def invalidator(url, invalid_status='INVALID'): mcm = McMClient(dev=False) invalids = mcm.getA('invalidations',query='status=new') print len(invalids),"Object to be invalidated" for invalid in invalids: acknowledge= False if invalid['type'] == 'request': wfn = invalid['object'] print "need to invalidate the workflow",wfn wfo = session.query(Workflow).filter(Workflow.name == wfn).first() if wfo: ## set forget of that thing (although checkor will recover from it) wfo.status = 'forget' session.commit() wfi = workflowInfo(url, wfn) success = "not rejected" if wfi.request['RequestStatus'] in ['assignment-approved','new','completed']: success = reqMgrClient.rejectWorkflow(url, wfn) pass else: success = reqMgrClient.abortWorkflow(url, wfn) pass print success acknowledge= True elif invalid['type'] == 'dataset': dataset = invalid['object'] if 'None-' in dataset: continue if 'FAKE-' in dataset: continue print "setting",dataset,"to",invalid_status success = "not invalidated" success = setDatasetStatusDBS3.setStatusDBS3('https://cmsweb.cern.ch/dbs/prod/global/DBSWriter', dataset, invalid_status, None) print success ## make a delete request from everywhere we can find ? acknowledge= True else: print "\t\t",invalid['type']," type not recognized" if acknowledge: ## acknoldge invalidation in mcm, provided we can have the api print "No acknowledgment api yet available"
if node_info['decided'] and (info['checked'] - node_info['decided'])>(transfer_timeout*24.*60*60): ## not completed 7 days after approval lagging_custodial[dataset] = {'size' : ds_size, 'since' : (info['checked'] - node_info['decided'])/(24.*60*60), 'nodes' : info['nodes'], 'nmissing': info['nmissing']} else: ## there was no information about missing blocks ## last time you checked this was a lost dataset sendEmail('dataset waiting for custodial with no block','%s looks very odd'% dataset) #unlock = True pass else: unlock = True if not bad_ds and unlock and tier in tiers_keep_on_disk: ## now check with mcm if possible to relock the dataset if use_mcm: requests_using = mcm.getA('requests',query='input_dataset=%s'%dataset) pending_requests_using = filter(lambda req: req['status'] not in ['submitted','done'], requests_using) if len(pending_requests_using): print "relocking",dataset,"because of",len(requests_using),"using it",",".join( [req['prepid'] for req in pending_requests_using] ) unlock=False elif len(requests_using): print "unlocking",dataset,"because no pending request is using it in mcm" ## no one is using it unlock=True else: print "cannot unlock",dataset,"because no request seems to be using it" unlock=False else: ## relocking outs = session.query(Output).filter(Output.datasetname==dataset).all() delay_days = 30
## just using this to lock all valid secondaries newly_locking.update(CI.allSecondaries()) ## check on the one left out, which would seem to get unlocked for dataset in already_locked - newly_locking - also_locking_from_reqmgr: try: if not dataset: continue unlock = True time_point("Checking %s" % dataset) tier = dataset.split('/')[-1] if tier in tiers_keep_on_disk: ## now check with mcm if possible to relock the dataset if use_mcm: requests_using = mcm.getA('requests', query='input_dataset=%s' % dataset) pending_requests_using = filter( lambda req: req['status'] not in ['submitted', 'done'], requests_using) if len(pending_requests_using): print "relocking", dataset, "because of", len( requests_using), "using it", ",".join( [req['prepid'] for req in pending_requests_using]) unlock = False elif len(requests_using): print "unlocking", dataset, "because no pending request is using it in mcm" ## no one is using it unlock = True else: #print "cannot unlock",dataset,"because no request seems to be using it" #unlock=False
def collector(url, specific, options): up = componentInfo(mcm=False, soft=['mcm']) if not up.check(): return SI = siteInfo() dss = DSS() #NL = newLockInfo() mcm = McMClient(dev=False) fetch_in_campaigns = ['RunIISummer15GS'] mcm_statuses=['new']#,'validation','defined','approved'] will_be_used = defaultdict(list) secondary_used = defaultdict(list) for campaign,status in itertools.product( fetch_in_campaigns, mcm_statuses): queries=[] if campaign: print "getting for",campaign queries.append('member_of_campaign=%s'%campaign) if status: print "getting for",status queries.append('status=%s'%status) rs = mcm.getA('requests', query='&'.join(queries)) for r in rs: #if r['type'] != 'Prod': continue dataset = r['input_dataset'] if dataset: #print r['prepid'],dataset will_be_used[dataset].append( r ) pileup = r['pileup_dataset_name'] if pileup: secondary_used['pileup'].append( r ) all_transfers = defaultdict(list) print len(will_be_used),"datasets that can be pre-fetched" ## for secondary we really need to have the campaign right print len(secondary_used),"pileup will be used" datasets = will_be_used.keys() if options.limit: print "Restricting to randomly picked",options.limit random.shuffle( datasets ) datasets = datasets[:options.limit] for dataset in datasets: print "\tlooking at",dataset #presence = getDatasetPresence(url, dataset)#, within_sites=['T2_CH_CERN']) ## lock all those, and pre-fecth them #NL.lock( dataset ) ## we could get the reqmgr dictionnary from McM if it was implemented and use standard workflowInfo !!! for request in will_be_used[dataset]: print "will be used by",request['prepid'] campaign = request['member_of_campaign'] ## based on the campaign, pre-fetch a site list sites_allowed = SI.sites_T1s + SI.sites_with_goodIO if options.spread: ## pick up the number of copies from campaign copies_needed = 1 ## hard coded for now else: copies_needed = 1 ## hard coded for now print "Will look for",copies_needed,"of",dataset ## figure out where it is and going destinations, all_block_names = getDatasetDestinations(url, dataset, within_sites = [SI.CE_to_SE(site) for site in sites_allowed]) print json.dumps( destinations, indent=2) prim_location = [site for (site,info) in destinations.items() if info['completion']==100 and info['data_fraction']==1] prim_destination = [site for site in destinations.keys() if not site in prim_location] prim_destination = [site for site in prim_destination if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])] copies_needed = max(0,copies_needed - len(prim_location)) copies_being_made = [ sum([info['blocks'].keys().count(block) for site,info in destinations.items() if site in prim_destination]) for block in all_block_names] prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location] prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in prim_destination] ## take out the ones that cannot receive transfers prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])] copies_needed = max(0,copies_needed - min(copies_being_made)) spreading = {} if copies_needed: print "needing",copies_needed chops,sizes = getDatasetChops(dataset, chop_threshold = options.chopsize) spreading = distributeToSites( chops, prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges, sizes=sizes) else: print "no copy needed for",dataset for (site,items) in spreading.items(): all_transfers[site].extend( items ) print "accumulated transfers" print json.dumps(all_transfers, indent=2) if not options.test: sendEmail('dataset to be fetched', 'the following datasets and location were figured from mcm up-coming requests\n%s'%( json.dumps(all_transfers, indent=2) ), destination=['*****@*****.**']) ## now collect and make transfer request for (site,items_to_transfer) in all_transfers.iteritems(): print "Directing at",site items_to_transfer = list(set(items_to_transfer)) site_se = SI.CE_to_SE(site) blocks = [it for it in items_to_transfer if '#' in it] datasets = [it for it in items_to_transfer if not '#' in it] print "\t",len(blocks),"blocks" ## remove blocks if full dataset is send out blocks = [block for block in blocks if not block.split('#')[0] in datasets] blocks_dataset = list(set([block.split('#')[0] for block in blocks])) print "\t",len(blocks),"needed blocks for",blocks_dataset print "\t",len(datasets),"datasets" print "\t",datasets items_to_transfer = blocks + datasets total_size = 0 for dataset in datasets: ds_size,_ = dss.get_block_size( dataset ) total_size += ds_size for dataset in blocks_dataset: _,bs_size = dss.get_block_size( dataset ) total_size += sum([ s for b,s in bs_size if b in blocks ]) print "For a total of",total_size,"[GB]" if options.test: result= {'phedex':{'request_created' : []}} else: ##result = makeReplicaRequest(url, site_se, items_to_transfer, 'fetching pre-production', priority='normal', approve=True) ## should make sure there is something in it pass
def invalidator(url, invalid_status='INVALID'): use_mcm = True up = componentInfo(mcm=use_mcm) if not up.check(): return mcm = McMClient(dev=False) invalids = mcm.getA('invalidations',query='status=announced') print len(invalids),"Object to be invalidated" text_to_batch = defaultdict(str) text_to_request = defaultdict(str) for invalid in invalids: acknowledge= False pid = invalid['prepid'] batch_lookup = invalid['prepid'] text = "" if invalid['type'] == 'request': wfn = invalid['object'] print "need to invalidate the workflow",wfn wfo = session.query(Workflow).filter(Workflow.name == wfn).first() if wfo: ## set forget of that thing (although checkor will recover from it) print "setting the status of",wfo.status,"to forget" wfo.status = 'forget' session.commit() else: ## do not go on like this, do not acknoledge it print wfn,"is set to be rejected, but we do not know about it yet" #continue wfi = workflowInfo(url, wfn) success = "not rejected" ## to do, we should find a way to reject the workflow and any related acdc success = reqMgrClient.invalidateWorkflow(url, wfn, current_status = wfi.request['RequestStatus']) ## need to find the whole familly and reject the whole gang familly = getWorkflowById( url, wfi.request['PrepID'] , details=True) for fwl in familly: ## take out all acdc if fwl['RequestDate'] < wfi.request['RequestDate']:continue if fwl['RequestType']!='Resubmission': continue print "rejecting",fwl['RequestName'] success = reqMgrClient.invalidateWorkflow(url, fwl['RequestName'], current_status=fwl['RequestStatus']) print success wfi.sendLog('invalidator',"rejection is performed from McM invalidations request") acknowledge= True text = "The workflow %s (%s) was rejected due to invalidation in McM" % ( wfn, pid ) batch_lookup = wfn ##so that the batch id is taken as the one containing the workflow name elif invalid['type'] == 'dataset': dataset = invalid['object'] if '?' in dataset: continue if 'None' in dataset: continue if 'None-' in dataset: continue if 'FAKE-' in dataset: continue print "setting",dataset,"to",invalid_status success = setDatasetStatus(dataset , invalid_status ) if success: acknowledge= True text = "The dataset %s (%s) was set INVALID due to invalidation in McM" % ( dataset, pid ) else: print "invalidation of",dataset,"did not go so well" else: print "\t\t",invalid['type']," type not recognized" if acknowledge: ## acknoldge invalidation in mcm, provided we can have the api print "acknowledgment to mcm" mcm.get('/restapi/invalidations/acknowledge/%s'%( invalid['_id'] )) # prepare the text for batches batches = [] batches.extend(mcm.getA('batches',query='contains=%s'%batch_lookup)) batches = filter(lambda b : b['status'] in ['announced','done','reset'], batches) if len(batches): bid = batches[-1]['prepid'] print "batch nofication to",bid text_to_batch[bid] += text+"\n\n" # prepare the text for requests text_to_request[pid] += text+"\n\n" for bid,text in text_to_batch.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/batches/notify',{ "notes" : text, "prepid" : bid}) pass for pid,text in text_to_request.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/requests/notify',{ "message" : text, "prepids" : [pid]})
def invalidator(url, invalid_status='INVALID'): use_mcm = True up = componentInfo(soft=['wtc','jira']) if not up.check(): return mcm = McMClient(dev=False) invalids = mcm.getA('invalidations',query='status=announced') if not invalids: return print len(invalids),"Object to be invalidated" text_to_batch = defaultdict(str) text_to_request = defaultdict(str) for invalid in invalids: acknowledge= False pid = invalid['prepid'] batch_lookup = invalid['prepid'] text = "" if invalid['type'] == 'request': wfn = invalid['object'] print "need to invalidate the workflow",wfn wfo = session.query(Workflow).filter(Workflow.name == wfn).first() if wfo: ## set forget of that thing (although checkor will recover from it) print "setting the status of",wfo.status,"to forget" wfo.status = 'forget' session.commit() else: ## do not go on like this, do not acknoledge it print wfn,"is set to be rejected, but we do not know about it yet" #continue wfi = workflowInfo(url, wfn) success = "not rejected" ## to do, we should find a way to reject the workflow and any related acdc successes = invalidate(url, wfi, only_resub=True, with_output=False) wfi.sendLog('invalidator',"rejection is performed from McM invalidations request") acknowledge= all(successes) text = "The workflow %s (%s) was rejected due to invalidation in McM" % ( wfn, pid ) batch_lookup = wfn ##so that the batch id is taken as the one containing the workflow name elif invalid['type'] == 'dataset': dataset = invalid['object'] if '?' in dataset: continue if 'None' in dataset: continue if 'None-' in dataset: continue if 'FAKE-' in dataset: continue print "setting",dataset,"to",invalid_status success = setDatasetStatus(dataset , invalid_status ) if success: acknowledge= True text = "The dataset %s (%s) was set INVALID due to invalidation in McM" % ( dataset, pid ) else: msg = "Could not invalidate {}. Please consider contacting data management team for manual intervention.".format(dataset) print(msg) sendLog('invalidator', msg, level='critical') else: print "\t\t",invalid['type']," type not recognized" if acknowledge: ## acknoldge invalidation in mcm, provided we can have the api print "acknowledgment to mcm" ackno_url = '/restapi/invalidations/acknowledge/%s'%( invalid['_id'] ) print "at",ackno_url mcm.get(ackno_url) # prepare the text for batches batches = [] batches.extend(mcm.getA('batches',query='contains=%s'%batch_lookup)) batches = filter(lambda b : b['status'] in ['announced','done','reset'], batches) if len(batches): bid = batches[-1]['prepid'] print "batch nofication to",bid text_to_batch[bid] += text+"\n\n" # prepare the text for requests text_to_request[pid] += text+"\n\n" for bid,text in text_to_batch.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/batches/notify',{ "notes" : text, "prepid" : bid}) pass for pid,text in text_to_request.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/requests/notify',{ "message" : text, "prepids" : [pid]})
def transferor(url ,specific = None, talk=True, options=None): if userLock('transferor'): return if options and options.test: execute = False else: execute = True SI = siteInfo() CI = campaignInfo() mcm = McMClient(dev=False) dss = DSS() print "counting all being handled..." being_handled = len(session.query(Workflow).filter(Workflow.status == 'away').all()) being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('stag')).all()) being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance')).all()) max_to_handle = options.maxworkflows allowed_to_handle = max(0,max_to_handle - being_handled) wf_buffer = 5 if allowed_to_handle<=wf_buffer: ## buffer for having several wf per transfer print "Not allowed to run more than",max_to_handle,"at a time. Currently",being_handled,"and",wf_buffer,"buffer" else: print being_handled,"already being handled",max_to_handle,"max allowed,",allowed_to_handle,"remaining","and",wf_buffer,"buffer" print "... done" all_transfers=defaultdict(list) workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset wfs_and_wfh=[] print "getting all wf to consider ..." cache = getWorkflows(url, 'assignment-approved', details=True) for wfo in session.query(Workflow).filter(Workflow.status=='considered').all(): if specific and not specific in wfo.name: continue cache_r =filter(lambda d:d['RequestName']==wfo.name, cache) if len(cache_r): wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False, request = cache_r[0]) ) ) else: wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False) ) ) print "... done" input_sizes = {} ## list the size of those in transfer already in_transfer_priority=0 min_transfer_priority=100000000 print "getting all wf in staging ..." for wfo in session.query(Workflow).filter(Workflow.status=='staging').all(): wfh = workflowInfo( url, wfo.name, spec=False) (_,primary,_,_) = wfh.getIO() for prim in primary: input_sizes[prim] = dss.get( prim ) in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority'])) min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority'])) print "... done" print "Max priority in transfer already",in_transfer_priority print "Min priority in transfer already",min_transfer_priority in_transfer_already = sum(input_sizes.values()) #sort by priority higher first wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']) ), reverse=True) ## list the size of all inputs print "getting all input sizes ..." for (wfo,wfh) in wfs_and_wfh: (_,primary,_,_) = wfh.getIO() for prim in primary: input_sizes[prim] = dss.get( prim ) print "... done" grand_total = sum(input_sizes.values()) to_transfer = grand_total - in_transfer_already grand_transfer_limit = options.maxtransfer transfer_limit = grand_transfer_limit - in_transfer_already print "%15.4f GB already being transfered"%in_transfer_already print "%15.4f GB is the current requested transfer load"%to_transfer print "%15.4f GB is the global transfer limit"%grand_transfer_limit print "%15.4f GB is the available limit"%transfer_limit # the max priority value per dataset. max_priority = defaultdict(int) needs_transfer=0 ## so that we can count'em passing_along = 0 transfer_sizes={} went_over_budget=False for (wfo,wfh) in wfs_and_wfh: print wfh.request['RequestPriority'] print wfo.name,"to be transfered" #wfh = workflowInfo( url, wfo.name) (_,primary,_,_) = wfh.getIO() this_load=sum([input_sizes[prim] for prim in primary]) if ( this_load and (sum(transfer_sizes.values())+this_load > transfer_limit or went_over_budget ) ): if went_over_budget: print "Transfer has gone over bubget." else: print "Transfer will go over bubget." print "%15.4f GB this load"%this_load print "%15.4f GB already this round"%sum(transfer_sizes.values()) print "%15.4f GB is the available limit"%transfer_limit went_over_budget=True if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority: print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over budget" else: if not options.go: print min_transfer_priority,"minimum priority",wfh.request['RequestPriority'],"<",in_transfer_priority,"stop" continue ## throtlle by campaign go if not CI.go( wfh.request['Campaign'] ): print "No go for",wfh.request['Campaign'] if not options.go: continue ## check if the batch is announced announced=False is_real=False for b in mcm.getA('batches',query='contains=%s'% wfo.name): is_real = True if b['status']=='announced': announced=True break if not announced: print wfo.name,"does not look announced."# skipping?, rejecting?, reporting?" if not is_real: print wfo.name,"does not appear to be genuine." ## prevent any duplication. if the wf is not mentioned in any batch, regardless of status continue ## check on a grace period injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.) now = time.mktime(time.gmtime()) / (60.*60.) if float(now - injection_time) < 4.: if not options.go and not announced: print "It is too soon to start transfer: %3.2fH remaining"%(now - injection_time) continue passing_along += 1 if passing_along >= allowed_to_handle: if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority: print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over",max_to_handle else: print "Not allowed to pass more than",max_to_handle,"at a time. Currently",being_handled,"handled, and adding",passing_along break (lheinput,primary,parent,secondary) = wfh.getIO() if options and options.tosites: sites_allowed = options.tosites.split(',') else: sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) ) if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']): sites_allowed = CI.parameters(wfh.request['Campaign'])['SiteWhitelist'] blocks = [] if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']: blocks = wfh.request['BlockWhitelist'] can_go = True staging=False if primary: if talk: print wfo.name,'reads',', '.join(primary),'in primary' ## chope the primary dataset for prim in primary: max_priority[prim] = max(max_priority[prim],int(wfh.request['RequestPriority'])) sites_really_allowed = [site for site in sites_allowed if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])] print "Sites allowed minus the vetoed transfer" print sites_really_allowed copies_needed = int(0.35*len(sites_really_allowed))+1 ## should just go for a fixed number based if the white list grows that big print "Would make",copies_needed,"copies" if options.maxcopy>0: copies_needed = min(options.maxcopy,copies_needed) ## remove the sites that do not want transfers print "need",copies_needed workflow_dependencies[prim].add( wfo.id ) presence = getDatasetPresence( url, prim ) prim_location = [site for site,pres in presence.items() if pres[0]==True] if len(prim_location) >= copies_needed: print "The output is all fully in place at",len(prim_location),"sites" continue # reduce the number of copies required by existing full copies copies_needed = max(0,copies_needed - len(prim_location)) print "now need",copies_needed subscriptions = listSubscriptions( url , prim ) prim_destination = list(set([site for (site,(tid,decision)) in subscriptions.items() if decision and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])])) ## need to reject from that list the ones with a full copy already: i.e the transfer corresponds to the copy in place prim_destination = [site for site in prim_destination if not site in prim_location] ## add transfer dependencies latching_on_transfers = list(set([ tid for (site,(tid,decision)) in subscriptions.items() if decision and site in prim_destination and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])])) print latching_on_transfers for latching in latching_on_transfers: tfo = session.query(Transfer).filter(Transfer.phedexid == latching).first() if not tfo: tfo = Transfer( phedexid = latching) tfo.workflows_id = [] session.add(tfo) if not wfo.id in tfo.workflows_id: print "adding",wfo.id,"to",tfo.id,"with phedexid",latching l = copy.deepcopy( tfo.workflows_id ) l.append( wfo.id ) tfo.workflows_id = l if not options.test: session.commit() else: session.flush() ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query can_go = False transfer_sizes[prim] = input_sizes[prim] staging = True # reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ?? copies_needed = max(0,copies_needed - len(prim_destination)) print "then need",copies_needed if copies_needed == 0: print "The output is either fully in place or getting in full somewhere with",latching_on_transfers can_go = True continue prim_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in prim_location])] prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in prim_destination])] ## take out the ones that cannot receive transfers prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])] if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the if not options or options.chop: spreading = distributeToSites( getDatasetChops(prim), prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges) else: spreading = {} for site in prim_to_distribute: spreading[site]=[prim] can_go = False transfer_sizes[prim] = input_sizes[prim] for (site,items) in spreading.items(): all_transfers[site].extend( items ) if secondary: if talk: print wfo.name,'reads',', '.join(secondary),'in secondary' for sec in secondary: workflow_dependencies[sec].add( wfo.id ) presence = getDatasetPresence( url, sec ) sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites subscriptions = listSubscriptions( url ,sec ) sec_destination = [site for site in subscriptions] sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])] sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])] sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])] if len( sec_to_distribute )>0: for site in sec_to_distribute: all_transfers[site].append( sec ) can_go = False ## is that possible to do something more if can_go: ## no explicit transfer required this time if staging: ## but using existing ones print wfo.name,"latches on existing transfers, and nothing else" wfo.status = 'staging' else: print wfo.name,"should just be assigned NOW to",sites_allowed wfo.status = 'staged' print "setting status to",wfo.status session.commit() continue else: ## there is an explicit transfer required if staging: ## and also using an existing one print wfo.name,"latches on existing transfers" if not options.test: wfo.status = 'staging' print "setting status to",wfo.status session.commit() print wfo.name,"needs a transfer" needs_transfer+=1 #print json.dumps(all_transfers) fake_id=-1 wf_id_in_prestaging=set() for (site,items_to_transfer) in all_transfers.iteritems(): items_to_transfer = list(set(items_to_transfer)) ## convert to storage element site_se = SI.CE_to_SE(site) ## site that do not want input datasets if site in SI.sites_veto_transfer: print site,"does not want transfers" continue ## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured. ## massage a bit the items blocks = [it for it in items_to_transfer if '#' in it] datasets = [it for it in items_to_transfer if not '#' in it] if execute: print "Making a replica to",site,"(CE)",site_se,"(SE) for" else: print "Would make a replica to",site,"(CE)",site_se,"(SE) for" print "\t",len(blocks),"blocks" ## remove blocks if full dataset is send out blocks = [block for block in blocks if not block.split('#')[0] in datasets] print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks])) print "\t",len(datasets),"datasets" print "\t",datasets items_to_transfer = blocks + datasets ## operate the transfer if options and options.stop: ## ask to move-on answer = raw_input('Continue with that ?') if not answer.lower() in ['y','yes','go']: continue if execute: result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority='normal') ## make use of max_priority dataset:priority to set the subscriptions priority """ ## does not function once = True for item in items_to_transfer: bds = item.split('#')[0] if max_priority[bds] >= 90000: if once: w=10 print "waiting",w,"s before raising priority" time.sleep(w) once=False ## raise it to high priority print item,"subscription priority raised to high at",site_se #print "This does not work yet properly it seems" print updateSubscription(url, site_se, item, priority='high') """ else: #result= {'phedex':{'request_created' : [{'id' : fake_id}]}} result= {'phedex':{'request_created' : []}} fake_id-=1 if not result: print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging" continue for phedexid in [o['id'] for o in result['phedex']['request_created']]: new_transfer = session.query(Transfer).filter(Transfer.phedexid == phedexid).first() print phedexid,"transfer created" if not new_transfer: new_transfer = Transfer( phedexid = phedexid) session.add( new_transfer ) new_transfer.workflows_id = set() for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))): new_transfer.workflows_id.update( workflow_dependencies[transfering] ) new_transfer.workflows_id = list(new_transfer.workflows_id) wf_id_in_prestaging.update(new_transfer.workflows_id) session.commit() ## auto approve it if execute: approved = approveSubscription(url, phedexid, [site_se]) for wfid in wf_id_in_prestaging: tr_wf = session.query(Workflow).get(wfid) if tr_wf and tr_wf.status!='staging': if execute: tr_wf.status = 'staging' if talk: print "setting",tr_wf.name,"to staging" session.commit()
def invalidator(url, invalid_status='INVALID'): use_mcm = True up = componentInfo(soft=['wtc']) if not up.check(): return mcm = McMClient(dev=False) invalids = mcm.getA('invalidations', query='status=announced') if not invalids: return print len(invalids), "Object to be invalidated" text_to_batch = defaultdict(str) text_to_request = defaultdict(str) for invalid in invalids: acknowledge = False pid = invalid['prepid'] batch_lookup = invalid['prepid'] text = "" if invalid['type'] == 'request': wfn = invalid['object'] print "need to invalidate the workflow", wfn wfo = session.query(Workflow).filter(Workflow.name == wfn).first() if wfo: ## set forget of that thing (although checkor will recover from it) print "setting the status of", wfo.status, "to forget" wfo.status = 'forget' session.commit() else: ## do not go on like this, do not acknoledge it print wfn, "is set to be rejected, but we do not know about it yet" #continue wfi = workflowInfo(url, wfn) success = "not rejected" ## to do, we should find a way to reject the workflow and any related acdc successes = invalidate(url, wfi, only_resub=True, with_output=False) wfi.sendLog( 'invalidator', "rejection is performed from McM invalidations request") acknowledge = all(successes) text = "The workflow %s (%s) was rejected due to invalidation in McM" % ( wfn, pid) batch_lookup = wfn ##so that the batch id is taken as the one containing the workflow name elif invalid['type'] == 'dataset': dataset = invalid['object'] if '?' in dataset: continue if 'None' in dataset: continue if 'None-' in dataset: continue if 'FAKE-' in dataset: continue print "setting", dataset, "to", invalid_status success = setDatasetStatus(dataset, invalid_status) if success: acknowledge = True text = "The dataset %s (%s) was set INVALID due to invalidation in McM" % ( dataset, pid) else: print "invalidation of", dataset, "did not go so well" else: print "\t\t", invalid['type'], " type not recognized" if acknowledge: ## acknoldge invalidation in mcm, provided we can have the api print "acknowledgment to mcm" ackno_url = '/restapi/invalidations/acknowledge/%s' % ( invalid['_id']) print "at", ackno_url mcm.get(ackno_url) # prepare the text for batches batches = [] batches.extend( mcm.getA('batches', query='contains=%s' % batch_lookup)) batches = filter( lambda b: b['status'] in ['announced', 'done', 'reset'], batches) if len(batches): bid = batches[-1]['prepid'] print "batch nofication to", bid text_to_batch[bid] += text + "\n\n" # prepare the text for requests text_to_request[pid] += text + "\n\n" for bid, text in text_to_batch.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/batches/notify', {"notes": text, "prepid": bid}) pass for pid, text in text_to_request.items(): if not text: continue text += '\n This is an automated message' mcm.put('/restapi/requests/notify', { "message": text, "prepids": [pid] })
def checkor(url, spec=None, options=None): fDB = closeoutInfo() if userLock(): return if duplicateLock(): return UC = unifiedConfiguration() use_mcm = True up = componentInfo(mcm=use_mcm, soft=["mcm"]) if not up.check(): return use_mcm = up.status["mcm"] wfs = [] if options.fetch: ## get all in running and check wfs.extend(session.query(Workflow).filter(Workflow.status == "away").all()) wfs.extend(session.query(Workflow).filter(Workflow.status == "assistance").all()) if options.nofetch: ## than get all in need for assistance wfs.extend(session.query(Workflow).filter(Workflow.status.startswith("assistance-")).all()) custodials = defaultdict(list) # sites : dataset list transfers = defaultdict(list) # sites : dataset list invalidations = [] # a list of files SI = global_SI CI = campaignInfo() mcm = McMClient(dev=False) def get_campaign(output, wfi): campaign = None try: campaign = output.split("/")[2].split("-")[0] except: if "Campaign" in wfi.request: campaign = wfi.request["Campaign"] return campaign by_passes = [] holdings = [] for bypassor, email in [ ("jbadillo", "*****@*****.**"), ("vlimant", "*****@*****.**"), ("jen_a", "*****@*****.**"), ]: bypass_file = "/afs/cern.ch/user/%s/%s/public/ops/bypass.json" % (bypassor[0], bypassor) if not os.path.isfile(bypass_file): print "no file", bypass_file continue try: by_passes.extend(json.loads(open(bypass_file).read())) except: print "cannot get by-passes from", bypass_file, "for", bypassor sendEmail("malformated by-pass information", "%s is not json readable" % (bypass_file), destination=[email]) holding_file = "/afs/cern.ch/user/%s/%s/public/ops/onhold.json" % (bypassor[0], bypassor) if not os.path.isfile(holding_file): print "no file", holding_file continue try: holdings.extend(json.loads(open(holding_file).read())) except: print "cannot get holdings from", holding_file, "for", bypassor sendEmail( "malformated by-pass information", "%s is not json readable" % (holding_file), destination=[email] ) total_running_time = 5.0 * 60.0 sleep_time = max(0.5, total_running_time / len(wfs)) for wfo in wfs: if spec and not (spec in wfo.name): continue time.sleep(sleep_time) print "checking on", wfo.name ## get info wfi = workflowInfo(url, wfo.name) ## make sure the wm status is up to date. # and send things back/forward if necessary. wfo.wm_status = wfi.request["RequestStatus"] if wfo.wm_status == "closed-out": ## manually closed-out print wfo.name, "is already", wfo.wm_status wfo.status = "close" session.commit() continue elif wfo.wm_status in [ "failed", "aborted", "aborted-archived", "rejected", "rejected-archived", "aborted-completed", ]: ## went into trouble wfo.status = "trouble" print wfo.name, "is in trouble", wfo.wm_status session.commit() continue elif wfo.wm_status in ["assigned", "acquired"]: ## not worth checking yet print wfo.name, "not running yet" session.commit() continue if "-onhold" in wfo.status: if wfo.name in holdings and wfo.name not in by_passes: print wfo.name, "on hold" continue if wfo.name in holdings and wfo.name not in by_passes: wfo.status = "assistance-onhold" print "setting", wfo.name, "on hold" session.commit() continue if wfo.wm_status != "completed" and not wfo.name in by_passes: ## for sure move on with closeout check if in completed print "no need to check on", wfo.name, "in status", wfo.wm_status session.commit() continue session.commit() sub_assistance = "" # if that string is filled, there will be need for manual assistance is_closing = True ## get it from somewhere by_pass_checks = False if wfo.name in by_passes: print "we can bypass checks on", wfo.name by_pass_checks = True for bypass in by_passes: if bypass in wfo.name: print "we can bypass", wfo.name, "because of keyword", bypass by_pass_checks = True break if not CI.go(wfi.request["Campaign"]) and not by_pass_checks: print "No go for", wfo.name continue # tuck out DQMIO/DQM wfi.request["OutputDatasets"] = [out for out in wfi.request["OutputDatasets"] if not "/DQM" in out] ## anything running on acdc familly = getWorkflowById(url, wfi.request["PrepID"], details=True) acdc = [] acdc_inactive = [] has_recovery_going = False had_any_recovery = False for member in familly: if member["RequestType"] != "Resubmission": continue if member["RequestName"] == wfo.name: continue if member["RequestDate"] < wfi.request["RequestDate"]: continue if member["RequestStatus"] in [ "running-open", "running-closed", "assignment-approved", "assigned", "acquired", ]: print wfo.name, "still has an ACDC running", member["RequestName"] acdc.append(member["RequestName"]) # print json.dumps(member,indent=2) ## hook for just waiting ... is_closing = False has_recovery_going = True elif member["RequestStatus"] == None: print member["RequestName"], "is not real" pass else: acdc_inactive.append(member["RequestName"]) had_any_recovery = True ## completion check percent_completions = {} # print "let's see who is crashing", wfo.name # print wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis'] if not "TotalInputEvents" in wfi.request: event_expected, lumi_expected = 0, 0 if not "recovery" in wfo.status: sendEmail( "missing member of the request", "TotalInputEvents is missing from the workload of %s" % wfo.name, destination=["*****@*****.**"], ) else: event_expected, lumi_expected = wfi.request["TotalInputEvents"], wfi.request["TotalInputLumis"] if "RequestNumEvents" in wfi.request: event_expected = int(wfi.request["RequestNumEvents"]) elif "Task1" in wfi.request and "RequestNumEvents" in wfi.request["Task1"]: event_expected = int(wfi.request["Task1"]["RequestNumEvents"]) fractions_pass = {} for output in wfi.request["OutputDatasets"]: event_count, lumi_count = getDatasetEventsAndLumis(dataset=output) percent_completions[output] = 0.0 if lumi_expected: percent_completions[output] = lumi_count / float(lumi_expected) if event_expected: percent_completions[output] = max(percent_completions[output], event_count / float(event_expected)) fractions_pass[output] = 0.95 c = get_campaign(output, wfi) if c in CI.campaigns and "fractionpass" in CI.campaigns[c]: fractions_pass[output] = CI.campaigns[c]["fractionpass"] print "overriding fraction to", fractions_pass[output], "for", output if options.fractionpass: fractions_pass[output] = options.fractionpass print "overriding fraction to", fractions_pass[output], "by command line for", output if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]): print wfo.name, "is not completed" print json.dumps(percent_completions, indent=2) print json.dumps(fractions_pass, indent=2) ## hook for creating automatically ACDC ? if has_recovery_going: sub_assistance += "-recovering" elif had_any_recovery: ## we want to have this looked at sub_assistance += "-manual" else: sub_assistance += "-recovery" is_closing = False ## correct lumi < 300 event per lumi events_per_lumi = {} for output in wfi.request["OutputDatasets"]: events_per_lumi[output] = getDatasetEventsPerLumi(output) lumi_upper_limit = {} for output in wfi.request["OutputDatasets"]: upper_limit = 301.0 campaign = get_campaign(output, wfi) # if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request: # upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency'] # print "setting the upper limit of lumisize to",upper_limit,"by request configuration" if campaign in CI.campaigns and "lumisize" in CI.campaigns[campaign]: upper_limit = CI.campaigns[campaign]["lumisize"] print "overriding the upper lumi size to", upper_limit, "for", campaign if options.lumisize: upper_limit = options.lumisize print "overriding the upper lumi size to", upper_limit, "by command line" lumi_upper_limit[output] = upper_limit if any([events_per_lumi[out] >= lumi_upper_limit[out] for out in events_per_lumi]): print wfo.name, "has big lumisections" print json.dumps(events_per_lumi, indent=2) ## hook for rejecting the request ? sub_assistance += "-biglumi" is_closing = False any_presence = {} for output in wfi.request["OutputDatasets"]: any_presence[output] = getDatasetPresence(url, output, vetoes=[]) ## custodial copy custodial_locations = {} custodial_presences = {} for output in wfi.request["OutputDatasets"]: custodial_presences[output] = [s for s in any_presence[output] if "MSS" in s] custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output) if not custodial_locations[output]: custodial_locations[output] = [] ## presence in phedex phedex_presence = {} for output in wfi.request["OutputDatasets"]: phedex_presence[output] = phedexClient.getFileCountDataset(url, output) vetoed_custodial_tier = UC.get("tiers_with_no_custodial") out_worth_checking = [ out for out in custodial_locations.keys() if out.split("/")[-1] not in vetoed_custodial_tier ] size_worth_checking = sum( [getDatasetSize(out) / 1023.0 for out in out_worth_checking] ) ## size in TBs of all outputs if not all(map(lambda sites: len(sites) != 0, [custodial_locations[out] for out in out_worth_checking])): print wfo.name, "has not all custodial location" print json.dumps(custodial_locations, indent=2) ########## ## hook for making a custodial replica ? custodial = None ## get from other outputs for output in out_worth_checking: if len(custodial_locations[output]): custodial = custodial_locations[output][0] if custodial and float(SI.storage[custodial]) < size_worth_checking: print "cannot use the other output custodial:", custodial, "because of limited space" custodial = None ## try to get it from campaign configuration if not custodial: for output in out_worth_checking: campaign = get_campaign(output, wfi) if campaign in CI.campaigns and "custodial" in CI.campaigns[campaign]: custodial = CI.campaigns[campaign]["custodial"] print "Setting custodial to", custodial, "from campaign configuration" break if custodial and float(SI.storage[custodial]) < size_worth_checking: print "cannot use the unified configuration custodial:", custodial, "because of limited space" custodial = None ## get from the parent pick_custodial = True if not custodial and "InputDataset" in wfi.request: ## this is terribly dangerous to assume only parents_custodial = phedexClient.getCustodialSubscriptionRequestSite(wfi.request["InputDataset"]) ###parents_custodial = findCustodialLocation(url, wfi.request['InputDataset']) if not parents_custodial: parents_custodial = [] if len(parents_custodial): custodial = parents_custodial[0] else: print "the input dataset", wfi.request[ "InputDataset" ], "does not have custodial in the first place. abort" sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor" % wfi.request["InputDataset"], ) is_closing = False pick_custodial = False if custodial and float(SI.storage[custodial]) < size_worth_checking: print "cannot use the parent custodial:", custodial, "because of limited space" custodial = None if not custodial and pick_custodial: ## pick one at random custodial = SI.pick_SE(size=size_worth_checking) if not custodial: print "cannot find a custodial for", wfo.name sendEmail( "cannot find a custodial", "cannot find a custodial for %s probably because of the total output size %d" % (wfo.name, size_worth_checking), ) if custodial and ((not sub_assistance and not acdc) or by_pass_checks): ## remember how much you added this round already ; this stays locally SI.storage[custodial] -= size_worth_checking ## register the custodial request, if there are no other big issues for output in out_worth_checking: if not len(custodial_locations[output]): if phedex_presence[output] >= 1: custodials[custodial].append(output) else: print "no file in phedex for", output, " not good to add to custodial requests" is_closing = False ## disk copy disk_copies = {} for output in wfi.request["OutputDatasets"]: disk_copies[output] = [s for s in any_presence[output] if (not "MSS" in s) and (not "Buffer" in s)] if not all(map(lambda sites: len(sites) != 0, disk_copies.values())): print wfo.name, "has not all output on disk" print json.dumps(disk_copies, indent=2) ## presence in dbs dbs_presence = {} dbs_invalid = {} for output in wfi.request["OutputDatasets"]: dbs_presence[output] = dbs3Client.getFileCountDataset(output) dbs_invalid[output] = dbs3Client.getFileCountDataset(output, onlyInvalid=True) fraction_invalid = 0.01 if ( not all( [ dbs_presence[out] == (dbs_invalid[out] + phedex_presence[out]) for out in wfi.request["OutputDatasets"] ] ) and not options.ignorefiles ): print wfo.name, "has a dbs,phedex mismatch" print json.dumps(dbs_presence, indent=2) print json.dumps(dbs_invalid, indent=2) print json.dumps(phedex_presence, indent=2) ## hook for just waiting ... is_closing = False if ( not all( [ (dbs_invalid[out] <= int(fraction_invalid * dbs_presence[out])) for out in wfi.request["OutputDatasets"] ] ) and not options.ignorefiles ): print wfo.name, "has a dbs invalid file level too high" print json.dumps(dbs_presence, indent=2) print json.dumps(dbs_invalid, indent=2) print json.dumps(phedex_presence, indent=2) ## need to be going and taking an eye sub_assistance += "-invalidfiles" is_closing = False ## put that heavy part at the end ## duplication check duplications = {} if is_closing: print "starting duplicate checker for", wfo.name for output in wfi.request["OutputDatasets"]: print "\tchecking", output duplications[output] = True try: duplications[output] = dbs3Client.duplicateRunLumi(output) except: try: duplications[output] = dbs3Client.duplicateRunLumi(output) except: print "was not possible to get the duplicate count for", output is_closing = False if any(duplications.values()) and not options.ignoreduplicates: print wfo.name, "has duplicates" print json.dumps(duplications, indent=2) ## hook for making file invalidation ? sub_assistance += "-duplicates" is_closing = False ## for visualization later on if not wfo.name in fDB.record: # print "adding",wfo.name,"to close out record" fDB.record[wfo.name] = {"datasets": {}, "name": wfo.name, "closeOutWorkflow": None} fDB.record[wfo.name]["closeOutWorkflow"] = is_closing for output in wfi.request["OutputDatasets"]: if not output in fDB.record[wfo.name]["datasets"]: fDB.record[wfo.name]["datasets"][output] = {} rec = fDB.record[wfo.name]["datasets"][output] rec["percentage"] = float("%.2f" % (percent_completions[output] * 100)) rec["duplicate"] = duplications[output] if output in duplications else "N/A" rec["phedexReqs"] = ( float("%.2f" % any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output]) != 0 else "N/A" ) rec["closeOutDataset"] = is_closing rec["transPerc"] = ( float("%.2f" % any_presence[output][disk_copies[output][0]][1]) if len(disk_copies[output]) != 0 else "N/A" ) rec["correctLumis"] = ( int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True ) rec["missingSubs"] = ( False if len(custodial_locations[output]) == 0 else ",".join(list(set(custodial_locations[output]))) ) rec["dbsFiles"] = dbs_presence[output] rec["dbsInvFiles"] = dbs_invalid[output] rec["phedexFiles"] = phedex_presence[output] rec["acdc"] = "%d / %d" % (len(acdc), len(acdc + acdc_inactive)) if by_pass_checks: ## force closing is_closing = True ## and move on if is_closing: ## toggle status to closed-out in request manager print "setting", wfo.name, "closed-out" if not options.test: if wfo.wm_status in ["closed-out", "announced", "normal-archived"]: print wfo.name, "is already", wfo.wm_status, "not trying to closed-out and assuming it does" res = None else: res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name) print "close out answer", res if not res in ["None", None]: print "retrying to closing out" print res res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name) if res in [None, "None"]: wfo.status = "close" session.commit() else: print "could not close out", wfo.name, "will try again next time" else: ## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it new_status = "assistance" + sub_assistance print wfo.name, "needs assistance with", new_status if sub_assistance and wfo.status != new_status and "PrepID" in wfi.request and not "manual" in wfo.status: pid = wfi.getPrepIDs()[0].replace("task_", "") # pid = wfi.request['PrepID'].replace('task_','') ## notify messages = { "recovery": "Samples completed with missing statistics:\n%s " % ( "\n".join( [ "%.2f %% complete for %s" % (percent_completions[output] * 100, output) for output in wfi.request["OutputDatasets"] ] ) ), "biglumi": "Samples completed with large luminosity blocks:\n%s " % ( "\n".join( [ "%d > %d for %s" % (events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request["OutputDatasets"] if (events_per_lumi[output] > lumi_upper_limit[output]) ] ) ), "duplicate": "Samples completed with duplicated luminosity blocks:\n%s" % ( "\n".join( [ "%s" % output for output in wfi.request["OutputDatasets"] if output in duplications and duplications[output] ] ) ), } text = "The request %s (%s) is facing issue in production.\n" % (pid, wfo.name) content = "" for case in messages: if case in new_status: content += "\n" + messages[case] + "\n" text += content text += "You are invited to check, while this is being taken care of by Ops.\n" text += "This is an automated message." if use_mcm and content: print "Sending notification back to requestor" print text batches = mcm.getA("batches", query="contains=%s&status=announced" % pid) if len(batches): ## go notify the batch bid = batches[-1]["prepid"] print "batch nofication to", bid mcm.put("/restapi/batches/notify", {"notes": text, "prepid": bid}) ## go notify the request print "request notification to", pid mcm.put("/restapi/requests/notify", {"message": text, "prepids": [pid]}) ## case where the workflow was in manual from recoveror if not "manual" in wfo.status or new_status != "assistance-recovery": wfo.status = new_status if not options.test: print "setting", wfo.name, "to", wfo.status session.commit() else: print "current status is", wfo.status, "not changing to anything" fDB.html() ## custodial requests print "Custodials" print json.dumps(custodials, indent=2) for site in custodials: print ",".join(custodials[site]), "=>", site if not options.test: result = makeReplicaRequest( url, site, list(set(custodials[site])), "custodial copy at production close-out", custodial="y", priority="low", approve=(site in SI.sites_auto_approve), ) print result print "Transfers" print json.dumps(transfers, indent=2) ## replicas requests for site in transfers: print ",".join(transfers[site]), "=>", site if not options.test: result = None # result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out") print result print "File Invalidation" print invalidations
def collector(url, specific, options): up = componentInfo(mcm=False, soft=['mcm']) if not up.check(): return SI = siteInfo() dss = DSS() #NL = newLockInfo() mcm = McMClient(dev=False) fetch_in_campaigns = ['RunIISummer15GS'] mcm_statuses = ['new'] #,'validation','defined','approved'] will_be_used = defaultdict(list) secondary_used = defaultdict(list) for campaign, status in itertools.product(fetch_in_campaigns, mcm_statuses): queries = [] if campaign: print "getting for", campaign queries.append('member_of_campaign=%s' % campaign) if status: print "getting for", status queries.append('status=%s' % status) rs = mcm.getA('requests', query='&'.join(queries)) for r in rs: #if r['type'] != 'Prod': continue dataset = r['input_dataset'] if dataset: #print r['prepid'],dataset will_be_used[dataset].append(r) pileup = r['pileup_dataset_name'] if pileup: secondary_used['pileup'].append(r) all_transfers = defaultdict(list) print len(will_be_used), "datasets that can be pre-fetched" ## for secondary we really need to have the campaign right print len(secondary_used), "pileup will be used" datasets = will_be_used.keys() if options.limit: print "Restricting to randomly picked", options.limit random.shuffle(datasets) datasets = datasets[:options.limit] for dataset in datasets: print "\tlooking at", dataset #presence = getDatasetPresence(url, dataset)#, within_sites=['T2_CH_CERN']) ## lock all those, and pre-fecth them #NL.lock( dataset ) ## we could get the reqmgr dictionnary from McM if it was implemented and use standard workflowInfo !!! for request in will_be_used[dataset]: print "will be used by", request['prepid'] campaign = request['member_of_campaign'] ## based on the campaign, pre-fetch a site list sites_allowed = SI.sites_T1s + SI.sites_with_goodIO if options.spread: ## pick up the number of copies from campaign copies_needed = 1 ## hard coded for now else: copies_needed = 1 ## hard coded for now print "Will look for", copies_needed, "of", dataset ## figure out where it is and going destinations, all_block_names = getDatasetDestinations( url, dataset, within_sites=[SI.CE_to_SE(site) for site in sites_allowed]) print json.dumps(destinations, indent=2) prim_location = [ site for (site, info) in destinations.items() if info['completion'] == 100 and info['data_fraction'] == 1 ] prim_destination = [ site for site in destinations.keys() if not site in prim_location ] prim_destination = [ site for site in prim_destination if not any([ osite.startswith(site) for osite in SI.sites_veto_transfer ]) ] copies_needed = max(0, copies_needed - len(prim_location)) copies_being_made = [ sum([ info['blocks'].keys().count(block) for site, info in destinations.items() if site in prim_destination ]) for block in all_block_names ] prim_to_distribute = [ site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location ] prim_to_distribute = [ site for site in prim_to_distribute if not SI.CE_to_SE(site) in prim_destination ] ## take out the ones that cannot receive transfers prim_to_distribute = [ site for site in prim_to_distribute if not any([ osite.startswith(site) for osite in SI.sites_veto_transfer ]) ] copies_needed = max(0, copies_needed - min(copies_being_made)) spreading = {} if copies_needed: print "needing", copies_needed chops, sizes = getDatasetChops(dataset, chop_threshold=options.chopsize) spreading = distributeToSites(chops, prim_to_distribute, n_copies=copies_needed, weights=SI.cpu_pledges, sizes=sizes) else: print "no copy needed for", dataset for (site, items) in spreading.items(): all_transfers[site].extend(items) print "accumulated transfers" print json.dumps(all_transfers, indent=2) if not options.test: sendEmail( 'dataset to be fetched', 'the following datasets and location were figured from mcm up-coming requests\n%s' % (json.dumps(all_transfers, indent=2)), destination=['*****@*****.**']) ## now collect and make transfer request for (site, items_to_transfer) in all_transfers.iteritems(): print "Directing at", site items_to_transfer = list(set(items_to_transfer)) site_se = SI.CE_to_SE(site) blocks = [it for it in items_to_transfer if '#' in it] datasets = [it for it in items_to_transfer if not '#' in it] print "\t", len(blocks), "blocks" ## remove blocks if full dataset is send out blocks = [ block for block in blocks if not block.split('#')[0] in datasets ] blocks_dataset = list(set([block.split('#')[0] for block in blocks])) print "\t", len(blocks), "needed blocks for", blocks_dataset print "\t", len(datasets), "datasets" print "\t", datasets items_to_transfer = blocks + datasets total_size = 0 for dataset in datasets: ds_size, _ = dss.get_block_size(dataset) total_size += ds_size for dataset in blocks_dataset: _, bs_size = dss.get_block_size(dataset) total_size += sum([s for b, s in bs_size if b in blocks]) print "For a total of", total_size, "[GB]" if options.test: result = {'phedex': {'request_created': []}} else: ##result = makeReplicaRequest(url, site_se, items_to_transfer, 'fetching pre-production', priority='normal', approve=True) ## should make sure there is something in it pass
else: ## there was no information about missing blocks ## last time you checked this was a lost dataset # sendEmail('dataset waiting for custodial with no block','%s looks very odd'% dataset) sendLog( "dataset waiting for custodial with no block, %s looks very odd" % dataset, level="critical" ) # unlock = True pass else: unlock = True if not bad_ds and unlock and tier in tiers_keep_on_disk: ## now check with mcm if possible to relock the dataset if use_mcm: requests_using = mcm.getA("requests", query="input_dataset=%s" % dataset) pending_requests_using = filter(lambda req: req["status"] not in ["submitted", "done"], requests_using) if len(pending_requests_using): print "relocking", dataset, "because of", len(requests_using), "using it", ",".join( [req["prepid"] for req in pending_requests_using] ) unlock = False elif len(requests_using): print "unlocking", dataset, "because no pending request is using it in mcm" ## no one is using it unlock = True else: # print "cannot unlock",dataset,"because no request seems to be using it" # unlock=False print "Unlocking", dataset, "because no request is using it in input" unlock = True
def checkor(url, spec=None, options=None): fDB = closeoutInfo() use_mcm = True up = componentInfo(mcm=use_mcm, soft=['mcm']) if not up.check(): return use_mcm = up.status['mcm'] wfs=[] if options.fetch: ## get all in running and check wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() ) wfs.extend( session.query(Workflow).filter(Workflow.status== 'assistance').all() ) if options.nofetch: ## than get all in need for assistance wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all() ) custodials = defaultdict(list) #sites : dataset list transfers = defaultdict(list) #sites : dataset list invalidations = [] #a list of files SI = siteInfo() CI = campaignInfo() mcm = McMClient(dev=False) def get_campaign(output, wfi): campaign = None try: campaign = output.split('/')[2].split('-')[0] except: if 'Campaign' in wfi.request: campaign = wfi.request['Campaign'] return campaign by_passes = [] holdings = [] for bypassor,email in [('jbadillo','*****@*****.**'),('vlimant','*****@*****.**'),('jen_a','*****@*****.**')]: bypass_file = '/afs/cern.ch/user/%s/%s/public/ops/bypass.json'%(bypassor[0],bypassor) if not os.path.isfile(bypass_file): print "no file",bypass_file continue try: by_passes.extend( json.loads(open(bypass_file).read())) except: print "cannot get by-passes from",bypass_file,"for",bypassor sendEmail("malformated by-pass information","%s is not json readable"%(bypass_file), destination=[email]) holding_file = '/afs/cern.ch/user/%s/%s/public/ops/onhold.json'%(bypassor[0],bypassor) if not os.path.isfile(holding_file): print "no file",holding_file continue try: holdings.extend( json.loads(open(holding_file).read())) except: print "cannot get holdings from",holding_file,"for",bypassor sendEmail("malformated by-pass information","%s is not json readable"%(holding_file), destination=[email]) total_running_time = 5.*60. sleep_time = max(0.5, total_running_time / len(wfs)) for wfo in wfs: if spec and not (spec in wfo.name): continue time.sleep( sleep_time ) print "checking on",wfo.name ## get info wfi = workflowInfo(url, wfo.name) ## make sure the wm status is up to date. # and send things back/forward if necessary. wfo.wm_status = wfi.request['RequestStatus'] if wfo.wm_status == 'closed-out': ## manually closed-out print wfo.name,"is already",wfo.wm_status wfo.status = 'close' session.commit() continue elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']: ## went into trouble wfo.status = 'trouble' print wfo.name,"is in trouble",wfo.wm_status session.commit() continue elif wfo.wm_status in ['assigned','acquired']: ## not worth checking yet print wfo.name,"not running yet" session.commit() continue if '-onhold' in wfo.status: if wfo.name in holdings: print wfo.name,"on hold" continue if wfo.name in holdings: wfo.status = 'assistance-onhold' print "setting",wfo.name,"on hold" session.commit() continue if wfo.wm_status != 'completed': ## for sure move on with closeout check if in completed print "no need to check on",wfo.name,"in status",wfo.wm_status session.commit() continue session.commit() sub_assistance="" # if that string is filled, there will be need for manual assistance is_closing = True ## do the closed-out checks one by one ## get it from somewhere by_pass_checks = False if wfo.name in by_passes: print "we can bypass checks on",wfo.name by_pass_checks = True for bypass in by_passes: if bypass in wfo.name: print "we can bypass",wfo.name,"because of keyword",bypass by_pass_checks = True break # tuck out DQMIO/DQM wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not '/DQM' in out] ## anything running on acdc familly = getWorkflowById(url, wfi.request['PrepID'], details=True) acdc = [] acdc_inactive = [] has_recovery_going=False had_any_recovery = False for member in familly: if member['RequestType'] != 'Resubmission': continue if member['RequestName'] == wfo.name: continue if member['RequestDate'] < wfi.request['RequestDate']: continue if member['RequestStatus'] in ['running-open','running-closed','assignment-approved','assigned','acquired']: print wfo.name,"still has an ACDC running",member['RequestName'] acdc.append( member['RequestName'] ) #print json.dumps(member,indent=2) ## hook for just waiting ... is_closing = False has_recovery_going=True elif member['RequestStatus']==None: print member['RequestName'],"is not real" pass else: acdc_inactive.append( member['RequestName'] ) had_any_recovery = True ## completion check percent_completions = {} # print "let's see who is crashing", wfo.name # print wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis'] if not 'TotalInputEvents' in wfi.request: event_expected,lumi_expected = 0,0 if not 'recovery' in wfo.status: sendEmail("missing member of the request","TotalInputEvents is missing from the workload of %s"% wfo.name, destination=['*****@*****.**']) else: event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis'] fractions_pass = {} for output in wfi.request['OutputDatasets']: event_count,lumi_count = getDatasetEventsAndLumis(dataset=output) percent_completions[output] = 0. if lumi_expected: percent_completions[output] = lumi_count / float( lumi_expected ) fractions_pass[output] = 0.95 c = get_campaign(output, wfi) if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]: fractions_pass[output] = CI.campaigns[c]['fractionpass'] print "overriding fraction to",fractions_pass[output],"for",output if options.fractionpass: fractions_pass[output] = options.fractionpass print "overriding fraction to",fractions_pass[output],"by command line for",output if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]): print wfo.name,"is not completed" print json.dumps(percent_completions, indent=2) print json.dumps(fractions_pass, indent=2) ## hook for creating automatically ACDC ? if has_recovery_going: sub_assistance+='-recovering' elif had_any_recovery: ## we want to have this looked at sub_assistance+='-manual' else: sub_assistance+='-recovery' is_closing = False ## correct lumi < 300 event per lumi events_per_lumi = {} for output in wfi.request['OutputDatasets']: events_per_lumi[output] = getDatasetEventsPerLumi( output ) lumi_upper_limit = {} for output in wfi.request['OutputDatasets']: upper_limit = 301. campaign = get_campaign(output, wfi) #if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request: # upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency'] # print "setting the upper limit of lumisize to",upper_limit,"by request configuration" if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]: upper_limit = CI.campaigns[campaign]['lumisize'] print "overriding the upper lumi size to",upper_limit,"for",campaign if options.lumisize: upper_limit = options.lumisize print "overriding the upper lumi size to",upper_limit,"by command line" lumi_upper_limit[output] = upper_limit if any([ events_per_lumi[out] >= lumi_upper_limit[out] for out in events_per_lumi]): print wfo.name,"has big lumisections" print json.dumps(events_per_lumi, indent=2) ## hook for rejecting the request ? sub_assistance+='-biglumi' is_closing = False any_presence = {} for output in wfi.request['OutputDatasets']: any_presence[output] = getDatasetPresence(url, output, vetoes=[]) ## custodial copy custodial_locations = {} custodial_presences = {} for output in wfi.request['OutputDatasets']: custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s] custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output) if not custodial_locations[output]: custodial_locations[output] = [] ## presence in phedex phedex_presence ={} for output in wfi.request['OutputDatasets']: phedex_presence[output] = phedexClient.getFileCountDataset(url, output ) vetoed_custodial_tier = ['MINIAODSIM'] out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier] if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])): print wfo.name,"has not all custodial location" print json.dumps(custodial_locations, indent=2) ########## ## hook for making a custodial replica ? custodial = None ## get from other outputs for output in out_worth_checking: if len(custodial_locations[output]): custodial = custodial_locations[output][0] ## try to get it from campaign configuration if not custodial: for output in out_worth_checking: campaign = get_campaign(output, wfi) if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]: custodial = CI.campaigns[campaign]['custodial'] print "Setting custodial to",custodial,"from campaign configuration" break ## get from the parent pick_custodial = True if not custodial and 'InputDataset' in wfi.request: ## this is terribly dangerous to assume only parents_custodial = phedexClient.getCustodialSubscriptionRequestSite( wfi.request['InputDataset']) ###parents_custodial = findCustodialLocation(url, wfi.request['InputDataset']) if not parents_custodial: parents_custodial = [] if len(parents_custodial): custodial = parents_custodial[0] else: print "the input dataset",wfi.request['InputDataset'],"does not have custodial in the first place. abort" sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor"%wfi.request['InputDataset']) is_closing = False pick_custodial = False if not custodial and pick_custodial: ## pick one at random custodial = SI.pick_SE() if custodial and ((not sub_assistance and not acdc) or by_pass_checks): ## register the custodial request, if there are no other big issues for output in out_worth_checking: if not len(custodial_locations[output]): if phedex_presence[output]>=1: custodials[custodial].append( output ) else: print "no file in phedex for",output," not good to add to custodial requests" else: print "cannot find a custodial for",wfo.name is_closing = False ## disk copy disk_copies = {} for output in wfi.request['OutputDatasets']: disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)] if not all(map( lambda sites : len(sites)!=0, disk_copies.values())): print wfo.name,"has not all output on disk" print json.dumps(disk_copies, indent=2) ## presence in dbs dbs_presence = {} dbs_invalid = {} for output in wfi.request['OutputDatasets']: dbs_presence[output] = dbs3Client.getFileCountDataset( output ) dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True) fraction_invalid = 0.01 if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles: print wfo.name,"has a dbs,phedex mismatch" print json.dumps(dbs_presence, indent=2) print json.dumps(dbs_invalid, indent=2) print json.dumps(phedex_presence, indent=2) ## hook for just waiting ... is_closing = False if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles: print wfo.name,"has a dbs invalid file level too high" print json.dumps(dbs_presence, indent=2) print json.dumps(dbs_invalid, indent=2) print json.dumps(phedex_presence, indent=2) ## need to be going and taking an eye sub_assistance+="-invalidfiles" is_closing = False ## put that heavy part at the end ## duplication check duplications = {} if is_closing: print "starting duplicate checker for",wfo.name for output in wfi.request['OutputDatasets']: print "\tchecking",output duplications[output] = True try: duplications[output] = dbs3Client.duplicateRunLumi( output ) except: try: duplications[output] = dbs3Client.duplicateRunLumi( output ) except: print "was not possible to get the duplicate count for",output is_closing=False if any(duplications.values()) and not options.ignoreduplicates: print wfo.name,"has duplicates" print json.dumps(duplications,indent=2) ## hook for making file invalidation ? sub_assistance+='-duplicates' is_closing = False ## for visualization later on if not wfo.name in fDB.record: #print "adding",wfo.name,"to close out record" fDB.record[wfo.name] = { 'datasets' :{}, 'name' : wfo.name, 'closeOutWorkflow' : None, } fDB.record[wfo.name]['closeOutWorkflow'] = is_closing for output in wfi.request['OutputDatasets']: if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {} rec = fDB.record[wfo.name]['datasets'][output] rec['percentage'] = float('%.2f'%(percent_completions[output]*100)) rec['duplicate'] = duplications[output] if output in duplications else 'N/A' rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A' rec['closeOutDataset'] = is_closing rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A' rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output]))) rec['dbsFiles'] = dbs_presence[output] rec['dbsInvFiles'] = dbs_invalid[output] rec['phedexFiles'] = phedex_presence[output] rec['acdc'] = "%d / %d"%(len(acdc),len(acdc+acdc_inactive)) if by_pass_checks: ## force closing is_closing = True ## and move on if is_closing: ## toggle status to closed-out in request manager print "setting",wfo.name,"closed-out" if not options.test: res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name) print "close out answer",res if not res in ["None",None]: print "retrying to closing out" print res res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name) if res in [None,"None"]: wfo.status = 'close' session.commit() else: print "could not close out",wfo.name,"will try again next time" else: ## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it new_status = 'assistance'+sub_assistance print wfo.name,"needs assistance with",new_status if sub_assistance and wfo.status != new_status and 'PrepID' in wfi.request and not 'manual' in wfo.status: pid = wfi.getPrepIDs()[0].replace('task_','') #pid = wfi.request['PrepID'].replace('task_','') ## notify messages= { 'recovery' : 'Samples completed with missing lumi count:\n%s '%( '\n'.join(['%.2f %% complete for %s'%(percent_completions[output]*100, output) for output in wfi.request['OutputDatasets'] ] ) ), 'biglumi' : 'Samples completed with large luminosity blocks:\n%s '%('\n'.join(['%d > %d for %s'%(events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request['OutputDatasets'] if (events_per_lumi[output] > lumi_upper_limit[output])])), 'duplicate' : 'Samples completed with duplicated luminosity blocks:\n%s'%( '\n'.join(['%s'%output for output in wfi.request['OutputDatasets'] if output in duplications and duplications[output] ] ) ), } text ="The request %s (%s) is facing issue in production.\n" %( pid, wfo.name ) content = "" for case in messages: if case in new_status: content+= "\n"+messages[case]+"\n" text += content text += "You are invited to check, while this is being taken care of by Ops.\n" text += "This is an automated message." if use_mcm and content: print "Sending notification back to requestor" print text batches = mcm.getA('batches',query='contains=%s&status=announced'%pid) if len(batches): ## go notify the batch bid = batches[-1]['prepid'] print "batch nofication to",bid mcm.put('/restapi/batches/notify', { "notes" : text, "prepid" : bid}) ## go notify the request print "request notification to",pid mcm.put('/restapi/requests/notify',{ "message" : text, "prepids" : [pid] }) ## case where the workflow was in manual from recoveror if not 'manual' in wfo.status or new_status!='assistance-recovery': wfo.status = new_status if not options.test: print "setting",wfo.name,"to",wfo.status session.commit() else: print "current status is",wfo.status,"not changing to anything" fDB.html() ## custodial requests print "Custodials" print json.dumps(custodials, indent=2) for site in custodials: print ','.join(custodials[site]),'=>',site if not options.test: result = makeReplicaRequest(url, site, list(set(custodials[site])),"custodial copy at production close-out",custodial='y',priority='low', approve = (site in SI.sites_auto_approve) ) print result print "Transfers" print json.dumps(transfers, indent=2) ## replicas requests for site in transfers: print ','.join(transfers[site]),'=>',site if not options.test: result = None #result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out") print result print "File Invalidation" print invalidations
def transferor(url, specific=None, talk=True, options=None): if userLock('transferor'): return if options and options.test: execute = False else: execute = True SI = siteInfo() CI = campaignInfo() mcm = McMClient(dev=False) dss = DSS() print "counting all being handled..." being_handled = len( session.query(Workflow).filter(Workflow.status == 'away').all()) being_handled += len( session.query(Workflow).filter( Workflow.status.startswith('stag')).all()) being_handled += len( session.query(Workflow).filter( Workflow.status.startswith('assistance')).all()) max_to_handle = options.maxworkflows allowed_to_handle = max(0, max_to_handle - being_handled) wf_buffer = 5 if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer" else: print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer" print "... done" all_transfers = defaultdict(list) workflow_dependencies = defaultdict( set) ## list of wf.id per input dataset wfs_and_wfh = [] print "getting all wf to consider ..." cache = getWorkflows(url, 'assignment-approved', details=True) for wfo in session.query(Workflow).filter( Workflow.status == 'considered').all(): if specific and not specific in wfo.name: continue cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache) if len(cache_r): wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False, request=cache_r[0]))) else: wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False))) print "... done" input_sizes = {} ## list the size of those in transfer already in_transfer_priority = 0 min_transfer_priority = 100000000 print "getting all wf in staging ..." for wfo in session.query(Workflow).filter( Workflow.status == 'staging').all(): wfh = workflowInfo(url, wfo.name, spec=False) (_, primary, _, _) = wfh.getIO() for prim in primary: input_sizes[prim] = dss.get(prim) in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority'])) min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority'])) print "... done" print "Max priority in transfer already", in_transfer_priority print "Min priority in transfer already", min_transfer_priority in_transfer_already = sum(input_sizes.values()) #sort by priority higher first wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[ 'RequestPriority']), int(j[1].request['RequestPriority'])), reverse=True) ## list the size of all inputs print "getting all input sizes ..." for (wfo, wfh) in wfs_and_wfh: (_, primary, _, _) = wfh.getIO() for prim in primary: input_sizes[prim] = dss.get(prim) print "... done" grand_total = sum(input_sizes.values()) to_transfer = grand_total - in_transfer_already grand_transfer_limit = options.maxtransfer transfer_limit = grand_transfer_limit - in_transfer_already print "%15.4f GB already being transfered" % in_transfer_already print "%15.4f GB is the current requested transfer load" % to_transfer print "%15.4f GB is the global transfer limit" % grand_transfer_limit print "%15.4f GB is the available limit" % transfer_limit # the max priority value per dataset. max_priority = defaultdict(int) needs_transfer = 0 ## so that we can count'em passing_along = 0 transfer_sizes = {} went_over_budget = False for (wfo, wfh) in wfs_and_wfh: print wfh.request['RequestPriority'] print wfo.name, "to be transfered" #wfh = workflowInfo( url, wfo.name) (_, primary, _, _) = wfh.getIO() this_load = sum([input_sizes[prim] for prim in primary]) if (this_load and (sum(transfer_sizes.values()) + this_load > transfer_limit or went_over_budget)): if went_over_budget: print "Transfer has gone over bubget." else: print "Transfer will go over bubget." print "%15.4f GB this load" % this_load print "%15.4f GB already this round" % sum(transfer_sizes.values()) print "%15.4f GB is the available limit" % transfer_limit went_over_budget = True if int( wfh.request['RequestPriority'] ) >= in_transfer_priority and min_transfer_priority != in_transfer_priority: print "Higher priority sample", wfh.request[ 'RequestPriority'], ">=", in_transfer_priority, "go-on over budget" else: if not options.go: print min_transfer_priority, "minimum priority", wfh.request[ 'RequestPriority'], "<", in_transfer_priority, "stop" continue ## throtlle by campaign go if not CI.go(wfh.request['Campaign']): print "No go for", wfh.request['Campaign'] if not options.go: continue ## check if the batch is announced announced = False is_real = False for b in mcm.getA('batches', query='contains=%s' % wfo.name): is_real = True if b['status'] == 'announced': announced = True break if not announced: print wfo.name, "does not look announced." # skipping?, rejecting?, reporting?" if not is_real: print wfo.name, "does not appear to be genuine." ## prevent any duplication. if the wf is not mentioned in any batch, regardless of status continue ## check on a grace period injection_time = time.mktime( time.strptime('.'.join(map(str, wfh.request['RequestDate'])), "%Y.%m.%d.%H.%M.%S")) / (60. * 60.) now = time.mktime(time.gmtime()) / (60. * 60.) if float(now - injection_time) < 4.: if not options.go and not announced: print "It is too soon to start transfer: %3.2fH remaining" % ( now - injection_time) continue passing_along += 1 if passing_along >= allowed_to_handle: if int( wfh.request['RequestPriority'] ) >= in_transfer_priority and min_transfer_priority != in_transfer_priority: print "Higher priority sample", wfh.request[ 'RequestPriority'], ">=", in_transfer_priority, "go-on over", max_to_handle else: print "Not allowed to pass more than", max_to_handle, "at a time. Currently", being_handled, "handled, and adding", passing_along break (lheinput, primary, parent, secondary) = wfh.getIO() if options and options.tosites: sites_allowed = options.tosites.split(',') else: sites_allowed = getSiteWhiteList( (lheinput, primary, parent, secondary)) if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']): sites_allowed = CI.parameters( wfh.request['Campaign'])['SiteWhitelist'] blocks = [] if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']: blocks = wfh.request['BlockWhitelist'] can_go = True staging = False if primary: if talk: print wfo.name, 'reads', ', '.join(primary), 'in primary' ## chope the primary dataset for prim in primary: max_priority[prim] = max(max_priority[prim], int(wfh.request['RequestPriority'])) sites_really_allowed = [ site for site in sites_allowed if not any([ osite.startswith(site) for osite in SI.sites_veto_transfer ]) ] print "Sites allowed minus the vetoed transfer" print sites_really_allowed copies_needed = int( 0.35 * len(sites_really_allowed) ) + 1 ## should just go for a fixed number based if the white list grows that big print "Would make", copies_needed, "copies" if options.maxcopy > 0: copies_needed = min(options.maxcopy, copies_needed) ## remove the sites that do not want transfers print "need", copies_needed workflow_dependencies[prim].add(wfo.id) presence = getDatasetPresence(url, prim) prim_location = [ site for site, pres in presence.items() if pres[0] == True ] if len(prim_location) >= copies_needed: print "The output is all fully in place at", len( prim_location), "sites" continue # reduce the number of copies required by existing full copies copies_needed = max(0, copies_needed - len(prim_location)) print "now need", copies_needed subscriptions = listSubscriptions(url, prim) prim_destination = list( set([ site for (site, (tid, decision)) in subscriptions.items() if decision and not any([ site.endswith(veto) for veto in ['MSS', 'Export', 'Buffer'] ]) ])) ## need to reject from that list the ones with a full copy already: i.e the transfer corresponds to the copy in place prim_destination = [ site for site in prim_destination if not site in prim_location ] ## add transfer dependencies latching_on_transfers = list( set([ tid for (site, (tid, decision)) in subscriptions.items() if decision and site in prim_destination and not any([ site.endswith(veto) for veto in ['MSS', 'Export', 'Buffer'] ]) ])) print latching_on_transfers for latching in latching_on_transfers: tfo = session.query(Transfer).filter( Transfer.phedexid == latching).first() if not tfo: tfo = Transfer(phedexid=latching) tfo.workflows_id = [] session.add(tfo) if not wfo.id in tfo.workflows_id: print "adding", wfo.id, "to", tfo.id, "with phedexid", latching l = copy.deepcopy(tfo.workflows_id) l.append(wfo.id) tfo.workflows_id = l if not options.test: session.commit() else: session.flush( ) ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query can_go = False transfer_sizes[prim] = input_sizes[prim] staging = True # reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ?? copies_needed = max(0, copies_needed - len(prim_destination)) print "then need", copies_needed if copies_needed == 0: print "The output is either fully in place or getting in full somewhere with", latching_on_transfers can_go = True continue prim_to_distribute = [ site for site in sites_allowed if not any( [osite.startswith(site) for osite in prim_location]) ] prim_to_distribute = [ site for site in prim_to_distribute if not any( [osite.startswith(site) for osite in prim_destination]) ] ## take out the ones that cannot receive transfers prim_to_distribute = [ site for site in prim_to_distribute if not any([ osite.startswith(site) for osite in SI.sites_veto_transfer ]) ] if len( prim_to_distribute ) > 0: ## maybe that a parameter we can play with to limit the if not options or options.chop: spreading = distributeToSites(getDatasetChops(prim), prim_to_distribute, n_copies=copies_needed, weights=SI.cpu_pledges) else: spreading = {} for site in prim_to_distribute: spreading[site] = [prim] can_go = False transfer_sizes[prim] = input_sizes[prim] for (site, items) in spreading.items(): all_transfers[site].extend(items) if secondary: if talk: print wfo.name, 'reads', ', '.join(secondary), 'in secondary' for sec in secondary: workflow_dependencies[sec].add(wfo.id) presence = getDatasetPresence(url, sec) sec_location = [ site for site, pres in presence.items() if pres[1] > 90. ] ## more than 90% of the minbias at sites subscriptions = listSubscriptions(url, sec) sec_destination = [site for site in subscriptions] sec_to_distribute = [ site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location]) ] sec_to_distribute = [ site for site in sec_to_distribute if not any( [osite.startswith(site) for osite in sec_destination]) ] sec_to_distribute = [ site for site in sec_to_distribute if not any([ osite.startswith(site) for osite in SI.sites_veto_transfer ]) ] if len(sec_to_distribute) > 0: for site in sec_to_distribute: all_transfers[site].append(sec) can_go = False ## is that possible to do something more if can_go: ## no explicit transfer required this time if staging: ## but using existing ones print wfo.name, "latches on existing transfers, and nothing else" wfo.status = 'staging' else: print wfo.name, "should just be assigned NOW to", sites_allowed wfo.status = 'staged' print "setting status to", wfo.status session.commit() continue else: ## there is an explicit transfer required if staging: ## and also using an existing one print wfo.name, "latches on existing transfers" if not options.test: wfo.status = 'staging' print "setting status to", wfo.status session.commit() print wfo.name, "needs a transfer" needs_transfer += 1 #print json.dumps(all_transfers) fake_id = -1 wf_id_in_prestaging = set() for (site, items_to_transfer) in all_transfers.iteritems(): items_to_transfer = list(set(items_to_transfer)) ## convert to storage element site_se = SI.CE_to_SE(site) ## site that do not want input datasets if site in SI.sites_veto_transfer: print site, "does not want transfers" continue ## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured. ## massage a bit the items blocks = [it for it in items_to_transfer if '#' in it] datasets = [it for it in items_to_transfer if not '#' in it] if execute: print "Making a replica to", site, "(CE)", site_se, "(SE) for" else: print "Would make a replica to", site, "(CE)", site_se, "(SE) for" print "\t", len(blocks), "blocks" ## remove blocks if full dataset is send out blocks = [ block for block in blocks if not block.split('#')[0] in datasets ] print "\t", len(blocks), "needed blocks for", list( set([block.split('#')[0] for block in blocks])) print "\t", len(datasets), "datasets" print "\t", datasets items_to_transfer = blocks + datasets ## operate the transfer if options and options.stop: ## ask to move-on answer = raw_input('Continue with that ?') if not answer.lower() in ['y', 'yes', 'go']: continue if execute: result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority='normal') ## make use of max_priority dataset:priority to set the subscriptions priority """ ## does not function once = True for item in items_to_transfer: bds = item.split('#')[0] if max_priority[bds] >= 90000: if once: w=10 print "waiting",w,"s before raising priority" time.sleep(w) once=False ## raise it to high priority print item,"subscription priority raised to high at",site_se #print "This does not work yet properly it seems" print updateSubscription(url, site_se, item, priority='high') """ else: #result= {'phedex':{'request_created' : [{'id' : fake_id}]}} result = {'phedex': {'request_created': []}} fake_id -= 1 if not result: print "ERROR Could not make a replica request for", site, items_to_transfer, "pre-staging" continue for phedexid in [o['id'] for o in result['phedex']['request_created']]: new_transfer = session.query(Transfer).filter( Transfer.phedexid == phedexid).first() print phedexid, "transfer created" if not new_transfer: new_transfer = Transfer(phedexid=phedexid) session.add(new_transfer) new_transfer.workflows_id = set() for transfering in list( set(map(lambda it: it.split('#')[0], items_to_transfer))): new_transfer.workflows_id.update( workflow_dependencies[transfering]) new_transfer.workflows_id = list(new_transfer.workflows_id) wf_id_in_prestaging.update(new_transfer.workflows_id) session.commit() ## auto approve it if execute: approved = approveSubscription(url, phedexid, [site_se]) for wfid in wf_id_in_prestaging: tr_wf = session.query(Workflow).get(wfid) if tr_wf and tr_wf.status != 'staging': if execute: tr_wf.status = 'staging' if talk: print "setting", tr_wf.name, "to staging" session.commit()