def do_monitor(self, data_list):
for data in data_list:
if 'gpu' in data['tags'].keys():
device_name = data['measurement']
amd_gpu_temperature = data['fields']['temperature']
if amd_gpu_temperature > HEAT_LIMIT:
LOG.error(
'Current temperature of AMD GPU-%s is %d > %d! Killing all miners...',
device_name, amd_gpu_temperature, HEAT_LIMIT)
self.switch_off_miner_overheat(SLEEP_TIMEOUT_MINS)
def do_monitor(self, data_list):
for data in data_list:
device_name = data['measurement']
temperature = int(data['fields']['temperature'])
power_usage = int(data['fields']['power_usage'])
if temperature > HEAT_LIMIT:
LOG.error(
'Current temperature of Nvidia GPU-%s is %d > %d! Killing all miners...',
device_name, temperature, HEAT_LIMIT)
self.switch_off_miner_overheat(HEATAGE_SLEEP_TIMEOUT_MINS)
if power_usage < POWER_LIMIT:
LOG.error(
'Current power usage from Nvidia GPU-%s is %d < %d! Killing all miners...',
device_name, power_usage, POWER_LIMIT)
self.switch_off_miner_underpowered(WATTAGE_SLEEP_TIMEOUT_MINS)
def _receive_frame_info(self, fd):
client_socket = self.__fd_to_socket[fd]
header = client_socket.recv(4, socket.MSG_WAITALL)
LOG.info("the length of header is {}, header:{}", len(header), header)
if len(header) != 4:
LOG.error('header length is not 4')
return None
data_length = struct.unpack('i', header)[0]
data_bin = client_socket.recv(data_length, socket.MSG_WAITALL)
if len(data_bin) != data_length:
LOG.error('the length of dataBin is not equal to data length')
return None
# 前面16个字节是魔数、补偿信息、是否终止、帧序号，后面是帧数据
magic_number, reward, terminal, frame_index = struct.unpack(
'ifii', data_bin[0:16])
if magic_number != MAGIC_NUMBER:
LOG.error('magic number error')
return None
np_array = np.fromstring(data_bin[16:], np.uint8)
image = np.reshape(np_array, (IMAGE_WIDTH, IMAGE_HEIGHT))
done = bool(terminal == 1)
LOG.debug(
'receive frame information from fd {}: frame index = {}, reward = {}'
.format(client_socket.fileno(), frame_index, reward))
return image, reward, done, frame_index
def _load_server_config(self):
current_path = os.path.abspath(
os.path.dirname(os.path.dirname(__file__)))
config_path = os.path.join(current_path, SERVER_CONFIG_FILE)
LOG.info("the config path of server is {}, current_path:{}".format(
config_path, current_path))
if os.path.exists(config_path):
server_config = configparser.ConfigParser()
server_config.read(config_path)
else:
LOG.error('Config File not exist in {0}'.format(config_path))
return False
# 从配置文件中读取IP和端口
self.server_ip = server_config.get('SERVER', 'IP', fallback='0.0.0.0')
self.server_port = server_config.getint('SERVER',
'Port',
fallback=8888)
LOG.info("the server info as list, ip:{}, port:{}".format(
self.server_ip, self.server_port))
return True
