def __init__(self, config): # queue url used in WorkQueueManager self.thisAgentUrl = "http://" + config.Agent.hostName + ":5984" self.globalBackend = WorkQueueBackend(config.WorkloadSummary.couchurl) self.localBackend = WorkQueueBackend(config.WorkQueueManager.couchurl) self.dbsUtil = DBSBufferUtil() self.condorAPI = PyCondorAPI()
class DrainStatusAPI(object): """ Provides methods for querying dbs and condor for drain statistics """ def __init__(self): self.dbsUtil = DBSBufferUtil() self.condorAPI = PyCondorAPI() def collectDrainInfo(self): """ Call methods to check the drain status """ results = {} results['workflows_completed'] = self.checkWorkflows() # if workflows are completed, collect additional drain statistics if results['workflows_completed']: results['upload_status'] = self.checkFileUploadStatus() results['condor_status'] = self.checkCondorStates() return results def checkWorkflows(self): """ Check to see if all workflows have a 'completed' status """ results = self.dbsUtil.isAllWorkflowCompleted() return results def checkCondorStates(self): """ Check idle and running jobs in Condor """ results = {} queries = [["1", "idle"], ["2", "running"]] for query in queries: jobs = self.condorAPI.getCondorJobs("JobStatus=="+query[0], []) # if there is an error, report it instead of the length of an empty list if jobs is None: results[query[1]] = "unknown (schedd query error)" else: results[query[1]] = len(jobs) return results def checkFileUploadStatus(self): """ Check file upload status: Blocks open in DBS Files not uploaded in DBS Files not uploaded to Phedex """ results = {} results['dbs_open_blocks'] = self.dbsUtil.countOpenBlocks() results['dbs_notuploaded'] = self.dbsUtil.countFilesByStatus(status="NOTUPLOADED") results['phedex_notuploaded'] = self.dbsUtil.countPhedexNotUploaded() return results
def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() self.daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'copyBlock', False) self.copyPath = getattr(self.config.DBS3Upload, 'copyBlockPath', '/data/mnorman/block.json') self.timeoutWaiver = 1 return
def setup(self, parameters): """ set db connection(couchdb, wmbs) to prepare to gather information """ # set the connection to local queue if not hasattr(self.config, "Tier0Feeder"): self.localQueue = WorkQueueService( self.config.AnalyticsDataCollector.localQueueURL) # set the connection for local couchDB call self.localCouchDB = LocalCouchDBData( self.config.AnalyticsDataCollector.localCouchURL, self.config.JobStateMachine.summaryStatsDBName, self.summaryLevel) # interface to WMBS/BossAir db myThread = threading.currentThread() # set wmagent db data self.wmagentDB = WMAgentDBData(self.summaryLevel, myThread.dbi, myThread.logger) # set the connection for local couchDB call self.localSummaryCouchDB = WMStatsWriter( self.config.AnalyticsDataCollector.localWMStatsURL, appName="WMStatsAgent") # use local db for tier0 if hasattr(self.config, "Tier0Feeder"): centralRequestCouchDBURL = self.config.AnalyticsDataCollector.localT0RequestDBURL else: centralRequestCouchDBURL = self.config.AnalyticsDataCollector.centralRequestDBURL self.centralRequestCouchDB = RequestDBWriter( centralRequestCouchDBURL, couchapp=self.config.AnalyticsDataCollector.RequestCouchApp) self.centralWMStatsCouchDB = WMStatsWriter( self.config.AnalyticsDataCollector.centralWMStatsURL) #TODO: change the config to hold couch url self.localCouchServer = CouchMonitor( self.config.JobStateMachine.couchurl) self.dbsBufferUtil = DBSBufferUtil() if self.pluginName is not None: pluginFactory = WMFactory( "plugins", "WMComponent.AnalyticsDataCollector.Plugins") self.plugin = pluginFactory.loadObject(classname=self.pluginName)
def setUp(self): """ _setUp_ Setup the database and logging connection. Create some DBSBuffer tables and fake data for testing """ self.testInit = TestInit(__file__) self.testInit.setLogging() self.testInit.setDatabaseConnection() self.testInit.setSchema( customModules=["WMComponent.DBS3Buffer", "WMCore.WMBS"], useDefault=False) myThread = threading.currentThread() self.dbsbufferFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.wmbsFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi) self.dbsUtil = DBSBufferUtil() # Create two test dbsbuffer workflows insertWorkflow = self.dbsbufferFactory(classname="InsertWorkflow") insertWorkflow.execute("Test1", "Task1", 0, 0, 0, 0) insertWorkflow.execute("Test2", "Task2", 0, 0, 0, 0) # Update one workflow to "completed" state updateWorkflow = self.dbsbufferFactory( classname="UpdateWorkflowsToCompleted") updateWorkflow.execute(["Test1"]) # Create a test wmbs workflow testWorkflow = Workflow(spec="somespec.xml", owner="Erik", name="Test1", task="Task1") testWorkflow.create() # Create a test dbsbuffer file self.createTestFiles()
def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() self.daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'dumpBlock', False) self.copyPath = os.path.join(getattr(self.config.DBS3Upload, 'componentDir', '/data/srv/'), 'dbsuploader_block.json') self.timeoutWaiver = 1 return
def __init__(self, config, dbsconfig=None): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") #myThread = threading.currentThread() BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.maxBlockFiles = self.config.DBSUpload.DBSBlockMaxFiles self.maxBlockTime = self.config.DBSUpload.DBSBlockMaxTime self.maxBlockSize = self.config.DBSUpload.DBSBlockMaxSize self.dbsUrl = self.config.DBSUpload.dbsUrl self.dbsUtil = DBSBufferUtil() self.pool = [] self.input = None self.result = None self.nProc = getattr(self.config.DBSUpload, 'nProcesses', 4) self.wait = getattr(self.config.DBSUpload, 'dbsWaitTime', 1) self.nTries = getattr(self.config.DBSUpload, 'dbsNTries', 300) self.dbs3UploadOnly = getattr(self.config.DBSUpload, 'dbs3UploadOnly', False) self.physicsGroup = getattr(self.config.DBSUpload, 'physicsGroup', 'DBS3Test') self.blockCount = 0 # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.dasCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBSUpload, 'copyBlock', False) self.copyPath = getattr(self.config.DBSUpload, 'copyBlockPath', '/data/mnorman/block.json') return
def __init__(self, config, dbsconfig = None): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl self.dbsUtil = DBSBufferUtil() self.pool = [] self.blocksToCheck = [] self.input = None self.result = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.dbs3UploadOnly = getattr(self.config.DBS3Upload, "dbs3UploadOnly", False) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url = self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.dasCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'copyBlock', False) self.copyPath = getattr(self.config.DBS3Upload, 'copyBlockPath', '/data/mnorman/block.json') self.timeoutWaiver = 1 return
def setup(self, parameters): """ set db connection(couchdb, wmbs) to prepare to gather information """ # set the connection to local queue if not hasattr(self.config, "Tier0Feeder"): self.localQueue = WorkQueueService(self.config.AnalyticsDataCollector.localQueueURL) # set the connection for local couchDB call self.localCouchDB = LocalCouchDBData(self.config.AnalyticsDataCollector.localCouchURL, self.config.JobStateMachine.summaryStatsDBName, self.summaryLevel) # interface to WMBS/BossAir db myThread = threading.currentThread() # set wmagent db data self.wmagentDB = WMAgentDBData(self.summaryLevel, myThread.dbi, myThread.logger) # set the connection for local couchDB call self.localSummaryCouchDB = WMStatsWriter(self.config.AnalyticsDataCollector.localWMStatsURL, appName="WMStatsAgent") # use local db for tier0 if hasattr(self.config, "Tier0Feeder"): centralRequestCouchDBURL = self.config.AnalyticsDataCollector.localT0RequestDBURL else: centralRequestCouchDBURL = self.config.AnalyticsDataCollector.centralRequestDBURL self.centralRequestCouchDB = RequestDBWriter(centralRequestCouchDBURL, couchapp=self.config.AnalyticsDataCollector.RequestCouchApp) self.centralWMStatsCouchDB = WMStatsWriter(self.config.General.centralWMStatsURL) #TODO: change the config to hold couch url self.localCouchServer = CouchMonitor(self.config.JobStateMachine.couchurl) self.dbsBufferUtil = DBSBufferUtil() if self.pluginName is not None: pluginFactory = WMFactory("plugins", "WMComponent.AnalyticsDataCollector.Plugins") self.plugin = pluginFactory.loadObject(classname=self.pluginName)
def setUp(self): """ _setUp_ Setup the database and logging connection. Create some DBSBuffer tables and fake data for testing """ self.testInit = TestInit(__file__) self.testInit.setLogging() self.testInit.setDatabaseConnection() self.testInit.setSchema(customModules=["WMComponent.DBS3Buffer", "WMCore.WMBS"], useDefault=False) myThread = threading.currentThread() self.dbsbufferFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.wmbsFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi) self.dbsUtil = DBSBufferUtil() # Create two test dbsbuffer workflows insertWorkflow = self.dbsbufferFactory(classname="InsertWorkflow") insertWorkflow.execute("Test1", "Task1", 0, 0, 0, 0) insertWorkflow.execute("Test2", "Task2", 0, 0, 0, 0) # Update one workflow to "completed" state updateWorkflow = self.dbsbufferFactory(classname="UpdateWorkflowsToCompleted") updateWorkflow.execute(["Test1"]) # Create a test wmbs workflow testWorkflow = Workflow(spec="somespec.xml", owner="Erik", name="Test1", task="Task1") testWorkflow.create() # Create a test dbsbuffer file self.createTestFiles()
class DrainStatusAPI(object): """ Provides methods for querying dbs and condor for drain statistics """ def __init__(self, config): # queue url used in WorkQueueManager self.thisAgentUrl = "http://" + config.Agent.hostName + ":5984" self.globalBackend = WorkQueueBackend(config.WorkloadSummary.couchurl) self.localBackend = WorkQueueBackend(config.WorkQueueManager.couchurl) self.dbsUtil = DBSBufferUtil() self.condorAPI = PyCondorAPI() self.condorStates = ("Running", "Idle") def collectDrainInfo(self): """ Call methods to check the drain status """ results = {} results['workflows_completed'] = self.checkWorkflows() # if workflows are completed, collect additional drain statistics if results['workflows_completed']: results['upload_status'] = self.checkFileUploadStatus() results['condor_status'] = self.checkCondorStates() results['local_wq_status'] = self.checkLocalWQStatus( dbname="workqueue") results['local_wqinbox_status'] = self.checkLocalWQStatus( dbname="workqueue_inbox") results['global_wq_status'] = self.checkGlobalWQStatus() return results def checkWorkflows(self): """ Check to see if all workflows have a 'completed' status """ results = self.dbsUtil.isAllWorkflowCompleted() return results def checkCondorStates(self): """ Check idle and running jobs in Condor """ results = {} jobs = self.condorAPI.getCondorJobsSummary() for state in self.condorStates: # if there is an error, report it instead of the length of an empty list if not jobs: results[state.lower()] = None else: results[state.lower()] = int(jobs[0].get(state)) return results def checkFileUploadStatus(self): """ Check file upload status: Blocks open in DBS Files not uploaded in DBS Files not uploaded to Phedex """ results = {} results['dbs_open_blocks'] = self.dbsUtil.countOpenBlocks() results['dbs_notuploaded'] = self.dbsUtil.countFilesByStatus( status="NOTUPLOADED") results['phedex_notuploaded'] = self.dbsUtil.countPhedexNotUploaded() return results def checkLocalWQStatus(self, dbname): """ Query local WorkQueue workqueue/workqueue_inbox database to see whether there are any active elements in this agent. """ results = {} for st in ('Available', 'Negotiating', 'Acquired', 'Running'): if dbname == "workqueue": elements = self.localBackend.getElements(status=st, returnIdOnly=True) else: elements = self.localBackend.getInboxElements( status=st, returnIdOnly=True) results[st] = len(elements) return results def checkGlobalWQStatus(self): """ Query Global WorkQueue workqueue database to see whether there are any active elements set to this agent. """ results = {} for st in ("Acquired", "Running"): elements = self.globalBackend.getElements( status=st, returnIdOnly=True, ChildQueueUrl=self.thisAgentUrl) results[st] = len(elements) return results
def testBulkLoad(self): """ _testBulkLoad_ Can we load in bulk? """ addToBuffer = DBSBufferUtil() bulkLoad = self.daoFactory(classname = "DBSBufferFiles.LoadBulkFilesByID") testFileChildA = DBSBufferFile(lfn = "/this/is/a/child/lfnA", size = 1024, events = 20) testFileChildA.setAlgorithm(appName = "cmsRun", appVer = "CMSSW_2_1_8", appFam = "RECO", psetHash = "GIBBERISH", configContent = "MOREGIBBERISH") testFileChildA.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildB = DBSBufferFile(lfn = "/this/is/a/child/lfnB", size = 1024, events = 20) testFileChildB.setAlgorithm(appName = "cmsRun", appVer = "CMSSW_2_1_8", appFam = "RECO", psetHash = "GIBBERISH", configContent = "MOREGIBBERISH") testFileChildB.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildC = DBSBufferFile(lfn = "/this/is/a/child/lfnC", size = 1024, events = 20) testFileChildC.setAlgorithm(appName = "cmsRun", appVer = "CMSSW_2_1_8", appFam = "RECO", psetHash = "GIBBERISH", configContent = "MOREGIBBERISH") testFileChildC.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildA.create() testFileChildB.create() testFileChildC.create() testFileChildA.setLocation(["se1.fnal.gov", "se1.cern.ch"]) testFileChildB.setLocation(["se1.fnal.gov", "se1.cern.ch"]) testFileChildC.setLocation(["se1.fnal.gov", "se1.cern.ch"]) runSet = set() runSet.add(Run( 1, *[45])) runSet.add(Run( 2, *[67, 68])) testFileChildA.addRunSet(runSet) testFileChildB.addRunSet(runSet) testFileChildC.addRunSet(runSet) testFileChildA.save() testFileChildB.save() testFileChildC.save() setCksumAction = self.daoFactory(classname = "DBSBufferFiles.AddChecksumByLFN") binds = [{'lfn': "/this/is/a/child/lfnA", 'cktype': 'adler32', 'cksum': 201}, {'lfn': "/this/is/a/child/lfnA", 'cktype': 'cksum', 'cksum': 101}, {'lfn': "/this/is/a/child/lfnB", 'cktype': 'adler32', 'cksum': 201}, {'lfn': "/this/is/a/child/lfnB", 'cktype': 'cksum', 'cksum': 101}, {'lfn': "/this/is/a/child/lfnC", 'cktype': 'adler32', 'cksum': 201}, {'lfn': "/this/is/a/child/lfnC", 'cktype': 'cksum', 'cksum': 101}] setCksumAction.execute(bulkList = binds) testFile = DBSBufferFile(lfn = "/this/is/a/lfn", size = 1024, events = 10) testFile.setAlgorithm(appName = "cmsRun", appVer = "CMSSW_2_1_8", appFam = "RECO", psetHash = "GIBBERISH", configContent = "MOREGIBBERISH") testFile.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFile.create() testFileChildA.addParents([testFile["lfn"]]) testFileChildB.addParents([testFile["lfn"]]) testFileChildC.addParents([testFile["lfn"]]) binds = [{'id': testFileChildA.exists()}, {'id': testFileChildB.exists()}, {'id': testFileChildC.exists()}] listOfFiles = addToBuffer.loadDBSBufferFilesBulk(fileObjs = binds) #print listOfFiles compareList = ['locations', 'psetHash', 'configContent', 'appName', 'appVer', 'appFam', 'events', 'datasetPath', 'runs'] for f in listOfFiles: self.assertTrue(f['lfn'] in ["/this/is/a/child/lfnA", "/this/is/a/child/lfnB", "/this/is/a/child/lfnC"], "Unknown file in loaded results") self.assertEqual(f['checksums'], {'adler32': '201', 'cksum': '101'}) for parent in f['parents']: self.assertEqual(parent['lfn'], testFile['lfn']) for key in compareList: self.assertEqual(f[key], testFileChildA[key])
class AnalyticsPoller(BaseWorkerThread): """ Gether the summary data for request (workflow) from local queue, local job couchdb, wmbs/boss air and populate summary db for monitoring """ def __init__(self, config): """ initialize properties specified from config """ BaseWorkerThread.__init__(self) # set the workqueue service for REST call self.config = config # need to get campaign, user, owner info self.agentInfo = initAgentInfo(self.config) self.summaryLevel = (config.AnalyticsDataCollector.summaryLevel).lower() self.pluginName = getattr(config.AnalyticsDataCollector, "pluginName", None) self.plugin = None def setup(self, parameters): """ set db connection(couchdb, wmbs) to prepare to gather information """ # set the connection to local queue if not hasattr(self.config, "Tier0Feeder"): self.localQueue = WorkQueueService(self.config.AnalyticsDataCollector.localQueueURL) # set the connection for local couchDB call self.localCouchDB = LocalCouchDBData(self.config.AnalyticsDataCollector.localCouchURL, self.config.JobStateMachine.summaryStatsDBName, self.summaryLevel) # interface to WMBS/BossAir db myThread = threading.currentThread() # set wmagent db data self.wmagentDB = WMAgentDBData(self.summaryLevel, myThread.dbi, myThread.logger) # set the connection for local couchDB call self.localSummaryCouchDB = WMStatsWriter(self.config.AnalyticsDataCollector.localWMStatsURL, appName="WMStatsAgent") # use local db for tier0 if hasattr(self.config, "Tier0Feeder"): centralRequestCouchDBURL = self.config.AnalyticsDataCollector.localT0RequestDBURL else: centralRequestCouchDBURL = self.config.AnalyticsDataCollector.centralRequestDBURL self.centralRequestCouchDB = RequestDBWriter(centralRequestCouchDBURL, couchapp=self.config.AnalyticsDataCollector.RequestCouchApp) self.centralWMStatsCouchDB = WMStatsWriter(self.config.General.centralWMStatsURL) #TODO: change the config to hold couch url self.localCouchServer = CouchMonitor(self.config.JobStateMachine.couchurl) self.dbsBufferUtil = DBSBufferUtil() if self.pluginName is not None: pluginFactory = WMFactory("plugins", "WMComponent.AnalyticsDataCollector.Plugins") self.plugin = pluginFactory.loadObject(classname=self.pluginName) @timeFunction def algorithm(self, parameters): """ get information from wmbs, workqueue and local couch """ try: # jobs per request info logging.info("Getting Job Couch Data ...") jobInfoFromCouch = self.localCouchDB.getJobSummaryByWorkflowAndSite() # fwjr per request info logging.info("Getting FWJRJob Couch Data ...") fwjrInfoFromCouch = self.localCouchDB.getJobPerformanceByTaskAndSiteFromSummaryDB() skippedInfoFromCouch = self.localCouchDB.getSkippedFilesSummaryByWorkflow() logging.info("Getting Batch Job Data ...") batchJobInfo = self.wmagentDB.getBatchJobInfo() logging.info("Getting Finished Task Data ...") finishedTasks = self.wmagentDB.getFinishedSubscriptionByTask() logging.info("Getting DBS PhEDEx upload status ...") completedWfs = self.dbsBufferUtil.getPhEDExDBSStatusForCompletedWorkflows(summary=True) # get the data from local workqueue: # request name, input dataset, inWMBS, inQueue logging.info("Getting Local Queue Data ...") localQInfo = {} if not hasattr(self.config, "Tier0Feeder"): localQInfo = self.localQueue.getAnalyticsData() else: logging.debug("Tier-0 instance, not checking WorkQueue") # combine all the data from 3 sources logging.info("""Combining data from Job Couch(%s), FWJR(%s), WorkflowsWithSkippedFile(%s), Batch Job(%s), Finished Tasks(%s), Local Queue(%s) Completed workflows(%s).. ...""", len(jobInfoFromCouch), len(fwjrInfoFromCouch), len(skippedInfoFromCouch), len(batchJobInfo), len(finishedTasks), len(localQInfo), len(completedWfs)) tempCombinedData = combineAnalyticsData(jobInfoFromCouch, batchJobInfo) tempCombinedData2 = combineAnalyticsData(tempCombinedData, localQInfo) combinedRequests = combineAnalyticsData(tempCombinedData2, completedWfs) # set the uploadTime - should be the same for all docs uploadTime = int(time.time()) logging.info("%s requests Data combined,\n uploading request data...", len(combinedRequests)) requestDocs = convertToRequestCouchDoc(combinedRequests, fwjrInfoFromCouch, finishedTasks, skippedInfoFromCouch, self.agentInfo, uploadTime, self.summaryLevel) if self.plugin != None: self.plugin(requestDocs, self.localSummaryCouchDB, self.centralRequestCouchDB) existingDocs = self.centralWMStatsCouchDB.getAllAgentRequestRevByID(self.agentInfo["agent_url"]) self.centralWMStatsCouchDB.bulkUpdateData(requestDocs, existingDocs) logging.info("Request data upload success\n %s request, \nsleep for next cycle", len(requestDocs)) self.centralWMStatsCouchDB.updateAgentInfoInPlace(self.agentInfo["agent_url"], {"data_last_update": uploadTime, "data_error": "ok"}) except Exception as ex: msg = str(ex) logging.exception("Error occurred, will retry later: %s", msg) try: self.centralWMStatsCouchDB.updateAgentInfoInPlace(self.agentInfo["agent_url"], {"data_error": msg}) except: logging.error("upload Agent Info to central couch failed")
def testDualUpload(self): """ _testDualUpload_ Verify that the dual upload mode works correctly. """ self.dbsApi = DbsApi(url = self.dbsUrl) config = self.getConfig(dbs3UploadOnly = True) dbsUploader = DBSUploadPoller(config = config) dbsUtil = DBSBufferUtil() # First test verifies that uploader will poll and then not do anything # as the database is empty. dbsUploader.algorithm() acqEra = "Summer%s" % (int(time.time())) parentFiles = self.createParentFiles(acqEra) (moreParentFiles, childFiles) = \ self.createFilesWithChildren(parentFiles, acqEra) allFiles = parentFiles + moreParentFiles allBlocks = [] for i in range(4): blockName = parentFiles[0]["datasetPath"] + "#" + makeUUID() dbsBlock = DBSBlock(blockName, "malpaquet", 1) dbsBlock.status = "Open" dbsUtil.createBlocks([dbsBlock]) for file in allFiles[i * 5 : (i * 5) + 5]: dbsBlock.addFile(file) dbsUtil.setBlockFiles({"block": blockName, "filelfn": file["lfn"]}) if i < 2: dbsBlock.status = "InDBS" dbsUtil.updateBlocks([dbsBlock]) dbsUtil.updateFileStatus([dbsBlock], "InDBS") allBlocks.append(dbsBlock) blockName = childFiles[0]["datasetPath"] + "#" + makeUUID() dbsBlock = DBSBlock(blockName, "malpaquet", 1) dbsBlock.status = "InDBS" dbsUtil.createBlocks([dbsBlock]) for file in childFiles: dbsBlock.addFile(file) dbsUtil.setBlockFiles({"block": blockName, "filelfn": file["lfn"]}) dbsUtil.updateFileStatus([dbsBlock], "InDBS") dbsUploader.algorithm() time.sleep(5) dbsUploader.algorithm() time.sleep(5) self.verifyData(parentFiles[0]["datasetPath"], parentFiles) # Change the status of the rest of the parent blocks so we can upload # them and the children. for dbsBlock in allBlocks: dbsBlock.status = "InDBS" dbsUtil.updateBlocks([dbsBlock]) dbsUploader.algorithm() time.sleep(5) self.verifyData(parentFiles[0]["datasetPath"], parentFiles + moreParentFiles) # Run the uploader one more time to upload the children. dbsUploader.algorithm() time.sleep(5) self.verifyData(childFiles[0]["datasetPath"], childFiles) return
def testCloseSettingsPerWorkflow(self): """ _testCloseSettingsPerWorkflow_ Test the block closing mechanics in the DBS3 uploader, this uses a fake dbs api to avoid reliance on external services. """ # Signal trapExit that we are a friend os.environ["DONT_TRAP_EXIT"] = "True" try: # Monkey patch the imports of DbsApi from WMComponent.DBS3Buffer import DBSUploadPoller as MockDBSUploadPoller MockDBSUploadPoller.DbsApi = MockDbsApi # Set the poller and the dbsUtil for verification myThread = threading.currentThread() (_, dbsFilePath) = mkstemp(dir = self.testDir) self.dbsUrl = dbsFilePath config = self.getConfig() dbsUploader = MockDBSUploadPoller.DBSUploadPoller(config = config) dbsUtil = DBSBufferUtil() # First test is event based limits and timeout with no new files. # Set the files and workflow acqEra = "TropicalSeason%s" % (int(time.time())) workflowName = 'TestWorkload%s' % (int(time.time())) taskPath = '/%s/TestProcessing' % workflowName self.injectWorkflow(workflowName, taskPath, MaxWaitTime = 2, MaxFiles = 100, MaxEvents = 150) self.createParentFiles(acqEra, nFiles = 20, workflowName = workflowName, taskPath = taskPath) # The algorithm needs to be run twice. On the first iteration it will # create all the blocks and upload one with less than 150 events. # On the second iteration the second block is uploaded. dbsUploader.algorithm() dbsUploader.checkBlocks() openBlocks = dbsUtil.findOpenBlocks() self.assertEqual(len(openBlocks), 1) globalFiles = myThread.dbi.processData("SELECT id FROM dbsbuffer_file WHERE status = 'InDBS'")[0].fetchall() notUploadedFiles = myThread.dbi.processData("SELECT id FROM dbsbuffer_file WHERE status = 'NOTUPLOADED'")[0].fetchall() self.assertEqual(len(globalFiles), 14) self.assertEqual(len(notUploadedFiles), 6) # Check the fake DBS for data fakeDBS = open(self.dbsUrl, 'r') fakeDBSInfo = json.load(fakeDBS) fakeDBS.close() self.assertEqual(len(fakeDBSInfo), 2) for block in fakeDBSInfo: self.assertTrue('block_events' not in block['block']) self.assertEqual(block['block']['file_count'], 7) self.assertEqual(block['block']['open_for_writing'], 0) self.assertTrue('close_settings' not in block) time.sleep(3) dbsUploader.algorithm() dbsUploader.checkBlocks() openBlocks = dbsUtil.findOpenBlocks() self.assertEqual(len(openBlocks), 0) fakeDBS = open(self.dbsUrl, 'r') fakeDBSInfo = json.load(fakeDBS) fakeDBS.close() self.assertEqual(len(fakeDBSInfo), 3) for block in fakeDBSInfo: if block['block']['file_count'] != 6: self.assertEqual(block['block']['file_count'], 7) self.assertTrue('block_events' not in block['block']) self.assertEqual(block['block']['open_for_writing'], 0) self.assertTrue('close_settings' not in block) # Now check the limit by size and timeout with new files acqEra = "TropicalSeason%s" % (int(time.time())) workflowName = 'TestWorkload%s' % (int(time.time())) taskPath = '/%s/TestProcessing' % workflowName self.injectWorkflow(workflowName, taskPath, MaxWaitTime = 2, MaxFiles = 5, MaxEvents = 200000000) self.createParentFiles(acqEra, nFiles = 16, workflowName = workflowName, taskPath = taskPath) dbsUploader.algorithm() dbsUploader.checkBlocks() openBlocks = dbsUtil.findOpenBlocks() self.assertEqual(len(openBlocks), 1) fakeDBS = open(self.dbsUrl, 'r') fakeDBSInfo = json.load(fakeDBS) fakeDBS.close() self.assertEqual(len(fakeDBSInfo), 6) for block in fakeDBSInfo: if acqEra in block['block']['block_name']: self.assertEqual(block['block']['file_count'], 5) self.assertTrue('block_events' not in block['block']) self.assertTrue('close_settings' not in block) self.assertEqual(block['block']['open_for_writing'], 0) # Put more files, they will go into the same block and then it will be closed # after timeout time.sleep(3) self.createParentFiles(acqEra, nFiles = 3, workflowName = workflowName, taskPath = taskPath) dbsUploader.algorithm() dbsUploader.checkBlocks() openBlocks = dbsUtil.findOpenBlocks() self.assertEqual(len(openBlocks), 0) fakeDBS = open(self.dbsUrl, 'r') fakeDBSInfo = json.load(fakeDBS) fakeDBS.close() self.assertEqual(len(fakeDBSInfo), 7) for block in fakeDBSInfo: if acqEra in block['block']['block_name']: if block['block']['file_count'] < 5: self.assertEqual(block['block']['file_count'], 4) else: self.assertEqual(block['block']['file_count'], 5) self.assertTrue('block_events' not in block['block']) self.assertEqual(block['block']['open_for_writing'], 0) self.assertTrue('close_settings' not in block) # Finally test size limits acqEra = "TropicalSeason%s" % (int(time.time())) workflowName = 'TestWorkload%s' % (int(time.time())) taskPath = '/%s/TestProcessing' % workflowName self.injectWorkflow(workflowName, taskPath, MaxWaitTime = 1, MaxFiles = 500, MaxEvents = 200000000, MaxSize = 2048) self.createParentFiles(acqEra, nFiles = 7, workflowName = workflowName, taskPath = taskPath) dbsUploader.algorithm() dbsUploader.checkBlocks() time.sleep(2) dbsUploader.algorithm() dbsUploader.checkBlocks() self.assertEqual(len(openBlocks), 0) fakeDBS = open(self.dbsUrl, 'r') fakeDBSInfo = json.load(fakeDBS) fakeDBS.close() self.assertEqual(len(fakeDBSInfo), 11) for block in fakeDBSInfo: if acqEra in block['block']['block_name']: if block['block']['file_count'] != 1: self.assertEqual(block['block']['block_size'], 2048) self.assertEqual(block['block']['file_count'], 2) self.assertTrue('block_events' not in block['block']) self.assertEqual(block['block']['open_for_writing'], 0) self.assertTrue('close_settings' not in block) except: self.fail("We failed at some point in the test") finally: # We don't trust anyone else with _exit del os.environ["DONT_TRAP_EXIT"] return
class DBSUploadPoller(BaseWorkerThread): """ Handles poll-based DBSUpload """ def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl # Tier0 Agent don't need this if hasattr(self.config, "Tier0Feeder"): self.wmstatsServerSvc = None else: wmstatsSvcURL = self.config.General.centralWMStatsURL.replace( "couchdb/wmstats", "wmstatsserver") self.wmstatsServerSvc = WMStatsServer(wmstatsSvcURL) self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.updateBlocksDAO = daoFactory(classname="UpdateBlocks") self.updateFilesDAO = daoFactory(classname="UpdateFiles") self.createBlocksDAO = daoFactory(classname="CreateBlocks") self.setBlockFilesDAO = daoFactory(classname="SetBlockFiles") self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'dumpBlock', False) self.copyPath = os.path.join( getattr(self.config.DBS3Upload, 'componentDir', '/data/srv/'), 'dbsuploader_block.json') self.timeoutWaiver = 1 self.datasetParentageCache = {} return def setupPool(self): """ _setupPool_ Set up the processing pool for work """ if self.pool: # Then something already exists. Continue return self.workInput = multiprocessing.Queue() self.workResult = multiprocessing.Queue() # Starting up the pool: for _ in range(self.nProc): p = multiprocessing.Process(target=uploadWorker, args=(self.workInput, self.workResult, self.dbsUrl)) p.start() self.pool.append(p) return def __del__(self): """ __del__ Trigger a close of connections if necessary """ self.close() return def close(self): """ _close_ Kill all connections and terminate """ terminate = False for _ in self.pool: try: self.workInput.put('STOP') except Exception as ex: # Something very strange happens here # It's like it raises a blank exception # Upon being told to return msg = "Hit some exception in deletion\n" msg += str(ex) logging.debug(msg) terminate = True try: self.workInput.close() self.workResult.close() except Exception: # What are you going to do? pass for proc in self.pool: if terminate: proc.terminate() else: proc.join() self.pool = [] self.workInput = None self.workResult = None return def terminate(self, parameters): """ Do one more pass, then terminate """ logging.debug("terminating. doing one more pass before we die") self.algorithm(parameters) @timeFunction def algorithm(self, parameters=None): """ _algorithm_ First, check blocks that may be already uploaded Then, load blocks Then, load files Then, move files into blocks Then add new blocks in DBSBuffer Then add blocks to DBS Then mark blocks as done in DBSBuffer """ logging.info("Starting the DBSUpload Polling Cycle") # refreshing parentageCache every cycle if self.updateDatasetParentageCache() is False: return logging.debug("Dataset parentage map: %s", self.datasetParentageCache) try: self.checkBlocks() self.loadBlocks() self.loadFiles() self.checkBlockCompletion() self.inputBlocks() self.retrieveBlocks() except WMException: raise except Exception as ex: msg = "Unhandled Exception in DBSUploadPoller! Error: %s" % str(ex) logging.exception(msg) raise DBSUploadException(msg) def updateDatasetParentageCache(self): """ Return True to indicate it successfully fetched the parentage map. If there was an exception, return False """ myThread = threading.currentThread() success = True if not self.wmstatsServerSvc: self.datasetParentageCache = {} return success try: self.datasetParentageCache = self.wmstatsServerSvc.getChildParentDatasetMap( ) except Exception as ex: excReason = getattr(ex, 'reason', '') errorMsg = 'Failed to fetch parentage map from WMStats, skipping this cycle. ' errorMsg += 'Exception: {}. Reason: {}. Error: {}. '.format( type(ex).__name__, excReason, str(ex)) if isPassiveError(ex): logging.warning(errorMsg) else: errorMsg += 'Hit a terminal exception in DBSUploadPoller.' raise DBSUploadException(errorMsg) myThread.logdbClient.post("DBS3Upload_parentMap", errorMsg, "warning") success = False else: myThread.logdbClient.delete("DBS3Upload_parentMap", "warning", this_thread=True) return success def loadBlocks(self): """ _loadBlocks_ Find all blocks; make sure they're in the cache """ openBlocks = self.dbsUtil.findOpenBlocks() logging.info("Found %d open blocks.", len(openBlocks)) logging.debug("These are the openblocks: %s", openBlocks) # Load them if we don't have them blocksToLoad = [] for block in openBlocks: if block['blockname'] not in self.blockCache: blocksToLoad.append(block['blockname']) # Now load the blocks try: loadedBlocks = self.dbsUtil.loadBlocks(blocksToLoad) logging.info("Loaded %d blocks.", len(loadedBlocks)) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks to load: %s\n", blocksToLoad) raise DBSUploadException(msg) for blockInfo in loadedBlocks: block = DBSBufferBlock(name=blockInfo['block_name'], location=blockInfo['origin_site_name'], datasetpath=blockInfo['datasetpath']) parent = self.datasetParentageCache.get(blockInfo['datasetpath']) if parent: block.addDatasetParent(parent) logging.debug( "Load block: Child dataset %s, Parent dataset %s", blockInfo['datasetpath'], parent) block.FillFromDBSBuffer(blockInfo) blockname = block.getName() # Now we have to load files... try: files = self.dbsUtil.loadFilesByBlock(blockname=blockname) logging.info("Have %i files for block %s", len(files), blockname) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading files for existing blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks being loaded: %s\n", blockname) raise DBSUploadException(msg) # Add the loaded files to the block for f in files: block.addFile(f, self.datasetType, self.primaryDatasetType) # Add to the cache self.blockCache[blockInfo['block_name']] = block return def loadFiles(self): """ _loadFiles_ Load all files that need to be loaded. I will do this by DatasetPath to break the monstrous calls down into smaller chunks. """ dspList = self.dbsUtil.findUploadableDAS() readyBlocks = [] for dspInfo in dspList: datasetpath = dspInfo['DatasetPath'] # Get the files try: loadedFiles = self.dbsUtil.findUploadableFilesByDAS( datasetpath=datasetpath) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading uploadable files for DatasetPath.\n" msg += str(ex) logging.error(msg) logging.debug("DatasetPath being loaded: %s\n", datasetpath) raise DBSUploadException(msg) # Sort the files and blocks by location fileDict = sortListByKey(loadedFiles, 'locations') # Now add each file for location in fileDict.keys(): files = fileDict.get(location) if files: currentBlock = self.getBlock(files[0], location, True) currentBlock.setAcquisitionEra( era=dspInfo['AcquisitionEra']) currentBlock.setProcessingVer( procVer=dspInfo['ProcessingVer']) for newFile in files: if not newFile.get('block', 1) is None: # Then this file already has a block # It should be accounted for somewhere # Or loaded with the block continue # Check if we can put files in this block if not self.isBlockOpen(newFile=newFile, block=currentBlock): # Then we have to close the block and get a new one currentBlock.setPendingAndCloseBlock() readyBlocks.append(currentBlock) currentBlock = self.getBlock(newFile=newFile, location=location) currentBlock.setAcquisitionEra( era=dspInfo['AcquisitionEra']) currentBlock.setProcessingVer( procVer=dspInfo['ProcessingVer']) # Now deal with the file currentBlock.addFile(newFile, self.datasetType, self.primaryDatasetType) self.filesToUpdate.append({ 'filelfn': newFile['lfn'], 'block': currentBlock.getName() }) # Done with the location readyBlocks.append(currentBlock) for block in readyBlocks: self.blockCache[block.getName()] = block return def checkBlockCompletion(self): """ _checkBlockCompletion_ Mark Open blocks as Pending if they have timed out or their workflows have completed """ completedWorkflows = self.dbsUtil.getCompletedWorkflows() for block in self.blockCache.values(): if block.status == "Open": if (block.getTime() > block.getMaxBlockTime()) or any( key in completedWorkflows for key in block.workflows): block.setPendingAndCloseBlock() return def isBlockOpen(self, newFile, block, doTime=False): """ _isBlockOpen_ Check and see if a block is full This will check on time, but that's disabled by default The plan is to do a time check after we do everything else, so open blocks about to time out can still get more files put in them. """ if block.getMaxBlockFiles() is None or block.getMaxBlockNumEvents() is None or \ block.getMaxBlockSize() is None or block.getMaxBlockTime() is None: return True if block.status != 'Open': # Then somebody has dumped this already return False if block.getSize() + newFile['size'] > block.getMaxBlockSize(): return False if block.getNumEvents( ) + newFile['events'] > block.getMaxBlockNumEvents(): return False if block.getNFiles() >= block.getMaxBlockFiles(): # Then we have to dump it because this file # will put it over the limit. return False if block.getTime() > block.getMaxBlockTime() and doTime: return False return True def getBlock(self, newFile, location, skipOpenCheck=False): """ _getBlock_ Retrieve a block is one exists with matching datasetpath/location and is open. If no such block is found create and return a new one. """ datasetpath = newFile["datasetPath"] for block in self.blockCache.values(): if datasetpath == block.getDatasetPath( ) and location == block.getLocation(): if not self.isBlockOpen(newFile=newFile, block=block) and not skipOpenCheck: # Block isn't open anymore. Mark it as pending so that it gets uploaded. block.setPendingAndCloseBlock() else: return block # A suitable open block does not exist. Create a new one. blockname = "%s#%s" % (datasetpath, makeUUID()) newBlock = DBSBufferBlock(name=blockname, location=location, datasetpath=datasetpath) parent = self.datasetParentageCache.get(datasetpath) if parent: newBlock.addDatasetParent(parent) logging.debug("Get block: Child dataset %s, Parent dataset %s", datasetpath, parent) self.blockCache[blockname] = newBlock return newBlock def inputBlocks(self): """ _inputBlocks_ Loop through all of the "active" blocks and sort them so we can act appropriately on them. Everything will be sorted based on the following: Queued - Block is already being acted on by another process. We just ignore it. Pending, not in DBSBuffer - Block that has been closed and needs to be injected into DBS and also written to DBSBuffer. We'll do both. Pending, in DBSBuffer - Block has been closed and written to DBSBuffer. We just need to inject it into DBS. Open, not in DBSBuffer - Newly created block that needs to be written not DBSBuffer. Open, in DBSBuffer - Newly created block that has already been written to DBSBuffer. We don't have to do anything with it. """ if not self.blockCache: return myThread = threading.currentThread() createInDBS = [] createInDBSBuffer = [] updateInDBSBuffer = [] for block in self.blockCache.values(): if block.getName() in self.queuedBlocks: # Block is already being dealt with by another process. We'll # ignore it here. continue if block.status == 'Pending': # All pending blocks need to be injected into DBS. createInDBS.append(block) # If this is a new block it needs to be added to DBSBuffer # otherwise it just needs to be updated in DBSBuffer. if not block.inBuff: createInDBSBuffer.append(block) else: updateInDBSBuffer.append(block) if block.status == 'Open' and not block.inBuff: # New block that needs to be added to DBSBuffer. createInDBSBuffer.append(block) # First handle new and updated blocks if createInDBSBuffer or updateInDBSBuffer: try: myThread.transaction.begin() if createInDBSBuffer: self.createBlocksDAO.execute( blocks=createInDBSBuffer, conn=myThread.transaction.conn, transaction=True) if updateInDBSBuffer: self.updateBlocksDAO.execute( blocks=updateInDBSBuffer, conn=myThread.transaction.conn, transaction=True) except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while writing new blocks into DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for DBSBuffer: %s\n", createInDBSBuffer) logging.debug("Blocks for Update: %s\n", updateInDBSBuffer) raise DBSUploadException(msg) else: myThread.transaction.commit() # Update block status in the block cache. Mark the blocks that we have # added to DBSBuffer as being in DBSBuffer. for block in createInDBSBuffer: self.blockCache.get(block.getName()).inBuff = True # Record new file/block associations in DBSBuffer. if self.filesToUpdate: try: myThread.transaction.begin() self.setBlockFilesDAO.execute(binds=self.filesToUpdate, conn=myThread.transaction.conn, transaction=True) self.filesToUpdate = [] except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while setting blocks in files.\n" msg += str(ex) logging.error(msg) logging.debug("Files to Update: %s\n", self.filesToUpdate) raise DBSUploadException(msg) else: myThread.transaction.commit() if not createInDBS: # then there is nothing else to do return # Build the pool if it was closed if not self.pool: self.setupPool() # Finally upload blocks to DBS. for block in createInDBS: if not block.files: # What are we doing? logging.debug("Skipping empty block") continue if block.getDataset() is None: # Then we have to fix the dataset dbsFile = block.files[0] block.setDataset(datasetName=dbsFile['datasetPath'], primaryType=self.primaryDatasetType, datasetType=self.datasetType, physicsGroup=dbsFile.get( 'physicsGroup', None), prep_id=dbsFile.get('prep_id', None)) logging.debug("Found block %s in blocks", block.getName()) block.setPhysicsGroup(group=self.physicsGroup) encodedBlock = block.convertToDBSBlock() logging.info("About to insert block %s", block.getName()) self.workInput.put({ 'name': block.getName(), 'block': encodedBlock }) self.blockCount += 1 if self.produceCopy: with open(self.copyPath, 'w') as jo: json.dump(encodedBlock, jo, indent=2) self.queuedBlocks.append(block.getName()) # And all work is in and we're done for now return def retrieveBlocks(self): """ _retrieveBlocks_ Once blocks are in DBS, we have to retrieve them and see what's in them. What we do is get everything out of the result queue, and then update it in DBSBuffer. To do this, the result queue needs to pass back the blockname """ myThread = threading.currentThread() blocksToClose = [] emptyCount = 0 while self.blockCount > 0: if emptyCount > self.nTries: # When timeoutWaiver is 0 raise error. # It could take long time to get upload data to DBS # if there are a lot of files are cumulated in the buffer. # in first try but second try should be faster. # timeoutWaiver is set as component variable - only resets when component restarted. # The reason for that is only back log will occur when component is down # for a long time while other component still running and feeding the data to # dbsbuffer if self.timeoutWaiver == 0: msg = "Exceeded max number of waits while waiting for DBS to finish" raise DBSUploadException(msg) else: self.timeoutWaiver = 0 return try: # Get stuff out of the queue with a ridiculously # short wait time blockresult = self.workResult.get(timeout=self.wait) blocksToClose.append(blockresult) self.blockCount -= 1 logging.debug("Got a block to close") except queue.Empty: # This means the queue has no current results time.sleep(2) emptyCount += 1 continue loadedBlocks = [] for result in blocksToClose: # Remove from list of work being processed self.queuedBlocks.remove(result.get('name')) if result["success"] == "uploaded": block = self.blockCache.get(result.get('name')) block.status = 'InDBS' loadedBlocks.append(block) elif result["success"] == "check": block = result["name"] self.blocksToCheck.append(block) else: logging.error( "Error found in multiprocess during process of block %s", result.get('name')) logging.error(result['error']) # Continue to the next block # Block will remain in pending status until it is transferred if loadedBlocks: try: myThread.transaction.begin() self.updateFilesDAO.execute(blocks=loadedBlocks, status="InDBS", conn=myThread.transaction.conn, transaction=True) self.updateBlocksDAO.execute(blocks=loadedBlocks, conn=myThread.transaction.conn, transaction=True) except Exception as ex: myThread.transaction.rollback() # possible deadlock with PhEDExInjector, retry once after 10s logging.warning( "Oracle exception, possible deadlock due to race condition, retry after 10s sleep" ) time.sleep(10) try: myThread.transaction.begin() self.updateFilesDAO.execute(blocks=loadedBlocks, status="InDBS", conn=myThread.transaction.conn, transaction=True) self.updateBlocksDAO.execute( blocks=loadedBlocks, conn=myThread.transaction.conn, transaction=True) except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for Update: %s\n", loadedBlocks) raise DBSUploadException(msg) else: myThread.transaction.commit() else: myThread.transaction.commit() for block in loadedBlocks: # Clean things up name = block.getName() del self.blockCache[name] # Clean up the pool so we don't have stuff waiting around if self.pool: self.close() # And we're done return def checkBlocks(self): """ _checkBlocks_ Check with DBS3 if the blocks marked as check are uploaded or not. """ myThread = threading.currentThread() blocksUploaded = [] # See if there is anything to check for block in self.blocksToCheck: logging.debug("Checking block existence: %s", block) # Check in DBS if the block was really inserted try: result = self.dbsApi.listBlocks(block_name=block) # it is an empty list if block cannot be found if result: loadedBlock = self.blockCache.get(block) loadedBlock.status = 'InDBS' blocksUploaded.append(loadedBlock) except Exception as ex: msg = "Error trying to check block %s through DBS. Error: %s" % ( block, str(ex)) logging.exception(msg) # Update the status of those blocks that were truly inserted if blocksUploaded: try: myThread.transaction.begin() self.updateBlocksDAO.execute(blocks=blocksUploaded, conn=myThread.transaction.conn, transaction=True) self.updateFilesDAO.execute(blocks=blocksUploaded, status="InDBS", conn=myThread.transaction.conn, transaction=True) except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.exception(msg) logging.debug("Blocks for Update: %s\n", blocksUploaded) raise DBSUploadException(msg) else: myThread.transaction.commit() for block in blocksUploaded: # Clean things up name = block.getName() del self.blockCache[name] # Clean the check list self.blocksToCheck = [] # We're done return
class DBSBufferUtilTest(unittest.TestCase): def setUp(self): """ _setUp_ Setup the database and logging connection. Create some DBSBuffer tables and fake data for testing """ self.testInit = TestInit(__file__) self.testInit.setLogging() self.testInit.setDatabaseConnection() self.testInit.setSchema( customModules=["WMComponent.DBS3Buffer", "WMCore.WMBS"], useDefault=False) myThread = threading.currentThread() self.dbsbufferFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.wmbsFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi) self.dbsUtil = DBSBufferUtil() # Create two test dbsbuffer workflows insertWorkflow = self.dbsbufferFactory(classname="InsertWorkflow") insertWorkflow.execute("Test1", "Task1", 0, 0, 0, 0) insertWorkflow.execute("Test2", "Task2", 0, 0, 0, 0) # Update one workflow to "completed" state updateWorkflow = self.dbsbufferFactory( classname="UpdateWorkflowsToCompleted") updateWorkflow.execute(["Test1"]) # Create a test wmbs workflow testWorkflow = Workflow(spec="somespec.xml", owner="Erik", name="Test1", task="Task1") testWorkflow.create() # Create a test dbsbuffer file self.createTestFiles() def createTestFiles(self): """ _createTestFiles_ Create some dbsbuffer test files with different statuses :return: """ phedexStatus = self.dbsbufferFactory( classname="DBSBufferFiles.SetPhEDExStatus") for i in range(0, 4): lfn = "/path/to/some/lfn" + str(i) # Two files should be InDBS, two files should be NOTUPLOADED if i in [0, 2]: status = 'InDBS' else: status = 'NOTUPLOADED' testDBSFile = DBSBufferFile(lfn=lfn, size=600000, events=60000, status=status, workflowId=1) testDBSFile.setAlgorithm(appName="cmsRun", appVer="UNKNOWN", appFam="RECO", psetHash="SOMEHASH" + str(i), configContent="SOMECONTENT") testDBSFile.setDatasetPath("/path/to/some/dataset") testDBSFile.create() # Create all four combinations of status(InDBS,NOTUPLOADED) and in_phedex(0,1) if i in [0, 1]: phedexStatus.execute(lfn, 1) def tearDown(self): """ _tearDown_ Drop all the DBSBuffer tables. """ self.testInit.clearDatabase() # List of methods to potentially test # def loadDBSBufferFilesBulk(self, fileObjs): # def findUploadableDAS(self): # def testFindOpenBlocks(self): # def loadBlocksByDAS(self, das): # # def loadBlocks(self, blocknames): # # def findUploadableFilesByDAS(self, datasetpath): # # def loadFilesByBlock(self, blockname): def testGetPhEDExDBSStatusForCompletedWorkflows(self): """ _testGetPhEDExDBSStatusForCompletedWorkflows_ :return: """ results = self.dbsUtil.getPhEDExDBSStatusForCompletedWorkflows() self.assertEqual(results["Test1"]["InDBS"], 2, "ERROR: Files with InDBS status is incorrect.") self.assertEqual(results["Test1"]["InPhEDEx"], 2, "ERROR: Files with InPhEDEx status is incorrect.") self.assertEqual(results["Test1"]["NotInDBS"], 2, "ERROR: Files with NotInDBS status is incorrect.") self.assertEqual(results["Test1"]["NotInPhEDEx"], 2, "ERROR: Files with NotInPhEDEx status is incorrect.") return def testGetCompletedWorkflows(self): """ _testGetCompletedWorkflows_ :return: """ results = self.dbsUtil.getCompletedWorkflows() self.assertEqual( len(results), 1, "ERROR: GetCompletedWorkflows returned incorrect number of completed workflows." ) self.assertIn( "Test1", results, "ERROR: GetCompletedWorkflows returned incorrect workflow.") return
class DBSBufferUtilTest(unittest.TestCase): def setUp(self): """ _setUp_ Setup the database and logging connection. Create some DBSBuffer tables and fake data for testing """ self.testInit = TestInit(__file__) self.testInit.setLogging() self.testInit.setDatabaseConnection() self.testInit.setSchema(customModules=["WMComponent.DBS3Buffer", "WMCore.WMBS"], useDefault=False) myThread = threading.currentThread() self.dbsbufferFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.wmbsFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi) self.dbsUtil = DBSBufferUtil() # Create two test dbsbuffer workflows insertWorkflow = self.dbsbufferFactory(classname="InsertWorkflow") insertWorkflow.execute("Test1", "Task1", 0, 0, 0, 0) insertWorkflow.execute("Test2", "Task2", 0, 0, 0, 0) # Update one workflow to "completed" state updateWorkflow = self.dbsbufferFactory(classname="UpdateWorkflowsToCompleted") updateWorkflow.execute(["Test1"]) # Create a test wmbs workflow testWorkflow = Workflow(spec="somespec.xml", owner="Erik", name="Test1", task="Task1") testWorkflow.create() # Create a test dbsbuffer file self.createTestFiles() def createTestFiles(self): """ _createTestFiles_ Create some dbsbuffer test files with different statuses :return: """ phedexStatus = self.dbsbufferFactory(classname="DBSBufferFiles.SetPhEDExStatus") for i in range(0, 4): lfn = "/path/to/some/lfn" + str(i) # Two files should be InDBS, two files should be NOTUPLOADED if i in [0,2]: status = 'InDBS' else: status = 'NOTUPLOADED' testDBSFile = DBSBufferFile(lfn=lfn, size=600000, events=60000, status=status, workflowId=1) testDBSFile.setAlgorithm(appName="cmsRun", appVer="UNKNOWN", appFam="RECO", psetHash="SOMEHASH" + str(i), configContent="SOMECONTENT") testDBSFile.setDatasetPath("/path/to/some/dataset") testDBSFile.create() # Create all four combinations of status(InDBS,NOTUPLOADED) and in_phedex(0,1) if i in [0,1]: phedexStatus.execute(lfn, 1) def tearDown(self): """ _tearDown_ Drop all the DBSBuffer tables. """ self.testInit.clearDatabase() # List of methods to potentially test # def loadDBSBufferFilesBulk(self, fileObjs): # def findUploadableDAS(self): # def testFindOpenBlocks(self): # def loadBlocksByDAS(self, das): # # def loadBlocks(self, blocknames): # # def findUploadableFilesByDAS(self, datasetpath): # # def loadFilesByBlock(self, blockname): def testGetPhEDExDBSStatusForCompletedWorkflows(self): """ _testGetPhEDExDBSStatusForCompletedWorkflows_ :return: """ results = self.dbsUtil.getPhEDExDBSStatusForCompletedWorkflows() self.assertEqual(results["Test1"]["InDBS"], 2, "ERROR: Files with InDBS status is incorrect.") self.assertEqual(results["Test1"]["InPhEDEx"], 2, "ERROR: Files with InPhEDEx status is incorrect.") self.assertEqual(results["Test1"]["NotInDBS"], 2, "ERROR: Files with NotInDBS status is incorrect.") self.assertEqual(results["Test1"]["NotInPhEDEx"], 2, "ERROR: Files with NotInPhEDEx status is incorrect.") return def testGetCompletedWorkflows(self): """ _testGetCompletedWorkflows_ :return: """ results = self.dbsUtil.getCompletedWorkflows() self.assertEqual(len(results), 1, "ERROR: GetCompletedWorkflows returned incorrect number of completed workflows.") self.assertIn("Test1", results, "ERROR: GetCompletedWorkflows returned incorrect workflow.") return
def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl # Tier0 Agent don't need this if hasattr(self.config, "Tier0Feeder"): self.wmstatsServerSvc = None else: wmstatsSvcURL = self.config.General.centralWMStatsURL.replace("couchdb/wmstats", "wmstatsserver") self.wmstatsServerSvc = WMStatsServer(wmstatsSvcURL) self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.updateBlocksDAO = daoFactory(classname="UpdateBlocks") self.updateFilesDAO = daoFactory(classname="UpdateFiles") self.createBlocksDAO = daoFactory(classname="CreateBlocks") self.setBlockFilesDAO = daoFactory(classname="SetBlockFiles") self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'dumpBlock', False) self.copyPath = os.path.join(getattr(self.config.DBS3Upload, 'componentDir', '/data/srv/'), 'dbsuploader_block.json') self.timeoutWaiver = 1 self.datasetParentageCache = {} return
def testA_basicFunction(self): """ _basicFunction_ See if I can make the damn thing work. """ myThread = threading.currentThread() config = self.getConfig() from WMComponent.DBS3Buffer.DBSUploadPoller import DBSUploadPoller dbsUploader = DBSUploadPoller(config = config) dbsUtil = DBSBufferUtil() from dbs.apis.dbsClient import DbsApi dbsApi = DbsApi(url = config.DBSUpload.dbsUrl) # This should do nothing # Just making sure we don't crash try: dbsUploader.algorithm() except: dbsUploader.close() raise name = "ThisIsATest%s" % (int(time.time())) tier = "RECO" nFiles = 12 name = name.replace('-', '_') name = '%s-v0' % name files = self.getFiles(name = name, tier = tier, nFiles = nFiles) datasetPath = "/Cosmics/%s/%s" % (name, tier) try: dbsUploader.algorithm() except: dbsUploader.close() raise time.sleep(5) # Now look in DBS try: result = dbsApi.listDatasets(dataset = datasetPath, detail = True, dataset_access_type = 'PRODUCTION') self.assertEqual(len(result), 1) self.assertEqual(result[0]['data_tier_name'], 'RECO') self.assertEqual(result[0]['processing_version'], 0) self.assertEqual(result[0]['acquisition_era_name'], name.split('-')[0]) result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 11) except: dbsUploader.close() raise # All the blocks except for the last one should # now be there result = myThread.dbi.processData("SELECT id FROM dbsbuffer_block")[0].fetchall() self.assertEqual(len(result), 12) # The last block should still be open self.assertEqual(len(dbsUtil.findOpenBlocks()), 1) try: dbsUploader.algorithm() except: raise finally: dbsUploader.close() # All files should now be available result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 12) # The last block should now be closed self.assertEqual(len(dbsUtil.findOpenBlocks()), 0) result = myThread.dbi.processData("SELECT status FROM dbsbuffer_block")[0].fetchall() for res in result: self.assertEqual(res.values()[0], 'InDBS') return
def testBulkLoad(self): """ _testBulkLoad_ Can we load in bulk? """ addToBuffer = DBSBufferUtil() testFileChildA = DBSBufferFile(lfn="/this/is/a/child/lfnA", size=1024, events=20) testFileChildA.setAlgorithm(appName="cmsRun", appVer="CMSSW_2_1_8", appFam="RECO", psetHash="GIBBERISH", configContent="MOREGIBBERISH") testFileChildA.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildB = DBSBufferFile(lfn="/this/is/a/child/lfnB", size=1024, events=20) testFileChildB.setAlgorithm(appName="cmsRun", appVer="CMSSW_2_1_8", appFam="RECO", psetHash="GIBBERISH", configContent="MOREGIBBERISH") testFileChildB.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildC = DBSBufferFile(lfn="/this/is/a/child/lfnC", size=1024, events=20) testFileChildC.setAlgorithm(appName="cmsRun", appVer="CMSSW_2_1_8", appFam="RECO", psetHash="GIBBERISH", configContent="MOREGIBBERISH") testFileChildC.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFileChildA.create() testFileChildB.create() testFileChildC.create() testFileChildA.setLocation(["se1.fnal.gov", "se1.cern.ch"]) testFileChildB.setLocation(["se1.fnal.gov", "se1.cern.ch"]) testFileChildC.setLocation(["se1.fnal.gov", "se1.cern.ch"]) runSet = set() runSet.add(Run(1, *[45])) runSet.add(Run(2, *[67, 68])) testFileChildA.addRunSet(runSet) testFileChildB.addRunSet(runSet) testFileChildC.addRunSet(runSet) testFileChildA.save() testFileChildB.save() testFileChildC.save() setCksumAction = self.daoFactory( classname="DBSBufferFiles.AddChecksumByLFN") binds = [{ 'lfn': "/this/is/a/child/lfnA", 'cktype': 'adler32', 'cksum': 201 }, { 'lfn': "/this/is/a/child/lfnA", 'cktype': 'cksum', 'cksum': 101 }, { 'lfn': "/this/is/a/child/lfnB", 'cktype': 'adler32', 'cksum': 201 }, { 'lfn': "/this/is/a/child/lfnB", 'cktype': 'cksum', 'cksum': 101 }, { 'lfn': "/this/is/a/child/lfnC", 'cktype': 'adler32', 'cksum': 201 }, { 'lfn': "/this/is/a/child/lfnC", 'cktype': 'cksum', 'cksum': 101 }] setCksumAction.execute(bulkList=binds) testFile = DBSBufferFile(lfn="/this/is/a/lfn", size=1024, events=10) testFile.setAlgorithm(appName="cmsRun", appVer="CMSSW_2_1_8", appFam="RECO", psetHash="GIBBERISH", configContent="MOREGIBBERISH") testFile.setDatasetPath("/Cosmics/CRUZET09-PromptReco-v1/RECO") testFile.create() testFileChildA.addParents([testFile["lfn"]]) testFileChildB.addParents([testFile["lfn"]]) testFileChildC.addParents([testFile["lfn"]]) binds = [{ 'id': testFileChildA.exists() }, { 'id': testFileChildB.exists() }, { 'id': testFileChildC.exists() }] listOfFiles = addToBuffer.loadDBSBufferFilesBulk(fileObjs=binds) # print listOfFiles compareList = [ 'locations', 'psetHash', 'configContent', 'appName', 'appVer', 'appFam', 'events', 'datasetPath', 'runs' ] for f in listOfFiles: self.assertTrue( f['lfn'] in [ "/this/is/a/child/lfnA", "/this/is/a/child/lfnB", "/this/is/a/child/lfnC" ], "Unknown file in loaded results") self.assertEqual(f['checksums'], { 'adler32': '201', 'cksum': '101' }) for parent in f['parents']: self.assertEqual(parent['lfn'], testFile['lfn']) for key in compareList: self.assertEqual(f[key], testFileChildA[key])
def testA_basicFunction(self): """ _basicFunction_ See if I can make the damn thing work. """ return myThread = threading.currentThread() config = self.getConfig() dbsUploader = DBSUploadPoller(config = config) dbsUtil = DBSBufferUtil() dbsApi = DbsApi(url = config.DBSUpload.dbsUrl) # This should do nothing # Just making sure we don't crash try: dbsUploader.algorithm() except: dbsUploader.close() raise name = "ThisIsATest_%s" % (makeUUID()) tier = "RECO" nFiles = 12 files = self.getFiles(name = name, tier = tier, nFiles = nFiles) datasetPath = '/%s/%s/%s' % (name, name, tier) shortPath = '/%s/%s' % (name, name) try: dbsUploader.algorithm() except: dbsUploader.close() raise time.sleep(3) # Now look in DBS try: result = dbsApi.listPrimaryDatasets(dataset = name) self.assertEqual(len(result), 1) self.assertEqual(result[0]['primary_ds_name'], name) result = dbsApi.listDatasets(dataset = datasetPath, detail = True, dataset_access_type = 'PROCESSING') self.assertEqual(len(result), 1) self.assertEqual(result[0]['data_tier_name'], u'RECO') self.assertEqual(result[0]['processing_version'], u'V0') self.assertEqual(result[0]['acquisition_era_name'], u"DBS3TEST") result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 11) except: dbsUploader.close() raise # All the blocks except for the last one should # now be there result = myThread.dbi.processData("SELECT id FROM dbsbuffer_block")[0].fetchall() self.assertEqual(len(result), 12) # The last block should still be open self.assertEqual(len(dbsUtil.findOpenBlocks()), 1) try: dbsUploader.algorithm() except: raise finally: dbsUploader.close() # All files should now be available result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 12) # The last block should now be closed self.assertEqual(len(dbsUtil.findOpenBlocks()), 0) result = myThread.dbi.processData("SELECT status FROM dbsbuffer_block")[0].fetchall() for res in result: self.assertEqual(res.values()[0], 'InDBS') return
class DBSUploadPoller(BaseWorkerThread): """ Handles poll-based DBSUpload """ def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() self.daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'dumpBlock', False) self.copyPath = os.path.join(getattr(self.config.DBS3Upload, 'componentDir', '/data/srv/'), 'dbsuploader_block.json') self.timeoutWaiver = 1 return def setupPool(self): """ _setupPool_ Set up the processing pool for work """ if len(self.pool) > 0: # Then something already exists. Continue return self.workInput = multiprocessing.Queue() self.workResult = multiprocessing.Queue() # Starting up the pool: for _ in range(self.nProc): p = multiprocessing.Process(target=uploadWorker, args=(self.workInput, self.workResult, self.dbsUrl)) p.start() self.pool.append(p) return def __del__(self): """ __del__ Trigger a close of connections if necessary """ self.close() return def close(self): """ _close_ Kill all connections and terminate """ terminate = False for _ in self.pool: try: self.workInput.put('STOP') except Exception as ex: # Something very strange happens here # It's like it raises a blank exception # Upon being told to return msg = "Hit some exception in deletion\n" msg += str(ex) logging.debug(msg) terminate = True try: self.workInput.close() self.workResult.close() except Exception: # What are you going to do? pass for proc in self.pool: if terminate: proc.terminate() else: proc.join() self.pool = [] self.workInput = None self.workResult = None return def terminate(self, params): """ Do one more pass, then terminate """ logging.debug("terminating. doing one more pass before we die") self.algorithm(params) @timeFunction def algorithm(self, parameters=None): """ _algorithm_ First, check blocks that may be already uploaded Then, load blocks Then, load files Then, move files into blocks Then add new blocks in DBSBuffer Then add blocks to DBS Then mark blocks as done in DBSBuffer """ try: logging.info("Starting the DBSUpload Polling Cycle") self.checkBlocks() self.loadBlocks() self.loadFiles() self.checkBlockCompletion() self.inputBlocks() self.retrieveBlocks() except WMException: raise except Exception as ex: msg = "Unhandled Exception in DBSUploadPoller!\n" msg += str(ex) msg += str(str(traceback.format_exc())) logging.error(msg) raise DBSUploadException(msg) def loadBlocks(self): """ _loadBlocks_ Find all blocks; make sure they're in the cache """ openBlocks = self.dbsUtil.findOpenBlocks() logging.info("Found %d open blocks.", len(openBlocks)) logging.debug("These are the openblocks: %s", openBlocks) # Load them if we don't have them blocksToLoad = [] for block in openBlocks: if not block['blockname'] in self.blockCache.keys(): blocksToLoad.append(block['blockname']) # Now load the blocks try: loadedBlocks = self.dbsUtil.loadBlocks(blocksToLoad) logging.info("Loaded %d blocks.", len(loadedBlocks)) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks to load: %s\n", blocksToLoad) raise DBSUploadException(msg) for blockInfo in loadedBlocks: block = DBSBufferBlock(name=blockInfo['block_name'], location=blockInfo['origin_site_name'], datasetpath=blockInfo['datasetpath']) block.FillFromDBSBuffer(blockInfo) blockname = block.getName() # Now we have to load files... try: files = self.dbsUtil.loadFilesByBlock(blockname=blockname) logging.info("Have %i files for block %s", len(files), blockname) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading files for existing blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks being loaded: %s\n", blockname) raise DBSUploadException(msg) # Add the loaded files to the block for f in files: block.addFile(f, self.datasetType, self.primaryDatasetType) # Add to the cache self.blockCache[blockInfo['block_name']] = block return def loadFiles(self): """ _loadFiles_ Load all files that need to be loaded. I will do this by DatasetPath to break the monstrous calls down into smaller chunks. """ dspList = self.dbsUtil.findUploadableDAS() readyBlocks = [] for dspInfo in dspList: datasetpath = dspInfo['DatasetPath'] # Get the files try: loadedFiles = self.dbsUtil.findUploadableFilesByDAS(datasetpath=datasetpath) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading uploadable files for DatasetPath.\n" msg += str(ex) logging.error(msg) logging.debug("DatasetPath being loaded: %s\n", datasetpath) raise DBSUploadException(msg) # Sort the files and blocks by location fileDict = sortListByKey(loadedFiles, 'locations') # Now add each file for location in fileDict.keys(): files = fileDict.get(location) if len(files) > 0: currentBlock = self.getBlock(files[0], location, True) currentBlock.setAcquisitionEra(era=dspInfo['AcquisitionEra']) currentBlock.setProcessingVer(procVer=dspInfo['ProcessingVer']) for newFile in files: if not newFile.get('block', 1) is None: # Then this file already has a block # It should be accounted for somewhere # Or loaded with the block continue # Check if we can put files in this block if not self.isBlockOpen(newFile=newFile, block=currentBlock): # Then we have to close the block and get a new one currentBlock.setPendingAndCloseBlock() readyBlocks.append(currentBlock) currentBlock = self.getBlock(newFile=newFile, location=location) currentBlock.setAcquisitionEra(era=dspInfo['AcquisitionEra']) currentBlock.setProcessingVer(procVer=dspInfo['ProcessingVer']) # Now deal with the file currentBlock.addFile(newFile, self.datasetType, self.primaryDatasetType) self.filesToUpdate.append({'filelfn': newFile['lfn'], 'block': currentBlock.getName()}) # Done with the location readyBlocks.append(currentBlock) for block in readyBlocks: self.blockCache[block.getName()] = block return def checkBlockCompletion(self): """ _checkBlockCompletion_ Mark Open blocks as Pending if they have timed out or their workflows have completed """ completedWorkflows = self.dbsUtil.getCompletedWorkflows() for block in self.blockCache.values(): if block.status == "Open": if (block.getTime() > block.getMaxBlockTime()) or any( key in completedWorkflows for key in block.workflows): block.setPendingAndCloseBlock() return def isBlockOpen(self, newFile, block, doTime=False): """ _isBlockOpen_ Check and see if a block is full This will check on time, but that's disabled by default The plan is to do a time check after we do everything else, so open blocks about to time out can still get more files put in them. """ if block.getMaxBlockFiles() is None or block.getMaxBlockNumEvents() is None or \ block.getMaxBlockSize() is None or block.getMaxBlockTime() is None: return True if block.status != 'Open': # Then somebody has dumped this already return False if block.getSize() + newFile['size'] > block.getMaxBlockSize(): return False if block.getNumEvents() + newFile['events'] > block.getMaxBlockNumEvents(): return False if block.getNFiles() >= block.getMaxBlockFiles(): # Then we have to dump it because this file # will put it over the limit. return False if block.getTime() > block.getMaxBlockTime() and doTime: return False return True def getBlock(self, newFile, location, skipOpenCheck=False): """ _getBlock_ Retrieve a block is one exists with matching datasetpath/location and is open. If no such block is found create and return a new one. """ datasetpath = newFile["datasetPath"] for block in self.blockCache.values(): if datasetpath == block.getDatasetPath() and location == block.getLocation(): if not self.isBlockOpen(newFile=newFile, block=block) and not skipOpenCheck: # Block isn't open anymore. Mark it as pending so that it gets uploaded. block.setPendingAndCloseBlock() else: return block # A suitable open block does not exist. Create a new one. blockname = "%s#%s" % (datasetpath, makeUUID()) newBlock = DBSBufferBlock(name=blockname, location=location, datasetpath=datasetpath) self.blockCache[blockname] = newBlock return newBlock def inputBlocks(self): """ _inputBlocks_ Loop through all of the "active" blocks and sort them so we can act appropriately on them. Everything will be sorted based on the following: Queued - Block is already being acted on by another process. We just ignore it. Pending, not in DBSBuffer - Block that has been closed and needs to be injected into DBS and also written to DBSBuffer. We'll do both. Pending, in DBSBuffer - Block has been closed and written to DBSBuffer. We just need to inject it into DBS. Open, not in DBSBuffer - Newly created block that needs to be written not DBSBuffer. Open, in DBSBuffer - Newly created block that has already been written to DBSBuffer. We don't have to do anything with it. """ myThread = threading.currentThread() createBlocksDAO = self.daoFactory(classname="CreateBlocks") updateBlocksDAO = self.daoFactory(classname="UpdateBlocks") setBlockFilesDAO = self.daoFactory(classname="SetBlockFiles") createInDBS = [] createInDBSBuffer = [] updateInDBSBuffer = [] for block in self.blockCache.values(): if block.getName() in self.queuedBlocks: # Block is already being dealt with by another process. We'll # ignore it here. continue if block.status == 'Pending': # All pending blocks need to be injected into DBS. createInDBS.append(block) # If this is a new block it needs to be added to DBSBuffer # otherwise it just needs to be updated in DBSBuffer. if not block.inBuff: createInDBSBuffer.append(block) else: updateInDBSBuffer.append(block) if block.status == 'Open' and not block.inBuff: # New block that needs to be added to DBSBuffer. createInDBSBuffer.append(block) # Build the pool if it was closed if len(self.pool) == 0: self.setupPool() # First handle new and updated blocks if len(createInDBSBuffer) > 0 or len(updateInDBSBuffer) > 0: try: myThread.transaction.begin() if len(createInDBSBuffer) > 0: createBlocksDAO.execute(blocks=createInDBSBuffer, conn=myThread.transaction.conn, transaction=True) if len(updateInDBSBuffer) > 0: updateBlocksDAO.execute(blocks=updateInDBSBuffer, conn=myThread.transaction.conn, transaction=True) except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while writing new blocks into DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for DBSBuffer: %s\n", createInDBSBuffer) logging.debug("Blocks for Update: %s\n", updateInDBSBuffer) raise DBSUploadException(msg) else: myThread.transaction.commit() # Update block status in the block cache. Mark the blocks that we have # added to DBSBuffer as being in DBSBuffer. for block in createInDBSBuffer: self.blockCache.get(block.getName()).inBuff = True # Record new file/block associations in DBSBuffer. if len(self.filesToUpdate) > 0: try: myThread.transaction.begin() setBlockFilesDAO.execute(binds=self.filesToUpdate, conn=myThread.transaction.conn, transaction=True) self.filesToUpdate = [] except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while setting blocks in files.\n" msg += str(ex) logging.error(msg) logging.debug("Files to Update: %s\n", self.filesToUpdate) raise DBSUploadException(msg) else: myThread.transaction.commit() # Finally upload blocks to DBS. for block in createInDBS: if len(block.files) < 1: # What are we doing? logging.debug("Skipping empty block") continue if block.getDataset() == None: # Then we have to fix the dataset dbsFile = block.files[0] block.setDataset(datasetName=dbsFile['datasetPath'], primaryType=self.primaryDatasetType, datasetType=self.datasetType, physicsGroup=dbsFile.get('physicsGroup', None), prep_id=dbsFile.get('prep_id', None)) logging.debug("Found block %s in blocks", block.getName()) block.setPhysicsGroup(group=self.physicsGroup) encodedBlock = block.convertToDBSBlock() logging.info("About to insert block %s", block.getName()) self.workInput.put({'name': block.getName(), 'block': encodedBlock}) self.blockCount += 1 if self.produceCopy: with open(self.copyPath, 'w') as jo: json.dump(encodedBlock, jo, indent=2) self.queuedBlocks.append(block.getName()) # And all work is in and we're done for now return def retrieveBlocks(self): """ _retrieveBlocks_ Once blocks are in DBS, we have to retrieve them and see what's in them. What we do is get everything out of the result queue, and then update it in DBSBuffer. To do this, the result queue needs to pass back the blockname """ myThread = threading.currentThread() updateBlocksDAO = self.daoFactory(classname="UpdateBlocks") updateFilesDAO = self.daoFactory(classname="UpdateFiles") blocksToClose = [] emptyCount = 0 while self.blockCount > 0: if emptyCount > self.nTries: # When timeoutWaiver is 0 raise error. # It could take long time to get upload data to DBS # if there are a lot of files are cumulated in the buffer. # in first try but second try should be faster. # timeoutWaiver is set as component variable - only resets when component restarted. # The reason for that is only back log will occur when component is down # for a long time while other component still running and feeding the data to # dbsbuffer if self.timeoutWaiver == 0: msg = "Exceeded max number of waits while waiting for DBS to finish" raise DBSUploadException(msg) else: self.timeoutWaiver = 0 return try: # Get stuff out of the queue with a ridiculously # short wait time blockresult = self.workResult.get(timeout=self.wait) blocksToClose.append(blockresult) self.blockCount -= 1 logging.debug("Got a block to close") except Queue.Empty: # This means the queue has no current results time.sleep(2) emptyCount += 1 continue loadedBlocks = [] for result in blocksToClose: # Remove from list of work being processed self.queuedBlocks.remove(result.get('name')) if result["success"] == "uploaded": block = self.blockCache.get(result.get('name')) block.status = 'InDBS' loadedBlocks.append(block) elif result["success"] == "check": block = result["name"] self.blocksToCheck.append(block) else: logging.error("Error found in multiprocess during process of block %s", result.get('name')) logging.error(result['error']) # Continue to the next block # Block will remain in pending status until it is transferred if len(loadedBlocks) > 0: try: myThread.transaction.begin() updateFilesDAO.execute(blocks=loadedBlocks, status="InDBS", conn=myThread.transaction.conn, transaction=True) updateBlocksDAO.execute(blocks=loadedBlocks, conn=myThread.transaction.conn, transaction=True) except Exception as ex: myThread.transaction.rollback() # possible deadlock with PhEDExInjector, retry once after 10s logging.warning("Oracle exception, possible deadlock due to race condition, retry after 10s sleep") time.sleep(10) try: myThread.transaction.begin() updateFilesDAO.execute(blocks=loadedBlocks, status="InDBS", conn=myThread.transaction.conn, transaction=True) updateBlocksDAO.execute(blocks=loadedBlocks, conn=myThread.transaction.conn, transaction=True) except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for Update: %s\n", loadedBlocks) raise DBSUploadException(msg) else: myThread.transaction.commit() else: myThread.transaction.commit() for block in loadedBlocks: # Clean things up name = block.getName() del self.blockCache[name] # Clean up the pool so we don't have stuff waiting around if len(self.pool) > 0: self.close() # And we're done return def checkBlocks(self): """ _checkBlocks_ Check with DBS3 if the blocks marked as check are uploaded or not. """ myThread = threading.currentThread() updateBlocksDAO = self.daoFactory(classname="UpdateBlocks") updateFilesDAO = self.daoFactory(classname="UpdateFiles") blocksUploaded = [] # See if there is anything to check for block in self.blocksToCheck: logging.debug("Checking block existence: %s", block) # Check in DBS if the block was really inserted try: result = self.dbsApi.listBlocks(block_name=block) for blockResult in result: if blockResult['block_name'] == block: loadedBlock = self.blockCache.get(block) loadedBlock.status = 'InDBS' blocksUploaded.append(loadedBlock) break except Exception as ex: exString = str(ex) msg = "Error trying to check block %s through DBS.\n" % block msg += exString logging.error(msg) logging.error(str(traceback.format_exc())) # Update the status of those blocks that were truly inserted if len(blocksUploaded) > 0: try: myThread.transaction.begin() updateBlocksDAO.execute(blocks=blocksUploaded, conn=myThread.transaction.conn, transaction=True) updateFilesDAO.execute(blocks=blocksUploaded, status="InDBS", conn=myThread.transaction.conn, transaction=True) except WMException: myThread.transaction.rollback() raise except Exception as ex: myThread.transaction.rollback() msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for Update: %s\n", blocksUploaded) raise DBSUploadException(msg) else: myThread.transaction.commit() for block in blocksUploaded: # Clean things up name = block.getName() del self.blockCache[name] # Clean the check list self.blocksToCheck = [] # We're done return
def testA_basicFunction(self): """ _basicFunction_ See if I can make the damn thing work. """ myThread = threading.currentThread() config = self.getConfig() from WMComponent.DBS3Buffer.DBSUploadPoller import DBSUploadPoller dbsUploader = DBSUploadPoller(config=config) dbsUtil = DBSBufferUtil() from dbs.apis.dbsClient import DbsApi dbsApi = DbsApi(url=config.DBSUpload.dbsUrl) # This should do nothing # Just making sure we don't crash try: dbsUploader.algorithm() except: dbsUploader.close() raise name = "ThisIsATest%s" % (int(time.time())) tier = "RECO" nFiles = 12 name = name.replace('-', '_') name = '%s-v0' % name files = self.getFiles(name=name, tier=tier, nFiles=nFiles) datasetPath = "/Cosmics/%s/%s" % (name, tier) try: dbsUploader.algorithm() except: dbsUploader.close() raise time.sleep(5) # Now look in DBS try: result = dbsApi.listDatasets(dataset=datasetPath, detail=True, dataset_access_type='PRODUCTION') self.assertEqual(len(result), 1) self.assertEqual(result[0]['data_tier_name'], 'RECO') self.assertEqual(result[0]['processing_version'], 0) self.assertEqual(result[0]['acquisition_era_name'], name.split('-')[0]) result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 11) except: dbsUploader.close() raise # All the blocks except for the last one should # now be there result = myThread.dbi.processData( "SELECT id FROM dbsbuffer_block")[0].fetchall() self.assertEqual(len(result), 12) # The last block should still be open self.assertEqual(len(dbsUtil.findOpenBlocks()), 1) try: dbsUploader.algorithm() except: raise finally: dbsUploader.close() # All files should now be available result = dbsApi.listFiles(dataset=datasetPath) self.assertEqual(len(result), 12) # The last block should now be closed self.assertEqual(len(dbsUtil.findOpenBlocks()), 0) result = myThread.dbi.processData( "SELECT status FROM dbsbuffer_block")[0].fetchall() for res in result: self.assertEqual(res.values()[0], 'InDBS') return
def testDualUpload(self): """ _testDualUpload_ Verify that the dual upload mode works correctly. """ self.dbsApi = DbsApi(url = self.dbsUrl) config = self.getConfig() dbsUploader = DBSUploadPoller(config = config) dbsUtil = DBSBufferUtil() # First test verifies that uploader will poll and then not do anything # as the database is empty. dbsUploader.algorithm() acqEra = "Summer%s" % (int(time.time())) parentFiles = self.createParentFiles(acqEra) (moreParentFiles, childFiles) = \ self.createFilesWithChildren(parentFiles, acqEra) allFiles = parentFiles + moreParentFiles allBlocks = [] for i in range(4): DBSBufferDataset(parentFiles[0]["datasetPath"]).create() blockName = parentFiles[0]["datasetPath"] + "#" + makeUUID() dbsBlock = DBSBufferBlock(blockName, location = "malpaquet", datasetpath = None) dbsBlock.status = "Open" dbsBlock.setDataset(parentFiles[0]["datasetPath"], 'data', 'VALID') dbsUtil.createBlocks([dbsBlock]) for file in allFiles[i * 5 : (i * 5) + 5]: dbsBlock.addFile(file, 'data', 'VALID') dbsUtil.setBlockFiles({"block": blockName, "filelfn": file["lfn"]}) if i < 2: dbsBlock.status = "InDBS" dbsUtil.updateBlocks([dbsBlock]) dbsUtil.updateFileStatus([dbsBlock], "InDBS") allBlocks.append(dbsBlock) DBSBufferDataset(childFiles[0]["datasetPath"]).create() blockName = childFiles[0]["datasetPath"] + "#" + makeUUID() dbsBlock = DBSBufferBlock(blockName, location = "malpaquet", datasetpath = None) dbsBlock.status = "InDBS" dbsBlock.setDataset(childFiles[0]["datasetPath"], 'data', 'VALID') dbsUtil.createBlocks([dbsBlock]) for file in childFiles: dbsBlock.addFile(file, 'data', 'VALID') dbsUtil.setBlockFiles({"block": blockName, "filelfn": file["lfn"]}) dbsUtil.updateFileStatus([dbsBlock], "InDBS") dbsUploader.algorithm() time.sleep(5) dbsUploader.algorithm() time.sleep(5) self.verifyData(parentFiles[0]["datasetPath"], parentFiles) # Change the status of the rest of the parent blocks so we can upload # them and the children. for dbsBlock in allBlocks: dbsBlock.status = "InDBS" dbsUtil.updateBlocks([dbsBlock]) dbsUploader.algorithm() time.sleep(5) self.verifyData(parentFiles[0]["datasetPath"], parentFiles + moreParentFiles) # Run the uploader one more time to upload the children. dbsUploader.algorithm() time.sleep(5) self.verifyData(childFiles[0]["datasetPath"], childFiles) return
def __init__(self): self.dbsUtil = DBSBufferUtil() self.condorAPI = PyCondorAPI()
class DrainStatusAPI(object): """ Provides methods for querying dbs and condor for drain statistics """ def __init__(self, config): # queue url used in WorkQueueManager self.thisAgentUrl = "http://" + config.Agent.hostName + ":5984" self.globalBackend = WorkQueueBackend(config.WorkloadSummary.couchurl) self.localBackend = WorkQueueBackend(config.WorkQueueManager.couchurl) self.dbsUtil = DBSBufferUtil() self.condorAPI = PyCondorAPI() def collectDrainInfo(self): """ Call methods to check the drain status """ results = {} results['workflows_completed'] = self.checkWorkflows() # if workflows are completed, collect additional drain statistics if results['workflows_completed']: results['upload_status'] = self.checkFileUploadStatus() results['condor_status'] = self.checkCondorStates() results['local_wq_status'] = self.checkLocalWQStatus(dbname="workqueue") results['local_wqinbox_status'] = self.checkLocalWQStatus(dbname="workqueue_inbox") results['global_wq_status'] = self.checkGlobalWQStatus() return results def checkWorkflows(self): """ Check to see if all workflows have a 'completed' status """ results = self.dbsUtil.isAllWorkflowCompleted() return results def checkCondorStates(self): """ Check idle and running jobs in Condor """ results = {} queries = [["1", "idle"], ["2", "running"]] for query in queries: jobs = self.condorAPI.getCondorJobs("JobStatus=="+query[0], []) # if there is an error, report it instead of the length of an empty list if jobs is None: results[query[1]] = "unknown (schedd query error)" else: results[query[1]] = len(jobs) return results def checkFileUploadStatus(self): """ Check file upload status: Blocks open in DBS Files not uploaded in DBS Files not uploaded to Phedex """ results = {} results['dbs_open_blocks'] = self.dbsUtil.countOpenBlocks() results['dbs_notuploaded'] = self.dbsUtil.countFilesByStatus(status="NOTUPLOADED") results['phedex_notuploaded'] = self.dbsUtil.countPhedexNotUploaded() return results def checkLocalWQStatus(self, dbname): """ Query local WorkQueue workqueue/workqueue_inbox database to see whether there are any active elements in this agent. """ results = {} for st in ('Available', 'Negotiating', 'Acquired', 'Running'): if dbname == "workqueue": elements = self.localBackend.getElements(status=st, returnIdOnly=True) else: elements = self.localBackend.getInboxElements(status=st, returnIdOnly=True) results[st] = len(elements) return results def checkGlobalWQStatus(self): """ Query Global WorkQueue workqueue database to see whether there are any active elements set to this agent. """ results = {} for st in ("Acquired", "Running"): elements = self.globalBackend.getElements(status=st, returnIdOnly=True, ChildQueueUrl=self.thisAgentUrl) results[st] = len(elements) return results
def __init__(self, config): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl # Tier0 Agent don't need this if hasattr(self.config, "Tier0Feeder"): self.wmstatsServerSvc = None else: wmstatsSvcURL = self.config.General.centralWMStatsURL.replace( "couchdb/wmstats", "wmstatsserver") self.wmstatsServerSvc = WMStatsServer(wmstatsSvcURL) self.dbsUtil = DBSBufferUtil() myThread = threading.currentThread() daoFactory = DAOFactory(package="WMComponent.DBS3Buffer", logger=myThread.logger, dbinterface=myThread.dbi) self.updateBlocksDAO = daoFactory(classname="UpdateBlocks") self.updateFilesDAO = daoFactory(classname="UpdateFiles") self.createBlocksDAO = daoFactory(classname="CreateBlocks") self.setBlockFilesDAO = daoFactory(classname="SetBlockFiles") self.pool = [] self.blocksToCheck = [] self.workInput = None self.workResult = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url=self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'dumpBlock', False) self.copyPath = os.path.join( getattr(self.config.DBS3Upload, 'componentDir', '/data/srv/'), 'dbsuploader_block.json') self.timeoutWaiver = 1 self.datasetParentageCache = {} return
class DBSUploadPoller(BaseWorkerThread): """ Handles poll-based DBSUpload """ def __init__(self, config, dbsconfig = None): """ Initialise class members """ logging.info("Running __init__ for DBS3 Uploader") BaseWorkerThread.__init__(self) self.config = config # This is slightly dangerous, but DBSUpload depends # on DBSInterface anyway self.dbsUrl = self.config.DBS3Upload.dbsUrl self.dbsUtil = DBSBufferUtil() self.pool = [] self.blocksToCheck = [] self.input = None self.result = None self.nProc = getattr(self.config.DBS3Upload, 'nProcesses', 4) self.wait = getattr(self.config.DBS3Upload, 'dbsWaitTime', 2) self.nTries = getattr(self.config.DBS3Upload, 'dbsNTries', 300) self.dbs3UploadOnly = getattr(self.config.DBS3Upload, "dbs3UploadOnly", False) self.physicsGroup = getattr(self.config.DBS3Upload, "physicsGroup", "NoGroup") self.datasetType = getattr(self.config.DBS3Upload, "datasetType", "PRODUCTION") self.primaryDatasetType = getattr(self.config.DBS3Upload, "primaryDatasetType", "mc") self.blockCount = 0 self.dbsApi = DbsApi(url = self.dbsUrl) # List of blocks currently in processing self.queuedBlocks = [] # Set up the pool of worker processes self.setupPool() # Setting up any cache objects self.blockCache = {} self.dasCache = {} self.filesToUpdate = [] self.produceCopy = getattr(self.config.DBS3Upload, 'copyBlock', False) self.copyPath = getattr(self.config.DBS3Upload, 'copyBlockPath', '/data/mnorman/block.json') self.timeoutWaiver = 1 return def setupPool(self): """ _setupPool_ Set up the processing pool for work """ if len(self.pool) > 0: # Then something already exists. Continue return self.input = multiprocessing.Queue() self.result = multiprocessing.Queue() # Starting up the pool: for _ in range(self.nProc): p = multiprocessing.Process(target = uploadWorker, args = (self.input, self.result, self.dbsUrl)) p.start() self.pool.append(p) return def __del__(self): """ __del__ Trigger a close of connections if necessary """ self.close() return def close(self): """ _close_ Kill all connections and terminate """ terminate = False for _ in self.pool: try: self.input.put('STOP') except Exception as ex: # Something very strange happens here # It's like it raises a blank exception # Upon being told to return msg = "Hit some exception in deletion\n" msg += str(ex) logging.debug(msg) terminate = True try: self.input.close() self.result.close() except: # What are you going to do? pass for proc in self.pool: if terminate: proc.terminate() else: proc.join() self.pool = [] self.input = None self.result = None return def terminate(self, params): """ Do one more pass, then terminate """ logging.debug("terminating. doing one more pass before we die") self.algorithm(params) def algorithm(self, parameters = None): """ _algorithm_ First, check blocks that may be already uploaded Then, load blocks Then, load files Then, move files into blocks Then add new blocks in DBSBuffer Then add blocks to DBS Then mark blocks as done in DBSBuffer """ try: logging.info("Starting the DBSUpload Polling Cycle") self.checkBlocks() self.loadBlocks() # The following two functions will actually place new files into # blocks. In DBS3 upload mode we rely on something else to do that # for us and will skip this step. if not self.dbs3UploadOnly: self.loadFiles() self.checkTimeout() self.checkCompleted() self.inputBlocks() self.retrieveBlocks() except WMException: raise except Exception as ex: msg = "Unhandled Exception in DBSUploadPoller!\n" msg += str(ex) msg += str(str(traceback.format_exc())) logging.error(msg) raise DBSUploadException(msg) def loadBlocks(self): """ _loadBlocks_ Find all blocks; make sure they're in the cache """ openBlocks = self.dbsUtil.findOpenBlocks(self.dbs3UploadOnly) logging.info("These are the openblocks: %s" % openBlocks) # Load them if we don't have them blocksToLoad = [] for block in openBlocks: if not block['blockname'] in self.blockCache.keys(): blocksToLoad.append(block['blockname']) # Now load the blocks try: loadedBlocks = self.dbsUtil.loadBlocks(blocksToLoad, self.dbs3UploadOnly) logging.info("Loaded blocks: %s" % loadedBlocks) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks to load: %s\n" % blocksToLoad) raise DBSUploadException(msg) for blockInfo in loadedBlocks: das = blockInfo['DatasetAlgo'] loc = blockInfo['origin_site_name'] workflow = blockInfo['workflow'] block = DBSBlock(name = blockInfo['block_name'], location = loc, das = das, workflow = workflow) block.FillFromDBSBuffer(blockInfo) blockname = block.getName() # Now we have to load files... try: files = self.dbsUtil.loadFilesByBlock(blockname = blockname) logging.info("Have %i files for block %s" % (len(files), blockname)) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading files for existing blocks.\n" msg += str(ex) logging.error(msg) logging.debug("Blocks being loaded: %s\n" % blockname) raise DBSUploadException(msg) # Add the loaded files to the block for file in files: block.addFile(file, self.datasetType, self.primaryDatasetType) # Add to the cache self.addNewBlock(block = block) # All blocks should now be loaded and present # in both the block cache (which has all the info) # and the dasCache (which is a list of name pointers # to the keys in the block cache). return def loadFiles(self): """ _loadFiles_ Load all files that need to be loaded. I will do this by DAS for now to break the monstrous calls down into smaller chunks. """ # Grab all the Dataset-Algo combindations dasList = self.dbsUtil.findUploadableDAS() if len(dasList) < 1: # Then there's nothing to do return [] readyBlocks = [] for dasInfo in dasList: dasID = dasInfo['DAS_ID'] # Get the files try: loadedFiles = self.dbsUtil.findUploadableFilesByDAS(das = dasID) except WMException: raise except Exception as ex: msg = "Unhandled exception while loading uploadable files for DAS.\n" msg += str(ex) logging.error(msg) logging.debug("DAS being loaded: %s\n" % dasID) raise DBSUploadException(msg) # Sort the files and blocks by location fileDict = sortListByKey(input = loadedFiles, key = 'locations') # Now add each file for location in fileDict.keys(): files = fileDict.get(location) if len(files) < 1: # Nothing to do here continue currentBlock = self.getBlock(files[0], location, dasID, True) currentBlock.setAcquisitionEra(era = dasInfo['AcquisitionEra']) currentBlock.setProcessingVer(procVer = dasInfo['ProcessingVer']) for newFile in files: if not newFile.get('block', 1) == None: # Then this file already has a block # It should be accounted for somewhere # Or loaded with the block continue # Check if we can put files in this block if not self.isBlockOpen(newFile = newFile, block = currentBlock): # Then we have to close the block and get a new one currentBlock.setPendingAndCloseBlock() readyBlocks.append(currentBlock) currentBlock = self.getBlock(newFile = newFile, location = location, das = dasID) currentBlock.setAcquisitionEra(era = dasInfo['AcquisitionEra']) currentBlock.setProcessingVer(procVer = dasInfo['ProcessingVer']) # Now deal with the file currentBlock.addFile(newFile, self.datasetType, self.primaryDatasetType) self.filesToUpdate.append({'filelfn': newFile['lfn'], 'block': currentBlock.getName()}) # Done with the location readyBlocks.append(currentBlock) # Should be done with the DAS once we've added all files # Update the blockCache with what is now ready. for block in readyBlocks: self.blockCache[block.getName()] = block return def checkTimeout(self): """ _checkTimeout_ Loop all Open blocks and mark them as Pending if they have timed out. """ for block in self.blockCache.values(): if block.status == "Open" and block.getTime() > block.getMaxBlockTime(): block.setPendingAndCloseBlock() self.blockCache[block.getName()] = block def checkCompleted(self): """ _checkTimeout_ Loop all Open blocks and mark them as Pending if they have timed out. """ completedWorkflows = self.dbsUtil.getCompletedWorkflows() for block in self.blockCache.values(): if block.status == "Open": if block.workflow in completedWorkflows: block.setPendingAndCloseBlock() self.blockCache[block.getName()] = block def addNewBlock(self, block): """ _addNewBlock_ Add a new block everywhere it has to go """ name = block.getName() location = block.getLocation() das = block.das self.blockCache[name] = block if not das in self.dasCache.keys(): self.dasCache[das] = {} self.dasCache[das][location] = [] elif not location in self.dasCache[das].keys(): self.dasCache[das][location] = [] if name not in self.dasCache[das][location]: self.dasCache[das][location].append(name) return def isBlockOpen(self, newFile, block, doTime = False): """ _isBlockOpen_ Check and see if a block is full This will check on time, but that's disabled by default The plan is to do a time check after we do everything else, so open blocks about to time out can still get more files put in them. """ if block.getMaxBlockFiles() is None or block.getMaxBlockNumEvents() is None or \ block.getMaxBlockSize() is None or block.getMaxBlockTime() is None: return True if block.status != 'Open': # Then somebody has dumped this already return False if block.getSize() + newFile['size'] > block.getMaxBlockSize(): return False if block.getNumEvents() + newFile['events'] > block.getMaxBlockNumEvents(): return False if block.getNFiles() >= block.getMaxBlockFiles(): # Then we have to dump it because this file # will put it over the limit. return False if block.getTime() > block.getMaxBlockTime() and doTime: return False return True def getBlock(self, newFile, location, das, skipOpenCheck = False): """ _getBlock_ Retrieve a block is one exists and is open. If no open block is found create and return a new one. """ if das in self.dasCache.keys() and location in self.dasCache[das].keys(): for blockName in self.dasCache[das][location]: block = self.blockCache.get(blockName) if not self.isBlockOpen(newFile = newFile, block = block) and not skipOpenCheck: # Block isn't open anymore. Mark it as pending so that it gets # uploaded. block.setPendingAndCloseBlock() self.blockCache[blockName] = block else: return block # A suitable open block does not exist. Create a new one. blockname = "%s#%s" % (newFile["datasetPath"], makeUUID()) newBlock = DBSBlock(name = blockname, location = location, das = das, workflow = newFile["workflow"]) self.addNewBlock(block = newBlock) return newBlock def inputBlocks(self): """ _inputBlocks_ Loop through all of the "active" blocks and sort them so we can act appropriately on them. Everything will be sorted based on the following: Queued - Block is already being acted on by another process. We just ignore it. Pending, not in DBSBuffer - Block that has been closed and needs to be injected into DBS and also written to DBSBuffer. We'll do both. Pending, in DBSBuffer - Block has been closed and written to DBSBuffer. We just need to inject it into DBS. Open, not in DBSBuffer - Newly created block that needs to be written not DBSBuffer. Open, in DBSBuffer - Newly created block that has already been written to DBSBuffer. We don't have to do anything with it. """ myThread = threading.currentThread() createInDBS = [] createInDBSBuffer = [] updateInDBSBuffer = [] for block in self.blockCache.values(): if block.getName() in self.queuedBlocks: # Block is already being dealt with by another process. We'll # ignore it here. continue if block.status == 'Pending': # All pending blocks need to be injected into DBS. createInDBS.append(block) # If this is a new block it needs to be added to DBSBuffer # otherwise it just needs to be updated in DBSBuffer. if not block.inBuff: createInDBSBuffer.append(block) else: updateInDBSBuffer.append(block) if block.status == 'Open' and not block.inBuff: # New block that needs to be added to DBSBuffer. createInDBSBuffer.append(block) # Build the pool if it was closed if len(self.pool) == 0: self.setupPool() # First handle new and updated blocks try: myThread.transaction.begin() self.dbsUtil.createBlocks(blocks = createInDBSBuffer) self.dbsUtil.updateBlocks(blocks = updateInDBSBuffer, dbs3UploadOnly = self.dbs3UploadOnly) myThread.transaction.commit() except WMException: myThread.transaction.rollback() raise except Exception as ex: msg = "Unhandled exception while writing new blocks into DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for DBSBuffer: %s\n" % createInDBSBuffer) logging.debug("Blocks for Update: %s\n" % updateInDBSBuffer) myThread.transaction.rollback() raise DBSUploadException(msg) # Update block status in the block cache. Mark the blocks that we have # added to DBSBuffer as being in DBSBuffer. for block in createInDBSBuffer: self.blockCache.get(block.getName()).inBuff = True # Record new file/block associations in DBSBuffer. try: myThread.transaction.begin() self.dbsUtil.setBlockFiles(binds = self.filesToUpdate) self.filesToUpdate = [] myThread.transaction.commit() except WMException: myThread.transaction.rollback() raise except Exception as ex: msg = "Unhandled exception while setting blocks in files.\n" msg += str(ex) logging.error(msg) logging.debug("Files to Update: %s\n" % self.filesToUpdate) myThread.transaction.rollback() raise DBSUploadException(msg) # Finally upload blocks to DBS. for block in createInDBS: if len(block.files) < 1: # What are we doing? logging.debug("Skipping empty block") continue if block.getDataset() == None: # Then we have to fix the dataset dbsFile = block.files[0] block.setDataset(datasetName = dbsFile['datasetPath'], primaryType = self.primaryDatasetType, datasetType = self.datasetType, physicsGroup = dbsFile.get('physicsGroup', None), prep_id = dbsFile.get('prep_id', None)) logging.debug("Found block %s in blocks" % block.getName()) block.setPhysicsGroup(group = self.physicsGroup) encodedBlock = block.convertToDBSBlock() logging.info("About to insert block %s" % block.getName()) self.input.put({'name': block.getName(), 'block': encodedBlock}) self.blockCount += 1 if self.produceCopy: import json f = open(self.copyPath, 'w') f.write(json.dumps(encodedBlock)) f.close() self.queuedBlocks.append(block.getName()) # And all work is in and we're done for now return def retrieveBlocks(self): """ _retrieveBlocks_ Once blocks are in DBS, we have to retrieve them and see what's in them. What we do is get everything out of the result queue, and then update it in DBSBuffer. To do this, the result queue needs to pass back the blockname """ myThread = threading.currentThread() blocksToClose = [] emptyCount = 0 while self.blockCount > 0: if emptyCount > self.nTries: # When timeoutWaiver is 0 raise error. # It could take long time to get upload data to DBS # if there are a lot of files are cumulated in the buffer. # in first try but second try should be faster. # timeoutWaiver is set as component variable - only resets when component restarted. # The reason for that is only back log will occur when component is down # for a long time while other component still running and feeding the data to # dbsbuffer if self.timeoutWaiver == 0: msg = "Exceeded max number of waits while waiting for DBS to finish" raise DBSUploadException(msg) else: self.timeoutWaiver = 0 return try: # Get stuff out of the queue with a ridiculously # short wait time blockresult = self.result.get(timeout = self.wait) blocksToClose.append(blockresult) self.blockCount -= 1 logging.debug("Got a block to close") except Queue.Empty: # This means the queue has no current results time.sleep(2) emptyCount += 1 continue loadedBlocks = [] for result in blocksToClose: # Remove from list of work being processed self.queuedBlocks.remove(result.get('name')) if result["success"] == "uploaded": block = self.blockCache.get(result.get('name')) block.status = 'InDBS' loadedBlocks.append(block) elif result["success"] == "check": block = result["name"] self.blocksToCheck.append(block) else: logging.error("Error found in multiprocess during process of block %s" % result.get('name')) logging.error(result['error']) # Continue to the next block # Block will remain in pending status until it is transferred try: myThread.transaction.begin() self.dbsUtil.updateBlocks(loadedBlocks, self.dbs3UploadOnly) if not self.dbs3UploadOnly: self.dbsUtil.updateFileStatus(loadedBlocks, "InDBS") myThread.transaction.commit() except WMException: myThread.transaction.rollback() raise except Exception as ex: msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for Update: %s\n" % loadedBlocks) myThread.transaction.rollback() raise DBSUploadException(msg) for block in loadedBlocks: # Clean things up name = block.getName() location = block.getLocation() das = block.das self.dasCache[das][location].remove(name) del self.blockCache[name] # Clean up the pool so we don't have stuff waiting around if len(self.pool) > 0: self.close() # And we're done return def checkBlocks(self): """ _checkBlocks_ Check with DBS3 if the blocks marked as check are uploaded or not. """ myThread = threading.currentThread() blocksUploaded = [] # See if there is anything to check for block in self.blocksToCheck: logging.debug("Checking block existence: %s" % block) # Check in DBS if the block was really inserted try: result = self.dbsApi.listBlocks(block_name = block) for blockResult in result: if blockResult['block_name'] == block: loadedBlock = self.blockCache.get(block) loadedBlock.status = 'InDBS' blocksUploaded.append(loadedBlock) break except Exception as ex: exString = str(ex) msg = "Error trying to check block %s through DBS.\n" % block msg += exString logging.error(msg) logging.error(str(traceback.format_exc())) # Update the status of those blocks that were truly inserted try: myThread.transaction.begin() self.dbsUtil.updateBlocks(blocksUploaded, self.dbs3UploadOnly) if not self.dbs3UploadOnly: self.dbsUtil.updateFileStatus(blocksUploaded, "InDBS") myThread.transaction.commit() except WMException: myThread.transaction.rollback() raise except Exception as ex: msg = "Unhandled exception while finished closed blocks in DBSBuffer\n" msg += str(ex) logging.error(msg) logging.debug("Blocks for Update: %s\n" % blocksUploaded) myThread.transaction.rollback() raise DBSUploadException(msg) for block in blocksUploaded: # Clean things up name = block.getName() location = block.getLocation() das = block.das self.dasCache[das][location].remove(name) del self.blockCache[name] # Clean the check list self.blocksToCheck = [] # We're done return