price_perf_row["EntryName"] = r1["EntryName"].values[0]
fom_row["PerfTtbarTotal"] = r1["PerfTtbarTotal"].values[0]
fom_row["EntryName"] = r1["EntryName"].values[0]
if r2.empty:
running_vms = 0
else:
running_vms = r2["RunningVms"].values[0]
if r3.empty:
price_perf_row["MaxLimit"] = DEFAULT_MAX_LIMIT
fom_row["MaxLimit"] = DEFAULT_MAX_LIMIT
else:
price_perf_row["MaxLimit"] = r3["MaxLimit"].values[0]
fom_row["MaxLimit"] = r3["MaxLimit"].values[0]
price_perf = price_performance(row["SpotPrice"], price_perf_row["PerfTtbarTotal"])
price_perf_row["AWS_Price_Performance"] = price_perf
fom_row["AWS_Figure_Of_Merit"] = figure_of_merit(running_vms, price_perf_row["MaxLimit"], price_perf)
price_perf_rows.append(price_perf_row)
fom_rows.append(fom_row)
price_perf_df = pd.DataFrame(price_perf_rows)
price_perf_df.reindex(sorted(price_perf_df.columns), axis=1)
fom_df = pd.DataFrame(fom_rows)
fom_df.reindex(sorted(fom_df.columns), axis=1)
return {"AWS_Price_Performance": price_perf_df, "AWS_Figure_Of_Merit": fom_df}
Transform.describe(FigureOfMerit)
raise RuntimeError('Mismatch in manifest keys: %s, %s' %
(m_keys, g_keys))
for key in m_keys:
merged_manifests[key] = manifests[key].append(
group_manifests[key], ignore_index=True)
else:
merged_manifests = group_manifests
return merged_manifests
def read_fe_config(self):
if not os.path.isfile(self.de_frontend_configfile):
raise RuntimeError(
'Error reading Frontend config for DE %s. '
'Run configure_gwms_frontend.py to generate one and after every change to the frontend configuration.'
% self.de_frontend_configfile)
fe_cfg = eval(open(self.de_frontend_configfile, 'r').read())
if not isinstance(fe_cfg, dict):
raise ValueError('Frontend config for DE in %s is invalid' %
self.de_frontend_configfile)
return fe_cfg
def sanitize_glidein_cpus(row):
if str(row['GLIDEIN_CPUS']).lower() == 'auto':
row['GLIDEIN_CPUS'] = row['GLIDEIN_ESTIMATED_CPUS']
row['GLIDEIN_CPUS'] = int(row['GLIDEIN_CPUS'])
return row
Transform.describe(GlideinRequestManifests)
self.logger.info(
f"total number of slots on Nersc = {total_slots_nersc}")
# pull slot info from User pool ########
total_slots_userpool = 0
for _index, row in userpool_slots_df.iterrows():
total_slots_userpool += int(row["TotalCpus"])
self.logger.info(
f"total number of slots on userpool = {total_slots_userpool}")
# compute the relative difference between the two metrics ###
rel_diff = 0.0
diff = abs(total_slots_userpool - total_slots_nersc)
if diff != 0:
more = max(total_slots_nersc, total_slots_userpool)
rel_diff = diff / more
self.logger.info(f"diff = {diff:f}, rel diff = {rel_diff:f}")
# construct the result namespace ############################
results["nersc.count"] = total_slots_nersc
results["userpool.count"] = total_slots_userpool
results["relative_diff"] = rel_diff
return {"nersc_userpool_slots_comparison": results}
Transform.describe(CompareNerscUserpoolSlots)
def __init__(self, config):
super().__init__(config)
def transform(self, data_block):
"""
Fills ``self.data`` with spot price data.
:type spot_price_history: :obj:`list`
:arg spot_price_history: list of spotprice data (:class:`SpotPriceData`)
"""
self.logger.debug("in AWSOccupancy transform")
account_dict = data_block.get("spot_occupancy_config").to_dict()
self.logger.debug(f"account_dict {account_dict}")
occupancy_data = []
for account in account_dict:
for region in account_dict[account]:
occcupancy = OccupancyForRegion(region, profile_name=account)
instances = occcupancy.get_ec2_instances()
if instances:
data = occcupancy.capacity(instances)
if data:
occupancy_data += data
oc_list = [i.data for i in occupancy_data]
# to fix the test failure
return {"AWS_Occupancy": pd.DataFrame(oc_list)}
Transform.describe(AWSOccupancy)
"GlideinConfigPerEntryMaxGlideins"
], ]
df["GLIDEIN_Supported_VOs"] = vo
if limits_df is None:
limits_df = df
else:
limits_df = limits_df.append(df, ignore_index=True)
az_it = sub_df.loc[
sub_df["GLIDEIN_Supported_VOs"].str.contains(vo),
["INSTANCE_TYPE", "AVAILABILITY_ZONE"]]
regions = {
az[:-1]
for az in az_it.AVAILABILITY_ZONE.unique()
}
so_config_dict[vo] = {}
for region in regions:
it = az_it.loc[az_it["AVAILABILITY_ZONE"].str.contains(
region)].INSTANCE_TYPE.unique().tolist()
so_config_dict[vo][region] = it
limits_df = limits_df.rename(columns=_ATTR_TRANSLATION_MAP)
return {
"aws_instance_limits": limits_df,
"spot_occupancy_config": pd.DataFrame.from_dict(so_config_dict)
}
Transform.describe(AWSFactoryEntryData)
# SPDX-FileCopyrightText: 2017 Fermi Research Alliance, LLC
# SPDX-License-Identifier: Apache-2.0
from decisionengine.framework.modules import Transform
from decisionengine.framework.modules.Transform import Parameter
@Transform.supports_config(Parameter("data_product_name", type=str),
Parameter("consumes", type=list))
class DynamicTransform(Transform.Transform):
def __init__(self, config):
self.data_product_name = config["data_product_name"]
self._consumes = dict.fromkeys(config["consumes"], int)
self._produces = {self.data_product_name: int}
def transform(self, data_block):
all_values = [data_block[key] for key in self._consumes.keys()]
return {self.data_product_name: sum(all_values, 0)}
Transform.describe(DynamicTransform)
# SPDX-FileCopyrightText: 2017 Fermi Research Alliance, LLC
# SPDX-License-Identifier: Apache-2.0
import pandas as pd
from decisionengine.framework.modules import Transform
@Transform.consumes(foo=pd.DataFrame)
@Transform.produces(bar=pd.DataFrame)
class TransformNOP(Transform.Transform):
def __init__(self, config):
super().__init__(config)
def transform(self, data_block):
df_in = self.foo(data_block)
return {"bar": pd.DataFrame(df_in["key2"])}
Transform.describe(TransformNOP)
from decisionengine.framework.modules import Transform
@Transform.consumes(GCE_Instance_Performance=pd.DataFrame,
GCE_Occupancy=pd.DataFrame)
@Transform.produces(GCE_Burn_Rate=pd.DataFrame)
class GceBurnRate(Transform.Transform):
def __init__(self, config):
super().__init__(config)
def transform(self, data_block):
self.logger.debug("in GceBurnRate transform")
performance = self.GCE_Instance_Performance(data_block).fillna(0)
occupancy = self.GCE_Occupancy(data_block).fillna(0)
burn_df = pd.DataFrame([{"BurnRate": 0.0}])
if not occupancy.empty:
df = pd.merge(occupancy,
performance,
how="inner",
on=["AvailabilityZone", "InstanceType"])
if not df.empty:
df["BurnRate"] = pd.to_numeric(
df["Occupancy"]) * pd.to_numeric(df["PreemptiblePrice"])
burn_df = pd.DataFrame([{"BurnRate": df["BurnRate"].sum()}])
return {"GCE_Burn_Rate": burn_df}
Transform.describe(GceBurnRate)
self.logger.debug("in AWSSpotPrice transform")
account_dict = data_block.get("spot_occupancy_config").to_dict()
self.logger.debug(f"account_dict {account_dict}")
sp_data = []
for account in account_dict:
for region, instances in account_dict[account].items():
spot_price_info = AWSSpotPriceForRegion(region,
profile_name=account)
spot_price_info.init_query(instance_types=instances)
spot_price_history = spot_price_info.get_price(self.logger)
if spot_price_history:
sp_data += spot_price_info.spot_price_summary(
spot_price_history)
sp_list = [i.data for i in sp_data]
column_names = [
"AccountName",
"AvailabilityZone",
"InstanceType",
"ProductDescription",
"SpotPrice",
"Timestamp",
]
return {
"provisioner_resource_spot_prices":
pd.DataFrame(sp_list, columns=column_names)
}
Transform.describe(AWSSpotPrice)
@Transform.consumes(provisioner_resource_spot_prices=pd.DataFrame,
AWS_Occupancy=pd.DataFrame)
@Transform.produces(AWS_Burn_Rate=pd.DataFrame)
class AwsBurnRate(Transform.Transform):
def __init__(self, config):
super().__init__(config)
def transform(self, data_block):
spot_prices = self.provisioner_resource_spot_prices(data_block).fillna(
0)
occupancy = self.AWS_Occupancy(data_block).fillna(0)
burn_df = pd.DataFrame([{"BurnRate": 0.}])
if not occupancy.empty:
df = pd.merge(
occupancy,
spot_prices,
how="inner",
on=["AccountName", "AvailabilityZone", "InstanceType"])
if not df.empty:
df["BurnRate"] = pd.to_numeric(
df["RunningVms"]) * pd.to_numeric(df["SpotPrice"])
burn_df = pd.DataFrame([{"BurnRate": df["BurnRate"].sum()}])
return {'AWS_Burn_Rate': burn_df}
Transform.describe(AwsBurnRate)
factory_entries_lcf = self.Factory_Entries_LCF(data_block)
figures_of_merit = []
for _i, row in factory_entries_lcf.iterrows():
entry_name = row["EntryName"]
perf_df = performance[performance.EntryName == entry_name]
for _j, perf_row in perf_df.iterrows():
running = float(row["GlideinMonitorTotalStatusRunning"])
max_allowed = float(row["GlideinConfigPerEntryMaxGlideins"])
max_idle = float(row["GlideinConfigPerEntryMaxIdle"])
idle = float(row["GlideinMonitorTotalStatusIdle"])
figures_of_merit.append({
"EntryName":
entry_name,
"FigureOfMerit":
fom.figure_of_merit(perf_row["PricePerformance"], running,
max_allowed, idle, max_idle,
self.logger),
})
return {
"Nersc_Price_Performance":
performance.filter(["EntryName", "PricePerformance"]),
"Nersc_Figure_Of_Merit":
pd.DataFrame(figures_of_merit),
}
Transform.describe(NerscFigureOfMerit)
if num_with_id == 0:
# both and factory_only stay '0'
results['factory_only.count'] = len(nersc_df)
else:
# compare the exact job IDs
nersc_id_list = nersc_df.jobid.tolist()
factory_id_set = set(factory_id_list)
nersc_id_set = set(nersc_id_list)
in_both_set = factory_id_set & nersc_id_set
factory_only_set = factory_id_set - nersc_id_set
nersc_only_set = nersc_id_set - factory_id_set
num_in_both = len(in_both_set)
num_in_factory = len(factory_only_set)
num_in_nersc = len(nersc_only_set)
results['both.count'] = num_in_both
results['nersc_only.count'] = num_in_nersc
results['factory_only.count'] = num_in_factory
for index, row in nersc_df.iterrows():
if row['status'] == 'R':
results['nersc.running.count'] += 1
return {'nersc_factory_jobs_comparison': results}
Transform.describe(CompareNerscFactoryJobs)
self.logger = logging.getLogger()
self.price_performance = config.get('price_performance', 1)
def transform(self, datablock):
"""
Grid sites FOM are straight up assumed as 0 for now
"""
entries = self.Factory_Entries_Grid(datablock)
if entries is None:
entries = pandas.DataFrame({ATTR_ENTRYNAME: []})
foms = []
if not entries.empty:
for index, entry in entries.iterrows():
running = float(entry['GlideinMonitorTotalStatusRunning'])
max_allowed = float(entry['GlideinConfigPerEntryMaxGlideins'])
max_idle = float(entry['GlideinConfigPerEntryMaxIdle'])
idle = float(entry['GlideinMonitorTotalStatusIdle'])
f = {
ATTR_ENTRYNAME: entry[ATTR_ENTRYNAME],
ATTR_FOM: figure_of_merit(self.price_performance,
running, max_allowed,
idle, max_idle)
}
foms.append(f)
return {'Grid_Figure_Of_Merit': pandas.DataFrame(foms)}
Transform.describe(GridFigureOfMerit)
df_job_clusters = pd.DataFrame(totals,
columns=[
"Job_Bucket_Criteria_Expr",
"Site_Bucket_Criteria_Expr",
"Totals", "Frontend_Group"
])
self.logger.debug(f"Job category totals: {df_job_clusters}")
except KeyError:
self.logger.exception(
"Unable to calculate totals from job manifests, may have missing classads or incorrect classad names"
)
return {"job_clusters": self.EMPTY_JOB_CLUSTER}
except ValueError:
self.logger.exception(
"Unable to calculate totals from job manifests, may have missing classads or incorrect classad names"
)
return {"job_clusters": self.EMPTY_JOB_CLUSTER}
except pd.core.computation.ops.UndefinedVariableError:
self.logger.exception(
"Unable to calculate totals from job manifests, may have missing classads or incorrect classad names"
)
return {"job_clusters": self.EMPTY_JOB_CLUSTER}
self.logger.info("*** Ending job clustering ***")
return {"job_clusters": df_job_clusters}
Transform.describe(JobClustering)
