def phedex():
phedexIn = PhEDEx(dict = {'endpoint' : 'https://cmsweb.cern.ch/phedex/datasvc/json/dev/',
'logger' : logging},
responseType = "json")
# requests = phedex.getRequestList(dataset = ['/TauParked/Run2012C-LogError-22Jan2013-v1/RAW-RECO'],
# node = 'T2_RU_ITEP')['phedex']['request']
# for request in requests:
# requestId = request['id']
# request = phedex.getTransferRequests(request = requestId)['phedex']['request']
# if request:
# request = request[0]
# print request
# x = PhEDExSubscription('/TauParked/Run2012C-22Jan2013-v1/AOD',
# 'T1_US_FNAL_MSS', 'DataOps', 'dataset', 'low', 'n', 'n', 'n', 'y', subscriptionId = 1)
# print x.matchesExistingTransferRequest(phedex)
# print x.matchesExistingSubscription(phedex)
deletion = PhEDExDeletion('/071103be-7d80-11e0-90de-00163e010039/PromptReco-v1/RECO',
'T1_CH_CERN_Buffer',
level = 'block',
comments = 'Blocks automatically deleted from T2_CH_CERN as it has already been processed and transferred to a custodial location',
blocks = {'/071103be-7d80-11e0-90de-00163e010039/PromptReco-v1/RECO' :
['/071103be-7d80-11e0-90de-00163e010039/PromptReco-v1/RECO#075ea9e8-7d80-11e0-90de-00163e010039']})
xmlData = XMLDrop.makePhEDExXMLForBlocks('http://vocms09.cern.ch:8880/cms_dbs_int_local_yy_writer/servlet/DBSServlet',
deletion.getDatasetsAndBlocks())
print str(xmlData)
response = phedexIn.delete(deletion, xmlData)
print response
requestId = response['phedex']['request_created'][0]['id']
phedexIn.updateRequest(requestId, 'approve', 'T1_CH_CERN_Buffer')
class PhEDExInjectorPoller(BaseWorkerThread):
"""
_PhEDExInjectorPoller_
Poll the DBSBuffer database and inject files as they are created.
"""
def __init__(self, config):
"""
___init___
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.dbsUrl = config.DBSInterface.globalDBSUrl
self.pollCounter = 0
self.subFrequency = None
if getattr(config.PhEDExInjector, "subscribeDatasets", False):
pollInterval = config.PhEDExInjector.pollInterval
subInterval = config.PhEDExInjector.subscribeInterval
self.subFrequency = max(1, int(round(subInterval / pollInterval)))
logging.info(
"SubscribeDataset and deleteBlocks will run every %d polling cycles",
self.subFrequency)
# subscribe on first cycle
self.pollCounter = self.subFrequency - 1
# retrieving the node mappings is fickle and can fail quite often
self.phedex = PhEDEx({"endpoint": config.PhEDExInjector.phedexurl},
"json")
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(2)
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(4)
nodeMappings = self.phedex.getNodeMap()
# This will be used to map SE names which are stored in the DBSBuffer to
# PhEDEx node names. The first key will be the "kind" which consists
# of one of the following: MSS, Disk, Buffer. The next key will be the
# SE name.
self.seMap = {}
self.nodeNames = []
for node in nodeMappings["phedex"]["node"]:
if node["kind"] not in self.seMap:
self.seMap[node["kind"]] = {}
logging.info("Adding mapping %s -> %s", node["se"], node["name"])
self.seMap[node["kind"]][node["se"]] = node["name"]
self.nodeNames.append(node["name"])
self.phedexNodes = {'MSS': [], 'Disk': []}
for node in nodeMappings["phedex"]["node"]:
if node["kind"] in ["MSS", "Disk"]:
self.phedexNodes[node["kind"]].append(node["name"])
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName="PhEDExInjector")
self.blocksToRecover = []
return
def setup(self, parameters):
"""
_setup_
Create DAO Factory and setup some DAO.
"""
myThread = threading.currentThread()
daofactory = DAOFactory(package="WMComponent.PhEDExInjector.Database",
logger=self.logger,
dbinterface=myThread.dbi)
self.getUninjected = daofactory(classname="GetUninjectedFiles")
self.getMigrated = daofactory(classname="GetMigratedBlocks")
self.findDeletableBlocks = daofactory(classname="GetDeletableBlocks")
self.markBlocksDeleted = daofactory(classname="MarkBlocksDeleted")
self.getUnsubscribed = daofactory(classname="GetUnsubscribedDatasets")
self.markSubscribed = daofactory(classname="MarkDatasetSubscribed")
daofactory = DAOFactory(package="WMComponent.DBS3Buffer",
logger=self.logger,
dbinterface=myThread.dbi)
self.setStatus = daofactory(classname="DBSBufferFiles.SetPhEDExStatus")
self.setBlockClosed = daofactory(classname="SetBlockClosed")
return
def algorithm(self, parameters):
"""
_algorithm_
Poll the database for uninjected files and attempt to inject them into
PhEDEx.
"""
logging.info("Running PhEDEx injector poller algorithm...")
self.pollCounter += 1
if self.blocksToRecover:
logging.info("""PhEDExInjector Recovery:
previous injection call failed,
check if files were injected to PhEDEx anyway""")
self.recoverInjectedFiles()
self.injectFiles()
self.closeBlocks()
if self.pollCounter == self.subFrequency:
self.pollCounter = 0
self.deleteBlocks()
self.subscribeDatasets()
return
def createInjectionSpec(self, injectionData):
"""
_createInjectionSpec_
Transform the data structure returned from the database into an XML
string for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path. Each dataset path will map to a
list of blocks, each block being a dict. The block dicts will have
three keys: name, is-open and files. The files key will be a list of
dicts, each of which have the following keys: lfn, size and checksum.
The following is an example object:
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
"""
injectionSpec = XMLDrop.XMLInjectionSpec(self.dbsUrl)
for datasetPath in injectionData:
datasetSpec = injectionSpec.getDataset(datasetPath)
for fileBlockName, fileBlock in injectionData[
datasetPath].iteritems():
blockSpec = datasetSpec.getFileblock(fileBlockName,
fileBlock["is-open"])
for f in fileBlock["files"]:
blockSpec.addFile(f["lfn"], f["checksum"], f["size"])
return injectionSpec.save()
def createRecoveryFileFormat(self, unInjectedData):
"""
_createRecoveryFileFormat_
Transform the data structure returned from database in to the dict format
for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path.
unInjectedData format
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
returns
[{"block1": set(["lfn1", "lfn2"])}, {"block2": set(["lfn3", "lfn4"])]
"""
blocks = []
for datasetPath in unInjectedData:
for blockName, fileBlock in unInjectedData[datasetPath].items():
newBlock = {blockName: set()}
for fileDict in fileBlock["files"]:
newBlock[blockName].add(fileDict["lfn"])
blocks.append(newBlock)
return blocks
def injectFiles(self):
"""
_injectFiles_
Inject any uninjected files in PhEDEx.
"""
logging.info("Starting injectFiles method")
uninjectedFiles = self.getUninjected.execute()
for siteName in uninjectedFiles.keys():
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location == None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
self.sendAlert(7, msg=msg)
continue
maxDataset = 20
maxBlocks = 50
maxFiles = 5000
numberDatasets = 0
numberBlocks = 0
numberFiles = 0
injectData = {}
lfnList = []
for dataset in uninjectedFiles[siteName]:
numberDatasets += 1
injectData[dataset] = uninjectedFiles[siteName][dataset]
for block in injectData[dataset]:
numberBlocks += 1
numberFiles += len(injectData[dataset][block]['files'])
for fileInfo in injectData[dataset][block]['files']:
lfnList.append(fileInfo['lfn'])
if numberDatasets >= maxDataset or numberBlocks >= maxBlocks or numberFiles >= maxFiles:
self.injectFilesPhEDExCall(location, injectData, lfnList)
numberDatasets = 0
numberBlocks = 0
numberFiles = 0
injectData = {}
lfnList = []
if injectData:
self.injectFilesPhEDExCall(location, injectData, lfnList)
return
def injectFilesPhEDExCall(self, location, injectData, lfnList):
"""
_injectFilesPhEDExCall_
actual PhEDEx call for file injection
"""
xmlData = self.createInjectionSpec(injectData)
logging.debug("injectFiles XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
# HTTPException with status 400 assumed to be duplicate injection
# trigger later block recovery (investgation needed if not the case)
if ex.status == 400:
self.blocksToRecover.extend(
self.createRecoveryFileFormat(injectData))
logging.error(
"PhEDEx file injection failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx file injection failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
logging.info("Injection result: %s", injectRes)
if "error" in injectRes:
msg = "Error injecting data %s: %s" % (injectData,
injectRes["error"])
logging.error(msg)
self.sendAlert(6, msg=msg)
else:
try:
self.setStatus.execute(lfnList, 1)
except:
# possible deadlock with DBS3Upload, retry once after 5s
logging.warning(
"Oracle exception during file status update, possible deadlock due to race condition, retry after 5s sleep"
)
time.sleep(5)
self.setStatus.execute(lfnList, 1)
return
def closeBlocks(self):
"""
_closeBlocks_
Close any blocks that have been migrated to global DBS
"""
logging.info("Starting closeBlocks method")
migratedBlocks = self.getMigrated.execute()
for siteName in migratedBlocks.keys():
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location == None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
self.sendAlert(6, msg=msg)
continue
xmlData = self.createInjectionSpec(migratedBlocks[siteName])
logging.debug("closeBlocks XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
logging.error(
"PhEDEx block close failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx block close failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
logging.info("Block closing result: %s", injectRes)
if "error" not in injectRes:
for datasetName in migratedBlocks[siteName]:
for blockName in migratedBlocks[siteName][datasetName]:
logging.debug("Closing block %s", blockName)
self.setBlockClosed.execute(blockName)
else:
msg = "Error injecting data %s: %s" % (
migratedBlocks[siteName], injectRes["error"])
logging.error(msg)
self.sendAlert(6, msg=msg)
return
def recoverInjectedFiles(self):
"""
When PhEDEx inject call timed out, run this function.
Since there are 3 min reponse time out in cmsweb, some times
PhEDEx injection call times out even though the call succeeded
In that case run the recovery mode
1. first check whether files which injection status = 0 are in the PhEDEx.
2. if those file exist set the in_phedex status to 1
3. set self.blocksToRecover = []
Run this recovery one block at a time, with too many blocks
the call to the PhEDEx data service on cmsweb can time out
"""
# recover one block at a time
for block in self.blocksToRecover:
injectedFiles = self.phedex.getInjectedFiles(block)
if injectedFiles:
self.setStatus.execute(injectedFiles, 1)
self.blocksToRecover = []
return
def deleteBlocks(self):
"""
_deleteBlocks_
Find deletable blocks, then decide if to delete based on:
Is there an active subscription for dataset or block ?
If yes => set deleted=2
If no => next check
Has transfer to all destinations finished ?
If yes => request block deletion, approve request, set deleted=1
If no => do nothing (check again next cycle)
"""
logging.info("Starting deleteBlocks method")
blockDict = self.findDeletableBlocks.execute(transaction=False)
if not blockDict:
return
try:
subscriptions = self.phedex.getSubscriptionMapping(
*blockDict.keys())
except:
logging.error(
"Couldn't get subscription info from PhEDEx, retry next cycle")
return
skippableBlocks = []
deletableEntries = {}
for blockName in blockDict:
location = blockDict[blockName]['location']
# should never be triggered, better safe than sorry
if location.endswith('_MSS'):
logging.debug("Location %s for block %s is MSS, skip deletion",
location, blockName)
skippableBlocks.append(blockName)
continue
dataset = blockDict[blockName]['dataset']
sites = blockDict[blockName]['sites']
if blockName in subscriptions and location in subscriptions[
blockName]:
logging.debug("Block %s subscribed to %s, skip deletion",
blockName, location)
binds = {'DELETED': 2, 'BLOCKNAME': blockName}
self.markBlocksDeleted.execute(binds)
else:
blockInfo = []
try:
blockInfo = self.phedex.getReplicaInfoForBlocks(
block=blockName, complete='y')['phedex']['block']
except:
logging.error(
"Couldn't get block info from PhEDEx, retry next cycle"
)
else:
for entry in blockInfo:
if entry['name'] == blockName:
nodes = set([x['node'] for x in entry['replica']])
if location not in nodes:
logging.debug(
"Block %s not present on %s, mark as deleted",
blockName, location)
binds = {'DELETED': 1, 'BLOCKNAME': blockName}
self.markBlocksDeleted.execute(binds)
elif sites.issubset(nodes):
logging.debug(
"Deleting block %s from %s since it is fully transfered",
blockName, location)
if location not in deletableEntries:
deletableEntries[location] = {}
if dataset not in deletableEntries[location]:
deletableEntries[location][dataset] = set()
deletableEntries[location][dataset].add(
blockName)
binds = []
for blockName in skippableBlocks:
binds.append({'DELETED': 2, 'BLOCKNAME': blockName})
if binds:
self.markBlocksDeleted.execute(binds)
for location in deletableEntries:
chunkSize = 100
numberOfBlocks = 0
blocksToDelete = {}
for dataset in deletableEntries[location]:
blocksToDelete[dataset] = deletableEntries[location][dataset]
numberOfBlocks += len(blocksToDelete[dataset])
if numberOfBlocks > chunkSize:
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
numberOfBlocks = 0
blocksToDelete = {}
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
return
def deleteBlocksPhEDExCalls(self, location, blocksToDelete):
"""
_deleteBlocksPhEDExCalls_
actual PhEDEx calls for block deletion
"""
deletion = PhEDExDeletion(
blocksToDelete.keys(),
location,
level='block',
comments="WMAgent blocks auto-delete from %s" % location,
blocks=blocksToDelete)
xmlData = XMLDrop.makePhEDExXMLForBlocks(
self.dbsUrl, deletion.getDatasetsAndBlocks())
logging.debug("deleteBlocks XMLData: %s", xmlData)
try:
response = self.phedex.delete(deletion, xmlData)
requestId = response['phedex']['request_created'][0]['id']
# auto-approve deletion request
self.phedex.updateRequest(requestId, 'approve', location)
except HTTPException as ex:
logging.error(
"PhEDEx block delete/approval failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error(
"PhEDEx block delete/approval failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
binds = []
for dataset in blocksToDelete:
for blockName in blocksToDelete[dataset]:
binds.append({'DELETED': 1, 'BLOCKNAME': blockName})
self.markBlocksDeleted.execute(binds)
return
def subscribeDatasets(self):
"""
_subscribeDatasets_
Poll the database for datasets and subscribe them.
"""
logging.info("Starting subscribeDatasets method")
# Check for completely unsubscribed datasets
unsubscribedDatasets = self.getUnsubscribed.execute()
# Keep a list of subscriptions to tick as subscribed in the database
subscriptionsMade = []
# Create a list of subscriptions as defined by the PhEDEx data structures
subs = SubscriptionList()
# Create the subscription objects and add them to the list
# The list takes care of the sorting internally
for subInfo in unsubscribedDatasets:
site = subInfo['site']
if site not in self.phedexNodes[
'MSS'] and site not in self.phedexNodes['Disk']:
msg = "Site %s doesn't appear to be valid to PhEDEx, " % site
msg += "skipping subscription: %s" % subInfo['id']
logging.error(msg)
self.sendAlert(7, msg=msg)
continue
# Avoid custodial subscriptions to disk nodes
if site not in self.phedexNodes['MSS']:
subInfo['custodial'] = 'n'
# Avoid auto approval in T1 sites
elif site.startswith("T1"):
subInfo['request_only'] = 'y'
phedexSub = PhEDExSubscription(
subInfo['path'],
site,
subInfo['phedex_group'],
priority=subInfo['priority'],
move=subInfo['move'],
custodial=subInfo['custodial'],
request_only=subInfo['request_only'],
subscriptionId=subInfo['id'])
# Check if the subscription is a duplicate
if phedexSub.matchesExistingSubscription(self.phedex) or \
phedexSub.matchesExistingTransferRequest(self.phedex):
subscriptionsMade.append(subInfo['id'])
continue
# Add it to the list
subs.addSubscription(phedexSub)
# Compact the subscriptions
subs.compact()
for subscription in subs.getSubscriptionList():
xmlData = XMLDrop.makePhEDExXMLForDatasets(
self.dbsUrl, subscription.getDatasetPaths())
logging.debug("subscribeDatasets XMLData: %s", xmlData)
logging.info(
"Subscribing: %s to %s, with options: Move: %s, Custodial: %s, Request Only: %s",
subscription.getDatasetPaths(), subscription.getNodes(),
subscription.move, subscription.custodial,
subscription.request_only)
try:
self.phedex.subscribe(subscription, xmlData)
except HTTPException as ex:
logging.error(
"PhEDEx dataset subscribe failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error(
"PhEDEx dataset subscribe failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
subscriptionsMade.extend(subscription.getSubscriptionIds())
# Register the result in DBSBuffer
if subscriptionsMade:
self.markSubscribed.execute(subscriptionsMade)
return
class PhEDExInjectorPoller(BaseWorkerThread):
"""
_PhEDExInjectorPoller_
Poll the DBSBuffer database and inject files as they are created.
"""
def __init__(self, config):
"""
___init___
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.dbsUrl = config.DBSInterface.globalDBSUrl
self.pollCounter = 0
self.subFrequency = None
if getattr(config.PhEDExInjector, "subscribeDatasets", False):
pollInterval = config.PhEDExInjector.pollInterval
subInterval = config.PhEDExInjector.subscribeInterval
self.subFrequency = max(1, int(round(subInterval/pollInterval)))
logging.info("SubscribeDataset and deleteBlocks will run every %d polling cycles", self.subFrequency)
# subscribe on first cycle
self.pollCounter = self.subFrequency - 1
# retrieving the node mappings is fickle and can fail quite often
self.phedex = PhEDEx({"endpoint": config.PhEDExInjector.phedexurl}, "json")
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(2)
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(4)
nodeMappings = self.phedex.getNodeMap()
# This will be used to map SE names which are stored in the DBSBuffer to
# PhEDEx node names. The first key will be the "kind" which consists
# of one of the following: MSS, Disk, Buffer. The next key will be the
# SE name.
self.seMap = {}
self.nodeNames = []
for node in nodeMappings["phedex"]["node"]:
if node["kind"] not in self.seMap:
self.seMap[node["kind"]] = {}
logging.info("Adding mapping %s -> %s", node["se"], node["name"])
self.seMap[node["kind"]][node["se"]] = node["name"]
self.nodeNames.append(node["name"])
self.phedexNodes = {'MSS': [], 'Disk': []}
for node in nodeMappings["phedex"]["node"]:
if node["kind"] in ["MSS", "Disk"]:
self.phedexNodes[node["kind"]].append(node["name"])
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "PhEDExInjector")
self.blocksToRecover = []
return
def setup(self, parameters):
"""
_setup_
Create DAO Factory and setup some DAO.
"""
myThread = threading.currentThread()
daofactory = DAOFactory(package = "WMComponent.PhEDExInjector.Database",
logger = self.logger,
dbinterface = myThread.dbi)
self.getUninjected = daofactory(classname = "GetUninjectedFiles")
self.getMigrated = daofactory(classname = "GetMigratedBlocks")
self.findDeletableBlocks = daofactory(classname = "GetDeletableBlocks")
self.markBlocksDeleted = daofactory(classname = "MarkBlocksDeleted")
self.getUnsubscribed = daofactory(classname = "GetUnsubscribedDatasets")
self.markSubscribed = daofactory(classname = "MarkDatasetSubscribed")
daofactory = DAOFactory(package = "WMComponent.DBS3Buffer",
logger = self.logger,
dbinterface = myThread.dbi)
self.setStatus = daofactory(classname = "DBSBufferFiles.SetPhEDExStatus")
self.setBlockClosed = daofactory(classname = "SetBlockClosed")
return
def algorithm(self, parameters):
"""
_algorithm_
Poll the database for uninjected files and attempt to inject them into
PhEDEx.
"""
logging.info("Running PhEDEx injector poller algorithm...")
self.pollCounter += 1
if self.blocksToRecover:
logging.info("""PhEDExInjector Recovery:
previous injection call failed,
check if files were injected to PhEDEx anyway""")
self.recoverInjectedFiles()
self.injectFiles()
self.closeBlocks()
if self.pollCounter == self.subFrequency:
self.pollCounter = 0
self.deleteBlocks()
self.subscribeDatasets()
return
def createInjectionSpec(self, injectionData):
"""
_createInjectionSpec_
Transform the data structure returned from the database into an XML
string for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path. Each dataset path will map to a
list of blocks, each block being a dict. The block dicts will have
three keys: name, is-open and files. The files key will be a list of
dicts, each of which have the following keys: lfn, size and checksum.
The following is an example object:
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
"""
injectionSpec = XMLDrop.XMLInjectionSpec(self.dbsUrl)
for datasetPath in injectionData:
datasetSpec = injectionSpec.getDataset(datasetPath)
for fileBlockName, fileBlock in injectionData[datasetPath].iteritems():
blockSpec = datasetSpec.getFileblock(fileBlockName,
fileBlock["is-open"])
for f in fileBlock["files"]:
blockSpec.addFile(f["lfn"], f["checksum"], f["size"])
return injectionSpec.save()
def createRecoveryFileFormat(self, unInjectedData):
"""
_createRecoveryFileFormat_
Transform the data structure returned from database in to the dict format
for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path.
unInjectedData format
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
returns
[{"block1": set(["lfn1", "lfn2"])}, {"block2": set(["lfn3", "lfn4"])]
"""
blocks = []
for datasetPath in unInjectedData:
for blockName, fileBlock in unInjectedData[datasetPath].items():
newBlock = { blockName: set() }
for fileDict in fileBlock["files"]:
newBlock[blockName].add(fileDict["lfn"])
blocks.append(newBlock)
return blocks
def injectFiles(self):
"""
_injectFiles_
Inject any uninjected files in PhEDEx.
"""
logging.info("Starting injectFiles method")
uninjectedFiles = self.getUninjected.execute()
for siteName in uninjectedFiles.keys():
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location == None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
self.sendAlert(7, msg = msg)
continue
maxDataset = 20
maxBlocks = 50
maxFiles = 5000
numberDatasets = 0
numberBlocks = 0
numberFiles = 0
injectData = {}
lfnList = []
for dataset in uninjectedFiles[siteName]:
numberDatasets += 1
injectData[dataset] = uninjectedFiles[siteName][dataset]
for block in injectData[dataset]:
numberBlocks += 1
numberFiles += len(injectData[dataset][block]['files'])
for fileInfo in injectData[dataset][block]['files']:
lfnList.append(fileInfo['lfn'])
if numberDatasets >= maxDataset or numberBlocks >= maxBlocks or numberFiles >= maxFiles:
self.injectFilesPhEDExCall(location, injectData, lfnList)
numberDatasets = 0
numberBlocks = 0
numberFiles = 0
injectData = {}
lfnList = []
if injectData:
self.injectFilesPhEDExCall(location, injectData, lfnList)
return
def injectFilesPhEDExCall(self, location, injectData, lfnList):
"""
_injectFilesPhEDExCall_
actual PhEDEx call for file injection
"""
xmlData = self.createInjectionSpec(injectData)
logging.debug("injectFiles XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
# HTTPException with status 400 assumed to be duplicate injection
# trigger later block recovery (investgation needed if not the case)
if ex.status == 400:
self.blocksToRecover.extend( self.createRecoveryFileFormat(injectData) )
logging.error("PhEDEx file injection failed with HTTPException: %s %s", ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx file injection failed with Exception: %s", str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
logging.info("Injection result: %s", injectRes)
if "error" in injectRes:
msg = "Error injecting data %s: %s" % (injectData, injectRes["error"])
logging.error(msg)
self.sendAlert(6, msg = msg)
else:
try:
self.setStatus.execute(lfnList, 1)
except:
# possible deadlock with DBS3Upload, retry once after 5s
logging.warning("Oracle exception during file status update, possible deadlock due to race condition, retry after 5s sleep")
time.sleep(5)
self.setStatus.execute(lfnList, 1)
return
def closeBlocks(self):
"""
_closeBlocks_
Close any blocks that have been migrated to global DBS
"""
logging.info("Starting closeBlocks method")
migratedBlocks = self.getMigrated.execute()
for siteName in migratedBlocks.keys():
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location == None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
self.sendAlert(6, msg = msg)
continue
xmlData = self.createInjectionSpec(migratedBlocks[siteName])
logging.debug("closeBlocks XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
logging.error("PhEDEx block close failed with HTTPException: %s %s", ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx block close failed with Exception: %s", str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
logging.info("Block closing result: %s", injectRes)
if "error" not in injectRes:
for datasetName in migratedBlocks[siteName]:
for blockName in migratedBlocks[siteName][datasetName]:
logging.debug("Closing block %s", blockName)
self.setBlockClosed.execute(blockName)
else:
msg = "Error injecting data %s: %s" % (migratedBlocks[siteName], injectRes["error"])
logging.error(msg)
self.sendAlert(6, msg = msg)
return
def recoverInjectedFiles(self):
"""
When PhEDEx inject call timed out, run this function.
Since there are 3 min reponse time out in cmsweb, some times
PhEDEx injection call times out even though the call succeeded
In that case run the recovery mode
1. first check whether files which injection status = 0 are in the PhEDEx.
2. if those file exist set the in_phedex status to 1
3. set self.blocksToRecover = []
Run this recovery one block at a time, with too many blocks
the call to the PhEDEx data service on cmsweb can time out
"""
# recover one block at a time
for block in self.blocksToRecover:
injectedFiles = self.phedex.getInjectedFiles(block)
if injectedFiles:
self.setStatus.execute(injectedFiles, 1)
self.blocksToRecover = []
return
def deleteBlocks(self):
"""
_deleteBlocks_
Find deletable blocks, then decide if to delete based on:
Is there an active subscription for dataset or block ?
If yes => set deleted=2
If no => next check
Has transfer to all destinations finished ?
If yes => request block deletion, approve request, set deleted=1
If no => do nothing (check again next cycle)
"""
logging.info("Starting deleteBlocks method")
blockDict = self.findDeletableBlocks.execute(transaction = False)
if not blockDict:
return
try:
subscriptions = self.phedex.getSubscriptionMapping(*blockDict.keys())
except:
logging.error("Couldn't get subscription info from PhEDEx, retry next cycle")
return
skippableBlocks = []
deletableEntries = {}
for blockName in blockDict:
location = blockDict[blockName]['location']
# should never be triggered, better safe than sorry
if location.endswith('_MSS'):
logging.debug("Location %s for block %s is MSS, skip deletion", location, blockName)
skippableBlocks.append(blockName)
continue
dataset = blockDict[blockName]['dataset']
sites = blockDict[blockName]['sites']
if blockName in subscriptions and location in subscriptions[blockName]:
logging.debug("Block %s subscribed to %s, skip deletion", blockName, location)
binds = { 'DELETED': 2,
'BLOCKNAME': blockName }
self.markBlocksDeleted.execute(binds)
else:
blockInfo = []
try:
blockInfo = self.phedex.getReplicaInfoForBlocks(block = blockName, complete = 'y')['phedex']['block']
except:
logging.error("Couldn't get block info from PhEDEx, retry next cycle")
else:
for entry in blockInfo:
if entry['name'] == blockName:
nodes = set([x['node'] for x in entry['replica']])
if location not in nodes:
logging.debug("Block %s not present on %s, mark as deleted", blockName, location)
binds = { 'DELETED': 1,
'BLOCKNAME': blockName }
self.markBlocksDeleted.execute(binds)
elif sites.issubset(nodes):
logging.debug("Deleting block %s from %s since it is fully transfered", blockName, location)
if location not in deletableEntries:
deletableEntries[location] = {}
if dataset not in deletableEntries[location]:
deletableEntries[location][dataset] = set()
deletableEntries[location][dataset].add(blockName)
binds = []
for blockName in skippableBlocks:
binds.append( { 'DELETED': 2,
'BLOCKNAME': blockName } )
if binds:
self.markBlocksDeleted.execute(binds)
for location in deletableEntries:
chunkSize = 100
numberOfBlocks = 0
blocksToDelete = {}
for dataset in deletableEntries[location]:
blocksToDelete[dataset] = deletableEntries[location][dataset]
numberOfBlocks += len(blocksToDelete[dataset])
if numberOfBlocks > chunkSize:
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
numberOfBlocks = 0
blocksToDelete = {}
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
return
def deleteBlocksPhEDExCalls(self, location, blocksToDelete):
"""
_deleteBlocksPhEDExCalls_
actual PhEDEx calls for block deletion
"""
deletion = PhEDExDeletion(blocksToDelete.keys(), location,
level = 'block',
comments = "WMAgent blocks auto-delete from %s" % location,
blocks = blocksToDelete)
xmlData = XMLDrop.makePhEDExXMLForBlocks(self.dbsUrl,
deletion.getDatasetsAndBlocks())
logging.debug("deleteBlocks XMLData: %s", xmlData)
try:
response = self.phedex.delete(deletion, xmlData)
requestId = response['phedex']['request_created'][0]['id']
# auto-approve deletion request
self.phedex.updateRequest(requestId, 'approve', location)
except HTTPException as ex:
logging.error("PhEDEx block delete/approval failed with HTTPException: %s %s", ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx block delete/approval failed with Exception: %s", str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
binds = []
for dataset in blocksToDelete:
for blockName in blocksToDelete[dataset]:
binds.append( { 'DELETED': 1,
'BLOCKNAME': blockName } )
self.markBlocksDeleted.execute(binds)
return
def subscribeDatasets(self):
"""
_subscribeDatasets_
Poll the database for datasets and subscribe them.
"""
logging.info("Starting subscribeDatasets method")
# Check for completely unsubscribed datasets
unsubscribedDatasets = self.getUnsubscribed.execute()
# Keep a list of subscriptions to tick as subscribed in the database
subscriptionsMade = []
# Create a list of subscriptions as defined by the PhEDEx data structures
subs = SubscriptionList()
# Create the subscription objects and add them to the list
# The list takes care of the sorting internally
for subInfo in unsubscribedDatasets:
site = subInfo['site']
if site not in self.phedexNodes['MSS'] and site not in self.phedexNodes['Disk']:
msg = "Site %s doesn't appear to be valid to PhEDEx, " % site
msg += "skipping subscription: %s" % subInfo['id']
logging.error(msg)
self.sendAlert(7, msg = msg)
continue
# Avoid custodial subscriptions to disk nodes
if site not in self.phedexNodes['MSS']:
subInfo['custodial'] = 'n'
# Avoid auto approval in T1 sites
elif site.startswith("T1"):
subInfo['request_only'] = 'y'
phedexSub = PhEDExSubscription(subInfo['path'], site, subInfo['phedex_group'],
priority = subInfo['priority'],
move = subInfo['move'],
custodial = subInfo['custodial'],
request_only = subInfo['request_only'],
subscriptionId = subInfo['id'])
# Check if the subscription is a duplicate
if phedexSub.matchesExistingSubscription(self.phedex) or \
phedexSub.matchesExistingTransferRequest(self.phedex):
subscriptionsMade.append(subInfo['id'])
continue
# Add it to the list
subs.addSubscription(phedexSub)
# Compact the subscriptions
subs.compact()
for subscription in subs.getSubscriptionList():
xmlData = XMLDrop.makePhEDExXMLForDatasets(self.dbsUrl, subscription.getDatasetPaths())
logging.debug("subscribeDatasets XMLData: %s" , xmlData)
logging.info("Subscribing: %s to %s, with options: Move: %s, Custodial: %s, Request Only: %s",
subscription.getDatasetPaths(),
subscription.getNodes(),
subscription.move,
subscription.custodial,
subscription.request_only)
try:
self.phedex.subscribe(subscription, xmlData)
except HTTPException as ex:
logging.error("PhEDEx dataset subscribe failed with HTTPException: %s %s", ex.status, ex.result)
except Exception as ex:
logging.error("PhEDEx dataset subscribe failed with Exception: %s", str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
subscriptionsMade.extend(subscription.getSubscriptionIds())
# Register the result in DBSBuffer
if subscriptionsMade:
self.markSubscribed.execute(subscriptionsMade)
return
class PhEDExInjectorPoller(BaseWorkerThread):
"""
_PhEDExInjectorPoller_
Poll the DBSBuffer database and inject files as they are created.
"""
def __init__(self, config):
"""
___init___
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.enabled = getattr(config.PhEDExInjector, "enabled", True)
self.dbsUrl = config.DBSInterface.globalDBSUrl
self.phedexGroup = config.PhEDExInjector.phedexGroup
self.pollCounter = 0
self.subFrequency = None
if getattr(config.PhEDExInjector, "subscribeDatasets", False):
pollInterval = config.PhEDExInjector.pollInterval
subInterval = config.PhEDExInjector.subscribeInterval
self.subFrequency = max(1, int(round(subInterval / pollInterval)))
logging.info(
"SubscribeDataset and deleteBlocks will run every %d polling cycles",
self.subFrequency)
# subscribe on first cycle
self.pollCounter = self.subFrequency - 1
# retrieving the node mappings is fickle and can fail quite often
self.phedex = PhEDEx({"endpoint": config.PhEDExInjector.phedexurl},
"json",
dbsUrl=self.dbsUrl)
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(2)
try:
nodeMappings = self.phedex.getNodeMap()
except:
time.sleep(4)
nodeMappings = self.phedex.getNodeMap()
# This will be used to map SE names which are stored in the DBSBuffer to
# PhEDEx node names. The first key will be the "kind" which consists
# of one of the following: MSS, Disk, Buffer. The next key will be the
# SE name.
self.seMap = {}
self.nodeNames = []
for node in nodeMappings["phedex"]["node"]:
if node["kind"] not in self.seMap:
self.seMap[node["kind"]] = {}
logging.info("Adding mapping %s -> %s", node["se"], node["name"])
self.seMap[node["kind"]][node["se"]] = node["name"]
self.nodeNames.append(node["name"])
self.phedexNodes = {'MSS': [], 'Disk': []}
for node in nodeMappings["phedex"]["node"]:
if node["kind"] in ["MSS", "Disk"]:
self.phedexNodes[node["kind"]].append(node["name"])
self.blocksToRecover = []
# X-component configuration is BAD! But it will only be here during the
# Rucio commissioning within WM
self.listTiersToSkip = config.RucioInjector.listTiersToInject
logging.info(
"Component configured to skip data injection for data tiers: %s",
self.listTiersToSkip)
return
def setup(self, parameters):
"""
_setup_
Create DAO Factory and setup some DAO.
"""
myThread = threading.currentThread()
daofactory = DAOFactory(package="WMComponent.RucioInjector.Database",
logger=self.logger,
dbinterface=myThread.dbi)
self.getUninjected = daofactory(classname="GetUninjectedFiles")
self.getMigrated = daofactory(classname="GetMigratedBlocks")
self.getUnsubscribedBlocks = daofactory(
classname="GetUnsubscribedBlocks")
self.setBlockRules = daofactory(classname="SetBlocksRule")
self.findDeletableBlocks = daofactory(classname="GetDeletableBlocks")
self.markBlocksDeleted = daofactory(classname="MarkBlocksDeleted")
self.getUnsubscribed = daofactory(classname="GetUnsubscribedDatasets")
self.markSubscribed = daofactory(classname="MarkDatasetSubscribed")
daofactory = DAOFactory(package="WMComponent.DBS3Buffer",
logger=self.logger,
dbinterface=myThread.dbi)
self.setStatus = daofactory(classname="DBSBufferFiles.SetPhEDExStatus")
self.setBlockClosed = daofactory(classname="SetBlockClosed")
return
@timeFunction
def algorithm(self, parameters):
"""
_algorithm_
Poll the database for uninjected files and attempt to inject them into
PhEDEx.
"""
if not self.enabled:
logging.info(
"PhEDExInjector component is disabled in the configuration, exiting."
)
return
logging.info("Running PhEDEx injector poller algorithm...")
self.pollCounter += 1
try:
if self.blocksToRecover:
logging.info("""PhEDExInjector Recovery:
previous injection call failed,
checking if files were injected to PhEDEx anyway"""
)
self.recoverInjectedFiles()
self.injectFiles()
self.closeBlocks()
if self.pollCounter == self.subFrequency:
self.pollCounter = 0
self.deleteBlocks()
self.subscribeDatasets()
self.subscribeBlocks()
except HTTPException as ex:
if hasattr(ex, "status") and ex.status in [502, 503]:
# then either proxy error or service is unavailable
msg = "Caught HTTPException in PhEDExInjector. Retrying in the next cycle.\n"
msg += str(ex)
logging.error(msg)
else:
msg = "Caught unexpected HTTPException in PhEDExInjector.\n%s" % str(
ex)
logging.exception(msg)
raise
except Exception as ex:
msg = "Caught unexpected exception in PhEDExInjector. Details:\n%s" % str(
ex)
logging.exception(msg)
raise PhEDExInjectorException(msg)
return
def createInjectionSpec(self, injectionData):
"""
_createInjectionSpec_
Transform the data structure returned from the database into an XML
string for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path. Each dataset path will map to a
list of blocks, each block being a dict. The block dicts will have
three keys: name, is-open and files. The files key will be a list of
dicts, each of which have the following keys: lfn, size and checksum.
The following is an example object:
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
"""
injectionSpec = XMLDrop.XMLInjectionSpec(self.dbsUrl)
for datasetPath in injectionData:
datasetSpec = injectionSpec.getDataset(datasetPath)
for fileBlockName, fileBlock in injectionData[
datasetPath].iteritems():
blockSpec = datasetSpec.getFileblock(fileBlockName,
fileBlock["is-open"])
for f in fileBlock["files"]:
blockSpec.addFile(f["lfn"], f["checksum"], f["size"])
return injectionSpec.save()
def createRecoveryFileFormat(self, unInjectedData):
"""
_createRecoveryFileFormat_
Transform the data structure returned from database in to the dict format
for the PhEDEx Data Service. The injectionData parameter must be
a dictionary keyed by dataset path.
unInjectedData format
{"dataset1":
{"block1": {"is-open": "y", "files":
[{"lfn": "lfn1", "size": 10, "checksum": {"cksum": "1234"}},
{"lfn": "lfn2", "size": 20, "checksum": {"cksum": "4321"}}]}}}
returns
[{"block1": set(["lfn1", "lfn2"])}, {"block2": set(["lfn3", "lfn4"])]
"""
blocks = []
for datasetPath in unInjectedData:
for blockName, fileBlock in unInjectedData[datasetPath].items():
newBlock = {blockName: set()}
for fileDict in fileBlock["files"]:
newBlock[blockName].add(fileDict["lfn"])
blocks.append(newBlock)
return blocks
def injectFiles(self):
"""
_injectFiles_
Inject any uninjected files in PhEDEx.
"""
logging.info("Starting injectFiles method")
uninjectedFiles = self.getUninjected.execute()
# filter out datatiers to be processed by RucioInjector
uninjectedFiles = filterDataByTier(uninjectedFiles,
self.listTiersToSkip)
for siteName in uninjectedFiles.keys():
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location is None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
continue
for dataset in uninjectedFiles[siteName]:
injectData = {}
lfnList = []
injectData[dataset] = uninjectedFiles[siteName][dataset]
for block in injectData[dataset]:
for fileInfo in injectData[dataset][block]['files']:
lfnList.append(fileInfo['lfn'])
logging.info("About to inject %d files for block %s",
len(injectData[dataset][block]['files']),
block)
self.injectFilesPhEDExCall(location, injectData, lfnList)
return
def injectFilesPhEDExCall(self, location, injectData, lfnList):
"""
_injectFilesPhEDExCall_
actual PhEDEx call for file injection
"""
xmlData = self.createInjectionSpec(injectData)
logging.debug("injectFiles XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
# HTTPException with status 400 assumed to be duplicate injection
# trigger later block recovery (investigation needed if not the case)
if ex.status == 400:
self.blocksToRecover.extend(
self.createRecoveryFileFormat(injectData))
logging.error(
"PhEDEx file injection failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
msg = "PhEDEx file injection failed with Exception: %s" % str(ex)
logging.exception(msg)
else:
logging.debug("Injection result: %s", injectRes)
if "error" in injectRes:
msg = "Error injecting data %s: %s" % (injectData,
injectRes["error"])
logging.error(msg)
else:
try:
self.setStatus.execute(lfnList, 1)
except Exception as ex:
if 'Deadlock found' in str(
ex) or 'deadlock detected' in str(ex):
logging.error(
"Database deadlock during file status update. Retrying again in the next cycle."
)
self.blocksToRecover.extend(
self.createRecoveryFileFormat(injectData))
else:
msg = "Failed to update file status in the database, reason: %s" % str(
ex)
logging.error(msg)
raise PhEDExInjectorException(msg)
return
def closeBlocks(self):
"""
_closeBlocks_
Close any blocks that have been migrated to global DBS
"""
logging.info("Starting closeBlocks method")
migratedBlocks = self.getMigrated.execute()
# filter out datatiers to be processed by RucioInjector
migratedBlocks = filterDataByTier(migratedBlocks, self.listTiersToSkip)
for siteName in migratedBlocks:
# SE names can be stored in DBSBuffer as that is what is returned in
# the framework job report. We'll try to map the SE name to a
# PhEDEx node name here.
location = None
if siteName in self.nodeNames:
location = siteName
else:
if "Buffer" in self.seMap and siteName in self.seMap["Buffer"]:
location = self.seMap["Buffer"][siteName]
elif "MSS" in self.seMap and siteName in self.seMap["MSS"]:
location = self.seMap["MSS"][siteName]
elif "Disk" in self.seMap and siteName in self.seMap["Disk"]:
location = self.seMap["Disk"][siteName]
if location is None:
msg = "Could not map SE %s to PhEDEx node." % siteName
logging.error(msg)
continue
for dset, blocks in migratedBlocks[siteName].items():
xmlData = self.createInjectionSpec({dset: blocks})
logging.debug("closeBlocks XMLData: %s", xmlData)
try:
injectRes = self.phedex.injectBlocks(location, xmlData)
except HTTPException as ex:
logging.error(
"PhEDEx block close failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
msg = "PhEDEx block close failed with Exception: %s" % str(
ex)
logging.exception(msg)
else:
logging.debug("Block closing result: %s", injectRes)
if "error" in injectRes:
logging.error(
"Failed to close blocks due to: %s, for data: %s",
injectRes["error"], migratedBlocks[siteName][dset])
else:
for blockName in blocks:
logging.info("Block closed in PhEDEx: %s",
blockName)
self.setBlockClosed.execute(blockName)
return
def recoverInjectedFiles(self):
"""
When PhEDEx inject call timed out, run this function.
Since there are 3 min reponse time out in cmsweb, some times
PhEDEx injection call times out even though the call succeeded
In that case run the recovery mode
1. first check whether files which injection status = 0 are in the PhEDEx.
2. if those file exist set the in_phedex status to 1
3. set self.blocksToRecover = []
Run this recovery one block at a time, with too many blocks
the call to the PhEDEx data service on cmsweb can time out
"""
# recover one block at a time
for block in self.blocksToRecover:
injectedFiles = self.phedex.getInjectedFiles(block)
if injectedFiles:
self.setStatus.execute(injectedFiles, 1)
self.blocksToRecover = []
return
def deleteBlocks(self):
"""
_deleteBlocks_
Find deletable blocks, then decide if to delete based on:
Is there an active subscription for dataset or block ?
If yes => set deleted=2
If no => next check
Has transfer to all destinations finished ?
If yes => request block deletion, approve request, set deleted=1
If no => do nothing (check again next cycle)
"""
logging.info("Starting deleteBlocks method")
blockDict = self.findDeletableBlocks.execute(transaction=False)
if not blockDict:
return
### logic to stop doing things to be done by RucioInjector or by DM team
for block in list(blockDict):
if not self._isDataTierAllowed(block):
blockDict.pop(block)
try:
subscriptions = self.phedex.getSubscriptionMapping(
*blockDict.keys())
except:
logging.error(
"Couldn't get subscription info from PhEDEx, retry next cycle")
return
skippableBlocks = []
deletableEntries = {}
for blockName in blockDict:
location = blockDict[blockName]['location']
# should never be triggered, better safe than sorry
if location.endswith('_MSS'):
logging.debug("Location %s for block %s is MSS, skip deletion",
location, blockName)
skippableBlocks.append(blockName)
continue
dataset = blockDict[blockName]['dataset']
sites = blockDict[blockName]['sites']
if blockName in subscriptions and location in subscriptions[
blockName]:
logging.debug("Block %s subscribed to %s, skip deletion",
blockName, location)
binds = {'DELETED': 2, 'BLOCKNAME': blockName}
self.markBlocksDeleted.execute(binds)
else:
blockInfo = []
try:
blockInfo = self.phedex.getReplicaInfoForBlocks(
block=blockName, complete='y')['phedex']['block']
except:
logging.error(
"Couldn't get block info from PhEDEx, retry next cycle"
)
else:
nodes = set()
for entry in blockInfo:
if entry['name'] == blockName:
nodes = set([x['node'] for x in entry['replica']])
if location not in nodes:
logging.debug(
"Block %s not present on %s, mark as deleted",
blockName, location)
binds = {'DELETED': 1, 'BLOCKNAME': blockName}
self.markBlocksDeleted.execute(binds)
elif sites.issubset(nodes):
logging.debug(
"Deleting block %s from %s since it is fully transfered",
blockName, location)
if location not in deletableEntries:
deletableEntries[location] = {}
if dataset not in deletableEntries[location]:
deletableEntries[location][dataset] = set()
deletableEntries[location][dataset].add(blockName)
binds = []
for blockName in skippableBlocks:
binds.append({'DELETED': 2, 'BLOCKNAME': blockName})
if binds:
self.markBlocksDeleted.execute(binds)
for location in deletableEntries:
chunkSize = 100
numberOfBlocks = 0
blocksToDelete = {}
for dataset in deletableEntries[location]:
blocksToDelete[dataset] = deletableEntries[location][dataset]
numberOfBlocks += len(blocksToDelete[dataset])
if numberOfBlocks > chunkSize:
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
numberOfBlocks = 0
blocksToDelete = {}
self.deleteBlocksPhEDExCalls(location, blocksToDelete)
return
def deleteBlocksPhEDExCalls(self, location, blocksToDelete):
"""
_deleteBlocksPhEDExCalls_
actual PhEDEx calls for block deletion
"""
deletion = PhEDExDeletion(
blocksToDelete.keys(),
location,
level='block',
comments="WMAgent blocks auto-delete from %s" % location,
blocks=blocksToDelete)
try:
response = self.phedex.delete(deletion)
requestId = response['phedex']['request_created'][0]['id']
# auto-approve deletion request
self.phedex.updateRequest(requestId, 'approve', location)
except HTTPException as ex:
logging.error(
"PhEDEx block delete/approval failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error(
"PhEDEx block delete/approval failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
binds = []
for dataset in blocksToDelete:
for blockName in blocksToDelete[dataset]:
binds.append({'DELETED': 1, 'BLOCKNAME': blockName})
self.markBlocksDeleted.execute(binds)
return
def _isDataTierAllowed(self, dataName):
"""
Check whether data belongs to an allowed datatier to
be handled by this component (either to inject or to
subscribe into PhEDEx)
:param dataName: string with the block or the dataset name
:return: boolean, True if the tier is allowed, False otherwise
"""
endTier = dataName.rsplit('/', 1)[1]
endTier = endTier.split('#')[0] if '#' in endTier else endTier
if endTier in self.listTiersToSkip:
logging.debug(
"Skipping data: %s because it's listed in the tiers to skip",
dataName)
return False
return True
def subscribeDatasets(self):
"""
_subscribeDatasets_
Poll the database for datasets and subscribe them.
"""
logging.info("Starting subscribeDatasets method")
# Check for completely unsubscribed datasets
unsubscribedDatasets = self.getUnsubscribed.execute()
# Keep a list of subscriptions to tick as subscribed in the database
subscriptionsMade = []
# Create a list of subscriptions as defined by the PhEDEx data structures
subs = SubscriptionList()
# Create the subscription objects and add them to the list
# The list takes care of the sorting internally
for subInfo in unsubscribedDatasets:
### logic to stop doing things to be done by RucioInjector or by DM team
if not self._isDataTierAllowed(subInfo['path']):
continue
site = subInfo['site']
if site not in self.phedexNodes[
'MSS'] and site not in self.phedexNodes['Disk']:
msg = "Site %s doesn't appear to be valid to PhEDEx, " % site
msg += "skipping subscription: %s" % subInfo['id']
logging.error(msg)
continue
# Avoid custodial subscriptions to disk nodes
if site not in self.phedexNodes['MSS']:
subInfo['custodial'] = 'n'
# Avoid auto approval in T1 sites
elif site.startswith("T1"):
subInfo['request_only'] = 'y'
phedexSub = PhEDExSubscription(
subInfo['path'],
site,
subInfo['phedex_group'],
priority=subInfo['priority'],
move=subInfo['move'],
custodial=subInfo['custodial'],
request_only=subInfo['request_only'],
subscriptionId=subInfo['id'])
# Check if the subscription is a duplicate
if phedexSub.matchesExistingSubscription(self.phedex) or \
phedexSub.matchesExistingTransferRequest(self.phedex):
subscriptionsMade.append(subInfo['id'])
continue
# Add it to the list
subs.addSubscription(phedexSub)
# Compact the subscriptions
subs.compact()
for subscription in subs.getSubscriptionList():
logging.info(
"Subscribing: %s to %s, with options: Move: %s, Custodial: %s, Request Only: %s",
subscription.getDatasetPaths(), subscription.getNodes(),
subscription.move, subscription.custodial,
subscription.request_only)
try:
self.phedex.subscribe(subscription)
except HTTPException as ex:
logging.error(
"PhEDEx dataset subscribe failed with HTTPException: %s %s",
ex.status, ex.result)
except Exception as ex:
logging.error(
"PhEDEx dataset subscribe failed with Exception: %s",
str(ex))
logging.debug("Traceback: %s", str(traceback.format_exc()))
else:
subscriptionsMade.extend(subscription.getSubscriptionIds())
# Register the result in DBSBuffer
if subscriptionsMade:
self.markSubscribed.execute(subscriptionsMade)
return
def subscribeBlocks(self):
"""
_subscribeBlocks_
Poll the database and subscribe blocks not yet subscribed.
"""
logging.info("Starting subscribeBlocks method")
unsubBlocks = self.getUnsubscribedBlocks.execute()
# now organize those by location in order to minimize phedex requests
# also remove blocks that this component is meant to skip
unsubBlocks = self.organizeBlocksByLocation(unsubBlocks)
for location, blockDict in unsubBlocks.items():
phedexSub = PhEDExSubscription(blockDict.keys(),
location,
self.phedexGroup,
blocks=blockDict,
level="block",
priority="normal",
move="n",
custodial="n",
request_only="n",
comments="WMAgent production site")
try:
res = self.phedex.subscribe(phedexSub)
transferId = res['phedex']['request_created'][0]['id']
logging.info(
"Subscribed %d blocks for %d datasets, to location: %s, under request ID: %s",
len(phedexSub.getBlocks()),
len(phedexSub.getDatasetPaths()), phedexSub.getNodes(),
transferId)
except HTTPException as ex:
logging.error(
"PhEDEx block subscription failed with HTTPException: %s %s",
ex.status, ex.result)
logging.error("The subscription object was: %s",
str(phedexSub))
except Exception as ex:
logging.exception(
"PhEDEx block subscription failed with Exception: %s",
str(ex))
else:
binds = []
for blockname in phedexSub.getBlocks():
binds.append({
'RULE_ID': str(transferId),
'BLOCKNAME': blockname
})
self.setBlockRules.execute(binds)
return
def organizeBlocksByLocation(self, blocksLocation):
"""
Given a list of dictionaries (with block name and location). Organize
those blocks per location to make phedex subscription calls more
efficient.
Also drops blocks that we cannot subscribe, and check for valid
phedex node names.
:param blocksLocation: list of dictionaries
:return: a dict of dictionaries, such as:
{"locationA": {"datasetA": ["blockA", "blockB", ...],
"datasetB": ["blockA", "blockB", ...]
},
"locationB": {"datasetA": ["blockA"],
...
"""
dictByLocation = {}
for item in blocksLocation:
### logic to stop doing things to be done by RucioInjector or by DM team
if not self._isDataTierAllowed(item['blockname']):
continue
site = item['pnn']
if site not in self.phedexNodes[
'MSS'] and site not in self.phedexNodes['Disk']:
msg = "Site %s doesn't appear to be valid to PhEDEx, " % site
msg += "skipping block subscription for: %s" % item['blockname']
logging.error(msg)
continue
dictByLocation.setdefault(site, {})
dsetName = item['blockname'].split("#")[0]
dictByLocation[site].setdefault(dsetName, [])
dictByLocation[site][dsetName].append(item['blockname'])
return dictByLocation
