def step(self, knob):
"""step
"""
restart_time = utils.time_start()
flag = self._apply_knobs(knob)
restart_time = utils.time_end(restart_time)
if not flag:
return -100.0, np.array([0] *
self.num_metric), True, self.score - 100, [
0, 0, 0
], restart_time
s = self._get_state(knob, method=self.method)
if s is None:
return -100.0, np.array([0] *
self.num_metric), True, self.score - 100, [
0, 0, 0
], restart_time
external_metrics, internal_metrics = s
reward = self._get_reward(external_metrics)
self.last_external_metrics = external_metrics
next_state = internal_metrics
terminate = self._terminate()
knobs.save_knobs(
knob=knob,
metrics=external_metrics,
knob_file='%sAutoTuner/tuner/save_knobs/knob_metric.txt' %
PROJECT_DIR)
return reward, next_state, terminate, self.score, external_metrics, restart_time
def step(self, knob):
"""step
"""
restart_time = utils.time_start()
flag = self._apply_knobs(knob)
restart_time = utils.time_end(restart_time)
# ?
if not flag:
return -10000000.0, np.array([0] * self.num_metric), True, self.score - 10000000, [0, 0, 0], restart_time
s = self._get_state(knob, method=self.method)
# crash
if s is None:
return -10000000.0, np.array([0] * self.num_metric), True, self.score - 10000000, [0, 0, 0], restart_time
external_metrics, internal_metrics = s
reward = self._get_reward(external_metrics)
flag = self.record_best(external_metrics)
if flag == True:
logger.info('Better performance changed!')
else:
logger.info('Performance remained!')
#get the best performance so far to calculate the reward
best_now_performance = self._get_best_now(self.best_result)
self.last_external_metrics = best_now_performance
next_state = internal_metrics
terminate = self._terminate()
knobs.save_knobs(
knob = knob,
metrics = external_metrics,
instance=self.db_info,
task_id=self.task_id
)
return reward, next_state, terminate, self.score, external_metrics, restart_time
def step(self, knob):
"""step
"""
filename = 'bestnow.log'
restart_time = utils.time_start()
# time delay
restart_time = utils.time_end(restart_time)
# s conclude external_metrics and internal_metrics
s = self._get_state(knob, method=self.method)
if s is None:
return -10000000.0, np.array(
[0] * self.num_metric), True, self.score - 10000000, [
0, 0, 0
], restart_time
external_metrics, internal_metrics = s
# 在这里改变了score的记录
reward = self._get_reward(external_metrics)
# 记录最佳的reward
flag = self.record_best(external_metrics)
if flag == True:
print('Better performance changed!')
else:
print('Performance remained!')
# get the best performance so far to calculate the reward
best_now_performance = self._get_best_now(filename)
self.last_external_metrics = best_now_performance
next_state = internal_metrics
terminate = self._terminate()
knobs.save_knobs(
knob=knob,
metrics=external_metrics,
knob_file='%sAutoTuner1/tuner/save_knobs/knob_metric.txt' %
PROJECT_DIR)
return reward, next_state, terminate, self.score, external_metrics, restart_time
# time for every step
step_times = []
# time for training
train_step_times = []
# time for setup, restart, test
env_step_times = []
# restart time
env_restart_times = []
# choose_action_time
action_step_times = []
print("[Environment Intialize]Tps: {} Lat:{}".format(default_metrics[0], default_metrics[1]))
print("------------------- Starting to Test -----------------------")
while step_counter < opt.max_steps:
step_time = utils.time_start()
state = current_state
action_step_time = utils.time_start()
action = model.choose_action(state)
action_step_time = utils.time_end(action_step_time)
current_knob = generate_knob(action, 'ddpg')
logger.info("[ddpg] Action: {}".format(action))
env_step_time = utils.time_start()
reward, state_, done, score, metrics, restart_time = env.step(current_knob)
env_step_time = utils.time_end(env_step_time)
logger.info("[{}][Step: {}][Metric tps:{} lat:{}, qps: {}]Reward: {} Score: {} Done: {}".format(
