Python PilotAgentsDB.selectPilots方法代码示例

# 需要导入模块: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB import PilotAgentsDB [as 别名]
# 或者: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB.PilotAgentsDB import selectPilots [as 别名]
class PilotStatusAgent(AgentModule):
  """
      The specific agents must provide the following methods:
        - initialize() for initial settings
        - beginExecution()
        - execute() - the main method called in the agent cycle
        - endExecution()
        - finalize() - the graceful exit of the method, this one is usually used
                   for the agent restart
  """

  queryStateList = ['Ready', 'Submitted', 'Running', 'Waiting', 'Scheduled']
  finalStateList = ['Done', 'Aborted', 'Cleared', 'Deleted', 'Failed']

  def __init__(self, *args, **kwargs):
    """ c'tor
    """
    AgentModule.__init__(self, *args, **kwargs)

    self.jobDB = None
    self.pilotDB = None
    self.diracadmin = None

  #############################################################################
  def initialize(self):
    """Sets defaults
    """

    self.am_setOption('PollingTime', 120)
    self.am_setOption('GridEnv', '')
    self.am_setOption('PilotStalledDays', 3)
    self.pilotDB = PilotAgentsDB()
    self.diracadmin = DiracAdmin()
    self.jobDB = JobDB()
    self.clearPilotsDelay = self.am_getOption('ClearPilotsDelay', 30)
    self.clearAbortedDelay = self.am_getOption('ClearAbortedPilotsDelay', 7)
    self.WMSAdministrator = WMSAdministratorClient()

    return S_OK()

  #############################################################################
  def execute(self):
    """The PilotAgent execution method.
    """

    self.pilotStalledDays = self.am_getOption('PilotStalledDays', 3)
    self.gridEnv = self.am_getOption('GridEnv')
    if not self.gridEnv:
      # No specific option found, try a general one
      setup = gConfig.getValue('/DIRAC/Setup', '')
      if setup:
        instance = gConfig.getValue('/DIRAC/Setups/%s/WorkloadManagement' % setup, '')
        if instance:
          self.gridEnv = gConfig.getValue('/Systems/WorkloadManagement/%s/GridEnv' % instance, '')
    result = self.pilotDB._getConnection()
    if result['OK']:
      connection = result['Value']
    else:
      return result

    # Now handle pilots not updated in the last N days (most likely the Broker is no
    # longer available) and declare them Deleted.
    result = self.handleOldPilots(connection)

    connection.close()

    result = self.WMSAdministrator.clearPilots(self.clearPilotsDelay, self.clearAbortedDelay)
    if not result['OK']:
      self.log.warn('Failed to clear old pilots in the PilotAgentsDB')

    return S_OK()

  def clearWaitingPilots(self, condDict):
    """ Clear pilots in the faulty Waiting state
    """

    last_update = Time.dateTime() - MAX_WAITING_STATE_LENGTH * Time.hour
    clearDict = {'Status': 'Waiting',
                 'OwnerDN': condDict['OwnerDN'],
                 'OwnerGroup': condDict['OwnerGroup'],
                 'GridType': condDict['GridType'],
                 'Broker': condDict['Broker']}
    result = self.pilotDB.selectPilots(clearDict, older=last_update)
    if not result['OK']:
      self.log.warn('Failed to get the Pilot Agents for Waiting state')
      return result
    if not result['Value']:
      return S_OK()
    refList = result['Value']

    for pilotRef in refList:
      self.log.info('Setting Waiting pilot to Stalled: %s' % pilotRef)
      result = self.pilotDB.setPilotStatus(pilotRef, 'Stalled', statusReason='Exceeded max waiting time')

    return S_OK()

  def clearParentJob(self, pRef, pDict, connection):
    """ Clear the parameteric parent job from the PilotAgentsDB
    """

#.........这里部分代码省略.........

# 需要导入模块: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB import PilotAgentsDB [as 别名]
# 或者: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB.PilotAgentsDB import selectPilots [as 别名]
class PilotStatusAgent(AgentModule):
    """
      The specific agents must provide the following methods:
      - initialize() for initial settings
      - beginExecution()
      - execute() - the main method called in the agent cycle
      - endExecution()
      - finalize() - the graceful exit of the method, this one is usually used
                 for the agent restart
  """

    queryStateList = ["Ready", "Submitted", "Running", "Waiting", "Scheduled"]
    finalStateList = ["Done", "Aborted", "Cleared", "Deleted", "Failed"]
    identityFieldsList = ["OwnerDN", "OwnerGroup", "GridType", "Broker"]
    eligibleGridTypes = ["gLite"]

    #############################################################################
    def initialize(self):
        """Sets defaults
    """

        self.am_setOption("PollingTime", 120)
        self.am_setOption("GridEnv", "")
        self.am_setOption("PilotStalledDays", 3)
        self.pilotDB = PilotAgentsDB()
        return S_OK()

    #############################################################################
    def execute(self):
        """The PilotAgent execution method.
    """

        self.pilotStalledDays = self.am_getOption("PilotStalledDays", 3)
        self.gridEnv = self.am_getOption("GridEnv")
        if not self.gridEnv:
            # No specific option found, try a general one
            setup = gConfig.getValue("/DIRAC/Setup", "")
            if setup:
                instance = gConfig.getValue("/DIRAC/Setups/%s/WorkloadManagement" % setup, "")
                if instance:
                    self.gridEnv = gConfig.getValue("/Systems/WorkloadManagement/%s/GridEnv" % instance, "")
        result = self.pilotDB._getConnection()
        if result["OK"]:
            connection = result["Value"]
        else:
            return result

        result = self.pilotDB.getPilotGroups(self.identityFieldsList, {"Status": self.queryStateList})
        if not result["OK"]:
            self.log.error("Fail to get identities Groups", result["Message"])
            return result
        if not result["Value"]:
            return S_OK()

        pilotsToAccount = {}

        for ownerDN, ownerGroup, gridType, broker in result["Value"]:

            if not gridType in self.eligibleGridTypes:
                continue

            self.log.verbose("Getting pilots for %s:%s @ %s %s" % (ownerDN, ownerGroup, gridType, broker))

            condDict1 = {
                "Status": "Done",
                "StatusReason": "Report from JobAgent",
                "OwnerDN": ownerDN,
                "OwnerGroup": ownerGroup,
                "GridType": gridType,
                "Broker": broker,
            }

            condDict2 = {
                "Status": self.queryStateList,
                "OwnerDN": ownerDN,
                "OwnerGroup": ownerGroup,
                "GridType": gridType,
                "Broker": broker,
            }

            for condDict in [condDict1, condDict2]:
                result = self.clearWaitingPilots(condDict)
                if not result["OK"]:
                    self.log.warn("Failed to clear Waiting Pilot Jobs")

                result = self.pilotDB.selectPilots(condDict)
                if not result["OK"]:
                    self.log.warn("Failed to get the Pilot Agents")
                    return result
                if not result["Value"]:
                    continue
                refList = result["Value"]

                ret = gProxyManager.getPilotProxyFromDIRACGroup(ownerDN, ownerGroup)
                if not ret["OK"]:
                    self.log.error(ret["Message"])
                    self.log.error("Could not get proxy:", 'User "%s", Group "%s"' % (ownerDN, ownerGroup))
                    continue
                proxy = ret["Value"]

#.........这里部分代码省略.........

# 需要导入模块: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB import PilotAgentsDB [as 别名]
# 或者: from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB.PilotAgentsDB import selectPilots [as 别名]
class PilotStatusAgent( AgentModule ):
  """
      The specific agents must provide the following methods:
      - initialize() for initial settings
      - beginExecution()
      - execute() - the main method called in the agent cycle
      - endExecution()
      - finalize() - the graceful exit of the method, this one is usually used
                 for the agent restart
  """

  queryStateList = ['Ready', 'Submitted', 'Running', 'Waiting', 'Scheduled']
  finalStateList = [ 'Done', 'Aborted', 'Cleared', 'Deleted', 'Failed' ]
  identityFieldsList = [ 'OwnerDN', 'OwnerGroup', 'GridType', 'Broker' ]
  eligibleGridTypes = [ 'gLite' ]

  #############################################################################
  def initialize( self ):
    """Sets defaults
    """

    self.am_setOption( 'PollingTime', 120 )
    self.am_setOption( 'GridEnv', '' )
    self.am_setOption( 'PilotStalledDays', 3 )
    self.pilotDB = PilotAgentsDB()
    self.diracadmin = DiracAdmin()
    self.jobDB = JobDB()
    return S_OK()

  #############################################################################
  def execute( self ):
    """The PilotAgent execution method.
    """

    self.pilotStalledDays = self.am_getOption( 'PilotStalledDays', 3 )
    self.gridEnv = self.am_getOption( 'GridEnv' )
    if not self.gridEnv:
      # No specific option found, try a general one
      setup = gConfig.getValue( '/DIRAC/Setup', '' )
      if setup:
        instance = gConfig.getValue( '/DIRAC/Setups/%s/WorkloadManagement' % setup, '' )
        if instance:
          self.gridEnv = gConfig.getValue( '/Systems/WorkloadManagement/%s/GridEnv' % instance, '' )
    result = self.pilotDB._getConnection()
    if result['OK']:
      connection = result['Value']
    else:
      return result

    result = self.pilotDB.getPilotGroups( self.identityFieldsList,
                                         {'Status': self.queryStateList } )
    if not result['OK']:
      self.log.error( 'Fail to get identities Groups', result['Message'] )
      return result
    if not result['Value']:
      return S_OK()

    pilotsToAccount = {}

    for ownerDN, ownerGroup, gridType, broker in result['Value']:

      if not gridType in self.eligibleGridTypes:
        continue

      self.log.verbose( 'Getting pilots for %s:%s @ %s %s' % ( ownerDN, ownerGroup, gridType, broker ) )

      condDict1 = {'Status':'Done',
                   'StatusReason':'Report from JobAgent',
                   'OwnerDN':ownerDN,
                   'OwnerGroup':ownerGroup,
                   'GridType':gridType,
                   'Broker':broker}

      condDict2 = {'Status':self.queryStateList,
                   'OwnerDN':ownerDN,
                   'OwnerGroup':ownerGroup,
                   'GridType':gridType,
                   'Broker':broker}

      for condDict in [ condDict1, condDict2]:
        result = self.clearWaitingPilots( condDict )
        if not result['OK']:
          self.log.warn( 'Failed to clear Waiting Pilot Jobs' )

        result = self.pilotDB.selectPilots( condDict )
        if not result['OK']:
          self.log.warn( 'Failed to get the Pilot Agents' )
          return result
        if not result['Value']:
          continue
        refList = result['Value']

        ret = gProxyManager.getPilotProxyFromDIRACGroup( ownerDN, ownerGroup )
        if not ret['OK']:
          self.log.error( ret['Message'] )
          self.log.error( 'Could not get proxy:', 'User "%s", Group "%s"' % ( ownerDN, ownerGroup ) )
          continue
        proxy = ret['Value']

        self.log.verbose( "Getting status for %s pilots for owner %s and group %s" % ( len( refList ),
#.........这里部分代码省略.........