NikolajDanger/meow_base
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

moved benchmarking to its own meow_benchmarks package

Browse Source
This commit is contained in:
PatchOfScotland
2023-03-15 11:16:14 +01:00
parent ede29f3158
commit a7224430e9
10 changed files with 8 additions and 738 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
benchmarking/mrme.py
View File

@ -1,35 +0,0 @@
from meow_base.patterns import FileEventPattern
from meow_base.recipes import get_recipe_from_notebook
from shared import run_test, MRME
def multiple_rules_multiple_events(job_count:int, REPEATS:int, job_counter:int,
requested_jobs:int, runtime_start:float):
patterns = {}
for i in range(job_count):
pattern = FileEventPattern(
f"pattern_{i}",
f"testing/file_{i}.txt",
"recipe_one",
"input"
)
patterns[pattern.name] = pattern
recipe = get_recipe_from_notebook("recipe_one", "test.ipynb")
recipes = {
recipe.name: recipe
}
run_test(
patterns,
recipes,
job_count,
job_count,
REPEATS,
job_counter,
requested_jobs,
runtime_start,
signature=MRME
)

35
benchmarking/mrse.py
View File

@ -1,35 +0,0 @@
from meow_base.patterns import FileEventPattern
from meow_base.recipes import get_recipe_from_notebook
from shared import run_test, MRSE
def multiple_rules_single_event(job_count:int, REPEATS:int, job_counter:int,
requested_jobs:int, runtime_start:float):
patterns = {}
for i in range(job_count):
pattern = FileEventPattern(
f"pattern_{i}",
f"testing/*",
"recipe_one",
"input"
)
patterns[pattern.name] = pattern
recipe = get_recipe_from_notebook("recipe_one", "test.ipynb")
recipes = {
recipe.name: recipe
}
run_test(
patterns,
recipes,
1,
job_count,
REPEATS,
job_counter,
requested_jobs,
runtime_start,
signature=MRSE
)

201
benchmarking/run_all.py
View File

@ -1,201 +0,0 @@
import matplotlib.pyplot as pyplot
import numpy
import sys
import time
import os
from typing import List
from shared import JOBS_COUNTS, REPEATS, TESTS, MRME, MRSE, SRME, SRSEP, \
SRSES, RESULTS_DIR, BASE
from mrme import multiple_rules_multiple_events
from mrse import multiple_rules_single_event
from srme import single_rule_multiple_events
from srsep import single_rule_single_event_parallel
from srsps import single_rule_single_event_sequential
from meow_base.core.correctness.vars import DEFAULT_JOB_OUTPUT_DIR, \
DEFAULT_JOB_QUEUE_DIR
from meow_base.functionality.file_io import rmtree
LINE_KEYS = {
SRSES: ('x','#a1467e'),
SRME: ('.','#896cff'),
MRME: ('d','#5983b0'),
MRSE: ('P','#ff6cbe'),
SRSEP: ('*','#3faf46'),
}
def run_tests():
rmtree(RESULTS_DIR)
requested_jobs=0
for job_count in JOBS_COUNTS:
requested_jobs += job_count * REPEATS * len(TESTS)
print(f"requested_jobs: {requested_jobs}")
runtime_start=time.time()
job_counter=0
for job_count in JOBS_COUNTS:
for test in TESTS:
if test == MRME:
multiple_rules_multiple_events(job_count, REPEATS, job_counter, requested_jobs, runtime_start)
job_counter += job_count * REPEATS
elif test == MRSE:
multiple_rules_single_event(job_count, REPEATS, job_counter, requested_jobs, runtime_start)
job_counter += job_count * REPEATS
elif test == SRME:
single_rule_multiple_events(job_count, REPEATS, job_counter, requested_jobs, runtime_start)
job_counter += job_count * REPEATS
elif test == SRSEP:
single_rule_single_event_parallel(job_count, REPEATS, job_counter, requested_jobs, runtime_start)
job_counter += job_count * REPEATS
elif test == SRSES:
single_rule_single_event_sequential(job_count, REPEATS, job_counter, requested_jobs, runtime_start)
job_counter += job_count * REPEATS
print(f"All tests completed in: {str(time.time()-runtime_start)}")
def get_meow_graph(results_dir:str):
lines = []
for run_type in os.listdir(results_dir):
#if run_type == 'single_Pattern_single_file_sequential':
# continue
# lines.append((f'scheduling {run_type}', [], 'solid'))
lines.append((run_type, [], 'solid'))
run_path = os.path.join(results_dir, run_type)
for job_count in os.listdir(run_path):
results_path = os.path.join(run_path, job_count, "results.txt")
with open(results_path, 'r') as f_in:
data = f_in.readlines()
scheduling_duration = 0
for line in data:
if "Average schedule time: " in line:
scheduling_duration = float(line.replace(
"Average schedule time: ", ''))
lines[-1][1].append((job_count, scheduling_duration))
lines[-1][1].sort(key=lambda y: float(y[0]))
return lines
def make_plot(lines:List, graph_path:str, title:str, logged:bool):
w = 10
h = 4
linecount = 0
columns = 1
pyplot.figure(figsize=(w, h))
for l in range(len(lines)):
x_values = numpy.asarray([float(i[0]) for i in lines[l][1]])
y_values = numpy.asarray([float(i[1]) for i in lines[l][1]])
# Remove this check to always display lines
if lines[l][2] == 'solid':
pyplot.plot(x_values, y_values, label=lines[l][0], linestyle=lines[l][2], marker=LINE_KEYS[lines[l][0]][0], color=LINE_KEYS[lines[l][0]][1])
linecount += 1
columns = int(linecount/3) + 1
pyplot.xlabel("Number of jobs scheduled")
pyplot.ylabel("Time taken (seconds)")
pyplot.title(title)
handles, labels = pyplot.gca().get_legend_handles_labels()
# legend_order = [2, 4, 0, 1, 3]
# pyplot.legend([handles[i] for i in legend_order], [labels[i] for i in legend_order])
pyplot.legend(ncol=columns, prop={'size': 12})
if logged:
pyplot.yscale('log')
x_ticks = []
for tick in x_values:
label = int(tick)
if tick <= 100 and tick % 20 == 0:
label = f"\n{int(tick)}"
x_ticks.append(label)
pyplot.xticks(x_values, x_ticks)
pyplot.savefig(graph_path, format='pdf', bbox_inches='tight')
def make_both_plots(lines, path, title, log=True):
make_plot(lines, path, title, False)
if log:
logged_path = path[:path.index(".pdf")] + "_logged" \
+ path[path.index(".pdf"):]
make_plot(lines, logged_path, title, True)
def make_graphs():
lines = get_meow_graph(RESULTS_DIR)
make_both_plots(
lines,
"result.pdf",
"MiG scheduling overheads on the Threadripper"
)
average_lines = []
all_delta_lines = []
no_spsfs_delta_lines = []
for line_signature, line_values, lines_style in lines:
if lines_style == 'solid':
averages = [(i, v/float(i)) for i, v in line_values]
average_lines.append((line_signature, averages, lines_style))
if line_signature not in [
"total single_Pattern_single_file_sequential",
"scheduling single_Pattern_single_file_sequential_jobs",
"SPSFS"]:
deltas = []
for i in range(len(line_values)-1):
deltas.append((line_values[i+1][0],
(averages[i+1][1]-averages[i][1]) \
/ (float(averages[i+1][0])-float(averages[i][0]))))
no_spsfs_delta_lines.append((line_signature, deltas, lines_style))
deltas = []
for i in range(len(line_values)-1):
deltas.append((line_values[i+1][0],
(averages[i+1][1]-averages[i][1]) \
/ (float(averages[i+1][0])-float(averages[i][0]))))
all_delta_lines.append((line_signature, deltas, lines_style))
make_both_plots(
average_lines,
"result_averaged.pdf",
"Per-job MiG scheduling overheads on the Threadripper"
)
make_both_plots(
all_delta_lines,
"result_deltas.pdf",
"Difference in per-job MiG scheduling overheads on the Threadripper",
log=False
)
if __name__ == '__main__':
try:
run_tests()
make_graphs()
rmtree(DEFAULT_JOB_QUEUE_DIR)
rmtree(DEFAULT_JOB_OUTPUT_DIR)
rmtree(BASE)
except KeyboardInterrupt as ki:
try:
sys.exit(1)
except SystemExit:
os._exit(1)

69
benchmarking/sequential.ipynb
View File

@ -1,69 +0,0 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [],
"source": [
"INPUT_FILE = 'file_0.txt'\n",
"MAX_COUNT = 100"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [],
"source": [
"with open(INPUT_FILE, 'r') as f:\n",
" data = int(f.read())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(\"read in: \"+ str(data))\n",
"print(\"writing out: \"+ str(data+1))"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [],
"source": [
"if data+1 < MAX_COUNT:\n",
" with open(INPUT_FILE.replace(str(data), str(data+1)), 'w') as f:\n",
" f.write(str(data+1))"
]
}
],
"metadata": {
"interpreter": {
"hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
},
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.5"
}
},
"nbformat": 4,
"nbformat_minor": 4
}

242
benchmarking/shared.py
View File

@ -1,242 +0,0 @@
import datetime
import io
import os
import pathlib
import time
import yaml
from typing import Any, Dict, Tuple, List
from meow_base.core.correctness.vars import DEFAULT_JOB_OUTPUT_DIR, \
DEFAULT_JOB_QUEUE_DIR
from meow_base.core.base_pattern import BasePattern
from meow_base.core.base_recipe import BaseRecipe
from meow_base.core.runner import MeowRunner
from meow_base.patterns.file_event_pattern import WatchdogMonitor
from meow_base.recipes.jupyter_notebook_recipe import PapermillHandler
from meow_base.conductors import LocalPythonConductor
from meow_base.functionality.file_io import rmtree
RESULTS_DIR = "results"
BASE = "benchmark_base"
REPEATS = 10
JOBS_COUNTS = [
10, 20, 30, 40, 50,
60, 70, 80, 90, 100,
125, 150, 175, 200,
250, 300, 400, 500
]
SRME = "single_rule_multiple_events"
MRSE = "multiple_rules_single_event"
SRSEP = "single_rule_single_event_parallel"
MRME = "multiple_rules_multiple_events"
SRSES = "single_rule_single_event_sequential"
TESTS = [
SRME,
MRSE,
SRSEP,
MRME,
# This test will take approx 90% of total time
SRSES
]
class DummyConductor(LocalPythonConductor):
def valid_execute_criteria(self, job:Dict[str,Any])->Tuple[bool,str]:
return False, ">:("
def datetime_to_timestamp(date_time_obj:datetime):
return time.mktime(date_time_obj.timetuple()) \
+ float(date_time_obj.microsecond)/1000000
def generate(file_count:int, file_path:str, file_type:str=".txt"):
first_filename = ''
start = time.time()
for i in range(int(file_count)):
filename = file_path + str(i) + file_type
if not first_filename:
first_filename = filename
with open(filename, 'w') as f:
f.write('0')
return first_filename, time.time() - start
def cleanup(jobs:List[str], file_out:str, base_time:float, gen_time:float,
execution:bool=False):
if not jobs:
return
job_timestamps = []
for job in jobs:
if execution:
with open(f"{DEFAULT_JOB_OUTPUT_DIR}/{job}/job.yml", 'r') as f_in:
data = yaml.load(f_in, Loader=yaml.Loader)
else:
with open(f"{DEFAULT_JOB_QUEUE_DIR}/{job}/job.yml", 'r') as f_in:
data = yaml.load(f_in, Loader=yaml.Loader)
create_datetime = data['create']
create_timestamp = datetime_to_timestamp(create_datetime)
job_timestamps.append((create_timestamp, create_datetime))
job_timestamps.sort(key=lambda y: int(y[0]))
first = job_timestamps[0]
last = job_timestamps[-1]
#dt = datetime.datetime.fromtimestamp(os.path.getctime(base_time), datetime.timezone(datetime.timedelta(hours=0)))
dt = datetime.datetime.fromtimestamp(os.path.getctime(base_time))
# if execution:
# queue_times = []
# execution_times = []
# for j in jobs:
# mrsl_dict = load(os.path.join(mrsl_dir, j))#
#
# queue_times.append(time.mktime(mrsl_dict['EXECUTING_TIMESTAMP']) - time.mktime(mrsl_dict['QUEUED_TIMESTAMP']))
# execution_times.append(time.mktime(mrsl_dict['FINISHED_TIMESTAMP']) - time.mktime(mrsl_dict['EXECUTING_TIMESTAMP']))
pathlib.Path(os.path.dirname(file_out)).mkdir(parents=True, exist_ok=True)
with open(file_out, 'w') as f_out:
f_out.write("Job count: "+ str(len(jobs)) +"\n")
f_out.write("Generation time: "+ str(round(gen_time, 5)) +"\n")
f_out.write("First trigger: "+ str(dt) +"\n")
f_out.write("First scheduling datetime: "+ str(first[1]) +"\n")
f_out.write("Last scheduling datetime: "+ str(last[1]) +"\n")
f_out.write("First scheduling unixtime: "+ str(first[0]) +"\n")
f_out.write("First scheduling unixtime: "+ str(last[0]) +"\n")
f_out.write("Scheduling difference (seconds): "+ str(round(last[0] - first[0], 3)) +"\n")
f_out.write("Initial scheduling delay (seconds): "+ str(round(first[0] - os.path.getctime(base_time), 3)) +"\n")
total_time = round(last[0] - os.path.getctime(base_time), 3)
f_out.write("Total scheduling delay (seconds): "+ str(total_time) +"\n")
# if execution:
# f_out.write("Average execution time (seconds): "+ str(round(mean(execution_times), 3)) +"\n")
# f_out.write("Max execution time (seconds): "+ str(round(max(execution_times), 3)) +"\n")
# f_out.write("Min execution time (seconds): "+ str(round(min(execution_times), 3)) +"\n")
# f_out.write("Average queueing delay (seconds): "+ str(round(mean(queue_times), 3)) +"\n")
# f_out.write("Max queueing delay (seconds): "+ str(round(max(queue_times), 3)) +"\n")
# f_out.write("Min queueing delay (seconds): "+ str(round(min(queue_times), 3)) +"\n")
# queue_times.remove(max(queue_times))
# f_out.write("Average excluded queueing delay (seconds): "+ str(round(mean(queue_times), 3)) +"\n")
return total_time
def mean(l:List):
return sum(l)/len(l)
def collate_results(base_dir:str):
scheduling_delays = []
for run in os.listdir(base_dir):
if run != 'results.txt':
with open(os.path.join(base_dir, run, "results.txt"), 'r') as f:
d = f.readlines()
for l in d:
if "Total scheduling delay (seconds): " in l:
scheduling_delays.append(float(l.replace(
"Total scheduling delay (seconds): ", '')))
with open(os.path.join(base_dir, 'results.txt'), 'w') as f:
f.write(f"Average schedule time: {round(mean(scheduling_delays), 3)}\n")
f.write(f"Scheduling times: {scheduling_delays}")
def run_test(patterns:Dict[str,BasePattern], recipes:Dict[str,BaseRecipe],
files_count:int, expected_job_count:int, repeats:int, job_counter:int,
requested_jobs:int, runtime_start:float, signature:str="",
execution:bool=False, print_logging:bool=False):
if not os.path.exists(RESULTS_DIR):
os.mkdir(RESULTS_DIR)
# Does not work. left here as reminder
if execution:
os.system("export LC_ALL=C.UTF-8")
os.system("export LANG=C.UTF-8")
for run in range(repeats):
# Ensure complete cleanup from previous run
for f in [BASE, DEFAULT_JOB_QUEUE_DIR, DEFAULT_JOB_OUTPUT_DIR]:
if os.path.exists(f):
rmtree(f)
file_base = os.path.join(BASE, 'testing')
pathlib.Path(file_base).mkdir(parents=True, exist_ok=True)
runner_debug_stream = io.StringIO("")
if execution:
runner = MeowRunner(
WatchdogMonitor(BASE, patterns, recipes, settletime=1),
PapermillHandler(),
LocalPythonConductor(),
print=runner_debug_stream,
logging=3
)
else:
runner = MeowRunner(
WatchdogMonitor(BASE, patterns, recipes, settletime=1),
PapermillHandler(),
DummyConductor(),
print=runner_debug_stream,
logging=3
)
runner.start()
# Generate triggering files
first_filename, generation_duration = \
generate(files_count, file_base +"/file_")
idle_loops = 0
total_loops = 0
messages = 0
total_time = expected_job_count * 3
if execution:
total_time = expected_job_count * 5
while idle_loops < 10 and total_loops < total_time:
time.sleep(1)
runner_debug_stream.seek(0)
new_messages = len(runner_debug_stream.readlines())
if messages == new_messages:
idle_loops += 1
else:
idle_loops = 0
messages = new_messages
total_loops += 1
runner.stop()
if execution:
jobs = os.listdir(DEFAULT_JOB_OUTPUT_DIR)
else:
jobs = os.listdir(DEFAULT_JOB_QUEUE_DIR)
results_path = os.path.join(
RESULTS_DIR,
signature,
str(expected_job_count),
str(run),
"results.txt"
)
cleanup(
jobs,
results_path,
first_filename,
generation_duration,
execution=execution
)
print(f"Completed scheduling run {str(run + 1)} of {str(len(jobs))}"
f"/{str(expected_job_count)} jobs for '{signature}' "
f"{job_counter + expected_job_count*(run+1)}/{requested_jobs} "
f"({str(round(time.time()-runtime_start, 3))}s)")
collate_results(
os.path.join(RESULTS_DIR, signature, str(expected_job_count))
)

34
benchmarking/srme.py
View File

@ -1,34 +0,0 @@
from meow_base.patterns import FileEventPattern
from meow_base.recipes import get_recipe_from_notebook
from shared import run_test, SRME
def single_rule_multiple_events(job_count:int, REPEATS:int, job_counter:int,
requested_jobs:int, runtime_start:float):
patterns = {}
pattern = FileEventPattern(
f"pattern_one",
f"testing/*",
"recipe_one",
"input"
)
patterns[pattern.name] = pattern
recipe = get_recipe_from_notebook("recipe_one", "test.ipynb")
recipes = {
recipe.name: recipe
}
run_test(
patterns,
recipes,
job_count,
job_count,
REPEATS,
job_counter,
requested_jobs,
runtime_start,
signature=SRME
)

36
benchmarking/srsep.py
View File

@ -1,36 +0,0 @@
from meow_base.patterns import FileEventPattern
from meow_base.recipes import get_recipe_from_notebook
from meow_base.functionality.meow import create_parameter_sweep
from shared import run_test, SRSEP
def single_rule_single_event_parallel(job_count:int, REPEATS:int,
job_counter:int, requested_jobs:int, runtime_start:float):
patterns = {}
pattern = FileEventPattern(
f"pattern_one",
f"testing/*",
"recipe_one",
"input",
sweep=create_parameter_sweep("var", 1, job_count, 1)
)
patterns[pattern.name] = pattern
recipe = get_recipe_from_notebook("recipe_one", "test.ipynb")
recipes = {
recipe.name: recipe
}
run_test(
patterns,
recipes,
1,
job_count,
REPEATS,
job_counter,
requested_jobs,
runtime_start,
signature=SRSEP
)

39
benchmarking/srsps.py
View File

@ -1,39 +0,0 @@
from meow_base.patterns import FileEventPattern
from meow_base.recipes import get_recipe_from_notebook
from shared import run_test, SRSES
def single_rule_single_event_sequential(job_count:int, REPEATS:int,
job_counter:int, requested_jobs:int, runtime_start:float):
patterns = {}
pattern = FileEventPattern(
f"pattern_one",
f"testing/*",
"recipe_two",
"INPUT_FILE",
parameters={
"MAX_COUNT":job_count
}
)
patterns[pattern.name] = pattern
recipe = get_recipe_from_notebook("recipe_two", "sequential.ipynb")
recipes = {
recipe.name: recipe
}
run_test(
patterns,
recipes,
1,
job_count,
REPEATS,
job_counter,
requested_jobs,
runtime_start,
signature=SRSES,
execution=True,
print_logging=False
)

47
benchmarking/test.ipynb
View File

@ -1,47 +0,0 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"ename": "",
"evalue": "",
"output_type": "error",
"traceback": [
"\u001b[1;31mFailed to start the Kernel. \n",
"Failed to start the Kernel 'Python 3.6.9 64-bit'. \n",
"View Jupyter <a href='command:jupyter.viewOutput'>log</a> for further details. Kernel has not been started"
]
}
],
"source": [
"print('this is some outpug ')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"interpreter": {
"hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
},
"kernelspec": {
"display_name": "Python 3.6.9 64-bit",
"language": "python",
"name": "python3"
},
"language_info": {
"name": "python",
"version": "3.6.9"
}
},
"nbformat": 4,
"nbformat_minor": 4
}

8
patterns/file_event_pattern.py
View File

@ -559,6 +559,14 @@ class WatchdogEventHandler(PatternMatchingEventHandler):
else:
self._recent_jobs[event.src_path] = \
[event.time_stamp, {event.event_type}]
# If we have a closed event then short-cut the wait and send event
# immediately
if event.event_type == FILE_CLOSED_EVENT:
self.monitor.match(event)
self._recent_jobs_lock.release()
return
except Exception as ex:
self._recent_jobs_lock.release()
raise Exception(ex)