def dataframe(self, timeframe, seed, dformat, dbs_extra, newdata=None):
"""Form a dataframe from various CMS data-providers"""
dtypes, stypes, rtypes, tiers = self.data_types()
pop_datasets = 0
dbs_datasets = 0
popdb_results = [r for r in self.popdb.dataset_stat(timeframe[0], timeframe[1])]
if dformat == 'csv':
row = popdb_results[0]
dataset = row['dataset']
target = dict(naccess=row['naccess'],nusers=row['nusers'],totcpu=row['totcpu'],
rnaccess=row['rnaccess'],rnusers=row['rnusers'],rtotcpu=row['rtotcpu'])
# seed dataset to determine headers of the dataframe
rows = self.dataset_info(timeframe, seed, dtypes, stypes, rtypes,
tiers, 'headers', target)
headers = [r for r in rows][0]
yield ','.join(headers)
tstamp = time.strftime("%Y-%m-%d %H:%M:%S GMT", time.gmtime())
if newdata: # request new dataset
if self.verbose:
print("Generate dataframe for new datasets", tstamp)
n_days = 7
if timeframe:
n_days = ndays(yyyymmdd(timeframe[0]), yyyymmdd(timeframe[1]))
new_datasets = self.dbs.new_datasets(n_days)
target = dict(naccess=0,nusers=0,totcpu=0,
rnaccess=0,rnusers=0,rtotcpu=0)
for row in new_datasets:
dataset = row['dataset']
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
for row in rows:
yield row
return
# get list of popular datasets in certain time frame
# popdb_results = [r for r in self.popdb.dataset_stat(timeframe[0], timeframe[1])]
popdb_datasets = {} #
for row in popdb_results:
dataset = row['dataset']
if not DATASET_PAT.match(dataset):
continue
if self.verbose:
print("Generate dataframe for %s, timeframe: %s, %s" \
% (dataset, timeframe, tstamp))
target = dict(naccess=row['naccess'],nusers=row['nusers'],totcpu=row['totcpu'],
rnaccess=row['rnaccess'],rnusers=row['rnusers'],rtotcpu=row['rtotcpu'])
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
popdb_datasets[dataset] = row
for row in rows:
yield row
pop_datasets += 1
# get list of datasets from DBS and discard from this list
# those who were presented in popdb
all_dbs_datasets = self.dbs.datasets()
dbsdatasets = [d for d in all_dbs_datasets if d not in popdb_datasets.keys()]
target = dict(naccess=0,nusers=0,totcpu=0,
rnaccess=0,rnusers=0,rtotcpu=0)
for dataset in random.sample(dbsdatasets, dbs_extra):
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
for row in rows:
yield row
dbs_datasets += 1
if self.verbose:
print("DBS datasets : %s" % dbs_datasets)
print("PopDB datasets: %s out of %s" % (pop_datasets, len(popdb_results)))
def dataframe(self, timeframe, seed, dformat, dbs_extra, newdata=None):
"""Form a dataframe from various CMS data-providers"""
dtypes, stypes, rtypes, tiers = self.data_types()
pop_datasets = 0
dbs_datasets = 0
popdb_results = [
r for r in self.popdb.dataset_stat(timeframe[0], timeframe[1])
]
if dformat == 'csv':
dataset = None
for row in popdb_results:
if len(row['dataset'].split(
'/')) == 4: # dataset with 3 slashes
dataset = row['dataset']
break
if not dataset:
raise Exception(
"Unable to find valid dataset name in popdb output")
target = dict(naccess=row['naccess'],
nusers=row['nusers'],
totcpu=row['totcpu'],
rnaccess=row['rnaccess'],
rnusers=row['rnusers'],
rtotcpu=row['rtotcpu'])
# seed dataset to determine headers of the dataframe
rows = self.dataset_info(timeframe, seed, dtypes, stypes, rtypes,
tiers, 'headers', target)
headers = [r for r in rows][0]
yield ','.join(headers)
tstamp = time.strftime("%Y-%m-%d %H:%M:%S GMT", time.gmtime())
if newdata: # request new dataset
if self.verbose:
print("Generate dataframe for new datasets", tstamp)
n_days = 7
if timeframe:
n_days = ndays(yyyymmdd(timeframe[0]), yyyymmdd(timeframe[1]))
new_datasets = self.dbs.new_datasets(n_days)
target = dict(naccess=0,
nusers=0,
totcpu=0,
rnaccess=0,
rnusers=0,
rtotcpu=0)
for row in new_datasets:
dataset = row['dataset']
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
for row in rows:
yield row
return
# get list of popular datasets in certain time frame
# popdb_results = [r for r in self.popdb.dataset_stat(timeframe[0], timeframe[1])]
popdb_datasets = {} #
for row in popdb_results:
dataset = row['dataset']
if not DATASET_PAT.match(dataset):
continue
if self.verbose:
print("Generate dataframe for %s, timeframe: %s, %s" \
% (dataset, timeframe, tstamp))
target = dict(naccess=row['naccess'],
nusers=row['nusers'],
totcpu=row['totcpu'],
rnaccess=row['rnaccess'],
rnusers=row['rnusers'],
rtotcpu=row['rtotcpu'])
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
popdb_datasets[dataset] = row
for row in rows:
yield row
pop_datasets += 1
# get list of datasets from DBS and discard from this list
# those who were presented in popdb
all_dbs_datasets = self.dbs.datasets()
dbsdatasets = [
d for d in all_dbs_datasets if d not in popdb_datasets.keys()
]
target = dict(naccess=0,
nusers=0,
totcpu=0,
rnaccess=0,
rnusers=0,
rtotcpu=0)
for dataset in random.sample(dbsdatasets, dbs_extra):
rows = self.dataset_info(timeframe, dataset, dtypes, stypes, \
rtypes, tiers, dformat, target)
for row in rows:
yield row
dbs_datasets += 1
if self.verbose:
print("DBS datasets : %s" % dbs_datasets)
print("PopDB datasets: %s out of %s" %
(pop_datasets, len(popdb_results)))
def dataset_info(self, timeframe, dataset, dtypes, stypes, rtypes, tiers, dformat, target=0):
"Return common dataset info in specified data format"
dbsinst = self.dbs.dataset_dbsinst(dataset)
if not dbsinst:
return
row = self.dbs.dataset_info(dataset, dbsinst)
if row:
if self.multitask:
releases, sites, parents, summary, dashboard = \
self.dataset_info_all(dataset, dbsinst, timeframe)
else:
releases = [rname for rname in self.dbs.dataset_release_versions(dataset, dbsinst)]
sites = [sname for sname in self.phedex.sites(dataset)]
parents = [r for r in self.dbs.dataset_parents(dataset, dbsinst)]
summary = self.dbs.dataset_summary(dataset, dbsinst)
dashboard = self.dashboard.dataset_info(dataset, timeframe[0], timeframe[1])
nrels = len(releases)
series = rtypes['series']
majors = rtypes['majors']
minors = rtypes['minors']
relclf = rtypes['rtypes']
for rel in releases:
rserie, rmajor, rminor = rel_ver(rel)
if not cmssw_test(rserie, rmajor, rminor):
continue
rtype = rel_type(rel)
try:
series['rel1_%s'%rserie] += 1
except:
pass
try:
majors['rel2_%s'%rmajor] += 1
except:
pass
try:
minors['rel3_%s'%rminor] += 1
except:
pass
try:
relclf['relt_%s'%rtype] += 1
except:
pass
nsites = len(sites)
for site in sites:
stier = site_tier(site)
stypes['s_%s'%stier] += 1
dataset_id = row['rid']
era = genkey(row['acquisition_era_name'], self.salt, 5)
create_dn = self.sitedb.dnid(row['create_by'])
dbsinstid = row['dbs_instance']
dtype = row['primary_ds_type']
# number of data types should be small and simple
# list look-up shouldn't be a problem
if dtype not in dtypes:
dtypes.append(dtype)
dtype = dtypes.index(dtype)
_, prim, proc, tier = dataset.split('/')
prim = genkey(prim, self.salt, 5)
proc = genkey(proc, self.salt, 5)
if tier not in tiers:
tiers.append(tier)
tier = genkey(tier, self.salt, 5)
parent = parents[0] if len(parents) else 0
uid = genuid(yyyymmdd(timeframe[0]), dbsinstid, dataset_id)
size_norm = 2**30 # normalization factor for file size
rec = dict(id=uid, dataset=dataset_id, primds=prim, procds=proc, tier=tier,
dtype=dtype, creator=create_dn, nrel=nrels, nsites=nsites,
parent=parent, era=era, dbs=dbsinstid,
nfiles=summary.get('num_file', 0),
nlumis=summary.get('num_lumi', 0),
nblk=summary.get('num_block', 0),
nevt=summary.get('num_event', 0),
size=summary.get('file_size', 0)/size_norm,
cpu=dashboard.get('cpu', 0),
wct=dashboard.get('wct', 0),
proc_evts=dashboard.get('nevt', 0))
if isinstance(target, dict):
rec.update(target)
for key,val in series.items():
rec.update({key:val})
for key, val in majors.items():
rec.update({key:val})
for key, val in minors.items():
rec.update({key:val})
for key, val in relclf.items():
rec.update({key:val})
for key, val in stypes.items():
rec.update({key:val})
headers = rec.keys()
headers.sort()
headers.remove('id')
headers = ['id'] + headers # let dataset id be the first column
if dformat == 'headers':
yield headers
elif dformat == 'csv':
res = [str(rec[h]) for h in headers]
yield ','.join(res)
elif dformat == 'vw':
target_str = target.get('rnaccess')
vals = ' '.join([str(rec[h]) for h in headers])
uid = genkey(vals, self.salt, 5) # unique row identified
vwrow = "%s '%s |f %s" % (target_str, uid, vals)
yield vwrow
def dataset_info(self,
timeframe,
dataset,
dtypes,
stypes,
rtypes,
tiers,
dformat,
target=0):
"Return common dataset info in specified data format"
dbsinst = self.dbs.dataset_dbsinst(dataset)
if not dbsinst:
return
row = self.dbs.dataset_info(dataset, dbsinst)
if row:
if self.multitask:
releases, sites, parents, summary, dashboard = \
self.dataset_info_all(dataset, dbsinst, timeframe)
else:
releases = [
rname for rname in self.dbs.dataset_release_versions(
dataset, dbsinst)
]
sites = [sname for sname in self.phedex.sites(dataset)]
parents = [
r for r in self.dbs.dataset_parents(dataset, dbsinst)
]
summary = self.dbs.dataset_summary(dataset, dbsinst)
dashboard = self.dashboard.dataset_info(
dataset, timeframe[0], timeframe[1])
nrels = len(releases)
series = {}
for k in rtypes['series'].keys():
series[k] = 0
majors = {}
for k in rtypes['majors'].keys():
majors[k] = 0
minors = {}
for k in rtypes['minors'].keys():
minors[k] = 0
relclf = {}
for k in rtypes['rtypes'].keys():
relclf[k] = 0
for rel in releases:
rserie, rmajor, rminor = rel_ver(rel)
if not cmssw_test(rserie, rmajor, rminor):
continue
rtype = rel_type(rel)
try:
series['rel1_%s' % rserie] += 1
except:
pass
try:
majors['rel2_%s' % rmajor] += 1
except:
pass
try:
minors['rel3_%s' % rminor] += 1
except:
pass
try:
relclf['relt_%s' % rtype] += 1
except:
pass
nsites = len(sites)
for site in sites:
stier = site_tier(site)
stypes['s_%s' % stier] += 1
dataset_id = row['rid']
era = genkey(row['acquisition_era_name'], self.salt, 5)
create_dn = self.sitedb.dnid(row['create_by'])
dbsinstid = row['dbs_instance']
dtype = row['primary_ds_type']
# number of data types should be small and simple
# list look-up shouldn't be a problem
if dtype not in dtypes:
dtypes.append(dtype)
dtype = dtypes.index(dtype)
_, prim, proc, tier = dataset.split('/')
prim = genkey(prim, self.salt, 5)
proc = genkey(proc, self.salt, 5)
if tier not in tiers:
tiers.append(tier)
tier = genkey(tier, self.salt, 5)
parent = parents[0] if len(parents) else 0
uid = genuid(yyyymmdd(timeframe[0]), dbsinstid, dataset_id)
size_norm = 2**30 # normalization factor for file size
if not summary:
summary = {} # we need a dict type
rec = dict(id=uid,
dataset=dataset_id,
primds=prim,
procds=proc,
tier=tier,
dtype=dtype,
creator=create_dn,
nrel=nrels,
nsites=nsites,
parent=parent,
era=era,
dbs=dbsinstid,
nfiles=summary.get('num_file', 0),
nlumis=summary.get('num_lumi', 0),
nblk=summary.get('num_block', 0),
nevt=summary.get('num_event', 0),
size=summary.get('file_size', 0) / size_norm,
cpu=dashboard.get('cpu', 0),
wct=dashboard.get('wct', 0),
proc_evts=dashboard.get('nevt', 0))
if isinstance(target, dict):
rec.update(target)
for key, val in series.items():
rec.update({key: val})
for key, val in majors.items():
rec.update({key: val})
for key, val in minors.items():
rec.update({key: val})
for key, val in relclf.items():
rec.update({key: val})
for key, val in stypes.items():
rec.update({key: val})
headers = rec.keys()
headers.sort()
headers.remove('id')
headers = ['id'] + headers # let dataset id be the first column
if dformat == 'headers':
yield headers
elif dformat == 'csv':
res = [str(rec[h]) for h in headers]
yield ','.join(res)
elif dformat == 'vw':
target_str = target.get('rnaccess')
vals = ' '.join([str(rec[h]) for h in headers])
uid = genkey(vals, self.salt, 5) # unique row identified
vwrow = "%s '%s |f %s" % (target_str, uid, vals)
yield vwrow