def cli():
import argparse, csv
parser = argparse.ArgumentParser(prog='{} {}'.format(*(sys.argv[0], sys.argv[1])))
parser.add_argument('--config', required=True, help='Config file to load for Ariel configuration')
args = parser.parse_args(args=sys.argv[2:])
config = utils.load_config(args.config)
from ariel import get_account_names, get_account_instance_summary, get_ec2_pricing, get_reserved_instances
account_names = get_account_names.load(config)
instances = get_account_instance_summary.load(config)
ris = get_reserved_instances.load(config)
pricing = get_ec2_pricing.load(config)
reports = generate(config, instances, ris, pricing)
for key, report in reports.items():
LOGGER.info("Writing {} report to ./output_{}.csv".format(key, key.lower()))
# Decorate report
if 'accountid' in report.columns and 'accountname' not in report.columns:
accountname_column = report.columns.get_loc('accountid') + 1
input_column = 'Account ID' if 'Account ID' in report.columns else 'accountid'
accountname_value = report[input_column].apply(lambda x: account_names[x] if x in account_names else x)
report.insert(accountname_column, 'accountname', accountname_value)
store_index = type(report.index) != pd.RangeIndex
report.to_csv("output_{}.csv".format(key.lower()), index=store_index)
LOGGER.debug("Report {}:\n".format(key) + str(report))
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-account-names.yaml'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
pass
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
account, role = utils.get_master(config)
retries = utils.get_config_value(config, 'ACCOUNT_NAMES', 'RETRIES', 5)
file = utils.get_config_value(config, 'ACCOUNT_NAMES', 'FILE', '')
account_names = {}
if role != '':
# Organizations should be queried, load that first
session = utils.assume_role(boto3.Session(), role)
org = session.client('organizations', region_name='us-east-1')
rsp = org.list_accounts()
while True:
for account in rsp['Accounts']:
account_names[account['Id']] = account['Name']
if 'NextToken' in rsp:
for i in range(retries):
try:
rsp = org.list_accounts(NextToken=rsp['NextToken'])
break
except ClientError as e:
if i == retries:
raise e
sleep(0.5 + 0.1 * i)
continue
break
if file != '':
# Update account names with file contents
with utils.get_read_handle(file) as f:
account_names.update(yaml.load(f, Loader=yaml.FullLoader))
with open(cache_file, 'w') as outfile:
yaml.dump(account_names, outfile, default_flow_style=False)
return account_names
with utils.get_read_handle(cache_file) as input:
account_names = yaml.load(input, Loader=yaml.FullLoader)
return account_names
def load_config(filename):
with get_read_handle(filename) as f:
config = yaml.load(f, Loader=yaml.FullLoader)
# Initialize Logging Config
log_level = get_config_value(config, 'DEFAULTS', 'LOG_LEVEL', '')
if log_level != '':
from ariel import LOGGER
import logging
LOGGER.setLevel(logging.getLevelName(log_level))
return config
def cli():
import argparse, csv
parser = argparse.ArgumentParser(prog='{} {}'.format(*(sys.argv[0],
sys.argv[1])))
parser.add_argument('--config',
required=True,
help='Config file to load for Ariel configuration')
args = parser.parse_args(args=sys.argv[2:])
config = utils.load_config(args.config)
from ariel import get_account_instance_summary, get_ec2_pricing, get_reserved_instances
instances = get_account_instance_summary.load(config)
ris = get_reserved_instances.load(config)
pricing = get_ec2_pricing.load(config)
reports = generate(config, instances, ris, pricing)
for key, report in reports.items():
LOGGER.info("Writing {} report to ./output_{}.csv".format(
key, key.lower()))
store_index = type(report.index) != pd.RangeIndex
report.to_csv("output_{}.csv".format(key.lower()), index=store_index)
LOGGER.debug("Report {}:\n".format(key) + str(report))
def generate(config, instances, ris, pricing):
def get_units(instancetype):
try:
return pricing['units'][instancetype]
except KeyError as e:
if '.' in instancetype:
raise e
for key in pricing['units']:
if key.endswith('.' + instancetype):
return pricing['units'][key]
raise e
# Make sure we have a reasonable about of data
if instances['usagestartdate'].max() - instances['usagestartdate'].min(
) < timedelta(days=14):
raise ValueError('Insufficient Data')
# Preaggregate some data
timerange = instances['usagestartdate'].unique()
# Add some additional data to instances
hourofweek_column = instances.columns.get_loc('usagestartdate') + 1
hourofweek_value = instances[
'usagestartdate'].dt.dayofweek * 24 + instances[
'usagestartdate'].dt.hour
instances.insert(hourofweek_column, 'hourofweek', hourofweek_value)
region_column = instances.columns.get_loc('availabilityzone')
region_value = instances['availabilityzone'].str[:-1]
instances.insert(region_column, 'region', region_value)
family_column = instances.columns.get_loc('instancetype')
# meckstmd:07/29/2019 - Metal RIs are no different than regular RIs - they are a family with a normalization factor
# for example, i3.metal is equivalent to i3.16xlarge. See https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/apply_ri.html
#family_value = instances['instancetype'].apply(lambda x: x if x.endswith('.metal') else x.split('.')[0])
family_value = instances['instancetype'].apply(lambda x: x.split('.')[0])
instances.insert(family_column, 'instancetypefamily', family_value)
instance_units_column = instances.columns.get_loc('instances') + 2
units_value = instances['instancetype'].apply(
get_units) * instances['instances']
instances.insert(instance_units_column, 'instance_units', units_value)
reserved_units_column = instances.columns.get_loc('reserved') + 2
units_value = instances['instancetype'].apply(
get_units) * instances['reserved']
instances.insert(reserved_units_column, 'reserved_units', units_value)
# Add some additional data to ris
family_column = ris.columns.get_loc('instancetype') + 1
# meckstmd:07/29/2019 - Metal RIs are no different than regular RIs - they are a family with a normalization factor
# for example, i3.metal is equivalent to i3.16xlarge. See https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/apply_ri.html
#family_value = ris['instancetype'].apply(lambda x: x if x.endswith('.metal') else x.split('.')[0])
family_value = ris['instancetype'].apply(lambda x: x.split('.')[0])
ris.insert(family_column, 'instancetypefamily', family_value)
units_column = ris.columns.get_loc('quantity') + 1
units_value = ris['instancetype'].apply(get_units) * ris['quantity']
ris.insert(units_column, 'units', units_value)
# Create aggregates for faster processing
az_instance_groups = instances.groupby(
['availabilityzone', 'instancetype', 'tenancy', 'operatingsystem'])
az_account_instance_groups = instances.groupby(az_instance_groups.keys +
['usageaccountid'])
region_instance_groups = instances.groupby(
['region', 'instancetypefamily', 'tenancy', 'operatingsystem'])
region_account_instance_groups = instances.groupby(
region_instance_groups.keys + ['usageaccountid'])
ri_groups = ris.groupby(region_instance_groups.keys + ['scope'])
# Reference Lookup
all_sizes = instances['instancetype'].apply(
lambda x: x.split('.')[1]).unique()
reference_sizes = {}
for family in ris['instancetypefamily'].unique():
for size in all_sizes:
if "{}.{}".format(family, size) in pricing['us-east-1']:
reference_sizes[family] = size
break
# Reports
unused_az_ris = pd.DataFrame(
columns=az_instance_groups.keys +
['min_unused_qty', 'avg_unused_qty', 'max_unused_qty'])
ri_hourly_usage_report = pd.DataFrame(
columns=region_instance_groups.keys + ['hourofweek'] + [
'total_ri_units', 'total_instance_units', 'floating_ri_units',
'floating_instance_units', 'unused_ri_units', 'coverage_chance'
])
ri_purchases = pd.DataFrame(columns=[
'Account ID', 'Scope', 'Region', 'External AZ', 'Offering Class',
'Quantity', 'Term', 'Instance Type', 'Product Description', 'RI Type',
'Tenancy', 'accountid', 'family', 'units', 'ri upfront cost',
'ri total cost', 'ri savings', 'ondemand value', 'algorithm'
])
# NOTE: For usage values, AZ usage is booked by quantity (instances), Region usage is booked by units.
# Iterate by Union of (Region Instance Groups and RI Groups)
for group in sorted(
list(
set(region_instance_groups.groups.keys())
| set(ris.groupby(region_instance_groups.keys).groups.keys()))
):
# for group in [('ap-northeast-1', 'c4', 'Shared', 'Linux')]:
region, family, tenancy, operatingsystem = group
LOGGER.debug("Evaluting {:>14}:{:3} ({}, {})".format(
region, family, tenancy, operatingsystem))
# Account for In-Account AZ RI usage
# In-Account RI usage only needs to be counted against Regional Usage for accuracy
try:
az_ris = ri_groups.get_group(group + tuple(['Availability Zone']))
except KeyError:
az_ris = pd.DataFrame(columns=ris.columns)
az_account_hour_ri_usage = pd.DataFrame(
columns=az_account_instance_groups.keys +
['hourofweek', 'instances'])
region_account_hour_ri_usage = pd.DataFrame(
columns=region_account_instance_groups.keys +
['hourofweek', 'instance_units'])
for index, az_ri in az_ris.iterrows():
LOGGER.debug("Evaluating In-Account AZ RI: {}:{} {} x{}".format(
az_ri['accountid'], az_ri['availabilityzone'],
az_ri['instancetype'], az_ri['quantity']))
try:
group_key = (az_ri['availabilityzone'], az_ri['instancetype'],
tenancy, operatingsystem, az_ri['accountid'])
az_account_instance_group = az_account_instance_groups.get_group(
group_key)
except KeyError:
continue
# Straight to hourofweek average, since there should not be multiple rows per usagestartdate
in_account_usage = az_account_instance_group.groupby(
['hourofweek'])['instances'].mean()
# Account for already assigned usage from previously evaluated AZ RIs
in_account_assigned = az_account_hour_ri_usage[
(az_account_hour_ri_usage['availabilityzone'] ==
az_ri['availabilityzone'])
& (az_account_hour_ri_usage['instancetype'] ==
az_ri['instancetype']) &
(az_account_hour_ri_usage['usageaccountid'] ==
az_ri['accountid'])].groupby('hourofweek')['instances'].sum()
if len(in_account_assigned) > 0:
in_account_usage -= in_account_assigned
in_account_used = np.minimum(in_account_usage, az_ri['quantity'])
# Build assignment usage rows
usage_keys = pd.DataFrame([group_key],
columns=az_account_instance_groups.keys)
usage_data = pd.DataFrame({
'key': 1,
'hourofweek': in_account_used.index,
'instances': in_account_used.values
})
usage = usage_keys.assign(key=1).merge(usage_data,
on='key').drop('key', 1)
LOGGER.debug("In-Account Assigned AZ Usage:\n" + str(usage.head()))
az_account_hour_ri_usage = az_account_hour_ri_usage.append(
usage, ignore_index=True)
# Build regional usage rows
usage_keys = pd.DataFrame(
[group + tuple([az_ri['accountid']])],
columns=region_account_instance_groups.keys)
usage_data = pd.DataFrame({
'key':
1,
'hourofweek':
in_account_used.index,
'instance_units':
in_account_used.values * get_units(az_ri['instancetype'])
})
usage = usage_keys.assign(key=1).merge(usage_data,
on='key').drop('key', 1)
LOGGER.debug("In-Account Regional Assigned AZ Usage:\n" +
str(usage.head()))
region_account_hour_ri_usage = region_account_hour_ri_usage.append(
usage, ignore_index=True)
# Account for Cross-Account AZ RI Usage
# To simplify analysis, treat in-account and cross-account identically since we only report unused AZ RIs
az_ris = az_ris.groupby(['availabilityzone', 'instancetype'])
# for index, az_ri in az_ris.iterrows():
for az_group in az_ris.groups.keys():
availabilityzone, instancetype = az_group
quantity = az_ris.get_group(az_group)['quantity'].sum()
LOGGER.debug("Evaluating Cross-Account AZ RI: {} {} x{}".format(
availabilityzone, instancetype, quantity))
try:
group_key = (availabilityzone, instancetype, tenancy,
operatingsystem)
az_instance_group = az_instance_groups.get_group(group_key)
except KeyError:
continue
# Aggregate by hour before hourofweek
total_usage = az_instance_group.groupby(['usagestartdate', 'hourofweek'])['instances'].sum(). \
groupby(['hourofweek']).mean()
# No pre-assigned usage since individual RI subscriptions are getting bundled
total_used = np.minimum(total_usage, quantity)
# Add to regional usage for purchase recommendations
usage_keys = pd.DataFrame(
[group + tuple(['000000000000'])],
columns=region_account_instance_groups.keys)
usage_data = pd.DataFrame({
'key':
1,
'hourofweek':
total_used.index,
'instance_units':
total_used.values * get_units(az_ri['instancetype'])
})
usage = usage_keys.assign(key=1).merge(usage_data,
on='key').drop('key', 1)
LOGGER.debug("Cross-Account Regional Assigned AZ Usage:\n" +
str(usage.head()))
region_account_hour_ri_usage = region_account_hour_ri_usage.append(
usage, ignore_index=True)
unused = quantity - total_used
if unused.max() > 0:
unused_az_ri_row = {
'availabilityzone': availabilityzone,
'instancetype': instancetype,
'tenancy': tenancy,
'operatingsystem': operatingsystem,
'min_unused_qty': unused.min(),
'avg_unused_qty': unused.mean(),
'max_unused_qty': unused.max(),
}
unused_az_ris = unused_az_ris.append(unused_az_ri_row,
ignore_index=True)
LOGGER.debug("Unused AZ RIs:\n" + str(unused_az_ri_row))
# Account for In-Account Region RI Usage
# In-Account Region RI usage only needed to calculate RI Float
try:
region_ris = ri_groups.get_group(group + tuple(['Region']))
except KeyError:
region_ris = pd.DataFrame(columns=ris.columns)
region_hour_ri_usage = pd.DataFrame(
columns=region_instance_groups.keys + ['hourofweek', 'units'])
account_region_ris = region_ris.groupby(['accountid'])
for accountid in account_region_ris.groups.keys():
ri_units = account_region_ris.get_group(accountid)['units'].sum()
LOGGER.debug(
"Evaluating In-Account Region RI: {}:{} {} x{}".format(
accountid, region, family, ri_units))
try:
group_key = (region, family, tenancy, operatingsystem,
accountid)
region_account_instance_group = region_account_instance_groups.get_group(
group_key)
except KeyError:
continue
# Aggregate by hour before hourofweek
in_account_usage = region_account_instance_group.groupby(['usagestartdate', 'hourofweek']) \
['instance_units'].sum().groupby(['hourofweek']).mean()
# Account for already assigned usage from AZ RIs
in_account_assigned = region_account_hour_ri_usage[(
region_account_hour_ri_usage['usageaccountid'] == accountid
)].groupby('hourofweek')['instance_units'].sum()
if len(in_account_assigned) > 0:
in_account_usage -= in_account_assigned
in_account_used = np.minimum(in_account_usage, ri_units)
# Fix partial indexes
in_account_used = in_account_used.reindex(range(168),
copy=False,
fill_value=0.0)
# Build usage rows
usage_keys = pd.DataFrame([group],
columns=region_instance_groups.keys)
usage_data = pd.DataFrame({
'key': 1,
'hourofweek': in_account_used.index,
'units': in_account_used.values
})
usage = usage_keys.assign(key=1).merge(usage_data,
on='key').drop('key', 1)
LOGGER.debug("In-Account Assigned Region Usage:\n" +
str(usage.head()))
region_hour_ri_usage = region_hour_ri_usage.append(
usage, ignore_index=True)
try:
region_instance_group = region_instance_groups.get_group(group)
except:
# This is a bit heavy, but it shouldn't be called frequently.
# Create a new DataFrame that has the right structure
region_instance_group = region_instance_groups.get_group(
list(region_instance_groups.groups.keys())[0])
region_instance_group = region_instance_group.assign(instances=0)
region_instance_group = region_instance_group.assign(
instance_units=0)
region_instance_group = region_instance_group.assign(reserved=0)
# Account for Cross-Account Region RI Usage
if len(region_ris) == 0:
ri_units = 0
else:
ri_units = region_ris['units'].sum()
LOGGER.debug(
"Evaluating Cross-Account Region RI: {} {} x{}".format(
region, family, ri_units))
# Aggregate by hour before hourofweek
total_usage = region_instance_group.groupby(['usagestartdate', 'hourofweek']) \
['instance_units'].sum().groupby(['hourofweek']).mean()
# In-Account usage to calculate float
in_account_usage = region_hour_ri_usage.groupby(['hourofweek'
])['units'].sum()
if len(in_account_usage) == 0:
in_account_usage = pd.Series(0, index=total_usage.index)
# Floating RIs
floating_ri_units = ri_units - in_account_usage
# Instances eligible for float
floating_instance_units = total_usage - in_account_usage
# Unused RIs
unused_ri_units = np.maximum(ri_units - total_usage, 0)
# % Change a new instance will be covered
coverage_chance = np.minimum(
floating_ri_units / floating_instance_units * 100, 100)
# Build report rows
usage_keys = pd.DataFrame([group],
columns=region_instance_groups.keys)
usage_data = pd.DataFrame({
'key':
1,
'hourofweek':
total_usage.index,
'total_ri_units':
ri_units,
'total_instance_units':
total_usage.values,
'floating_ri_units':
floating_ri_units.values,
'floating_instance_units':
floating_instance_units.values,
'unused_ri_units':
unused_ri_units.values,
'coverage_chance':
coverage_chance.values,
})
usage = usage_keys.assign(key=1).merge(usage_data,
on='key').drop('key', 1)
LOGGER.debug("Cross-Account Region Usage Report:\n" +
str(usage.head()))
ri_hourly_usage_report = ri_hourly_usage_report.append(
usage, ignore_index=True)
# RI Utilization Evaluation complete. Evaluate Purchase recommendations
if region_instance_group['instance_units'].sum() > 0:
# Calculate usage slope to determine purchase aggressiveness
# First filter data to reduce noise.
region_hourly_usage = region_instance_group.groupby(
['usagestartdate', 'hourofweek'])['instance_units'].sum()
threshold = int(
utils.get_config_value(config, 'RI_PURCHASES',
'FILTER_THRESHOLD', 3))
signal = region_hourly_usage.values.copy()
delta = np.abs(signal - np.mean(signal))
median_delta = np.median(delta)
if median_delta > 0:
mask = (delta / float(median_delta)) > threshold
signal[mask] = np.median(signal)
# Least squares fit
ts = region_hourly_usage.reset_index()['usagestartdate'].apply(
lambda x: x.timestamp())
A = np.vstack([ts, np.ones(len(ts))]).T
y = signal
m, c = np.linalg.lstsq(A, y, rcond=None)[0]
slope = m * 86400
# Determine RI Algorithm
algorithm = 'DEFAULT'
aggressive_threshold = utils.get_config_value(
config, 'RI_PURCHASES', 'AGGRESSIVE_THRESHOLD', 'NONE')
try:
if slope >= float(aggressive_threshold):
algorithm = 'AGGRESSIVE'
except ValueError:
pass
conservative_threshold = utils.get_config_value(
config, 'RI_PURCHASES', 'CONSERVATIVE_THRESHOLD', 'NONE')
try:
if slope <= float(conservative_threshold):
algorithm = 'CONSERVATIVE'
except ValueError:
pass
# Subtract AZ RI Usage from instances since we for the most part can completely ignore them.
az_assigned = region_account_hour_ri_usage.groupby(
'hourofweek')['instance_units'].sum()
if len(az_assigned) > 0:
region_hourly_usage -= az_assigned
# Determine our purchase size for this family
types = [
key for key in pricing[region].keys()
if key.startswith(family + '.')
]
type_units = {key: get_units(key) for key in types}
desired_size = utils.get_config_value(config, 'RI_PURCHASES',
'RI_SIZE', 'largest')
if desired_size == 'largest':
purchase_size, purchase_size_units = max(
type_units.items(), key=operator.itemgetter(1))
elif desired_size == 'smallest':
purchase_size, purchase_size_units = min(
type_units.items(), key=operator.itemgetter(1))
else:
desired_size_units = get_units(desired_size)
filtered_units = {
k: v
for k, v in type_units.items() if v <= desired_size_units
}
if len(filtered_units) > 0:
purchase_size, purchase_size_units = max(
filtered_units.items(), key=operator.itemgetter(1))
else:
purchase_size, purchase_size_units = min(
type_units.items(), key=operator.itemgetter(1))
# Get RI Details
term = utils.get_config_value(config, 'RI_PURCHASES', 'RI_TERM')
term_h = int(term) * 730
term_y = '3yr' if term == 36 else '1yr'
if region not in pricing or purchase_size not in pricing[region] or \
tenancy not in pricing[region][purchase_size] or \
operatingsystem not in pricing[region][purchase_size][tenancy]:
LOGGER.error('Missing RI Pricing data for {}:{}:{}:{}'.format(
region, purchase_size, tenancy, operatingsystem))
continue
for offering in ['standard', 'convertible']:
# Get RI Pricing Data
offering2 = 'classic' if offering == 'standard' else 'convertible'
rates = pricing[region][purchase_size][tenancy][
operatingsystem]
od_rate = rates['onDemandRate']
ri_rate = None
option = utils.get_config_value(config, 'RI_PURCHASES',
'RI_OPTION')
for o in (option, 'No Upfront', 'Partial Upfront',
'All Upfront'):
ri_key = '{}-{}-{}'.format(term_y, offering, o)
if ri_key in rates['reserved']:
ri_rate = rates['reserved'][ri_key]
option = o
break
if ri_rate is None:
LOGGER.error('Missing RI Pricing data(2) for {}:{}'.format(
region, purchase_size))
continue
for slush in [False, True]:
utilization_key = "{}_{}_{}UTIL_TARGET".format(
offering.upper(), algorithm, 'SLUSH_' if slush else '')
target_utilization = utils.get_config_value(
config, 'RI_PURCHASES', utilization_key, 'NONE')
if target_utilization == 'BREAK_EVEN':
target_utilization = (ri_rate['upfront'] +
ri_rate['hourly'] * term_h) / (
od_rate * term_h) * 100
# RI Total Cost / OnDemand Cost = Break Even Utilization
LOGGER.debug(
"Purchase: {:>14}:{:3} {:11} {:5}: slope={} algo={} target={}"
.format(region, family, offering,
'slush' if slush else 'acct', slope, algorithm,
target_utilization))
if target_utilization == 'NONE':
continue
# Subtract existing RIs from usage to determine demand
demand_hourly_usage = region_hourly_usage - ri_units
# Edge case fix here...
# If usage only exists for a part of the timerange, it's percentile will be incorrect
# unless we fill it with zeros.
demand_hourly_usage = demand_hourly_usage.reset_index(
level=1, drop=True).reindex(timerange, fill_value=0.0)
# Subtract previously recommended RIs from usage
prior_ri_units = ri_purchases[
(ri_purchases['Region'] == region)
& (ri_purchases['family'] == family) &
(ri_purchases['Product Description']
== operatingsystem) &
(ri_purchases['Tenancy'] == tenancy)]['units'].sum()
demand_hourly_usage -= prior_ri_units
# Evalute Demand
demand_units = sorted(demand_hourly_usage.values)[int(
len(demand_hourly_usage.values) *
(100 - target_utilization) / 100)]
demand_units -= demand_units % purchase_size_units
if demand_units < purchase_size_units:
LOGGER.debug(
"Purchase: {:>14}:{:3} : No additional RIs required"
.format(region, family))
else:
# Recommend purchases in accounts with the most uncovered demand
# Calculate per-account demand (single number at percentile)
if slush:
account_demand = pd.DataFrame(
{
'accountid':
utils.get_config_value(
config, 'RI_PURCHASES',
'SLUSH_ACCOUNT'),
'units':
demand_units,
},
index=[0])
else:
# Edge case fix here...
# If an account only has usage for a part of the window, it's percentile will be incorrect
# unless we fill the timerange with zeros.
idx = pd.merge(
pd.DataFrame({
'key':
1,
'usageaccountid':
region_instance_group['usageaccountid'].
unique()
}),
pd.DataFrame({
'key': 1,
'usagestartdate': timerange
}),
on='key')[['usageaccountid', 'usagestartdate']]
account_demand = region_instance_group.groupby(['usageaccountid', 'usagestartdate']) \
['instance_units'].sum().reindex(idx, fill_value=0.0).groupby('usageaccountid'). \
agg(lambda x: np.percentile(x, q=100 - target_utilization))
# subtract in-account RIs
account_ris = region_ris.groupby(
['accountid'])['units'].sum()
account_demand = account_demand.subtract(
account_ris, fill_value=0)
account_demand = pd.DataFrame({
'accountid':
account_demand.index,
'units':
account_demand.values
})
# Noramlize to purchase units
account_demand['units'] -= account_demand[
'units'] % purchase_size_units
# Filter for positive demand
account_demand = account_demand[
account_demand['units'] > 0]
# subtract from bottom to allow equal float opportunity
while account_demand['units'].sum(
) > demand_units + len(
account_demand) * purchase_size_units:
excess_qty_per_account = int(
(account_demand['units'].sum() -
demand_units) / purchase_size_units /
len(account_demand))
account_demand[
'units'] -= excess_qty_per_account * purchase_size_units
account_demand = account_demand[
account_demand['units'] > 0]
# Consistently distribute stragglers
if account_demand['units'].sum() > demand_units:
excess_qty = int(
(account_demand['units'].sum() -
demand_units) / purchase_size_units)
sorted_accounts = account_demand.sort_values(
['units', 'accountid'])
delta = pd.Series(
[purchase_size_units] * excess_qty + [0] *
(len(account_demand) - excess_qty),
index=sorted_accounts.index)
account_demand['units'] -= delta
account_demand = account_demand[
account_demand['units'] > 0]
# Build report rows
quantity = (account_demand['units'] /
purchase_size_units).astype(int)
purchases = pd.DataFrame({
'Account ID':
account_demand['accountid'].apply(
lambda x: '="{0:012}"'.format(x)),
'Scope':
'Region',
'Region':
region,
'External AZ':
'',
'Offering Class':
offering2,
'Quantity':
quantity,
'Term':
term,
'Instance Type':
purchase_size,
'Product Description':
operatingsystem,
'RI Type':
option,
'Tenancy':
tenancy,
'accountid':
account_demand['accountid'].apply(
lambda x: '{0:012}'.format(x)),
'family':
family,
'units':
account_demand['units'].astype(int),
'ri upfront cost':
quantity * ri_rate['upfront'],
'ri total cost':
quantity *
(ri_rate['upfront'] + ri_rate['hourly'] * term_h),
'ri savings':
quantity *
((od_rate - ri_rate['hourly']) * term_h -
ri_rate['upfront']),
'ondemand value':
quantity * od_rate * term_h,
'algorithm':
algorithm
})
LOGGER.debug("Purchases:\n" + str(purchases.head()))
ri_purchases = ri_purchases.append(purchases,
ignore_index=True)
# Assign to top until filly assigned
LOGGER.debug(
"Purchase: {:>14}:{:3} : type={} demand={}, recommend={} in {} accounts"
.format(region, family, purchase_size,
demand_units,
account_demand['units'].sum(),
len(account_demand)))
# GroupBy to assign appropriate index columns
unused_az_ris = unused_az_ris.groupby(az_instance_groups.keys).sum()
ri_hourly_usage_report = ri_hourly_usage_report.groupby(
region_instance_groups.keys + ['hourofweek']).sum()
instances = instances.drop('hourofweek', 1)
# https://github.com/yahoo/ariel/issues/8: this is necessary if the accounts have not purchased any RIs
if (len(ri_hourly_usage_report) == 0):
ri_hourly_usage_report = pd.DataFrame(columns=[
'region', 'instancetypefamily', 'tenancy', 'operatingsystem',
'hourofweek', 'total_ri_units', 'total_instance_units',
'floating_ri_units', 'floating_instance_units', 'unused_ri_units',
'coverage_chance'
])
ri_usage_report = pd.DataFrame(columns=[
'region', 'instancetypefamily', 'tenancy', 'operatingsystem',
'total_ri_units', 'total_instance_units', 'floating_ri_units',
'floating_instance_units', 'unused_ri_units', 'coverage_chance',
'xl_effective_rate', 'monthly_ri_cost', 'monthly_od_cost',
'monthly_ri_savings'
])
else:
# Build RI Usage report with Actual cost benefit
ri_usage_report = ri_hourly_usage_report.groupby(
region_instance_groups.keys).mean()
ri_cost = ris.groupby(
region_instance_groups.keys)['amortizedupfrontprice'].sum(
) + ris.groupby(
region_instance_groups.keys)['amortizedrecurringfee'].sum()
od_cost = ri_usage_report.apply(lambda x: 720 * pricing[x.name[0]][
"{}.{}".format(x.name[1], reference_sizes[x.name[1]])][
x.name[2]][x.name[3]]['onDemandRate'] * min(
x['total_ri_units'], x['total_instance_units']
) / get_units(x.name[1] + '.' + reference_sizes[x.name[1]]),
axis=1)
xl_effective_rate = (
(od_cost - ri_cost) *
(100 - ri_usage_report['coverage_chance']) / 100 +
ri_cost) / 720 / ri_usage_report['total_ri_units'] * 8
ri_usage_report.insert(len(ri_usage_report.columns),
'xl_effective_rate', xl_effective_rate)
ri_usage_report.insert(len(ri_usage_report.columns), 'monthly_ri_cost',
ri_cost)
ri_usage_report.insert(len(ri_usage_report.columns), 'monthly_od_cost',
od_cost)
ri_usage_report.insert(len(ri_usage_report.columns),
'monthly_ri_savings', od_cost - ri_cost)
# Apply some column formats
ri_hourly_usage_report['total_ri_units'] = ri_hourly_usage_report[
'total_ri_units'].map('{:.0f}'.format)
ri_hourly_usage_report[
'total_instance_units'] = ri_hourly_usage_report[
'total_instance_units'].map('{:.0f}'.format)
ri_hourly_usage_report['floating_ri_units'] = ri_hourly_usage_report[
'floating_ri_units'].map('{:.0f}'.format)
ri_hourly_usage_report[
'floating_instance_units'] = ri_hourly_usage_report[
'floating_instance_units'].map('{:.0f}'.format)
ri_hourly_usage_report['unused_ri_units'] = ri_hourly_usage_report[
'unused_ri_units'].map('{:.0f}'.format)
ri_hourly_usage_report['coverage_chance'] = ri_hourly_usage_report[
'coverage_chance'].map('{:.2f}'.format)
ri_usage_report['total_ri_units'] = ri_usage_report[
'total_ri_units'].map('{:.0f}'.format)
ri_usage_report['total_instance_units'] = ri_usage_report[
'total_instance_units'].map('{:.0f}'.format)
ri_usage_report['floating_ri_units'] = ri_usage_report[
'floating_ri_units'].map('{:.0f}'.format)
ri_usage_report['floating_instance_units'] = ri_usage_report[
'floating_instance_units'].map('{:.0f}'.format)
ri_usage_report['unused_ri_units'] = ri_usage_report[
'unused_ri_units'].map('{:.0f}'.format)
ri_usage_report['coverage_chance'] = ri_usage_report[
'coverage_chance'].map('{:.2f}'.format)
ri_usage_report['xl_effective_rate'] = ri_usage_report[
'xl_effective_rate'].map('${:,.4f}'.format)
ri_usage_report['monthly_ri_cost'] = ri_usage_report[
'monthly_ri_cost'].map('${:,.2f}'.format)
ri_usage_report['monthly_od_cost'] = ri_usage_report[
'monthly_od_cost'].map('${:,.2f}'.format)
ri_usage_report['monthly_ri_savings'] = ri_usage_report[
'monthly_ri_savings'].map('${:,.2f}'.format)
ri_purchases['ri upfront cost'] = ri_purchases['ri upfront cost'].map(
'${:,.2f}'.format)
ri_purchases['ri total cost'] = ri_purchases['ri total cost'].map(
'${:,.2f}'.format)
ri_purchases['ri savings'] = ri_purchases['ri savings'].map(
'${:,.2f}'.format)
ri_purchases['ondemand value'] = ri_purchases['ondemand value'].map(
'${:,.2f}'.format)
reports = {
"ACCOUNT_INSTANCE_SUMMARY": instances,
"RI_PURCHASES": ri_purchases,
"RI_USAGE": ri_usage_report,
"RI_HOURLY_USAGE": ri_hourly_usage_report,
"UNUSED_AZ_RIS": unused_az_ris,
}
return reports
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-unlimited-summary.csv'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
mtime = 0
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
account, role = utils.get_master(config)
region = utils.get_config_value(
config, 'ATHENA', 'AWS_REGION',
utils.get_config_value(config, 'DEFAULTS', 'AWS_REGION',
os.environ.get('AWS_DEFAULT_REGION')))
database = utils.get_config_value(config, 'ATHENA', 'CUR_DATABASE')
staging = utils.get_config_value(
config, 'ATHENA', 'STAGING',
's3://aws-athena-query-results-{0}-{1}/ariel-cur-output/'.format(
account, region))
days = utils.get_config_value(config, 'ATHENA', 'DAYS', 28)
offset = utils.get_config_value(config, 'ATHENA', 'OFFSET', 1)
session = boto3.Session()
proto, empty, staging_bucket, staging_prefix = staging.split('/', 3)
# Assume role if needed
if role is not None:
session = utils.assume_role(session, role)
# Connect to Athena
athena = session.client('athena', region_name=region)
# Validate database is usable
status_id = utils.execute_athena_query(
athena, staging,
'SELECT status FROM ' + database + '.cost_and_usage_data_status')
# Row 0 is header
status = athena.get_query_results(
QueryExecutionId=status_id
)['ResultSet']['Rows'][1]['Data'][0]['VarCharValue']
if status != 'READY':
raise Exception('Athena database not in READY status')
# Identify start to end range query
today = datetime.datetime.combine(datetime.datetime.today(),
datetime.time.min)
endtime = today - datetime.timedelta(days=offset)
starttime = endtime - datetime.timedelta(days=days)
# Download Instance and RI usage
query = ' '.join((
"" + "SELECT line_item_usage_account_id AS accountid ,"
" product_region AS region, "
" lower(product_instance) AS instancetypefamily, "
" sum(line_item_usage_amount) AS unlimitedusageamount, "
" sum(line_item_unblended_cost) AS unlimitedusagecost " +
" FROM " + database + ".cur " +
" WHERE line_item_usage_type like '%CPUCredits:%' " +
" AND line_item_usage_start_date >= cast('{}' as timestamp) ".
format(starttime.isoformat(' ')) +
" AND line_item_usage_start_date < cast('{}' as timestamp) ".
format(endtime.isoformat(' ')) +
" GROUP BY line_item_usage_account_id, product_region, lower(product_instance) "
+
" ORDER BY line_item_usage_account_id, product_region, lower(product_instance) "
).split())
query_id = utils.execute_athena_query(athena, staging, query)
session.client('s3').download_file(
staging_bucket, '{0}{1}.csv'.format(staging_prefix, query_id),
cache_file)
result = pd.read_csv(cache_file)
if len(result) == 0:
result = pd.DataFrame(columns=[
'accountid', 'region', 'instancetypefamily',
'unlimitedusageamount', 'unlimitedusagecost'
])
result['accountid'] = result['accountid'].map('{:012}'.format)
result['unlimitedusageamount'] = result['unlimitedusageamount'].map(
'{:.2f}'.format)
result['unlimitedusagecost'] = result['unlimitedusagecost'].map(
'${:,.2f}'.format)
LOGGER.info("Loaded {} unlimited rows".format(len(result)))
return result
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-reserved-instances.csv'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
pass
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
ris = pd.read_csv(cache_file)
else:
account, role = utils.get_master(config)
region = utils.get_config_value(
config, 'RESERVED_INSTANCES', 'AWS_REGION',
utils.get_config_value(config, 'DEFAULTS', 'AWS_REGION',
os.environ.get('AWS_DEFAULT_REGION')))
# start date cannot be after 2 days ago for GetReservationUtilization
monthend = datetime.date.today()
monthstart = (datetime.date.today() - datetime.timedelta(days=31))
ris = []
if role != '':
session = utils.assume_role(boto3.Session(), role)
ce = session.client('ce', region_name=region)
rsp = ce.get_reservation_utilization(TimePeriod={
"Start": str(monthstart),
"End": str(monthend)
},
GroupBy=[{
"Type":
"DIMENSION",
"Key":
"SUBSCRIPTION_ID"
}])
while True:
groups = rsp['UtilizationsByTime'][0]['Groups']
for row in groups:
# Make sure to only capture active RIs
endDate = datetime.datetime.strptime(
row['Attributes']['endDateTime'],
"%Y-%m-%dT%H:%M:%S.000Z")
if endDate.date() > datetime.date.today():
operatingSystem = 'Linux' if row['Attributes'][
'platform'] == 'Linux/UNIX' else row['Attributes'][
'platform'] # for CUR compatibility
ri = {
'accountid':
int(row['Attributes']['accountId']),
'accountname':
row['Attributes']['accountName'],
'reservationid':
row['Attributes']['leaseId'],
'subscriptionid':
row['Attributes']['subscriptionId'],
'startdate':
row['Attributes']['startDateTime'],
'enddate':
row['Attributes']['endDateTime'],
'state':
row['Attributes']['subscriptionStatus'],
'quantity':
int(row['Attributes']['numberOfInstances']),
'availabilityzone':
row['Attributes']['availabilityZone'],
'region':
row['Attributes']['region'],
'instancetype':
row['Attributes']['instanceType'],
'paymentoption':
row['Attributes']['subscriptionType'],
'tenancy':
row['Attributes']['tenancy'],
'operatingsystem':
operatingSystem,
'amortizedhours':
int(row['Utilization']['PurchasedHours']),
'amortizedupfrontprice':
float(row['Utilization']['AmortizedUpfrontFee']),
'amortizedrecurringfee':
float(row['Utilization']['AmortizedRecurringFee']),
'offeringclass':
row['Attributes']['offeringType'],
'scope':
row['Attributes']['scope'],
}
ris.append(ri)
if 'NextToken' in rsp:
rsp = ce.get_reservation_utilization(
NextToken=rsp['NextToken'])
continue
break
ris = pd.DataFrame.from_records(ris)
ris.to_csv(cache_file, index=False)
LOGGER.info("Loaded {} reserved instances".format(len(ris)))
return ris
def load(config, locations=LOCATIONS):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-ec2-pricing.json'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
pass
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
pricing_url = utils.get_config_value(config, 'PRICING', 'URL', 'https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/current/index.csv')
csv_reader = csv.reader(codecs.iterdecode(urlopen(pricing_url), 'utf-8'), utils.CsvDialect())
# Find header row
header_map = None
while header_map is None:
header = next(csv_reader)
if header[0] == "SKU":
header_map = utils.parse_header(header)
rowcount = 0
prices = {}
units = {}
for row in csv_reader:
if not check_row(header_map, row):
continue
sku = row[header_map['SKU']]
location = row[header_map['Location']]
instanceType = row[header_map['Instance Type']]
tenancy = row[header_map['Tenancy']]
operatingsystem = row[header_map['Operating System']]
# Resolve the AWS Region from location information.
if location not in locations:
locations[location] = utils.get_config_value(config, 'LOCATIONS', location, '')
if locations[location] == '':
LOGGER.info('Skipping unknown location: {}'.format(location))
if locations[location] == '':
continue
region = locations[location]
# Populate result set
if region not in prices:
prices[region] = {}
if instanceType not in prices[region]:
prices[region][instanceType] = {}
if tenancy not in prices[region][instanceType]:
prices[region][instanceType][tenancy] = {}
if operatingsystem not in prices[region][instanceType][tenancy]:
prices[region][instanceType][tenancy][operatingsystem] = {
"sku": sku,
"reserved": {}
}
units[instanceType] = float(row[header_map['Normalization Size Factor']])
price = prices[region][instanceType][tenancy][operatingsystem]
if price['sku'] != sku:
print('WARNING: Duplicate sku: {}:{} -> {} != {}'.format(region, instanceType, sku, price['sku']))
continue
# Add pricing data
if row[header_map['TermType']] == 'OnDemand':
if row[header_map['Unit']] in ('Hrs', 'Hours'):
price['onDemandRate'] = float(row[header_map['PricePerUnit']])
elif row[header_map['TermType']] == 'Reserved':
id = '{}-{}-{}'.format(row[header_map['LeaseContractLength']], row[header_map['OfferingClass']],
row[header_map['PurchaseOption']])
if id not in price['reserved']:
price['reserved'][id] = { 'upfront': 0.0, 'hourly': 0.0 }
if row[header_map['Unit']] in ('Hrs', 'Hours'):
price['reserved'][id]['hourly'] = float(row[header_map['PricePerUnit']])
elif row[header_map['Unit']] in ('Quantity'):
price['reserved'][id]['upfront'] = float(row[header_map['PricePerUnit']])
rowcount += 1
LOGGER.info("Loaded {} pricing rows".format(rowcount))
# Trim useless data
rowcount = 0
remove = []
for region in prices:
for instanceType in prices[region]:
for tenancy in prices[region][instanceType]:
for operatingsystem in prices[region][instanceType][tenancy]:
if 'onDemandRate' not in prices[region][instanceType][tenancy][operatingsystem] or \
len(prices[region][instanceType][tenancy][operatingsystem]['reserved']) == 0:
remove.append([region, instanceType, tenancy, operatingsystem])
else:
rowcount += 1
for keys in remove:
del prices[keys[0]][keys[1]][keys[2]][keys[3]]
prices['units'] = units
LOGGER.info("Loaded prices for {} instance types".format(rowcount))
with open(cache_file, 'w') as outfile:
json.dump(prices, outfile, indent=4)
return prices
with utils.get_read_handle(cache_file) as input:
prices = json.load(input)
return prices
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-locations.yaml'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
pass
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
account, role = utils.get_master(config)
region = utils.get_config_value(
config, 'ATHENA', 'AWS_REGION',
utils.get_config_value(config, 'DEFAULTS', 'AWS_REGION',
os.environ.get('AWS_DEFAULT_REGION')))
database = utils.get_config_value(config, 'ATHENA', 'CUR_DATABASE')
staging = utils.get_config_value(
config, 'ATHENA', 'STAGING',
's3://aws-athena-query-results-{0}-{1}/ariel-cur-output/'.format(
account, region))
days = utils.get_config_value(config, 'ATHENA', 'DAYS', 28)
offset = utils.get_config_value(config, 'ATHENA', 'OFFSET', 1)
session = boto3.Session()
proto, empty, staging_bucket, staging_prefix = staging.split('/', 3)
# Assume role if needed
if role is not None:
session = utils.assume_role(session, role)
# Connect to Athena
athena = session.client('athena', region_name=region)
# Validate database is usable
status_id = utils.execute_athena_query(
athena, staging,
'SELECT status FROM ' + database + '.cost_and_usage_data_status')
# Row 0 is header
status = athena.get_query_results(
QueryExecutionId=status_id
)['ResultSet']['Rows'][1]['Data'][0]['VarCharValue']
if status != 'READY':
raise Exception('Athena database not in READY status')
# Identify start to end range query
today = datetime.datetime.combine(datetime.datetime.today(),
datetime.time.min)
endtime = today - datetime.timedelta(days=offset)
starttime = endtime - datetime.timedelta(days=days)
# Retrieve location to region mapping for use with ec2 pricing data
query = ' '.join(
("" + "SELECT DISTINCT product_location, product_region " +
" FROM " + database + ".cur " +
" WHERE line_item_usage_start_date >= cast('{}' as timestamp) ".
format(starttime.isoformat(' ')) +
" AND line_item_usage_start_date < cast('{}' as timestamp) ".
format(endtime.isoformat(' ')) +
" AND product_operation = 'RunInstances' ").split())
map_id = utils.execute_athena_query(athena, staging, query)
map_result = athena.get_query_results(
QueryExecutionId=map_id)['ResultSet']['Rows']
locations = {}
for i in range(1, len(map_result)):
row = map_result[i]['Data']
location = row[0]['VarCharValue']
region = row[1]['VarCharValue']
locations[location] = region
with open(cache_file, 'w') as outfile:
yaml.dump(locations, outfile, default_flow_style=False)
return locations
with utils.get_read_handle(cache_file) as input:
locations = yaml.load(input, Loader=yaml.FullLoader)
return locations
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-account-instance-summary.csv'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
mtime = 0
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
account, role = utils.get_master(config)
region = utils.get_config_value(
config, 'ATHENA', 'AWS_REGION',
utils.get_config_value(config, 'DEFAULTS', 'AWS_REGION',
os.environ.get('AWS_DEFAULT_REGION')))
database = utils.get_config_value(config, 'ATHENA', 'CUR_DATABASE')
table_name = utils.get_config_value(config, 'ATHENA', 'CUR_TABLE_NAME',
'cur')
staging = utils.get_config_value(
config, 'ATHENA', 'STAGING',
's3://aws-athena-query-results-{0}-{1}/ariel-cur-output/'.format(
account, region))
days = utils.get_config_value(config, 'ATHENA', 'DAYS', 28)
offset = utils.get_config_value(config, 'ATHENA', 'OFFSET', 1)
session = boto3.Session()
proto, empty, staging_bucket, staging_prefix = staging.split('/', 3)
# Assume role if needed
if role is not None:
session = utils.assume_role(session, role)
# Connect to Athena
athena = session.client('athena', region_name=region)
# Validate database is usable
status_id = utils.execute_athena_query(
athena, staging,
'SELECT status FROM ' + database + '.cost_and_usage_data_status')
# Row 0 is header
status = athena.get_query_results(
QueryExecutionId=status_id
)['ResultSet']['Rows'][1]['Data'][0]['VarCharValue']
if status != 'READY':
raise Exception('Athena database not in READY status')
# Identify start to end range query
today = datetime.datetime.combine(datetime.datetime.today(),
datetime.time.min)
endtime = today - datetime.timedelta(days=offset)
starttime = endtime - datetime.timedelta(days=days)
# Download Instance and RI usage
query = ' '.join((
"" + " WITH preprocess AS ( " +
" SELECT line_item_usage_start_date AS usagestartdate, " +
" line_item_usage_account_id AS usageaccountid, " +
" line_item_availability_zone AS availabilityzone, " +
" CASE WHEN line_item_usage_type LIKE '%:%' THEN SPLIT(line_item_usage_type, ':')[2] "
+
" WHEN line_item_line_item_description LIKE '%m1.small%' THEN 'm1.small' "
+
" WHEN line_item_line_item_description LIKE '%m1.medium%' THEN 'm1.medium' "
+
" WHEN line_item_line_item_description LIKE '%m1.large%' THEN 'm1.large' "
+
" WHEN line_item_line_item_description LIKE '%m1.xlarge%' THEN 'm1.xlarge' "
+ " ELSE 'm1.error' " +
" END AS instancetype, " +
" product_tenancy AS tenancy, " +
" product_operating_system AS operatingsystem, " +
" CAST(line_item_usage_amount AS double) as usageamount, "
+
" CASE WHEN line_item_line_item_type = 'DiscountedUsage' THEN CAST(line_item_usage_amount AS DOUBLE) ELSE 0 END as reservedamount "
+ " FROM " + database + "." + table_name +
" WHERE product_operation = 'RunInstances' " +
" AND line_item_availability_zone != '' " +
" AND line_item_availability_zone NOT LIKE '%-wlz-%' " # Filter out Wavelength Instances. They're not available for RIs.
+ " AND product_tenancy = 'Shared' " + " ) " +
"SELECT usagestartdate, usageaccountid, availabilityzone, instancetype, tenancy, operatingsystem, SUM(usageamount) as instances, SUM(reservedamount) as reserved "
+ " FROM preprocess " +
" WHERE usagestartdate >= cast('{}' as timestamp) ".format(
starttime.isoformat(' ')) +
" AND usagestartdate < cast('{}' as timestamp) ".format(
endtime.isoformat(' ')) +
" GROUP BY usagestartdate, usageaccountid, availabilityzone, instancetype, tenancy, operatingsystem "
+
" ORDER BY usagestartdate, usageaccountid, availabilityzone, instancetype, tenancy, operatingsystem "
).split())
query_id = utils.execute_athena_query(athena, staging, query)
session.client('s3').download_file(
staging_bucket, '{0}{1}.csv'.format(staging_prefix, query_id),
cache_file)
result = pd.read_csv(cache_file, parse_dates=['usagestartdate'])
LOGGER.info("Loaded {} instance summary rows".format(len(result)))
return result
def load(config):
# If local files exists and is less than a day old, just use it.
cache_file = '/tmp/cached-unlimited-summary.csv'
caching = utils.get_config_value(config, 'DEFAULTS', 'CACHING', False)
mtime = 0
if caching:
try:
mtime = os.stat(cache_file).st_mtime
except FileNotFoundError:
mtime = 0
if mtime > time.time() - 86400:
LOGGER.info("Using existing cache file: " + cache_file)
else:
account, role = utils.get_master(config)
region = utils.get_config_value(
config, 'ATHENA', 'AWS_REGION',
utils.get_config_value(config, 'DEFAULTS', 'AWS_REGION',
os.environ.get('AWS_DEFAULT_REGION')))
database = utils.get_config_value(config, 'ATHENA', 'CUR_DATABASE')
table_name = utils.get_config_value(config, 'ATHENA', 'CUR_TABLE_NAME',
'cur')
staging = utils.get_config_value(
config, 'ATHENA', 'STAGING',
's3://aws-athena-query-results-{0}-{1}/ariel-cur-output/'.format(
account, region))
days = utils.get_config_value(config, 'ATHENA', 'DAYS', 28)
offset = utils.get_config_value(config, 'ATHENA', 'OFFSET', 1)
session = boto3.Session()
proto, empty, staging_bucket, staging_prefix = staging.split('/', 3)
# Assume role if needed
if role is not None:
session = utils.assume_role(session, role)
# Connect to Athena
athena = session.client('athena', region_name=region)
# Validate database is usable
status_id = utils.execute_athena_query(
athena, staging,
'SELECT status FROM ' + database + '.cost_and_usage_data_status')
# Row 0 is header
status = athena.get_query_results(
QueryExecutionId=status_id
)['ResultSet']['Rows'][1]['Data'][0]['VarCharValue']
if status != 'READY':
raise Exception('Athena database not in READY status')
# Identify start to end range query
today = datetime.datetime.combine(datetime.datetime.today(),
datetime.time.min)
endtime = today - datetime.timedelta(days=offset)
starttime = endtime - datetime.timedelta(days=days)
# Download Instance and RI usage
# meckstmd: 07/30/2019 - Something must have changed with the way AWS is exposing CPU Credits.
# There is a line_item_line_item_type column of Tax for each account which has CPU Credits which
# does not have a product_region or product_instance. Because these fields are empty, Ariel
# fails when trying to insert this report data into the unlimited_usage DB table because it does
# not allow nulls. The line_item_line_item_type column of Usage in this report has the per-instance
# CPU credits for unlimited and does have product_region and product_instance. I am guessing the
# Tax one was just added to this report and that is what broke Ariel.
# See https://github.com/yahoo/ariel/issues/5
query = ' '.join((
"" + "SELECT line_item_usage_account_id AS accountid ,"
" product_region AS region, "
" lower(product_instance) AS instancetypefamily, "
" sum(line_item_usage_amount) AS unlimitedusageamount, "
" sum(line_item_unblended_cost) AS unlimitedusagecost " +
" FROM " + database + "." + table_name +
" WHERE line_item_usage_type like '%CPUCredits:%' " +
" AND line_item_usage_start_date >= cast('{}' as timestamp) ".
format(starttime.isoformat(' ')) +
" AND line_item_usage_start_date < cast('{}' as timestamp) ".
format(endtime.isoformat(' ')) +
" AND product_region <> '' AND product_instance <> ''" +
" GROUP BY line_item_usage_account_id, product_region, lower(product_instance) "
+
" ORDER BY line_item_usage_account_id, product_region, lower(product_instance) "
).split())
query_id = utils.execute_athena_query(athena, staging, query)
session.client('s3').download_file(
staging_bucket, '{0}{1}.csv'.format(staging_prefix, query_id),
cache_file)
result = pd.read_csv(cache_file)
if len(result) == 0:
result = pd.DataFrame(columns=[
'accountid', 'region', 'instancetypefamily',
'unlimitedusageamount', 'unlimitedusagecost'
])
result['accountid'] = result['accountid'].map('{:012}'.format)
result['unlimitedusageamount'] = result['unlimitedusageamount'].map(
'{:.2f}'.format)
result['unlimitedusagecost'] = result['unlimitedusagecost'].map(
'${:,.2f}'.format)
LOGGER.info("Loaded {} unlimited rows".format(len(result)))
return result