락 성능 분석 결과 리포트

라이브러리 import, 데이터 형태 확인용 함수 작성

import pandas as pd
import os
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

def validateDir(path: str):
    assert os.path.exists(path) and os.path.isdir(path), '경로에 디렉토리가 존재하지 않습니다.'

def convert_to_numeric(df):
    numeric_feature = df.columns[(df.dtypes==int) | (df.dtypes== float)]
    categorical_feature = df.columns[df.dtypes=='O']
    
    print("수치형 컬럼 : ", list(numeric_feature))
    print("카테고리형 컬럼 : ", list(categorical_feature))
    
    for col in categorical_feature:
        if isinstance(df[col][0], int) or (isinstance(df[col][0],str) and df[col][0].isnumeric()):
            df[col] = pd.to_numeric(df[col])
    
    print("숫자값을 수치형 컬럼으로 변환합니다.")
    print(df.dtypes)


def check_missing_col(dataframe):
    missing_col = []
    counted_missing_col = 0
    for i, col in enumerate(dataframe.columns):
        missing_values = sum(dataframe[col].isna())
        is_missing = True if missing_values >= 1 else False
        if is_missing:
            counted_missing_col += 1
            print(f'결측치가 있는 컬럼: {col}')
            print(f'해당 컬럼의 결측치 수: {missing_values}')
            missing_col.append([col, dataframe[col].dtype])
    if counted_missing_col == 0:
        print('결측치가 존재하지 않습니다')
    return missing_col
    

특정 그룹별 duration 통계 (vus, conflict_rate로 필터링) 함수

def show_mean_duration_for(df, min_vus = 0, max_vus = 10000000, min_conflict_rate=0, max_conflict_rate=10000000, groupby_columns = []):
    print(f'\n(min_vus : {min_vus}, max_vus : {max_vus}, min_conflict_rate : {min_conflict_rate:.2f}, max_conflict_rate : {max_conflict_rate:.2f}) 조건에서의 통계 분석')
    df = df.loc[(df.vus >= min_vus)&(df.vus <= max_vus)&(df.conflict_rate >= min_conflict_rate)&(df.conflict_rate <= max_conflict_rate),:]
    
    print('\nduration 순으로 상위 5개')
    with pd.option_context('display.max_rows', None, 'display.max_columns', None):
        print(df.sort_values(['duration']).head(5))
    
    if groupby_columns:
        print('\n',groupby_columns, '별 duration 평균값')
        print(df.loc[:, ['duration']+groupby_columns].groupby(groupby_columns).mean())
        print('\n',groupby_columns, '별 duration 최소값')
        print(df.loc[:, ['duration']+groupby_columns].groupby(groupby_columns).min())
    else:
        print('\n평균 duration')
        print(df.mean())
        print('\n최소 duration')
        print(df.min())

변수 간 상관관계 그래프 함수

def corr_between_feature_target(df, feature, target):
    plt.figure(figsize=(15,10))

    heat_table = df.corr(method="kendall")
    mask = np.zeros_like(heat_table)
    mask[np.triu_indices_from(mask)] = True
    heatmap_ax = sns.heatmap(heat_table, annot=True, mask=mask, cmap='coolwarm')
    heatmap_ax.set_xticklabels(heatmap_ax.get_xticklabels(), fontsize=15, rotation=45)
    heatmap_ax.set_yticklabels(heatmap_ax.get_yticklabels(), fontsize=15)
    plt.title('correlation between features', fontsize=20)
    plt.show()
    
    '''
    numeric_feature = ['conflict_rate']

    plt.figure(figsize=(20,15))
    plt.suptitle("Pearson Scatter Plot", fontsize=15)
    
    for i in range(len(numeric_feature)):
        plt.subplot(2,3,i+1)
        plt.xlabel(numeric_feature[i])
        plt.ylabel(target)
        corr_score = df[[numeric_feature[i], target]].corr().iloc[0,1].round(2)
        c = 'red' if corr_score > 0 else 'blue'
        plt.scatter(df[numeric_feature[i]], df[target], color=c, label=f"corr = {corr_score}")
        plt.legend(fontsize=15)
    plt.tight_layout(rect=[0, 0.03, 1, 0.95])
    plt.show()
    
    
    categorical_feature = [col for col in feature if col not in ['conflict_rate']]
    print(categorical_feature)

    plt.figure(figsize=(20,15))
    plt.suptitle("Violin Plot", fontsize=15)
    
    for i in range(len(categorical_feature)):
        print(i+1, categorical_feature[i])
        plt.subplot(3,2,i+1)
        plt.xlabel(categorical_feature[i])
        plt.ylabel(target)
        sns.violinplot(x= df[categorical_feature[i]], y= df[target])
    plt.tight_layout(rect=[0, 0.03, 1, 0.95])
    plt.show()
    '''

1. 성능 분석 시작

1.1 성능 데이터 로드

path = os.getcwd()+'./result'
validateDir(path)

df = pd.DataFrame(columns=['lock', 'vus', 'tickets', 'minBackoff', 'maxBackoff', 'retry', 'waitTime', 'leaseTime', 'duration', 'passes', 'fails'])
subdir_names=os.listdir(path)
for file_name in subdir_names:
    df1 = pd.read_csv(path+"\\"+file_name, index_col = 0)
    df= pd.concat([df, df1])

convert_to_numeric(df)

df['pass_rate'] = df['passes'] /np.minimum(df['vus'], df['tickets'])
df['conflict_rate'] = df['vus'] / df['tickets']
df = df.round(2)

수치형 컬럼 :  ['duration', 'passes', 'fails']
카테고리형 컬럼 :  ['lock', 'vus', 'tickets', 'minBackoff', 'maxBackoff', 'retry', 'waitTime', 'leaseTime']
숫자값을 수치형 컬럼으로 변환합니다.
lock           object
vus             int64
tickets         int64
minBackoff      int64
maxBackoff      int64
retry           int64
waitTime        int64
leaseTime       int64
duration      float64
passes        float64
fails         float64
dtype: object

1.2 pass_rate가 1 미만인 항목 출력

print(df.loc[df.pass_rate < 1, ['lock','vus','tickets','passes', 'pass_rate']])
df = df.drop(['passes', 'fails'], axis=1)

      lock  vus  tickets  passes  pass_rate
27  d-lock   30     1000   29.67       0.99
64  d-lock   50      100   49.00       0.98
65  d-lock   50      100   48.33       0.97

1.3 최소 pass_rate 설정

pass_rate = 0.98

1.4 olock, plock, dlock 별 데이터 분리

common_columns = ['vus', 'tickets', 'duration', 'conflict_rate']
olock_columns = ['minBackoff', 'maxBackoff','retry']
plock_columns = []
dlock_columns =['waitTime', 'leaseTime']
o_df = df.loc[(df.lock == 'o-lock') & (df['pass_rate'] >= pass_rate), common_columns + olock_columns]
p_df = df.loc[df.lock == 'p-lock', common_columns + plock_columns]
d_df = df.loc[df.lock == 'd-lock', common_columns + dlock_columns]

2. 낙관적 락

2.1 낙관적 락 데이터 확인

print(o_df)

     vus  tickets  duration  conflict_rate  minBackoff  maxBackoff  retry
81    10     1000    145.33           0.01         200         300     30
82    10     1000    142.30           0.01         200         300     60
83    10     1000    143.21           0.01         200         300     90
84    10     1000    153.64           0.01         200         400     30
85    10     1000    145.37           0.01         200         400     60
..   ...      ...       ...            ...         ...         ...    ...
157   50       10    193.27           5.00         200         400     60
158   50       10    157.20           5.00         200         400     90
159   50       10    152.01           5.00         200         500     30
160   50       10    159.37           5.00         200         500     60
161   50       10    201.63           5.00         200         500     90

[81 rows x 7 columns]

2.2 minBackoff, maxBackoff, retry 별 duration 확인

# print(o_df.groupby(['maxBackoff', 'retry']).mean())
show_mean_duration_for(o_df,max_conflict_rate=1, groupby_columns=olock_columns)
# show_mean_duration_for(o_df, max_vus=50, groupby_columns=olock_columns)

feature = [col for col in o_df.columns if col not in ['duration']]
target = 'duration'
corr_between_feature_target(o_df, feature, target)

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 0.00, max_conflict_rate : 1.00) 조건에서의 통계 분석

duration 순으로 상위 5개
    vus  tickets  duration  conflict_rate  minBackoff  maxBackoff  retry
96   10      100     40.12            0.1         200         500     30
95   10      100     42.88            0.1         200         400     90
98   10      100     48.18            0.1         200         500     90
90   10      100     50.02            0.1         200         300     30
97   10      100     50.19            0.1         200         500     60

 ['minBackoff', 'maxBackoff', 'retry'] 별 duration 평균값
                               duration
minBackoff maxBackoff retry            
200        300        30     102.675714
                      60     103.818571
                      90     103.302857
           400        30     105.884286
                      60     109.222857
                      90     111.395714
           500        30     104.068571
                      60     106.268571
                      90     112.061429

 ['minBackoff', 'maxBackoff', 'retry'] 별 duration 최소값
                             duration
minBackoff maxBackoff retry          
200        300        30        50.02
                      60        68.41
                      90        55.85
           400        30        59.81
                      60        57.01
                      90        42.88
           500        30        40.12
                      60        50.19
                      90        48.18

3. 비관적 락

3.1 비관적 락 데이터 확인

print(p_df)

     vus  tickets  duration  conflict_rate
162   10     1000    234.65           0.01
163   10      100     90.02           0.10
164   10       10     72.14           1.00
165   30     1000    218.05           0.03
166   30      100    183.79           0.30
167   30       10    127.14           3.00
168   50     1000    313.14           0.05
169   50      100    289.23           0.50
170   50       10    156.00           5.00

3.2 비관적 락 상위 duration 5개 확인

# print(p_df.mean())
show_mean_duration_for(p_df,min_conflict_rate=1, groupby_columns=plock_columns)
# show_mean_duration_for(p_df, max_vus=50, groupby_columns=plock_columns)
feature = [col for col in p_df.columns if col not in ['duration']]
target = 'duration'
corr_between_feature_target(p_df, feature, target)

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 1.00, max_conflict_rate : 10000000.00) 조건에서의 통계 분석

duration 순으로 상위 5개
     vus  tickets  duration  conflict_rate
164   10       10     72.14            1.0
167   30       10    127.14            3.0
170   50       10    156.00            5.0

평균 duration
vus               30.000000
tickets           10.000000
duration         118.426667
conflict_rate      3.000000
dtype: float64

최소 duration
vus              10.00
tickets          10.00
duration         72.14
conflict_rate     1.00
dtype: float64

4. 분산 락

4.1 분산 락 데이터 확인

print(d_df)

    vus  tickets  duration  conflict_rate  waitTime  leaseTime
0    10     1000    276.67           0.01       300       1000
1    10     1000    274.56           0.01       300       1500
2    10     1000    274.80           0.01       300        500
3    10     1000    271.58           0.01       500       1000
4    10     1000    274.55           0.01       500       1500
..  ...      ...       ...            ...       ...        ...
76   50       10    398.90           5.00       500       1500
77   50       10    394.82           5.00       500        500
78   50       10    383.79           5.00       700       1000
79   50       10    390.45           5.00       700       1500
80   50       10    400.02           5.00       700        500

[81 rows x 6 columns]

4.2 waitTime, leaseTime별 duration 확인

show_mean_duration_for(d_df,min_conflict_rate=1, groupby_columns=dlock_columns)
# show_mean_duration_for(d_df, max_vus=50, groupby_columns=dlock_columns)
feature = [col for col in d_df.columns if col not in ['duration']]
target = 'duration'
corr_between_feature_target(d_df, feature, target)

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 1.00, max_conflict_rate : 10000000.00) 조건에서의 통계 분석

duration 순으로 상위 5개
    vus  tickets  duration  conflict_rate  waitTime  leaseTime
23   10       10    117.91            1.0       500        500
26   10       10    118.56            1.0       700        500
24   10       10    119.14            1.0       700       1000
22   10       10    120.57            1.0       500       1500
25   10       10    121.89            1.0       700       1500

 ['waitTime', 'leaseTime'] 별 duration 평균값
                      duration
waitTime leaseTime            
300      500        281.843333
         1000       272.203333
         1500       271.050000
500      500        263.770000
         1000       269.293333
         1500       267.786667
700      500        271.746667
         1000       268.016667
         1500       265.320000

 ['waitTime', 'leaseTime'] 별 duration 최소값
                    duration
waitTime leaseTime          
300      500          159.45
         1000         122.19
         1500         122.71
500      500          117.91
         1000         130.02
         1500         120.57
700      500          118.56
         1000         119.14
         1500         121.89

5. 전체 데이터 통계

5.1 상위 5개 & vus, tickets 별 duration 확인

show_mean_duration_for(df.loc[df.pass_rate >= pass_rate, :], groupby_columns=['lock', 'vus', 'tickets'])

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 0.00, max_conflict_rate : 10000000.00) 조건에서의 통계 분석

duration 순으로 상위 5개
      lock  vus  tickets  minBackoff  maxBackoff  retry  waitTime  leaseTime  \
96  o-lock   10      100         200         500     30         0          0   
95  o-lock   10      100         200         400     90         0          0   
98  o-lock   10      100         200         500     90         0          0   
90  o-lock   10      100         200         300     30         0          0   
97  o-lock   10      100         200         500     60         0          0   

    duration  pass_rate  conflict_rate  
96     40.12        1.0            0.1  
95     42.88        1.0            0.1  
98     48.18        1.0            0.1  
90     50.02        1.0            0.1  
97     50.19        1.0            0.1  

 ['lock', 'vus', 'tickets'] 별 duration 평균값
                      duration
lock   vus tickets            
d-lock 10  10       125.826667
           100      134.270000
           1000     271.670000
       30  10       292.123333
           100      338.795556
           1000     358.750000
       50  10       392.393333
           100      511.880000
           1000     540.513333
o-lock 10  10       135.137778
           100       53.398889
           1000     147.421111
       30  10       143.456667
           100       83.848889
           1000     102.275556
       50  10       170.506667
           100      104.290000
           1000     119.282222
p-lock 10  10        72.140000
           100       90.020000
           1000     234.650000
       30  10       127.140000
           100      183.790000
           1000     218.050000
       50  10       156.000000
           100      289.230000
           1000     313.140000

 ['lock', 'vus', 'tickets'] 별 duration 최소값
                    duration
lock   vus tickets          
d-lock 10  10         117.91
           100        130.12
           1000       262.67
       30  10         278.58
           100        313.83
           1000       351.98
       50  10         382.09
           100        487.06
           1000       512.25
o-lock 10  10         107.40
           100         40.12
           1000       142.30
       30  10         127.36
           100         68.41
           1000        98.59
       50  10         152.01
           100         93.84
           1000       107.87
p-lock 10  10          72.14
           100         90.02
           1000       234.65
       30  10         127.14
           100        183.79
           1000       218.05
       50  10         156.00
           100        289.23
           1000       313.14

5.2 conflict_rate가 0.5 이하인 데이터에 대해 5.1

show_mean_duration_for(df.loc[df.pass_rate >= pass_rate, :], max_conflict_rate=0.5, groupby_columns=['lock', 'vus', 'tickets'])

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 0.00, max_conflict_rate : 0.50) 조건에서의 통계 분석

duration 순으로 상위 5개
      lock  vus  tickets  minBackoff  maxBackoff  retry  waitTime  leaseTime  \
96  o-lock   10      100         200         500     30         0          0   
95  o-lock   10      100         200         400     90         0          0   
98  o-lock   10      100         200         500     90         0          0   
90  o-lock   10      100         200         300     30         0          0   
97  o-lock   10      100         200         500     60         0          0   

    duration  pass_rate  conflict_rate  
96     40.12        1.0            0.1  
95     42.88        1.0            0.1  
98     48.18        1.0            0.1  
90     50.02        1.0            0.1  
97     50.19        1.0            0.1  

 ['lock', 'vus', 'tickets'] 별 duration 평균값
                      duration
lock   vus tickets            
d-lock 10  100      134.270000
           1000     271.670000
       30  100      338.795556
           1000     358.750000
       50  100      511.880000
           1000     540.513333
o-lock 10  100       53.398889
           1000     147.421111
       30  100       83.848889
           1000     102.275556
       50  100      104.290000
           1000     119.282222
p-lock 10  100       90.020000
           1000     234.650000
       30  100      183.790000
           1000     218.050000
       50  100      289.230000
           1000     313.140000

 ['lock', 'vus', 'tickets'] 별 duration 최소값
                    duration
lock   vus tickets          
d-lock 10  100        130.12
           1000       262.67
       30  100        313.83
           1000       351.98
       50  100        487.06
           1000       512.25
o-lock 10  100         40.12
           1000       142.30
       30  100         68.41
           1000        98.59
       50  100         93.84
           1000       107.87
p-lock 10  100         90.02
           1000       234.65
       30  100        183.79
           1000       218.05
       50  100        289.23
           1000       313.14

5.3 conflict_rate가 1 이상인 데이터에 대해 5.1

show_mean_duration_for(df.loc[df.pass_rate >= pass_rate, :], min_conflict_rate=1, groupby_columns=['lock', 'vus', 'tickets'])    

(min_vus : 0, max_vus : 10000000, min_conflict_rate : 1.00, max_conflict_rate : 10000000.00) 조건에서의 통계 분석

duration 순으로 상위 5개
       lock  vus  tickets  minBackoff  maxBackoff  retry  waitTime  leaseTime  \
164  p-lock   10       10           0           0      0         0          0   
99   o-lock   10       10         200         300     30         0          0   
23   d-lock   10       10           0           0      0       500        500   
101  o-lock   10       10         200         300     90         0          0   
26   d-lock   10       10           0           0      0       700        500   

     duration  pass_rate  conflict_rate  
164     72.14        1.0            1.0  
99     107.40        1.0            1.0  
23     117.91        1.0            1.0  
101    118.50        1.0            1.0  
26     118.56        1.0            1.0  

 ['lock', 'vus', 'tickets'] 별 duration 평균값
                      duration
lock   vus tickets            
d-lock 10  10       125.826667
       30  10       292.123333
       50  10       392.393333
o-lock 10  10       135.137778
       30  10       143.456667
       50  10       170.506667
p-lock 10  10        72.140000
       30  10       127.140000
       50  10       156.000000

 ['lock', 'vus', 'tickets'] 별 duration 최소값
                    duration
lock   vus tickets          
d-lock 10  10         117.91
       30  10         278.58
       50  10         382.09
o-lock 10  10         107.40
       30  10         127.36
       50  10         152.01
p-lock 10  10          72.14
       30  10         127.14
       50  10         156.00