
Refactoring some code to make use of the library + Splitting into methods + Making methods independent with akrzos suggestions Change-Id: I7a18d4c506a5dcd28d96872226e5001d4603cf48
193 lines
9.4 KiB
Python
193 lines
9.4 KiB
Python
from Connmon import Connmon
|
|
from Tools import Tools
|
|
from collections import OrderedDict
|
|
from Grafana import Grafana
|
|
import datetime
|
|
import glob
|
|
import logging
|
|
import os
|
|
import shutil
|
|
import subprocess
|
|
import time
|
|
|
|
|
|
class Rally:
|
|
def __init__(self, config, hosts=None):
|
|
self.logger = logging.getLogger('browbeat.Rally')
|
|
self.config = config
|
|
self.tools = Tools(self.config)
|
|
self.connmon = Connmon(self.config)
|
|
self.grafana = Grafana(self.config)
|
|
self.error_count = 0
|
|
self.test_count = 0
|
|
self.scenario_count = 0
|
|
|
|
def run_scenario(self, task_file, scenario_args, result_dir, test_name):
|
|
self.logger.debug("--------------------------------")
|
|
self.logger.debug("task_file: {}".format(task_file))
|
|
self.logger.debug("scenario_args: {}".format(scenario_args))
|
|
self.logger.debug("result_dir: {}".format(result_dir))
|
|
self.logger.debug("test_name: {}".format(test_name))
|
|
self.logger.debug("--------------------------------")
|
|
|
|
from_ts = int(time.time() * 1000)
|
|
if 'sleep_before' in self.config['rally']:
|
|
time.sleep(self.config['rally']['sleep_before'])
|
|
task_args = str(scenario_args).replace("'", "\"")
|
|
cmd = "source {}; ".format(self.config['rally']['venv'])
|
|
cmd += "rally task start {} --task-args \'{}\' 2>&1 | tee {}.log".format(task_file,
|
|
task_args, test_name)
|
|
self.tools.run_cmd(cmd)
|
|
if 'sleep_after' in self.config['rally']:
|
|
time.sleep(self.config['rally']['sleep_after'])
|
|
to_ts = int(time.time() * 1000)
|
|
|
|
self.grafana.print_dashboard_url(from_ts, to_ts, test_name)
|
|
self.grafana.log_snapshot_playbook_cmd(from_ts, to_ts, result_dir, test_name)
|
|
self.grafana.run_playbook(from_ts, to_ts, result_dir, test_name)
|
|
|
|
def workload_logger(self, result_dir):
|
|
base = result_dir.split('/')
|
|
if not os.path.isfile("{}/{}/browbeat-rally-run.log".format(base[0], base[1])):
|
|
file = logging.FileHandler("{}/{}/browbeat-rally-run.log".format(base[0], base[1]))
|
|
file.setLevel(logging.DEBUG)
|
|
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)5s - %(message)s')
|
|
file.setFormatter(formatter)
|
|
self.logger.addHandler(file)
|
|
return None
|
|
|
|
def get_test_count(self):
|
|
return self.test_count
|
|
|
|
def get_error_count(self):
|
|
return self.error_count
|
|
|
|
def get_scenario_count(self):
|
|
return self.scenario_count
|
|
|
|
def get_task_id(self, test_name):
|
|
cmd = "grep \"rally task results\" {}.log | awk '{{print $4}}'".format(test_name)
|
|
return self.tools.run_cmd(cmd)
|
|
|
|
def _get_details(self):
|
|
self.logger.info("Current number of scenarios executed:{}".format(self.get_scenario_count()))
|
|
self.logger.info("Current number of test(s) executed:{}".format(self.get_test_count()))
|
|
self.logger.info("Current number of test failures:{}".format(self.get_error_count()))
|
|
|
|
def gen_scenario_html(self, task_ids, test_name):
|
|
all_task_ids = ' '.join(task_ids)
|
|
cmd = "source {}; ".format(self.config['rally']['venv'])
|
|
cmd += "rally task report --task {} --out {}.html".format(all_task_ids, test_name)
|
|
return self.tools.run_cmd(cmd)
|
|
|
|
def gen_scenario_json(self, task_id, test_name):
|
|
cmd = "source {}; ".format(self.config['rally']['venv'])
|
|
cmd += "rally task results {} > {}.json".format(task_id, test_name)
|
|
return self.tools.run_cmd(cmd)
|
|
|
|
# Iterate through all the Rally scenarios to run.
|
|
# If rerun is > 1, execute the test the desired number of times.
|
|
def start_workloads(self):
|
|
results = OrderedDict()
|
|
self.logger.info("Starting Rally workloads")
|
|
time_stamp = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
|
|
self.logger.debug("Time Stamp (Prefix): {}".format(time_stamp))
|
|
benchmarks = self.config.get('rally')['benchmarks']
|
|
if len(benchmarks) > 0:
|
|
for benchmark in benchmarks:
|
|
if benchmarks[benchmark]['enabled']:
|
|
self.logger.info("Benchmark: {}".format(benchmark))
|
|
|
|
scenarios = benchmarks[benchmark]['scenarios']
|
|
def_concurrencies = benchmarks[benchmark]['concurrency']
|
|
def_times = benchmarks[benchmark]['times']
|
|
self.logger.debug("Default Concurrencies: {}".format(def_concurrencies))
|
|
self.logger.debug("Default Times: {}".format(def_times))
|
|
for scenario in sorted(scenarios):
|
|
if scenarios[scenario]['enabled']:
|
|
self.scenario_count += 1
|
|
self.logger.info("Running Scenario: {}".format(scenario))
|
|
self.logger.debug("Scenario File: {}".format(
|
|
scenarios[scenario]['file']))
|
|
|
|
scenario_args = dict(scenarios[scenario])
|
|
del scenario_args['enabled']
|
|
del scenario_args['file']
|
|
if len(scenario_args) > 0:
|
|
self.logger.debug("Overriding Scenario Args: {}".format(
|
|
scenario_args))
|
|
|
|
result_dir = self.tools.create_results_dir(
|
|
self.config['browbeat']['results'], time_stamp, benchmark, scenario)
|
|
self.logger.debug("Created result directory: {}".format(result_dir))
|
|
self.workload_logger(result_dir)
|
|
|
|
# Override concurrency/times
|
|
if 'concurrency' in scenario_args:
|
|
concurrencies = scenario_args['concurrency']
|
|
del scenario_args['concurrency']
|
|
else:
|
|
concurrencies = def_concurrencies
|
|
if 'times' not in scenario_args:
|
|
scenario_args['times'] = def_times
|
|
|
|
for concurrency in concurrencies:
|
|
scenario_args['concurrency'] = concurrency
|
|
for run in range(self.config['browbeat']['rerun']):
|
|
if run not in results:
|
|
results[run] = []
|
|
self.test_count += 1
|
|
test_name = "{}-browbeat-{}-{}-iteration-{}".format(time_stamp,
|
|
scenario, concurrency, run)
|
|
|
|
if not result_dir:
|
|
self.logger.error("Failed to create result directory")
|
|
exit(1)
|
|
|
|
# Start connmon before rally
|
|
if self.config['browbeat']['connmon']:
|
|
self.connmon.start_connmon()
|
|
|
|
self.run_scenario(scenarios[scenario]['file'], scenario_args,
|
|
result_dir, test_name)
|
|
|
|
# Stop connmon at end of rally task
|
|
if self.config['browbeat']['connmon']:
|
|
self.connmon.stop_connmon()
|
|
try:
|
|
self.connmon.move_connmon_results(result_dir, test_name)
|
|
except:
|
|
self.logger.error("Connmon Result data missing, Connmon never started")
|
|
return False
|
|
self.connmon.connmon_graphs(result_dir, test_name)
|
|
|
|
# Find task id (if task succeeded in running)
|
|
task_id = self.get_task_id(test_name)
|
|
if task_id:
|
|
self.logger.info("Generating Rally HTML for task_id : {}".format(task_id))
|
|
self.gen_scenario_html([task_id], test_name)
|
|
self.gen_scenario_json(task_id, test_name)
|
|
results[run].append(task_id)
|
|
else:
|
|
self.logger.error("Cannot find task_id")
|
|
self.error_count += 1
|
|
|
|
for data in glob.glob("./{}*".format(test_name)):
|
|
shutil.move(data, result_dir)
|
|
|
|
self._get_details()
|
|
|
|
else:
|
|
self.logger.info("Skipping {} scenario enabled: false".format(scenario))
|
|
else:
|
|
self.logger.info("Skipping {} benchmarks enabled: false".format(benchmark))
|
|
self.logger.debug("Creating Combined Rally Reports")
|
|
for run in results:
|
|
combined_html_name = 'all-rally-run-{}'.format(run)
|
|
self.gen_scenario_html(results[run], combined_html_name)
|
|
shutil.move('{}.html'.format(combined_html_name),
|
|
'{}/{}'.format(self.config['browbeat']['results'], time_stamp))
|
|
|
|
else:
|
|
self.logger.error("Config file contains no rally benchmarks.")
|