def main():
if '--calc' in sys.argv:
if len(sys.argv) != 4:
print "[*]Useage: python main.py group_id clusterFileName"
exit()
else:
cross_cluster(sys.argv[1], sys.argv[2])
exit()
if '--status' in sys.argv:
init_options()
show_database_status()
exit()
if len(sys.argv) < 3 or '-h' in sys.argv:
helpmsg()
cpanel.group = sys.argv[1]
init_options()
read_packets(cpanel.group)
split_flow()
split_cflow()
scale_cflow()
if '--save' in sys.argv:
save_calc_results()
if '--debug' in sys.argv:
test()
if '--outfile' in sys.argv:
output_path = './output/results.txt'
save(output_path)
def execution_path(df_train_log, df_test_log):
# 提取normal日志键执行流
train_log_key_sequence_str = " ".join([str(EventId) for EventId in df_train_log["EventId"]])
# 将日志流通过滑动窗口分为4个日志为一个日志序列
X_train, Y_train = standard_log_key(train_log_key_sequence_str)
# 对日志序列进行lstm训练和测试
lg = log_key_model()
execution_path_model_filepath = "tmpdata/ExecutePathModel/model.pkl"
if os.path.isfile(execution_path_model_filepath):
model = load(execution_path_model_filepath)
else:
model = lg.train(X_train, Y_train)
save(execution_path_model_filepath, model)
# 提取test日志键执行流
test_log_key_sequence_str = " ".join([str(EventId) for EventId in df_test_log["EventId"]])
# 将日志流通过滑动窗口分为4个日志为一个日志序列
X_test, Y_test = standard_log_key(test_log_key_sequence_str)
anomaly_sequence = lg.predcit(model, X_test, Y_test)
print('the length of anomaly_sequence {} is {}'.format(anomaly_sequence, len(anomaly_sequence)))
# anomaly_sequence为异常的序列
anormal_lineid_list = [i+3 for i in anomaly_sequence]
df_anormal = df_test_log.loc[df_test_log["LineId"].isin(anormal_lineid_list)]
df_anormal.to_csv("tmpdata/execute_path_anormal.csv", index=False)
def login():
print("登录")
if user_info['name']:
print('您已经登录')
return
count = 0
user_dic = common.reads()
while True:
name = input('请输入名字:').strip()
if name == 'q':
break
password = input('请输入密码：').strip()
if name in user_dic:
if password == user_dic[name] and user_dic['locked'] == False:
user_info['name'] = name
print('登录成功！')
user_logger.info('%s 登录成功' % name)
else:
count += 1
print('输入错误或者已被锁定！')
if count == 3:
user_dic['locked'] = True
common.save(user_dic)
print('尝试过多，锁定')
user_logger.info('%s 已被锁定' % name)
break
def register():
print('注册')
if user_info['name']:
print('您已经登陆')
return
user_dic = common.reads()
while True:
name = input('请输入名字:').strip()
if name == 'q':
break
if name in user_dic:
print('该用户名已存在！')
break
password = input('请输入密码').strip()
conf_password = input('请确认密码').strip()
if password == conf_password:
account = input('请输入余额').strip()
user_dic = {
'name': name,
'password': password,
'account': account,
'locked': False,
'shoppingcart': {}
}
common.save(user_dic)
user_logger.info('%s 注册成功' % name)
else:
print('两次密码不一致')
def model_generate(key_para_dict):
# 对每个日志键进行训练，模型放在model_dict中，eventID为键，model为值
model_dict = {}
# 依次加载每个日志键的日志参数组成的矩阵，依次检测
for eventID, params in key_para_dict.items():
model_file = f"tmpdata/ParamModel/{eventID}.pkl"
if os.path.exists(model_file):
model_dict[eventID] = load(model_file)
continue
# 如果这个日志键对应的日志条目少于8个，则跳过对该日志键的模型生成
if len(params) <= 8:
continue
else:
# 设置滑动窗口大小为3
n_steps = 3
X, Y = training_data_generate(params, n_steps)
model = LSTM_model(X, Y)
model_dict[eventID] = model
save(model_file, model)
# yhat = model.predict(test_x)
# print("the predicted y shapeis:", yhat.shape) # (4, 2)
# print("the test y shape is:", test_y.shape) # (4, 2)
# # 测量实际值和预测值的均方误差
# mses = mean_squared_error_modified(test_y, yhat)
# print(f"mses: {mses}")
# sys.exit()
return model_dict
def spell_log(df_log, df_type="train"):
spell_result_path = "tmpdata/SpellResult/spell.pkl"
if os.path.isfile(spell_result_path):
slm = load(spell_result_path)
# 加载保存好的结果
else:
# 首先需要训练一遍，找出所有的日志健，保存在spell_result_path中
# 要选取可以涵盖所有日志类型的数据用来训练
slm = spell.lcsmap('[\\s]+')
for i in range(len(df_log)):
log_message = df_log["Content"][i]
# print(log_message)
sub = log_message.strip('\n')
slm.insert(sub)
# 将spell的训练结果保存在这里
save(spell_result_path, slm)
# 对每条日志进行训练一遍，然后保存在spell_result.txt中
templates = [0] * df_log.shape[0]
ids = [0] * df_log.shape[0]
ParameterList = [0] * df_log.shape[0]
time_interval = [0] * df_log.shape[0]
for i in range(len(df_log)):
log_message = df_log["Content"][i].strip()
obj = slm.insert(log_message)
# seq = re.split('[\\s]+', log_message)
# ParameterList[i] = obj.param(seq) # 取出log中的参数
# if param != []:
# param = reduce(operator.add, param) # 多维数组变一维数组
obj_json = obj.tojson(log_message)
templates[i] = obj_json["lcsseq"] # 获取该日志条目的日志键
ids[i] = obj_json["lcsseq_id"] # 获取日志键id 也就是事件编号
ParameterList[i] = obj_json["param"] # 取出log中的参数
# 生成两个日志时间差，加入param参数中
# print(df_log.shape)
# print(len(df_log))
for id in range(len(df_log)):
if id == 0:
time_interval[id] = "0"
else:
time_last = df_log["Time"][id - 1]
time_now = df_log["Time"][id]
elapsed = time_elapsed(time_last, time_now)
time_interval[id] = elapsed
ParameterList[id].append(time_interval[id])
# 将结果保存在df_log里面
df_log['EventId'] = ids # 事件向量
df_log['EventTemplate'] = templates # 日志模板 日志键
df_log["ParameterList"] = ParameterList
df_log.to_csv(f"tmpdata/struct/{df_type}_structured.csv", index=False)
return df_log
def shopping():
print('购物')
prod_list = [['coffee', 10], ['chicken', 20], ['iphone', 8000],
['macPro', 15000], ['car', 100000]]
shoppingcart = {}
cost = 0
user_dic = common.reads()
account = user_dic['account']
while True:
for i, prod in enumerate(prod_list):
print('%s : %s ' % (i, prod))
choice = input('请输入要购买的商品：').strip()
if choice.isdigit():
choice = int(choice)
if choice >= len(prod_list):
continue
prod_name = prod_list[choice][0]
prod_price = prod_list[choice][1]
if account >= prod_price:
if prod_name in shoppingcart:
shoppingcart[prod_name]['count'] += 1
else:
shoppingcart[prod_name] = {'price': prod_price, 'count': 1}
account -= prod_price
cost += prod_price
else:
print('余额不足')
elif choice == 'q':
if cost == 0:
break
print(shoppingcart)
buy = input('确认购买吗？（y/n）').strip()
if buy == 'y':
user_dic['account'] = account,
user_dic['shoppingcart'] = shoppingcart
common.save(user_dic)
else:
print('您没买东西')
break
else:
print('输入非法')