSSB()
if not sitesSSB:
return
logging.debug("Info from SSB: %s", sitesSSB)
# Check which site states need to be updated in the database
sitesRC = self.checkStatusChanges(sitesRC, sitesSSB)
# get number of agents working in the same team (not in DrainMode)
self.getAgentsByTeam()
# Check which site slots need to be updated in the database
self.checkSlotsChanges(sitesRC, sitesSSB, self.agentsNumByTeam)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s", traceback.format_exc())
logging.info(
"Resource control cycle finished updating site state and thresholds."
)
def getAgentsByTeam(self):
"""
_getAgentsByTeam_
Get the WMStats view about agents and teams
"""
if isDrainMode(self.config):
# maximize pending thresholds to get this agent drained ASAP
self.agentsNumByTeam = 1
return
agentsByTeam = {}
try:
agentsByTeam = self.centralCouchDBReader.agentsByTeam(
filterDrain=True)
except Exception:
logging.error("WMStats is not available or is unresponsive.")
if not agentsByTeam:
logging.warning(
"agentInfo couch view is not available, use default value %s",
self.agentsNumByTeam)
else:
self.agentsNumByTeam = agentsByTeam.get(self.teamName,
self.agentsNumByTeam)
logging.debug(
"Agents connected to the same team (not in DrainMode): %d",
self.agentsNumByTeam)
return
def getInfoFromSSB(self):
"""
_getInfoFromSSB_
Get site status, CPU bound and IO bound from dashboard (SSB).
Returns a dict of dicts where the first key is the site name.
"""
# urls from site status board
urlSiteState = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.siteStatusMetric)
urlCpuBound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.cpuBoundMetric)
urlIoBound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.ioBoundMetric)
# get info from dashboard
sites = urllib2.urlopen(urlSiteState).read()
cpuBound = urllib2.urlopen(urlCpuBound).read()
ioBound = urllib2.urlopen(urlIoBound).read()
# parse from json format to dictionary, get only 'csvdata'
ssbSiteState = json.loads(sites)['csvdata']
ssbCpuSlots = json.loads(cpuBound)['csvdata']
ssbIoSlots = json.loads(ioBound)['csvdata']
# dict updated by these methods with status/thresholds info keyed by the site name
ssbSiteSlots = {}
self.siteStatusByVOName(ssbSiteState, ssbSiteSlots)
self.thresholdsByVOName(ssbCpuSlots,
ssbSiteSlots,
slotsType='slotsCPU')
self.thresholdsByVOName(ssbIoSlots, ssbSiteSlots, slotsType='slotsIO')
# Now remove sites with state only, such that no updates are applied to them
ssbSiteSlots = {
k: v
for k, v in ssbSiteSlots.iteritems() if len(v) == 3
}
if not ssbSiteSlots:
logging.error(
"One or more of the SSB metrics is down. Please contact the Dashboard team."
)
return ssbSiteSlots
return ssbSiteSlots
def checkStatusChanges(self, infoRC, infoSSB):
"""
_checkStatusChanges_
Checks which sites need to have their site state updated in
resource control, based on:
1. settings defined for the component (config.py)
2. site state changes between SSB and RC
Returns the new infoRC dict (where a few key/value pairs were
deleted - no need to update slots information)
"""
# First sets list of forced sites to down (HLT @FNAL is an example)
for site in self.forceSiteDown:
if site in infoRC and infoRC[site]['state'] != 'Down':
logging.info("Forcing site %s to Down", site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# if onlySSB sites, force all the sites not in SSB to down
if self.onlySSB:
for site in set(infoRC).difference(set(infoSSB)):
if infoRC[site]['state'] != 'Down':
logging.info('Only SSBsites, forcing site %s to Down',
site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# this time don't update infoRC since we still want to update slots info
for site in set(infoRC).intersection(set(infoSSB)):
if infoRC[site]['state'] != infoSSB[site]['state']:
logging.info('Changing %s state from %s to %s', site,
infoRC[site]['state'], infoSSB[site]['state'])
self.updateSiteState(site, infoSSB[site]['state'])
return infoRC
def checkSlotsChanges(self, infoRC, infoSSB, agentsCount):
"""
_checkSlotsChanges_
Checks which sites need to have their running and/or pending
slots updated in resource control database, based on:
1. number of agents connected to the same team
2. and slots provided by the Dashboard team (SSB)
If site slots are updated, then also updates its tasks.
"""
tasksCPU = ['Processing', 'Production']
tasksIO = ['Merge', 'Cleanup', 'Harvesting', 'LogCollect', 'Skim']
minCPUSlots, minIOSlots = 50, 25
logging.debug(
"Settings for site and task pending slots: %s%% and %s%%",
self.pendingSlotsSitePercent, self.pendingSlotsTaskPercent)
for site in set(infoRC).intersection(set(infoSSB)):
if self.tier0Mode and 'T1_' in site:
# T1 cores utilization for Tier0
infoSSB[site]['slotsCPU'] = infoSSB[site][
'slotsCPU'] * self.t1SitesCores / 100
infoSSB[site]['slotsIO'] = infoSSB[site][
'slotsIO'] * self.t1SitesCores / 100
# round very small sites to the bare minimum
if infoSSB[site]['slotsCPU'] < minCPUSlots:
infoSSB[site]['slotsCPU'] = minCPUSlots
if infoSSB[site]['slotsIO'] < minIOSlots:
infoSSB[site]['slotsIO'] = minIOSlots
CPUBound = infoSSB[site]['slotsCPU']
IOBound = infoSSB[site]['slotsIO']
sitePending = max(
int(CPUBound / agentsCount * self.pendingSlotsSitePercent /
100), minCPUSlots)
taskCPUPending = max(
int(CPUBound / agentsCount * self.pendingSlotsTaskPercent /
100), minCPUSlots)
taskIOPending = max(
int(IOBound / agentsCount * self.pendingSlotsTaskPercent /
100), minIOSlots)
if infoRC[site]['running_slots'] != CPUBound or infoRC[site][
'pending_slots'] != sitePending:
# Update site running and pending slots
logging.info(
"Updating %s site thresholds for pend/runn: %d/%d", site,
sitePending, CPUBound)
self.resourceControl.setJobSlotsForSite(
site,
pendingJobSlots=sitePending,
runningJobSlots=CPUBound)
# Update site CPU tasks running and pending slots (large running slots)
logging.debug(
"Updating %s tasksCPU thresholds for pend/runn: %d/%d",
site, taskCPUPending, CPUBound)
for task in tasksCPU:
self.resourceControl.insertThreshold(
site,
taskType=task,
maxSlots=CPUBound,
pendingSlots=taskCPUPending)
# Update site IO tasks running and pending slots
logging.debug(
"Updating %s tasksIO thresholds for pend/runn: %d/%d",
site, taskIOPending, IOBound)
for task in tasksIO:
self.resourceControl.insertThreshold(
site,
taskType=task,
maxSlots=IOBound,
pendingSlots=taskIOPending)
if self.tier0Mode:
# Set task thresholds for Tier0
logging.debug(
"Updating %s Express and Repack task thresholds.", site)
expressSlots = int(CPUBound * self.runningExpressPercent / 100)
pendingExpress = int(expressSlots *
self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(site, 'Express',
expressSlots,
pendingExpress)
repackSlots = int(CPUBound * self.runningRepackPercent / 100)
pendingRepack = int(repackSlots *
self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(site, 'Repack',
repackSlots,
pendingRepack)
def thresholdsByVOName(self, sites, ssbSiteSlots, slotsType):
"""
_thresholdsByVOName_
Updates the dict with CPU and IO slots, only for sites with a valid state
"""
for site in sites:
voname = site['VOName']
value = site['Value']
if voname in ssbSiteSlots:
if value is None:
logging.warn(
'Site %s does not have thresholds in SSB. Taking no action',
voname)
# then we better remove this site from our final dict
ssbSiteSlots.pop(voname)
else:
ssbSiteSlots[voname][slotsType] = int(value)
else:
logging.warn(
'Found %s thresholds for site %s which has no state in SSB',
slotsType, voname)
return
def siteStatusByVOName(self, sites, ssbSiteSlots):
"""
_siteStatusByVOName_
Creates an inner dictionary for each site that will contain
the site state and the number of slots
"""
for site in sites:
voname = site['VOName']
status = site['Status']
if voname not in ssbSiteSlots:
statusAgent = self.getState(str(status))
if not statusAgent:
logging.error(
"Unkwown status '%s' for site %s, please check SSB",
status, voname)
else:
ssbSiteSlots[voname] = {'state': statusAgent}
else:
logging.error('I have a duplicated status entry in SSB for %s',
voname)
return
def getState(self, stateSSB):
"""
_getState_
Translates SSB states into resource control state
"""
ssb2agent = {
'enabled': 'Normal',
'drain': 'Draining',
'disabled': 'Down',
'test': 'Draining'
}
# 'test' state behaviour varies between production and tier0 agents
ssb2agent['test'] = 'Normal' if self.tier0Mode else "Draining"
return ssb2agent.get(stateSSB)
def updateSiteState(self, siteName, state):
"""
_updateSiteState_
Update only the site state in the resource control database.
"""
try:
self.resourceControl.changeSiteState(siteName, state)
except Exception as ex:
logging.error("Failed to update %s state to %s:", siteName, state)
logging.error(str(ex))
logging.error("Traceback: \n%s", traceback.format_exc())
return
def testList(self):
"""
_testList_
Test the functions that list thresholds for creating jobs and submitting
jobs.
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 10, 20, "testSE1", "testCE1", "T1_US_FNAL", "LsfPlugin")
myResourceControl.insertSite("testSite2", 20, 40, "testSE2", "testCE2")
myResourceControl.insertThreshold("testSite1", "Processing", 20, 10)
myResourceControl.insertThreshold("testSite1", "Merge", 200, 100)
myResourceControl.insertThreshold("testSite2", "Processing", 50, 25)
myResourceControl.insertThreshold("testSite2", "Merge", 135, 65)
testWorkflow = Workflow(spec = makeUUID(), owner = "Steve",
name = makeUUID(), task = "Test")
testWorkflow.create()
testFilesetA = Fileset(name = "TestFilesetA")
testFilesetA.create()
testFilesetB = Fileset(name = "TestFilesetB")
testFilesetB.create()
testFilesetC = Fileset(name = "TestFilesetC")
testFilesetC.create()
testFileA = File(lfn = "testFileA", locations = set(["testSE1", "testSE2"]))
testFileA.create()
testFilesetA.addFile(testFileA)
testFilesetA.commit()
testFilesetB.addFile(testFileA)
testFilesetB.commit()
testFilesetC.addFile(testFileA)
testFilesetC.commit()
testSubscriptionA = Subscription(fileset = testFilesetA,
workflow = testWorkflow,
type = "Processing")
testSubscriptionA.create()
testSubscriptionA.addWhiteBlackList([{"site_name": "testSite1", "valid": True}])
testSubscriptionB = Subscription(fileset = testFilesetB,
workflow = testWorkflow,
type = "Processing")
testSubscriptionB.create()
testSubscriptionB.addWhiteBlackList([{"site_name": "testSite1", "valid": False}])
testSubscriptionC = Subscription(fileset = testFilesetC,
workflow = testWorkflow,
type = "Merge")
testSubscriptionC.create()
testJobGroupA = JobGroup(subscription = testSubscriptionA)
testJobGroupA.create()
testJobGroupB = JobGroup(subscription = testSubscriptionB)
testJobGroupB.create()
testJobGroupC = JobGroup(subscription = testSubscriptionC)
testJobGroupC.create()
# Site1, Has been assigned a location and is complete.
testJobA = Job(name = "testJobA", files = [testFileA])
testJobA["couch_record"] = makeUUID()
testJobA.create(group = testJobGroupA)
testJobA["state"] = "success"
# Site 1, Has been assigned a location and is incomplete.
testJobB = Job(name = "testJobB", files = [testFileA])
testJobB["couch_record"] = makeUUID()
testJobB.create(group = testJobGroupA)
testJobB["state"] = "executing"
runJobB = RunJob()
runJobB.buildFromJob(testJobB)
runJobB["status"] = "PEND"
# Does not have a location, white listed to site 1
testJobC = Job(name = "testJobC", files = [testFileA])
testJobC["couch_record"] = makeUUID()
testJobC.create(group = testJobGroupA)
testJobC["state"] = "new"
# Site 2, Has been assigned a location and is complete.
testJobD = Job(name = "testJobD", files = [testFileA])
testJobD["couch_record"] = makeUUID()
testJobD.create(group = testJobGroupB)
testJobD["state"] = "success"
# Site 2, Has been assigned a location and is incomplete.
testJobE = Job(name = "testJobE", files = [testFileA])
testJobE["couch_record"] = makeUUID()
testJobE.create(group = testJobGroupB)
testJobE["state"] = "executing"
runJobE = RunJob()
runJobE.buildFromJob(testJobE)
runJobE["status"] = "RUN"
# Does not have a location, site 1 is blacklisted.
testJobF = Job(name = "testJobF", files = [testFileA])
testJobF["couch_record"] = makeUUID()
testJobF.create(group = testJobGroupB)
testJobF["state"] = "new"
# Site 3, Has been assigned a location and is complete.
testJobG = Job(name = "testJobG", files = [testFileA])
testJobG["couch_record"] = makeUUID()
testJobG.create(group = testJobGroupC)
testJobG["state"] = "cleanout"
# Site 3, Has been assigned a location and is incomplete.
testJobH = Job(name = "testJobH", files = [testFileA])
testJobH["couch_record"] = makeUUID()
testJobH.create(group = testJobGroupC)
testJobH["state"] = "new"
# Site 3, Does not have a location.
testJobI = Job(name = "testJobI", files = [testFileA])
testJobI["couch_record"] = makeUUID()
testJobI.create(group = testJobGroupC)
testJobI["state"] = "new"
# Site 3, Does not have a location and is in cleanout.
testJobJ = Job(name = "testJobJ", files = [testFileA])
testJobJ["couch_record"] = makeUUID()
testJobJ.create(group = testJobGroupC)
testJobJ["state"] = "cleanout"
changeStateAction = self.daoFactory(classname = "Jobs.ChangeState")
changeStateAction.execute(jobs = [testJobA, testJobB, testJobC, testJobD,
testJobE, testJobF, testJobG, testJobH,
testJobI, testJobJ])
self.insertRunJob.execute([runJobB, runJobE])
setLocationAction = self.daoFactory(classname = "Jobs.SetLocation")
setLocationAction.execute(testJobA["id"], "testSite1")
setLocationAction.execute(testJobB["id"], "testSite1")
setLocationAction.execute(testJobD["id"], "testSite1")
setLocationAction.execute(testJobE["id"], "testSite1")
setLocationAction.execute(testJobG["id"], "testSite1")
setLocationAction.execute(testJobH["id"], "testSite1")
createThresholds = myResourceControl.listThresholdsForCreate()
submitThresholds = myResourceControl.listThresholdsForSubmit()
self.assertEqual( len(createThresholds.keys()), 2,
"Error: Wrong number of sites in create thresholds" )
self.assertEqual( createThresholds["testSite1"]["total_slots"], 10,
"Error: Wrong number of slots for site 1" )
self.assertEqual( createThresholds["testSite2"]["total_slots"], 20,
"Error: Wrong number of slots for site 2" )
# We should have two running jobs with locations at site one,
# two running jobs without locations at site two, and one running
# job without a location at site one and two.
self.assertEqual( createThresholds["testSite1"]["pending_jobs"], 4,
"Error: Wrong number of pending jobs for site 1" )
# We should have one running job with a location at site 2 and
# another running job without a location.
self.assertEqual( createThresholds["testSite2"]["pending_jobs"], 2,
"Error: Wrong number of pending jobs for site 2" )
# We should also have a phedex_name
self.assertEqual(createThresholds["testSite1"]["cms_name"], "T1_US_FNAL")
self.assertEqual(createThresholds["testSite2"]["cms_name"], None)
mergeThreshold1 = None
mergeThreshold2 = None
procThreshold1 = None
procThreshold2 = None
self.assertEqual(submitThresholds["testSite1"]['cms_name'], 'T1_US_FNAL')
for threshold in submitThresholds["testSite1"]["thresholds"]:
if threshold['task_type'] == "Merge":
mergeThreshold1 = threshold
elif threshold['task_type'] == "Processing":
procThreshold1 = threshold
self.assertEqual(submitThresholds["testSite2"]['cms_name'], None)
for threshold in submitThresholds["testSite2"]["thresholds"]:
if threshold['task_type'] == "Merge":
mergeThreshold2 = threshold
elif threshold['task_type'] == "Processing":
procThreshold2 = threshold
self.assertEqual(submitThresholds["testSite1"]["total_running_jobs"], 1,
"Error: Wrong number of running jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite2"]["total_running_jobs"], 0,
"Error: Wrong number of running jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite1"]["total_pending_jobs"], 1,
"Error: Wrong number of pending jobs for submit thresholds.")
self.assertEqual(submitThresholds["testSite2"]["total_pending_jobs"], 0,
"Error: Wrong number of pending jobs for submit thresholds.")
self.assertEqual(mergeThreshold1["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold1["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold1["task_running_jobs"], 1,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold1["task_pending_jobs"], 1,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold2["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(mergeThreshold2["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold2["task_running_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
self.assertEqual(procThreshold2["task_pending_jobs"], 0,
"Error: Wrong number of task running jobs for submit thresholds.")
return
def testThresholdPriority(self):
"""
_testThresholdPriority_
Test that we get things back in priority order
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, 40, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3)
myResourceControl.changeTaskPriority("Merge", 3)
myResourceControl.changeTaskPriority("Processing", 1)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds'][0]['task_type'], 'Merge')
self.assertEqual(result['testSite1']['thresholds'][1]['task_type'], 'Processing')
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3)
myResourceControl.changeTaskPriority("Merge", 1)
myResourceControl.changeTaskPriority("Processing", 3)
# Should now be in reverse order
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds'][1]['task_type'], 'Merge')
self.assertEqual(result['testSite1']['thresholds'][0]['task_type'], 'Processing')
myResourceControl.insertSite("testSite2", 20, 40, "testSE2", "testCE2")
myResourceControl.insertThreshold("testSite2", "Processing", 10, 8)
myResourceControl.insertThreshold("testSite2", "Merge", 5, 3)
# Should be in the same order for site 1 and 2
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite2']['thresholds'][0]['task_type'], result['testSite1']['thresholds'][0]['task_type'])
self.assertEqual(result['testSite2']['thresholds'][1]['task_type'], result['testSite1']['thresholds'][1]['task_type'])
myResourceControl.changeTaskPriority("Merge", 4)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite2']['thresholds'][0]['priority'], 4)
return
def setupForKillTest(self, baAPI = None):
"""
_setupForKillTest_
Inject a workflow into WMBS that has a processing task, a merge task and
a cleanup task. Inject files into the various tasks at various
processing states (acquired, complete, available...). Also create jobs
for each subscription in various states.
"""
myThread = threading.currentThread()
daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
locationAction = daoFactory(classname = "Locations.New")
changeStateAction = daoFactory(classname = "Jobs.ChangeState")
resourceControl = ResourceControl()
resourceControl.insertSite(siteName = 'site1', seName = 'goodse.cern.ch',
ceName = 'site1', plugin = "TestPlugin")
resourceControl.insertThreshold(siteName = 'site1', taskType = 'Processing', \
maxSlots = 10000)
inputFileset = Fileset("input")
inputFileset.create()
inputFileA = File("lfnA", locations = "goodse.cern.ch")
inputFileB = File("lfnB", locations = "goodse.cern.ch")
inputFileC = File("lfnC", locations = "goodse.cern.ch")
inputFileA.create()
inputFileB.create()
inputFileC.create()
inputFileset.addFile(inputFileA)
inputFileset.addFile(inputFileB)
inputFileset.addFile(inputFileC)
inputFileset.commit()
unmergedOutputFileset = Fileset("unmerged")
unmergedOutputFileset.create()
unmergedFileA = File("ulfnA", locations = "goodse.cern.ch")
unmergedFileB = File("ulfnB", locations = "goodse.cern.ch")
unmergedFileC = File("ulfnC", locations = "goodse.cern.ch")
unmergedFileA.create()
unmergedFileB.create()
unmergedFileC.create()
unmergedOutputFileset.addFile(unmergedFileA)
unmergedOutputFileset.addFile(unmergedFileB)
unmergedOutputFileset.addFile(unmergedFileC)
unmergedOutputFileset.commit()
mainProcWorkflow = Workflow(spec = "spec1", owner = "Steve",
name = "Main", task = "Proc")
mainProcWorkflow.create()
mainProcMergeWorkflow = Workflow(spec = "spec1", owner = "Steve",
name = "Main", task = "ProcMerge")
mainProcMergeWorkflow.create()
mainCleanupWorkflow = Workflow(spec = "spec1", owner = "Steve",
name = "Main", task = "Cleanup")
mainCleanupWorkflow.create()
self.mainProcSub = Subscription(fileset = inputFileset,
workflow = mainProcWorkflow,
type = "Processing")
self.mainProcSub.create()
self.mainProcSub.acquireFiles(inputFileA)
self.mainProcSub.completeFiles(inputFileB)
procJobGroup = JobGroup(subscription = self.mainProcSub)
procJobGroup.create()
self.procJobA = Job(name = "ProcJobA")
self.procJobA["state"] = "new"
self.procJobA["location"] = "site1"
self.procJobB = Job(name = "ProcJobB")
self.procJobB["state"] = "executing"
self.procJobB["location"] = "site1"
self.procJobC = Job(name = "ProcJobC")
self.procJobC["state"] = "complete"
self.procJobC["location"] = "site1"
self.procJobA.create(procJobGroup)
self.procJobB.create(procJobGroup)
self.procJobC.create(procJobGroup)
self.mainMergeSub = Subscription(fileset = unmergedOutputFileset,
workflow = mainProcMergeWorkflow,
type = "Merge")
self.mainMergeSub.create()
self.mainMergeSub.acquireFiles(unmergedFileA)
self.mainMergeSub.failFiles(unmergedFileB)
mergeJobGroup = JobGroup(subscription = self.mainMergeSub)
mergeJobGroup.create()
self.mergeJobA = Job(name = "MergeJobA")
self.mergeJobA["state"] = "exhausted"
self.mergeJobA["location"] = "site1"
self.mergeJobB = Job(name = "MergeJobB")
self.mergeJobB["state"] = "cleanout"
self.mergeJobB["location"] = "site1"
self.mergeJobC = Job(name = "MergeJobC")
self.mergeJobC["state"] = "new"
self.mergeJobC["location"] = "site1"
self.mergeJobA.create(mergeJobGroup)
self.mergeJobB.create(mergeJobGroup)
self.mergeJobC.create(mergeJobGroup)
self.mainCleanupSub = Subscription(fileset = unmergedOutputFileset,
workflow = mainCleanupWorkflow,
type = "Cleanup")
self.mainCleanupSub.create()
self.mainCleanupSub.acquireFiles(unmergedFileA)
self.mainCleanupSub.completeFiles(unmergedFileB)
cleanupJobGroup = JobGroup(subscription = self.mainCleanupSub)
cleanupJobGroup.create()
self.cleanupJobA = Job(name = "CleanupJobA")
self.cleanupJobA["state"] = "new"
self.cleanupJobA["location"] = "site1"
self.cleanupJobB = Job(name = "CleanupJobB")
self.cleanupJobB["state"] = "executing"
self.cleanupJobB["location"] = "site1"
self.cleanupJobC = Job(name = "CleanupJobC")
self.cleanupJobC["state"] = "complete"
self.cleanupJobC["location"] = "site1"
self.cleanupJobA.create(cleanupJobGroup)
self.cleanupJobB.create(cleanupJobGroup)
self.cleanupJobC.create(cleanupJobGroup)
jobList = [self.procJobA, self.procJobB, self.procJobC,
self.mergeJobA, self.mergeJobB, self.mergeJobC,
self.cleanupJobA, self.cleanupJobB, self.cleanupJobC]
changeStateAction.execute(jobList)
if baAPI:
for job in jobList:
job['plugin'] = 'TestPlugin'
job['userdn'] = 'Steve'
job['custom']['location'] = 'site1'
baAPI.createNewJobs(wmbsJobs = jobList)
# We'll create an unrelated workflow to verify that it isn't affected
# by the killing code.
bogusFileset = Fileset("dontkillme")
bogusFileset.create()
bogusFileA = File("bogus/lfnA", locations = "goodse.cern.ch")
bogusFileA.create()
bogusFileset.addFile(bogusFileA)
bogusFileset.commit()
bogusWorkflow = Workflow(spec = "spec2", owner = "Steve",
name = "Bogus", task = "Proc")
bogusWorkflow.create()
self.bogusSub = Subscription(fileset = bogusFileset,
workflow = bogusWorkflow,
type = "Processing")
self.bogusSub.create()
self.bogusSub.acquireFiles(bogusFileA)
return
def testThresholdPriority(self):
"""
_testThresholdPriority_
Test that we get things back in priority order
"""
myResourceControl = ResourceControl()
myResourceControl.insertSite("testSite1", 20, 40, "testSE1", "testCE1")
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8, priority = 1)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3, priority = 2)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds'][0]['task_type'], 'Merge')
self.assertEqual(result['testSite1']['thresholds'][1]['task_type'], 'Processing')
myResourceControl.insertThreshold("testSite1", "Processing", 10, 8, priority = 2)
myResourceControl.insertThreshold("testSite1", "Merge", 5, 3, priority = 1)
# Should now be in reverse order
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite1']['thresholds'][1]['task_type'], 'Merge')
self.assertEqual(result['testSite1']['thresholds'][0]['task_type'], 'Processing')
myResourceControl.insertSite("testSite2", 20, 40, "testSE2", "testCE2")
myResourceControl.insertThreshold("testSite2", "Processing", 10, 8, priority = 1)
myResourceControl.insertThreshold("testSite2", "Merge", 5, 3, priority = 2)
# Should be in proper order for site 2
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite2']['thresholds'][0]['task_type'], 'Merge')
self.assertEqual(result['testSite2']['thresholds'][1]['task_type'], 'Processing')
# Should now be in reverse order for site 1
self.assertEqual(result['testSite1']['thresholds'][1]['task_type'], 'Merge')
self.assertEqual(result['testSite1']['thresholds'][0]['task_type'], 'Processing')
myResourceControl.insertThreshold("testSite2", "Merge", 20, 10)
result = myResourceControl.listThresholdsForSubmit()
self.assertEqual(result['testSite2']['thresholds'][0]['priority'], 2)
return
def setUp(self):
"""
setup for test.
"""
myThread = threading.currentThread()
self.testInit = TestInit(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
#self.tearDown()
self.testInit.setSchema(customModules = ["WMCore.WMBS", "WMCore.BossAir", "WMCore.ResourceControl", "WMCore.Agent.Database"],
useDefault = False)
self.testInit.setupCouch("bossair_t/jobs", "JobDump")
self.testInit.setupCouch("bossair_t/fwjrs", "FWJRDump")
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
locationAction = self.daoFactory(classname = "Locations.New")
locationSlots = self.daoFactory(classname = "Locations.SetJobSlots")
#Create sites in resourceControl
resourceControl = ResourceControl()
for site in self.sites:
resourceControl.insertSite(siteName = site, seName = 'se.%s' % (site),
ceName = site, plugin = "CondorPlugin", jobSlots = 1000)
resourceControl.insertThreshold(siteName = site, taskType = 'Processing', \
maxSlots = 1000)
resourceControl.insertSite(siteName = 'Xanadu', seName = 'se.Xanadu',
ceName = 'Xanadu', plugin = "TestPlugin")
resourceControl.insertThreshold(siteName = 'Xanadu', taskType = 'Processing', \
maxSlots = 10000)
resourceControl.insertSite(siteName = 'jade-cms.hip.fi', seName = 'madhatter.csc.fi',
ceName = 'jade-cms.hip.fi', plugin = "ARCPlugin")
resourceControl.insertThreshold(siteName = 'jade-cms.hip.fi', taskType = 'Processing', \
maxSlots = 100)
# using this for glite submissions
resourceControl.insertSite(siteName = 'grid-ce-01.ba.infn.it', seName = 'storm-se-01.ba.infn.it',
ceName = 'grid-ce-01.ba.infn.it', plugin = 'gLitePlugin')
resourceControl.insertThreshold(siteName = 'grid-ce-01.ba.infn.it', taskType = 'Processing', \
maxSlots = 50)
# Create user
newuser = self.daoFactory(classname = "Users.New")
newuser.execute(dn = "mnorman", group_name = "phgroup", role_name = "cmsrole")
# We actually need the user name
self.user = getpass.getuser()
self.testDir = self.testInit.generateWorkDir()
# Set heartbeat
componentName = 'test'
self.heartbeatAPI = HeartbeatAPI(componentName)
self.heartbeatAPI.registerComponent()
componentName = 'JobTracker'
self.heartbeatAPI2 = HeartbeatAPI(componentName)
self.heartbeatAPI2.registerComponent()
return
class ResourceControlUpdater(BaseWorkerThread):
"""
Update site status and thresholds from SSB
"""
def __init__(self, config):
"""
Initialize
"""
BaseWorkerThread.__init__(self)
self.config = config
# get dashboard url, set metric columns from config
self.dashboard = config.AgentStatusWatcher.dashboard
self.siteStatusMetric = config.AgentStatusWatcher.siteStatusMetric
self.cpuBoundMetric = config.AgentStatusWatcher.cpuBoundMetric
self.ioBoundMetric = config.AgentStatusWatcher.ioBoundMetric
# set pending percentages from config
self.pendingSlotsSitePercent = config.AgentStatusWatcher.pendingSlotsSitePercent
self.pendingSlotsTaskPercent = config.AgentStatusWatcher.pendingSlotsTaskPercent
self.runningExpressPercent = config.AgentStatusWatcher.runningExpressPercent
self.runningRepackPercent = config.AgentStatusWatcher.runningRepackPercent
# sites forced to down
self.forceSiteDown = getattr(config.AgentStatusWatcher, 'forceSiteDown', [])
# agent team (for dynamic threshold) and queueParams (drain mode)
self.teamName = config.Agent.teamName
self.agentsNumByTeam = getattr(config.AgentStatusWatcher, 'defaultAgentsNumByTeam', 5)
# only SSB sites
self.onlySSB = config.AgentStatusWatcher.onlySSB
# tier mode
self.tier0Mode = hasattr(config, "Tier0Feeder")
self.t1SitesCores = config.AgentStatusWatcher.t1SitesCores
# switch this component on/off
self.enabled = getattr(config.AgentStatusWatcher, 'enabled', True)
# set resource control
self.resourceControl = ResourceControl(config=self.config)
# wmstats connection
self.centralCouchDBReader = WMStatsReader(self.config.AgentStatusWatcher.centralWMStatsURL)
def algorithm(self, parameters):
"""
_algorithm_
Update site state and thresholds, based on differences between resource
control database and info available in SSB.
1. Get info from Resource Control database
2. Get info from SSB
3. Get information about teams and number of agents from WMStats
4. Change site state when needed (this triggers a condor clasAd fetch)
5. Change site thresholds when needed (and task thresholds)
Sites from SSB are validated with PhEDEx node names
"""
if not self.enabled:
logging.info("This component is not enabled in the configuration. Doing nothing.")
return
try:
sitesRC = self.resourceControl.listSitesSlots()
logging.debug("Info from resource control: %s", sitesRC)
sitesSSB = self.getInfoFromSSB()
if not sitesSSB:
return
logging.debug("Info from SSB: %s", sitesSSB)
# Check which site states need to be updated in the database
sitesRC = self.checkStatusChanges(sitesRC, sitesSSB)
# get number of agents working in the same team (not in DrainMode)
self.getAgentsByTeam()
# Check which site slots need to be updated in the database
self.checkSlotsChanges(sitesRC, sitesSSB, self.agentsNumByTeam)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s", traceback.format_exc())
logging.info("Resource control cycle finished updating site state and thresholds.")
def getAgentsByTeam(self):
"""
_getAgentsByTeam_
Get the WMStats view about agents and teams
"""
if isDrainMode(self.config):
# maximize pending thresholds to get this agent drained ASAP
self.agentsNumByTeam = 1
return
agentsByTeam = {}
try:
agentsByTeam = self.centralCouchDBReader.agentsByTeam(filterDrain=True)
except Exception:
logging.error("WMStats is not available or is unresponsive.")
if not agentsByTeam:
logging.warning("agentInfo couch view is not available, use default value %s", self.agentsNumByTeam)
else:
self.agentsNumByTeam = agentsByTeam.get(self.teamName, self.agentsNumByTeam)
logging.debug("Agents connected to the same team (not in DrainMode): %d", self.agentsNumByTeam)
return
def getInfoFromSSB(self):
"""
_getInfoFromSSB_
Get site status, CPU bound and IO bound from dashboard (SSB).
Returns a dict of dicts where the first key is the site name.
"""
# urls from site status board
urlSiteState = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.siteStatusMetric)
urlCpuBound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.cpuBoundMetric)
urlIoBound = self.dashboard + '/request.py/getplotdata?columnid=%s&batch=1&lastdata=1' % str(
self.ioBoundMetric)
# get info from dashboard
sites = urllib2.urlopen(urlSiteState).read()
cpuBound = urllib2.urlopen(urlCpuBound).read()
ioBound = urllib2.urlopen(urlIoBound).read()
# parse from json format to dictionary, get only 'csvdata'
ssbSiteState = json.loads(sites)['csvdata']
ssbCpuSlots = json.loads(cpuBound)['csvdata']
ssbIoSlots = json.loads(ioBound)['csvdata']
# dict updated by these methods with status/thresholds info keyed by the site name
ssbSiteSlots = {}
self.siteStatusByVOName(ssbSiteState, ssbSiteSlots)
self.thresholdsByVOName(ssbCpuSlots, ssbSiteSlots, slotsType='slotsCPU')
self.thresholdsByVOName(ssbIoSlots, ssbSiteSlots, slotsType='slotsIO')
# Now remove sites with state only, such that no updates are applied to them
ssbSiteSlots = {k: v for k, v in ssbSiteSlots.iteritems() if len(v) == 3}
if not ssbSiteSlots:
logging.error("One or more of the SSB metrics is down. Please contact the Dashboard team.")
return ssbSiteSlots
return ssbSiteSlots
def checkStatusChanges(self, infoRC, infoSSB):
"""
_checkStatusChanges_
Checks which sites need to have their site state updated in
resource control, based on:
1. settings defined for the component (config.py)
2. site state changes between SSB and RC
Returns the new infoRC dict (where a few key/value pairs were
deleted - no need to update slots information)
"""
# First sets list of forced sites to down (HLT @FNAL is an example)
for site in self.forceSiteDown:
if site in infoRC and infoRC[site]['state'] != 'Down':
logging.info("Forcing site %s to Down", site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# if onlySSB sites, force all the sites not in SSB to down
if self.onlySSB:
for site in set(infoRC).difference(set(infoSSB)):
if infoRC[site]['state'] != 'Down':
logging.info('Only SSBsites, forcing site %s to Down', site)
self.updateSiteState(site, 'Down')
infoRC.pop(site, None)
# this time don't update infoRC since we still want to update slots info
for site in set(infoRC).intersection(set(infoSSB)):
if infoRC[site]['state'] != infoSSB[site]['state']:
logging.info('Changing %s state from %s to %s', site, infoRC[site]['state'], infoSSB[site]['state'])
self.updateSiteState(site, infoSSB[site]['state'])
return infoRC
def checkSlotsChanges(self, infoRC, infoSSB, agentsCount):
"""
_checkSlotsChanges_
Checks which sites need to have their running and/or pending
slots updated in resource control database, based on:
1. number of agents connected to the same team
2. and slots provided by the Dashboard team (SSB)
If site slots are updated, then also updates its tasks.
"""
tasksCPU = ['Processing', 'Production']
tasksIO = ['Merge', 'Cleanup', 'Harvesting', 'LogCollect', 'Skim']
minCPUSlots, minIOSlots = 50, 25
logging.debug("Settings for site and task pending slots: %s%% and %s%%", self.pendingSlotsSitePercent,
self.pendingSlotsTaskPercent)
for site in set(infoRC).intersection(set(infoSSB)):
if self.tier0Mode and 'T1_' in site:
# T1 cores utilization for Tier0
infoSSB[site]['slotsCPU'] = infoSSB[site]['slotsCPU'] * self.t1SitesCores / 100
infoSSB[site]['slotsIO'] = infoSSB[site]['slotsIO'] * self.t1SitesCores / 100
# round very small sites to the bare minimum
if infoSSB[site]['slotsCPU'] < minCPUSlots:
infoSSB[site]['slotsCPU'] = minCPUSlots
if infoSSB[site]['slotsIO'] < minIOSlots:
infoSSB[site]['slotsIO'] = minIOSlots
CPUBound = infoSSB[site]['slotsCPU']
IOBound = infoSSB[site]['slotsIO']
sitePending = max(int(CPUBound / agentsCount * self.pendingSlotsSitePercent / 100), minCPUSlots)
taskCPUPending = max(int(CPUBound / agentsCount * self.pendingSlotsTaskPercent / 100), minCPUSlots)
taskIOPending = max(int(IOBound / agentsCount * self.pendingSlotsTaskPercent / 100), minIOSlots)
if infoRC[site]['running_slots'] != CPUBound or infoRC[site]['pending_slots'] != sitePending:
# Update site running and pending slots
logging.info("Updating %s site thresholds for pend/runn: %d/%d", site, sitePending, CPUBound)
self.resourceControl.setJobSlotsForSite(site, pendingJobSlots=sitePending,
runningJobSlots=CPUBound)
# Update site CPU tasks running and pending slots (large running slots)
logging.debug("Updating %s tasksCPU thresholds for pend/runn: %d/%d", site, taskCPUPending,
CPUBound)
for task in tasksCPU:
self.resourceControl.insertThreshold(site, taskType=task, maxSlots=CPUBound,
pendingSlots=taskCPUPending)
# Update site IO tasks running and pending slots
logging.debug("Updating %s tasksIO thresholds for pend/runn: %d/%d", site, taskIOPending,
IOBound)
for task in tasksIO:
self.resourceControl.insertThreshold(site, taskType=task, maxSlots=IOBound,
pendingSlots=taskIOPending)
if self.tier0Mode:
# Set task thresholds for Tier0
logging.debug("Updating %s Express and Repack task thresholds.", site)
expressSlots = int(CPUBound * self.runningExpressPercent / 100)
pendingExpress = int(expressSlots * self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(site, 'Express', expressSlots, pendingExpress)
repackSlots = int(CPUBound * self.runningRepackPercent / 100)
pendingRepack = int(repackSlots * self.pendingSlotsTaskPercent / 100)
self.resourceControl.insertThreshold(site, 'Repack', repackSlots, pendingRepack)
def thresholdsByVOName(self, sites, ssbSiteSlots, slotsType):
"""
_thresholdsByVOName_
Updates the dict with CPU and IO slots, only for sites with a valid state
"""
for site in sites:
voname = site['VOName']
value = site['Value']
if voname in ssbSiteSlots:
if value is None:
logging.warn('Site %s does not have thresholds in SSB. Taking no action', voname)
# then we better remove this site from our final dict
ssbSiteSlots.pop(voname)
else:
ssbSiteSlots[voname][slotsType] = int(value)
else:
logging.warn('Found %s thresholds for site %s which has no state in SSB', slotsType, voname)
return
def siteStatusByVOName(self, sites, ssbSiteSlots):
"""
_siteStatusByVOName_
Creates an inner dictionary for each site that will contain
the site state and the number of slots
"""
for site in sites:
voname = site['VOName']
status = site['Status']
if voname not in ssbSiteSlots:
statusAgent = self.getState(str(status))
if not statusAgent:
logging.error("Unkwown status '%s' for site %s, please check SSB", status, voname)
else:
ssbSiteSlots[voname] = {'state': statusAgent}
else:
logging.error('I have a duplicated status entry in SSB for %s', voname)
return
def getState(self, stateSSB):
"""
_getState_
Translates SSB states into resource control state
"""
ssb2agent = {'enabled': 'Normal',
'drain': 'Draining',
'disabled': 'Down',
'test': 'Draining'}
# 'test' state behaviour varies between production and tier0 agents
ssb2agent['test'] = 'Normal' if self.tier0Mode else "Draining"
return ssb2agent.get(stateSSB)
def updateSiteState(self, siteName, state):
"""
_updateSiteState_
Update only the site state in the resource control database.
"""
try:
self.resourceControl.changeSiteState(siteName, state)
except Exception as ex:
logging.error("Failed to update %s state to %s:", siteName, state)
logging.error(str(ex))
logging.error("Traceback: \n%s", traceback.format_exc())
return
