hawk-utils-scripts/monitoring/db/scripts/get_detailed_power.py
2024-02-12 17:37:40 +01:00

261 lines
7.2 KiB
Python
Executable file

#!/usr/bin/env python3
import argparse
import numpy as np
from collections import OrderedDict
import os.path
def parse_arguments(args):
parser = argparse.ArgumentParser(
description='Produce detailed power usage data for a list of jobids.')
parser.add_argument('-v', '--verbose', action='store_true',
help='Show database querries, etc.')
parser.add_argument('-t', '--interval', action='store', type=float, default=5.0,
help="Interval between power values in seconds")
parser.add_argument('--hawk-ai', action='store_true',
help="Job did run on Hawk-AI")
parser.add_argument('jobid', type=parse_jobid,
nargs='+',
help='Job ID such as "2260215" or "2260215.hawk-pbs5"')
return parser.parse_args(args)
def parse_jobid(s):
import re
hawkpbs = r'.hawk-pbs5'
jobid = re.sub(hawkpbs, '', s)
if not jobid.isdigit():
raise argparse.ArgumentTypeError(f'invalid job ID "{s}"')
return jobid
class Power:
def __init__(self, nodes):
self.nodes = nodes
self.epochs = OrderedDict()
self.first_ts = None
self.last_ts = None
@classmethod
def from_list(cls, data):
"""
Returns a Power instance from a list of tuples (timestamp, node, value).
Assumptions:
- data is sorted by timestamp ascending
- for each timestamp, there is the same set of nodes and in the same order
"""
idx_ts = 0; idx_node = 1; idx_value = 2
nodes = list(OrderedDict.fromkeys([line[idx_node] for line in data])) # preserves order of nodes
cls = Power(nodes)
values = {}
for l in data:
ts = l[idx_ts]
if ts not in values:
values[ts] = []
power = l[idx_value]
values[ts].append(power)
epochs = values.keys()
for epoch in epochs:
cls.insert_epoch(epoch, values[epoch])
# check implicit assumptions: 1) ts/epochs are sorted
e = list(epochs)
k = list(values.keys())
if not e == k:
print("Warning: Unexpected unsorted timestamps.")
# check implicit assumptions: 2) each line has #nodes values
nnodes = len(nodes)
for epoch in epochs:
actual = len(values[epoch])
if actual != nnodes:
print("Warning: Unexpected number of nodes ({actual}/{expected})".format(actual=actual, expected=nnodes))
break
return cls
@classmethod
def from_db(cls, db, jobid, interval, hawk_ai):
all_list = db.db_to_list(jobid, interval, hawk_ai)
if not all_list:
raise RuntimeError
power = cls.from_list(all_list)
return power
def to_file(self, jobid, header=""):
"""Dumps power data to file. Returns filename is succesfull and None if unsucessfull."""
fname = self.filename(jobid)
if os.path.exists(fname):
print("Error: cowardly refusing to overwrite file ", fname)
return None
try:
with open(fname, "w+") as f:
f.write(header + self.header())
f.write(self.body())
except IOError:
print("Error: could not write to file ", fname)
fname = None
return fname
def insert_epoch(self, ts, values):
self.epochs[ts] = values
if not self.first_ts:
self.first_ts = ts
self.last_ts = ts
def header(self):
hd = "# all timestamp have unit miliseconds since unix epoch\n"
hd += "# all power values have unit Watt\n"
hd += "timestamp,RESERVED,head_node_power,avg_node_power,median_node_power,min_node_power,max_node_power,std_dev_node_power"
# add node names here instead
hd += "," + ",".join(self.nodes)
hd += "\n"
return hd
def body(self):
_body = ""
for epoch in self.epochs.items():
_body += self.pretty_print(self.summarize_epoch(epoch))
return _body
def summarize_time(self, ts):
return ts, ""
@staticmethod
def summarize_values(val):
values = np.asarray(val)
head = values[0]
min, max = values.min(), values.max()
avg, stddev = values.mean(), values.std()
median = np.median(values)
return head, avg, median, min, max, stddev
def summarize_epoch(self, epoch):
ts, values = epoch
return self.summarize_time(ts) \
+ self.summarize_values(values) \
+ tuple(values)
@staticmethod
def pretty_print(args):
return ",".join(str(a) for a in args) + '\n'
def filename(self, jobid):
fname = "detailed_power_{jobid}.hawk-pbs5.{first}-{last}.csv".format(
jobid=jobid, first=self.first_ts, last=self.last_ts
)
return fname
class MonitoringDB:
QUERY_STRING_HAWK = """
-- For description of get_job_data(), see https://kb.hlrs.de/monitoring/index.php/TimescaleDB_-_Query_Guidelines#Function:_get_job_data_and_get_ai_job_data
select * from get_job_data(
'{jobid}.hawk-pbs5',
'cmc_power_racktraynodepoweravg', -- power data source
'{interval} seconds',
array['avg'], -- aggregation: average across samples in bucket
array['time','node'] -- sort by time first than node (ascending)
)
as t(time bigint, name varchar, avg double precision);
"""
QUERY_STRING_HAWK_AI = """
-- For description of get_ai_job_data(), see https://kb.hlrs.de/monitoring/index.php/TimescaleDB_-_Query_Guidelines#Function:_get_job_data_and_get_ai_job_data
select * from get_ai_job_data(
'{jobid}.hawk-pbs5',
'telegraf_ipmi_power_meter', -- power data source
'{interval} seconds',
array['avg'], -- aggregation: average across samples in bucket
array['time','node'] -- sort by time first than node (ascending)
)
as t(time bigint, name varchar, avg double precision);
"""
def __init__(self, verbose):
self.connection = self.init_db(verbose)
@staticmethod
def init_db(verbose):
import sqlalchemy as db
engine = db.create_engine('postgresql://hpc@hawk-monitor4:5432/coe_mon', echo=verbose)
connection = engine.connect()
return connection
def close_db(self):
return self.connection.close()
@classmethod
def build_query(cls, jobid, interval, hawk_ai):
import sqlalchemy as db
if hawk_ai:
query_string = cls.QUERY_STRING_HAWK_AI
else:
query_string = cls.QUERY_STRING_HAWK
return db.text(query_string.format(jobid=jobid, interval=interval))
def db_to_list(self, jobid, interval, hawk_ai):
query = self.build_query(jobid, interval, hawk_ai)
return self.connection.execute(query).fetchall()
def db_to_pf(self, jobid, interval, hawk_ai):
import pandas as pd
query = self.build_query(jobid, interval, hawk_ai)
return pd.read_sql(query, con=self.connection)
class App:
def __init__(self, config):
self.config = config
self.db = MonitoringDB(self.config.verbose)
@staticmethod
def warnings(config):
warn = ""
if not config.hawk_ai and config.interval < 5:
warn += '# Warning: interval<5 is very small and may lead to data gaps.'
if config.hawk_ai and config.interval < 60:
warn += '# Warning: interval<60 is very small for Hawk-AI nodes and may lead to data gaps.'
return warn
def run_all(self):
warnings = self.warnings(self.config)
if warnings:
print(warnings)
header = f"# {config.datetime}: {config.cmd}\n"
if warnings:
header += f"{warnings}\n"
header += "#\n"
for jobid in self.config.jobid:
try:
power = Power.from_db(self.db, jobid, self.config.interval, self.config.hawk_ai)
except RuntimeError:
print('No data found for job ID "{}"'.format(jobid))
continue
fn = power.to_file(jobid, header)
if fn:
print('Created file {fn}'.format(fn=fn))
if __name__ == "__main__":
import sys
from datetime import datetime
config = parse_arguments(sys.argv[1:])
config.cmd = " ".join(sys.argv)
config.datetime = f"{datetime.now()}"
main = App(config)
main.run_all()