restgym-TrackB / src /run.py
minhhungg's picture
feat: initial upload for restgym Track B datasets and logs
e675fa3
Raw
History Blame Contribute Delete
17.5 kB
import common
import socket
import random
import threading
import time
import psutil
import sys
import os
import yaml
from rich.progress import Progress
from rich import print
MINIMUM_CPUS = 4
MINIMUM_RAM_GB = 4
TIME_BUDGET_MINS = 60
specification_volumes = {}
# Read configuration form file to override default config
def read_config():
# Override config from file, if available
with open(f'{common.RESTGYM_BASE_DIR}/restgym-config.yml') as stream:
try:
config = yaml.safe_load(stream)
global MINIMUM_CPUS, MINIMUM_RAM_GB, TIME_BUDGET_MINS
MINIMUM_RAM_GB = int(config['minimum_ram_gb'])
MINIMUM_CPUS = int(config['minimum_cpus'])
TIME_BUDGET_MINS = int(config['time_budget_mins'])
except yaml.YAMLError as exc:
print("Could not parse RESTgym configuration file. Continuing with default configuration.")
print(exc)
# Check if a TCP port is free
def check_tcp_port(port):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
return not s.connect_ex(('localhost', port)) == 0
# Get random free TCP port
def get_random_free_tcp_port():
remaining_attempts = 1000
while remaining_attempts > 0:
candidate_port = random.randrange(10_000, 60_000)
if check_tcp_port(candidate_port):
return candidate_port
remaining_attempts -= 1
print("[red]ERROR[/red]: could not find a free TCP port for the API.")
sys.exit(1)
# Compute the remaining runs to execute
def compute_remaining_runs(desired_runs):
remaining_runs = []
apis = common.get_apis()
print("Enabled APIs: ", apis)
tools = common.get_tools()
print("Enabled tools: ", tools)
for api in apis:
for tool in tools:
try:
base_dir = f"{common.RESTGYM_BASE_DIR}/results/{api}/{tool}"
subdirs = os.listdir(base_dir)
count = 0
for subdir in subdirs:
if os.path.exists(f"{base_dir}/{subdir}/completed.txt"):
count += 1
while count < desired_runs:
remaining_runs.append({'api': api, 'tool': tool})
count += 1
except:
for _ in range(desired_runs):
remaining_runs.append({'api': api, 'tool': tool})
return remaining_runs
# Verify Docker images have been built
def check_docker_images(remaining_runs):
images = set()
missing_images = []
for remaining_run in remaining_runs:
images.add(remaining_run['tool'])
images.add(remaining_run['api'])
for image in images:
try:
common.DOCKER_CLIENT.images.get(common.DOCKER_PREFIX + image)
except:
missing_images.append(image)
return missing_images
# Filter out runs with missing images
def filter_runs_with_missing_images(remaining_runs, missing_images):
filtered_remaining_runs = []
for remaining_run in remaining_runs:
if remaining_run['tool'] not in missing_images and remaining_run['api'] not in missing_images:
filtered_remaining_runs.append(remaining_run)
return filtered_remaining_runs
# Remove excluded runs
def remove_matching_dicts(list1, list2):
# Convert list2 dicts to hashable tuples (sorted for key order invariance)
to_remove = {tuple(sorted(d.items())) for d in list2}
# Filter list1 to exclude exact matches
return [d for d in list1 if tuple(sorted(d.items())) not in to_remove]
# Check if there are enough resources for another concurrent run
def check_resources():
available_ram = getattr(psutil.virtual_memory(), 'available')
available_cpus = (1 - (psutil.cpu_percent() / 100)) * psutil.cpu_count()
return available_ram > MINIMUM_RAM_GB * 1024 * 1024 * 1024 and available_cpus > MINIMUM_CPUS
# Deeper check of resources (10 checks each second)
def deep_check_resources():
for _ in range(10):
if not check_resources():
return False
time.sleep(1)
return True
# Execute an experiment run
def launch_run(api, tool, run_count, total_runs, progress, experiment_task):
attempts = 5
successfully_completed = False
progress.update(experiment_task, advance=1)
while attempts > 0 and not successfully_completed:
attempts -= 1
error_occurred = False
run = 'run-' + time.strftime('%Y%m%d-%H%M%S')
results_path = f'{common.RESTGYM_BASE_DIR}/results/{api}/{tool}/{run}'
ports = {'9090/tcp': get_random_free_tcp_port()}
env = {
'API': api,
'TOOL': tool,
'RUN': run,
'TIME_BUDGET': TIME_BUDGET_MINS,
'HOST': 'localhost'
}
os.makedirs(results_path, exist_ok=True, mode=0o777)
os.makedirs(f'{results_path}{common.LOGS_PATH}', exist_ok=True, mode=0o777)
message = '[green]START[/green]' if attempts == 4 else '[yellow]RETRY[/yellow]'
with open(f'{results_path}/time-budget.txt', 'a') as f:
f.write(f'Time budget: {TIME_BUDGET_MINS} minutes.\n')
print(f" => [{message}] ({run_count}/{total_runs}) Running {tool} on {api} ({run}).")
with open(f'{results_path}/started.txt', 'a') as f:
f.write(f'Run started on {time.ctime()}.\n')
api_container_name = f'{common.DOCKER_PREFIX}-{api}-for-{tool}--{run}'
tool_container_name = f'{common.DOCKER_PREFIX}-{tool}-for-{api}--{run}'
# Verify Docker images have been built
try:
common.DOCKER_CLIENT.images.get(common.DOCKER_PREFIX + remaining_run['api'])
common.DOCKER_CLIENT.images.get(common.DOCKER_PREFIX + remaining_run['tool'])
except:
print(
f" => [[red]ERROR[/red]] ({run_count}/{total_runs}) Execution failed for {remaining_run['tool']} on {remaining_run['api']}. Missing Docker image(s). Have you built them?")
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(
f"Docker image(s) not found for API ({remaining_run['api']}) or tool ({remaining_run['tool']}).\n\n")
error_occurred = True
# Start API
if not error_occurred:
try:
api_container = common.DOCKER_CLIENT.containers.run(
image=f'{common.DOCKER_PREFIX}{api}',
name=api_container_name,
environment=env,
ports=ports,
volumes=[f'{common.RESTGYM_BASE_DIR_HOST}/results/:/results/'],
mem_limit='16g',
nano_cpus=8_000_000_000,
user=f'{os.getuid()}:{os.getgid()}',
detach=True
)
except Exception as e:
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f"Could not start API ({api}) container.\n{e}\n\n")
error_occurred = True
# Wait 45 seconds for the API to start
if not error_occurred:
time.sleep(45)
else:
time.sleep(2)
# Get the host port assigned by docker
if not error_occurred:
try:
api_container.reload()
env['PORT'] = str(int(api_container.attrs['NetworkSettings']['Ports']['9090/tcp'][0]['HostPort']))
except Exception as e:
print(
f" => [[red]ERROR[/red]] ({run_count}/{total_runs}) Execution failed for {remaining_run['tool']} on {remaining_run['api']}. The port 9090 is not published. (API container has crashed or API/proxy are not exposed).")
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(
f"Could not start API ({api}) container. Docker did not publish port 9090 as expected. The API container crashed on startup or the API/proxy is not exposed.\n{e}\n\n")
# Start tool
if not error_occurred:
try:
tool_container = common.DOCKER_CLIENT.containers.run(
image=f'{common.DOCKER_PREFIX}{tool}',
name=tool_container_name,
environment=env,
privileged=True,
volumes=specification_volumes,
network_mode='host',
mem_limit='16gb',
nano_cpus=8_000_000_000,
detach=True
)
time.sleep(1)
except Exception as e:
api_container.stop()
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f"Could not start tool ({tool}) container.\n{e}\n\n")
error_occurred = True
# Perform a health check of containers each minute, for 60 times
if not error_occurred:
for minute in range(1, TIME_BUDGET_MINS + 1):
time.sleep(60)
progress.update(experiment_task, advance=1)
try:
api_container.reload()
if api_container.status == 'exited':
raise Exception("API container exited.")
# If the container was removed or it exited
except Exception as e:
print(f" => [[red]ERROR[/red]] ({run_count}/{total_runs}) API container stopped.")
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f"API container not running at minute {minute}. Aborting.\n{e}\n\n")
try:
tool_container.stop()
except:
pass
error_occurred = True
break
try:
tool_container.reload()
if tool_container.status == 'exited':
raise Exception("Container exited")
# If the container was removed or it exited
except Exception as e:
print(f" => [[red]ERROR[/red]] ({run_count}/{total_runs}) Tool container stopped.")
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f"Tool container not running at minute {minute}. Aborting.\n{e}\n\n")
try:
api_container.stop()
except:
pass
error_occurred = True
break
# Stop tool container
if not error_occurred:
try:
tool_container.stop()
tool_container.wait()
except Exception as e:
error_occurred = True
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f'Could not stop tool ({tool}) container. It possibly crashed.\n{e}\n\n')
try:
api_container.stop()
except:
pass
# Save log of tool container and remove it
try:
with open(f'{results_path}{common.LOGS_PATH}/{tool}-stdout.log', 'wb') as f_out, open(f'{results_path}{common.LOGS_PATH}/{tool}-stderr.log', 'wb') as f_err:
f_out.write(tool_container.logs(stdout=True, stderr=False))
f_err.write(tool_container.logs(stdout=False, stderr=True))
tool_container.remove()
except Exception as e:
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f'Could not write log or remove tool ({tool}) container.\n{e}\n\n')
# Wait 5 seconds to let the API container store the database
if not error_occurred:
time.sleep(5)
# Stop API container
if not error_occurred:
try:
api_container.stop()
api_container.wait()
except Exception as e:
error_occurred = True
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f'Could not stop API ({api}) container. It possibly crashed.\n{e}\n\n')
# Save log of API container and remove it
try:
with open(f'{results_path}{common.LOGS_PATH}/{api}-stdout.log', 'wb') as f_out, open(
f'{results_path}{common.LOGS_PATH}/{api}-stderr.log', 'wb') as f_err:
f_out.write(api_container.logs(stdout=True, stderr=False))
f_err.write(api_container.logs(stdout=False, stderr=True))
api_container.remove()
except Exception as e:
with open(f'{results_path}/errors.txt', 'a') as f:
f.write(f'Could not write log or remove API ({api}) container. \n{e}\n\n')
# Final stages
if not error_occurred:
successfully_completed = True
with open(f'{results_path}/completed.txt', 'a') as f:
f.write(f'Run completed on {time.ctime()}.\n')
print(f" => [[green]-END-[/green]] ({run_count}/{total_runs}) Run of {tool} on {api} ({run}) completed.")
else:
time.sleep(2)
if attempts == 0:
print(f" => [[red]ERROR[/red]] ({run_count}/{total_runs}) Run of {tool} on {api} ({run}) terminated with errors.")
# Main
if __name__ == "__main__":
common.welcome()
read_config()
print("This is the run module. It will run up to 20 repetitions of the experiment for each tool and API.")
print(f"Time budget: {TIME_BUDGET_MINS} minutes. Minimum CPUs: {MINIMUM_CPUS}. Minimum RAM: {MINIMUM_RAM_GB} GB.")
try:
desired_runs = int(input("How many runs? [1-20]: "))
except:
print("Please specify an whole number.")
sys.exit(1)
if desired_runs < 1 or desired_runs > 20:
print("Please specify a number in the range 1-20.")
sys.exit(1)
remaining_runs = compute_remaining_runs(desired_runs)
# Do not execute the following configurations (e.g., a tool does not support an API, or it is known to crash)
skip_runs = [
{'api': 'pet-clinic', 'tool': 'schemathesis'}, # Crashes while parsing regex
{'api': 'flight-search', 'tool': 'autoresttest'}, # Crashes after 5 minutes
{'api': 'flight-search', 'tool': 'restest'}, # Crashes
{'api': 'features-service', 'tool': 'evomaster'}, # API Crashes
{'api': 'kafka-rest-proxy', 'tool': 'resttestgen'}, # API crashes
{'api': 'erc20', 'tool': 'cats'} # Crashes after 1 minute
]
remaining_runs = remove_matching_dicts(remaining_runs, skip_runs)
# Uncomment next line to launch a manual subset of runs
# remaining_runs = [{'api': 'market', 'tool': 'restler'}]
missing_images = check_docker_images(remaining_runs)
if len(missing_images) > 0:
filtered_remaining_runs = filter_runs_with_missing_images(remaining_runs, missing_images)
print(
f"Some Docker images required for the experiment have not been built. Skipping experiment runs that involve these images. Only {len(filtered_remaining_runs)} out of {len(remaining_runs)} runs can be launched.")
remaining_runs = filtered_remaining_runs
else:
print(f"Runs planned for execution: {len(remaining_runs)}.")
total_runs = len(remaining_runs)
run_count = 0
for api in common.get_apis():
if common.check_enabled('api', api):
# JSON specification
specification_volumes[f'{common.RESTGYM_BASE_DIR_HOST}/apis/{api}/specifications/{api}-openapi.json'] = {
'bind': f'/specifications/{api}-openapi.json',
'mode': 'ro'
}
# YAML specification
specification_volumes[f'{common.RESTGYM_BASE_DIR_HOST}/apis/{api}/specifications/{api}.yaml'] = {
'bind': f'/specifications/{api}.yaml',
'mode': 'ro'
}
input("Press ENTER to start the execution of the experiment (or CTRL+C to cancel)...")
with Progress() as progress:
experiment_task = progress.add_task("Running experiment...", total=total_runs*(TIME_BUDGET_MINS+1))
threads = []
while len(remaining_runs) > 0:
run_count += 1
# Pick random run
remaining_run = remaining_runs.pop(random.randrange(len(remaining_runs)))
# Stop until resources are available
notify_no_resources = True
while not deep_check_resources():
if notify_no_resources:
print(f" => [[blue]-WAIT[/blue]] ({run_count}/{total_runs}) Waiting for system resources to be released.")
notify_no_resources = False
time.sleep(30)
# Launch run in separate thread
run_thread = threading.Thread(
target=launch_run,
args=(remaining_run['api'], remaining_run['tool'], run_count, total_runs, progress, experiment_task)
)
run_thread.start()
threads.append(run_thread)
# If not last run, wait 60 seconds before launching next
if len(remaining_runs) > 0:
time.sleep(60)
# Wait for all the threads to complete
for t in threads:
t.join()
print("Execution completed.")