def loadConfig(db = "ProdAgentDB"):
"""
loadConfig method that loads the configuration of database base from ProdAgentConfig.xml and return it asdictionary
object
Argument:
db : database configuration block to load
Return:
Dictionary containing the required configuration
"""
try:
global first_access
cfg = loadProdAgentConfiguration()
dbConfig = cfg.getConfig(db)
if (first_access == True):
defaultConfig.update(dbConfig)
first_access = False
return dbConfig
except StandardError, ex:
msg = "%s.Config:" % db
msg += "Unable to load ProdAgent Config for " + db
msg += "%s\n" % ex
logging.warning(msg)
def __init__(self):
# TODO: The ResourceMonitor recreates this object in every call
# Is there a way to reuse the DB interface?
"""
_TaskQueueThresholdMonitor_
This enhanced constructor creates an API to the TaskQueue (not to
recreate it every time it is called).
"""
# Call our parent
MonitorInterface.__init__(self)
logging.debug("<<<<< Creating TaskQueueThresholdMonitor>>>>>")
self.me = "TaskQueueThresholdMonitor"
# Try to get the TaskQueue conf file from ProdAgentConfig.xml
what = "TaskQueue"
try:
cfg = loadProdAgentConfiguration()
self.tqConfig = cfg.getConfig(what)
except StandardError, ex:
msg = "%s.Config:" % what
msg += "Unable to load ProdAgent Config for " + what
msg += "%s\n" % ex
logging.critical(msg)
raise ProdAgentException(msg)
def getSiteInfoFromBase(sites, urls, ldaphost, dbp=None, samUrl=None):
"""
Collect all site info starting from LCG BDII Site names. Returns dictionary.
"""
use_phed=True
try:
if not dbp:
from ProdAgentCore.Configuration import loadProdAgentConfiguration
PAcfg = loadProdAgentConfiguration()
PhedCfg = PAcfg['PhEDExConfig']
dbp = PhedCfg['DBPARAM']
DBParamFile,Section = dbp.split(':')
if not os.path.isfile(DBParamFile):
logging.error("Configured DBParamFile %s not found" % DBParamFile)
use_phed=False
except:
print "Something went wrong with the Phedex config. Not using TMDB."
use_phed=False
#strip of text after "#", and remove site duplicates
#bdiiNames = list(set([x['SiteName'].split("#")[0] for x in sites]))
tmp_list = getBdii(sites, ldaphost)
#tmp_list=getBdii(bdiiNames,ldaphost)
#fcr info
#ceuids_in_fcr=getFcr(urls)
# for ceuid in tmp_list.keys():
# if ceuid in ceuids_in_fcr:
# tmp_list[ceuid]['in_fcr']=True
#tmp_list = [x for x in tmp_list if x['CEName'] in ceuids_in_fcr]
tmp_list = getFcr(urls, tmp_list)
#sam test info
tmp_list = samCheck(tmp_list, samUrl)
# if use_phed:
# #TODO: Out of date, Do we want to track links here -
# # esp if we are going to auto-create subscriptions
# Phmap=getPhedexSEmap(DBParamFile,Section)
#
# for site in tmp_list:
# #se_in_phedex=None
# for se in site.values(): #['SEs'][:]: #tmp_list[ceuid]['SEs']:
# se = se['SE']
# if not se in Phmap.values():
# ## this can happen more than once,
# ## but should not cause problems
# #se_in_phedex=se
# site['SEs'].remove(se)
# else:
# print "None of the published SEs can be found in Phedex."
# print " This is probably a bug somewhere."
# print "Currnet SEs: %s, CEUID: %s" % (
# str(tmp_list[ceuid]['SEs']),ceuid)
return tmp_list
def loadLSFConfig():
try:
config = loadProdAgentConfiguration()
except StandardError, ex :
msg = "Error reading configuration: \n"
msg += str(ex)
logging.error(msg)
raise RuntimeError, msg
def getResourceMonitorPlugin():
try:
config = loadProdAgentConfiguration()
compCfg = config.getConfig("ResourceMonitor")
except StandardError, ex:
msg = "Error reading configuration:\n"
msg += str(ex)
raise RuntimeError, msg
def alcaNewSelection():
if GlobalValues._alcaNewSelection == None:
GlobalValues._alcaNewSelection = False
paConfig = loadProdAgentConfiguration()
runConfig = paConfig.getConfig( "RunConfig" )
parameter = "AlcaSubscription"
if parameter in runConfig:
GlobalValues._alcaSubName = runConfig[parameter]
if GlobalValues._alcaDefSubName != GlobalValues._alcaSubName:
GlobalValues._alcaNewSelection = True
return GlobalValues._alcaNewSelection
def index(self,logFile) :
config = loadProdAgentConfiguration()
componentName = logFile.split("/")[0].strip()
componentDir = config.getConfig(componentName)['ComponentDir']
componentDir = os.path.expandvars(componentDir)
file = logFile.split("/")[1].strip()
html = """<html><body><h2>%s for %s </h2>\n """ % (file, componentName)
componentLog = open(componentDir+'/'+file).read().replace('\n','<br>')
html += componentLog
html += """</body></html>"""
return html
def getGlobalDBS(self):
"""getGlobalDBS
Gets Global DBS url form the ProdAgent Configuration
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error reading configuration:\n"
msg += str(ex)
raise RuntimeError, msg
def getSuccessArchiveDir():
"""
get the SuccessArchive dir from JobCleanup Component config
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error: error reading configuration:\n"
msg += str(ex)
print msg
sys.exit(1)
def getSuccessArchiveDir():
"""
get the SuccessArchive dir from JobCleanup Component config
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error: error reading configuration:\n"
msg += str(ex)
print msg
sys.exit(1)
def getRequestInjectorConfig():
"""
get the RequestInjector Component dir and the optional FirstRun
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error: error reading configuration:\n"
msg += str(ex)
print msg
sys.exit(1)
def getJobQueueConfig():
"""
find out if we should use the jobQueue
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error: error reading configuration:\n"
msg += str(ex)
print msg
sys.exit(1)
def getDbox(self):
from ProdAgentCore.Configuration import loadProdAgentConfiguration
dbox = None
try:
config = loadProdAgentConfiguration()
compCfg = config.getConfig("CrabServerConfigurations")
dbox = compCfg["CacheDir"]
except Exception, exc:
logging.error(str(exc))
raise Exception("Problem loading server configuration info " + str(exc))
def __call__(self):
"""
_operator()_
Invoke the Bot to check each component
"""
cfgObject = loadProdAgentConfiguration()
components = cfgObject.listComponents()
restartedComponents = []
for component in components:
logging.info("RestartBot: Checking %s" % component)
compCfg = cfgObject.getConfig(component)
compDir = compCfg['ComponentDir']
compDir = os.path.expandvars(compDir)
daemonXml = os.path.join(compDir, "Daemon.xml")
doRestart = False
if not os.path.exists(daemonXml):
doRestart = True
daemon = DaemonDetails(daemonXml)
if not daemon.isAlive():
doRestart = True
if doRestart:
restartedComponents.append(component)
modRef = __import__(component, globals(), locals(), [])
srcFile = inspect.getsourcefile(modRef)
srcDir = os.path.dirname(srcFile)
startup = os.path.join(srcDir, "Startup.py")
if not os.path.exists(startup):
msg = "Error starting component: %s\n" % component
msg += "Startup file is not found:\n"
msg += " %s \n" % startup
logging.warning(msg)
continue
logging.info( "Starting Component %s:" % component)
logging.info( "With: %s" % startup)
os.system("%s %s " % (sys.executable, startup))
else:
logging.info("RestartBot: Component %s Running" % component)
# //
# // Mail notification
#//
if len(restartedComponents) > 0 :
mailMsg = "RestartBot. The following components were not running. The have been restarted:\n"
for component in restartedComponents :
mailMsg += " %s" % (component)
mailMsg += "\nYours, RestartBot."
self.mail(mailMsg)
return
def getGroup():
"""
Read LSF group name from config file
Should I cache this ?
"""
try:
config = loadProdAgentConfiguration()
except StandardError, ex:
msg = "Error reading configuration:\n"
msg += str(ex)
logging.error(msg)
raise RuntimeError, msg
def __init__(self):
self.pluginConfig = None
self.prodAgentConfig = loadProdAgentConfiguration()
try:
# //
# // Always searches in JobCreator Config Block
#// for parameter called CreatorPluginConfig
self.pluginConfig = loadPluginConfig("JobCreator",
"Creator")
except StandardError, ex:
msg = "Failed to load Plugin Config for Creator Plugin:\n"
msg += "Plugin Name: %s\n" % self.__class__.__name__
msg += str(ex)
logging.warning(msg)
def main( argv ):
dbName = "PromptCalibDB"
try:
opts, args = getopt.getopt(argv, "h", ["help"])
except getopt.GetoptError:
print "Unable to parse argument list"
usage()
sys.exit(2)
for o, a in opts:
if o in ("-h", "--help"):
usage()
sys.exit()
if len(args) > 1:
print "Only expected one non-option argument"
usage()
sys.exit(2)
elif len(args) == 1:
dbName = args[0]
paConfig = loadProdAgentConfiguration()
t0astDBConfig = paConfig.getConfig( dbName )
print "Going to install the WMBS schema:"
print " File: %s" % t0astDBConfig["dbName"]
print ""
print "Is this OK? (yes/no): ",
line = sys.stdin.readline()
if line != "yes\n":
sys.exit(0)
print ""
clearDBInstance(t0astDBConfig["dbName"])
t0astDBConn = DBFactory.DBFactory(logging,"sqlite:///"+ t0astDBConfig['dbName'])
dbInterface = t0astDBConn.connect()
myThread = threading.currentThread()
myThread.dialect = "sqlite"
myThread.dbi = dbInterface
myThread.logger = logging
installWMBS(t0astDBConn, dbInterface)
setupWMBS()
sys.exit(0)
def __init__(self, cacheDepth = _DEFAULT_CACHE_DEPTH):
self.depth = cacheDepth
self.fileBackend = None
self.t0astDBConn = None
self.accessTimes = {}
self.runConfigs = {}
# The path to the Offline ConfDB file and the directory to cache
# framework configs is stored in the RunConfig section of the ProdAgent
# config file.
paConfig = loadProdAgentConfiguration().getConfig("RunConfig")
self.offlineConfDB = paConfig["OfflineConfDB"]
self.configCache = paConfig["ConfigCacheDir"]
logging.debug("RunConfigCache instantiated with depth=%s" % self.depth)
def verifySites():
"""
_verifySites_
(Safely) Extract the VerifySites parameter from the JobQueue
configuration block. Convert this to a True/False value
and return it, default is True (verify sites against site DB)
"""
try:
paConfig = loadProdAgentConfiguration()
except Exception, ex:
msg = "Unable to load PA Config: %s\n" % str(ex)
msg += "VerifySites defaulting to True"
logging.debug(msg)
return True
def installPADetails(dashboardInfo):
"""
_installPADetails_
Add PA IDs & meta information to dashboard info
"""
prodAgentConfig = loadProdAgentConfiguration()
paBlock = prodAgentConfig.get('ProdAgent', {})
prodMonBlock = prodAgentConfig.get("ProdMon", {})
paName = paBlock.get("ProdAgentName", "ProdAgent")
team = prodMonBlock.get("Team", "NoTeam")
dashboardInfo['ProdAgent'] = paName
dashboardInfo['ProductionTeam'] = team
return
def StartComponent():
"""
_StartComponent_
Function that Starts this Component
"""
try:
config = loadProdAgentConfiguration()
componentConfig = config.getConfig('AlertHandler')
except Exception, ex:
msg = 'Failed to load ProdAgent Configuration'
msg +='\nDetails: %s' % str(ex)
raise RuntimeError, msg
def __init__(self):
paConfig = loadProdAgentConfiguration()
t0astDBConfig = paConfig.getConfig("Tier0DB")
self.t0astDBConn = Tier0DB.Tier0DB(t0astDBConfig, manageGlobal = True)
self.t0astDBConn.connect()
localDBSConfig = paConfig.getConfig("LocalDBS")
globalDBSConfig = paConfig.getConfig("GlobalDBSDLS")
# phedexConfig = paConfig.getConfig("PhEDExConfig")
self.localDBSUrl = localDBSConfig["DBSURL"]
self.globalDBSUrl = globalDBSConfig["DBSURL"]
# self.phedexDSUrl = phedexConfig["DataServiceURL"]
self.localDbsApi = DbsApi({'url':localDBSConfig["DBSURL"]})
self.globalDbsApi = DbsApi({'url':globalDBSConfig["DBSURL"]})
def __init__(self):
"""
Overload constructor to add TaskQueue API initialization.
"""
BulkSubmitterInterface.__init__(self)
logging.debug("<<<<< Creating TaskQueueSubmitter>>>>>")
# Try to get the TaskQueue conf file from ProdAgentConfig.xml
what = "TaskQueue"
try:
cfg = loadProdAgentConfiguration()
self.tqConfig = cfg.getConfig(what)
except StandardError, ex:
msg = "%s.Config:" % what
msg += "Unable to load ProdAgent Config for " + what
msg += "%s\n" % ex
logging.warning(msg)
def __init__(self, args):
"""
"""
logging.debug("<<<<< Creating TaskQueueKiller>>>>>")
# Try to get the TaskQueue conf file from ProdAgentConfig.xml
what = "TaskQueue"
try:
cfg = loadProdAgentConfiguration()
self.tqConfig = cfg.getConfig(what)
except StandardError, ex:
msg = "%s.Config:" % what
msg += "Unable to load ProdAgent Config for " + what
msg += "%s\n" % ex
logging.critical(msg)
raise ProdAgentException(msg)
def initialize():
global maxRetries, requestJobFactory, jobSpecDir
if not requestJobFactory:
try:
config = loadProdAgentConfiguration()
compCfg = config.getConfig("ProdMgrInterface")
jobSpecDir=compCfg['JobSpecDir']
jobStatesCfg = config.getConfig("JobStates")
if jobStatesCfg.has_key('maxRetries'):
maxRetries=jobStatesCfg['maxRetries']
logging.debug("Initializing RequestJobFactory")
requestJobFactory = RequestJobFactory(None,'',0)
requestJobFactory.workflowSpec = WorkflowSpec()
requestJobFactory.workingDir = compCfg['JobSpecDir']
logging.debug("Writing job cut specs to: "+str(jobSpecDir))
except StandardError, ex:
msg = "Error reading configuration:\n"
msg += str(ex)
logging.info("WARNING: "+msg)
def configureScheduler(self, schedSession) :
"""
_configureScheduler_
perform any scheduler specific operation
Specific implementation in the Scheduler specific part
(e.g. BlGLiteBulkSubmitter)
"""
from ProdAgentCore.Configuration import loadProdAgentConfiguration
config = loadProdAgentConfiguration()
compCfg = config.getConfig("JobSubmitter")
filename = os.path.join(compCfg['ComponentDir'], 'lastDelegation')
format = '%Y-%m-%d %H:%M:%S'
timestamp = 0
if os.path.exists(filename) :
tsfile = open(filename, 'r')
tsentry = tsfile.read().strip()
try :
timestamp = time.time() - \
time.mktime(time.strptime(tsentry, format))
timestamp /= 43200
except ValueError:
pass
tsfile.close()
if timestamp > 0 and timestamp < 1:
return
logging.info('delegating proxy to wms after %s hours' % \
(timestamp * 12))
schedSession.getSchedulerInterface().delegateProxy()
tsfile = open(filename, 'w')
logging.info ('%s' % time.strftime(format))
tsfile.write(time.strftime(format))
tsfile.close()
return
def loadPluginConfig(component, plugin):
"""
_loadPluginConfig_
Get the PluginConfiguration object for the component plugin
using the name specified.
The Component will be looked up from the main ProdAgentConfig using
the PRODAGENT_CONFIG env var.
This Config will be searched for the plugin file using the convention:
<pluginName>PluginConfig
This is assumed to be a file which is a saved PluginConfiguration
object
"""
prodAgentConfig = loadProdAgentConfiguration()
compConfig = prodAgentConfig.getConfig(component)
keyname = "%sPluginConfig" % plugin
filename = compConfig.get(keyname, None)
if filename == None:
msg = "Cannot find entry for Plugin: %s\n" % plugin
msg += "In Config for component: %s\n" % component
msg += "Expected parameter: %s\n " % keyname
raise RuntimeError, msg
filename = os.path.expandvars(filename)
if not os.path.exists(filename):
msg = "Plugin Configuration file Not Found:\n"
msg += "%s\n" % filename
msg += "For component %s plugin %s\n" % (component, plugin)
raise RuntimeError, msg
plugCfg = PluginConfiguration()
try:
plugCfg.loadFromFile(filename)
except StandardError, ex:
msg = "Error loading Plugin Configuration for:\n"
msg += "Component %s Plugin %s\n" % (component, plugin)
msg += "From File:\n%s\n" % filename
msg += str(ex)
raise RuntimeError, msg
def index(self, componentName) :
config = loadProdAgentConfiguration()
componentDir = config.getConfig(componentName)['ComponentDir']
componentDir = os.path.expandvars(componentDir)
logs = []
for file in os.listdir(componentDir) :
if file.find('ComponentLog') > -1 :
logs.append(file)
logs.sort()
html = """<html><body><h2>List of Available Logs for Component %s </h2>\n """ % componentName
if len(logs) == 0 :
html = """<html><body><h2>No Available Logs for Component %s! </h2>\n """ % componentName
else :
html += "<ul>\n"
for log in logs:
logFile = os.path.join(componentName,log)
html += "<li><a href=\"%s?logFile=%s\">%s</a></li>\n" % (
self.writelog, logFile, log)
html += "</ul>\n"
html += """</body></html>"""
return html
def __init__(self, config):
"""
__init__
"""
self.jobavaliableflag = False
self.tarPath = config.PilotManagerComponent.tarPath
self.pilotcodeDir = config.PilotManagerComponent.pilotCode
self.tqServer = config.PilotManagerComponent.tqAddress
self.emulationMode = config.PilotManagerComponent.emulationMode
self.pilotParams = None
try:
cfg = loadProdAgentConfiguration()
pilot = cfg.getConfig("Pilot")
self.pilotParams = pilot
except:
logging.debug("could not load PRODAGENT_CONFIG file")
if ( not self.pilotParams ):
self.pilotParams={'badAttempts':6,'noTaskAttempts':6}
Harness.__init__(self, config)
def __init__(self, argv):
"""
initialize test instance
"""
# initialize super class
super(jobUnitTests, self).__init__(argv)
# read configuration
self.configure()
# MySQL: get DB configuration from config file
if self.database == "MySQL":
from ProdAgentCore.Configuration import loadProdAgentConfiguration
cfg = loadProdAgentConfiguration()
dbConfig = cfg.getConfig("ProdAgentDB")
dbConfig['dbName'] += '_BOSS'
# create DB instance
self.dbInstance = MysqlInstance(dbConfig)
else:
dbConfig = {}
dbConfig['dbName'] = 'ProdAgentDB_BOSS'
# create DB instance
self.dbInstance = SqliteInstance(dbConfig)
# remove all information from database
session = SafeSession(dbInstance = self.dbInstance)
try:
session.execute("delete from bl_runningjob")
session.execute("delete from bl_job")
session.execute("delete from bl_task")
session.commit()
except DbError, msg:
print "Cannot erase database contents: " + str(msg)
sys.exit(1)
def __init__(self, argv):
"""
initialize test instance
"""
# initialize super class
super(jobUnitTests, self).__init__(argv)
# read configuration
self.configure()
# MySQL: get DB configuration from config file
if self.database == "MySQL":
from ProdAgentCore.Configuration import loadProdAgentConfiguration
cfg = loadProdAgentConfiguration()
dbConfig = cfg.getConfig("ProdAgentDB")
dbConfig['dbName'] += '_BOSS'
# create DB instance
self.dbInstance = MysqlInstance(dbConfig)
else:
dbConfig = {}
dbConfig['dbName'] = 'ProdAgentDB_BOSS'
# create DB instance
self.dbInstance = SqliteInstance(dbConfig)
# remove all information from database
session = SafeSession(dbInstance=self.dbInstance)
try:
session.execute("delete from bl_runningjob")
session.execute("delete from bl_job")
session.execute("delete from bl_task")
session.commit()
except DbError, msg:
print "Cannot erase database contents: " + str(msg)
sys.exit(1)
def index(self):
config = loadProdAgentConfiguration()
components = config.listComponents()
html = """<head>\n"""
html += "<meta http-equiv=\"refresh\" content=\"60\" />\n\n"
html += """</head>"""
html += """<html><body><h2>ProdAgent Component Status</h2>\n """
html += "<table>\n"
html += "<tr><th>Component</th><th>Status</th><th>Logs</th><th>PostMortens</th></tr>\n"
for component in components :
componentDir = config.getConfig(component)['ComponentDir']
componentDir = os.path.expandvars(componentDir)
daemonXML = os.path.join(componentDir, "Daemon.xml")
daemon = DaemonDetails(daemonXML)
if not daemon.isAlive():
html += "<tr><td>%s</td><td><font color=\"FF0000\">Not Running</font></td>" % component
else :
html += "<tr><td>%s</td><td>PID : %s</td>" % (component, daemon['ProcessID'])
html += "<td><a href=\"%s?componentName=%s\">Show</a></td>\n" % (self.complog, component)
html += "<td><a href=\"%s?componentName=%s\">Show</a></td></tr>\n" % (self.postmorten, component)
html += "</table>\n"
html += """</body></html>"""
return html
