1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
import ast
import os
import re
import sys
import time
import traceback
import pandas as pd
from pyspark.sql.types import ArrayType, FloatType
os.environ["PYARROW_IGNORE_TIMEZONE"] = "1"
sys.path.append(os.path.dirname(sys.path[0])) # 上级目录
from utils.templates_mysql import TemplatesMysql
from utils.templates import Templates
# from ..utils.templates import Templates
from py4j.java_gateway import java_import
from sqlalchemy import text
from pyspark.sql import functions as F
import pyarrow as pa
import pyarrow.parquet as pq
from multiprocessing import Process
from multiprocessing import Pool
import multiprocessing
class PicturesIdIndex(Templates):
def __init__(self, site_name='us'):
super(PicturesIdIndex, self).__init__()
self.site_name = site_name
self.engine_pg = TemplatesMysql().engine_pg
self.db_save = f'pictures_dwd_id_index'
self.spark = self.create_spark_object(app_name=f"{self.db_save}: {self.site_name}")
self.df_features = self.spark.sql(f"select 1+1;")
self.df_save = self.spark.sql(f"select 1+1;")
self.df_save_local = self.spark.sql(f"select 1+1;")
self.partitions_by = ['site_name']
self.partitions_num = 1
# mysql表获取的相关变量
self.tn_pics_hdfs_index = f"{self.site_name}_pictures_hdfs_index"
self.id = int()
self.current_counts = int()
self.all_counts = int()
self.hdfs_path = str()
self.hdfs_block_name = str()
self.local_path = str()
def read_data(self):
# 读取mysql的hdfs_path路径信息
with self.engine_pg.begin() as conn:
sql_read = text(f"SELECT * FROM {self.tn_pics_hdfs_index} WHERE state=1 LIMIT 1 FOR UPDATE;")
print("sql_read:", sql_read)
result = conn.execute(sql_read)
df = pd.DataFrame(result.fetchall())
if df.shape[0]:
df.columns = result.keys()
self.id = list(df.id)[0] if list(df.id) else None
self.current_counts = list(df.current_counts)[0] if list(df.current_counts) else None
self.all_counts = list(df.all_counts)[0] if list(df.all_counts) else None
self.hdfs_path = list(df.hdfs_path)[0] if list(df.hdfs_path) else None
self.hdfs_block_name = re.findall("(part-\d+)-", self.hdfs_path)[0] if self.hdfs_path else None
self.local_path = rf"/mnt/image_search/pictures_id_index/{self.site_name}/{self.hdfs_block_name}"
print(f"df.shape:{df.shape}, self.id:{self.id}, self.current_counts:{self.current_counts}, self.all_counts:{self.all_counts}, self.hdfs_path:{self.hdfs_path}")
if self.id:
os.system(f"hdfs dfs -rm -r /home/big_data_selection/dwd/pictures_dwd_id_index/site_name={self.site_name}/block={self.hdfs_block_name}")
sql_update = text(
f"UPDATE {self.tn_pics_hdfs_index} SET state=2 WHERE id={self.id};")
print("sql_update:", sql_update)
conn.execute(sql_update)
else:
quit()
# 读取hdfs路径下的parquet文件
self.df_features = self.spark.read.text(self.hdfs_path).cache()
def save_data_local(self):
print("当前存储到本地:", self.local_path)
if os.path.exists(self.local_path):
os.system(f"rm -rf {self.local_path}")
os.makedirs(self.local_path)
# Convert DataFrame to Arrow Table
df = self.df_save_local.toPandas()
table = pa.Table.from_pandas(df)
# Save to Parquet
pq.write_table(table, f"{self.local_path}/{self.hdfs_block_name}.parquet")
def handle_data(self):
# 创建一个新的 DataFrame，其中每个字段都是一个独立的列
split_df = self.df_features.select(F.split(self.df_features['value'], '\t').alias('split_values'))
# 假设你知道你的数据有三个字段
# 你可以这样创建每个字段的独立列
final_df = split_df.select(
split_df['split_values'].getItem(0).alias('id'),
split_df['split_values'].getItem(1).alias('asin'),
split_df['split_values'].getItem(2).alias('embedding')
)
print("分块前分区数量:", final_df.rdd.getNumPartitions())
# 从hdfs读取parquet文件，进行split切分的时候是字符串类型-->转换成数值类型
final_df = final_df.withColumn('id', final_df['id'].cast('bigint')) # 然后你可以安全地转换
# 添加索引列
final_df = final_df.withColumn("index", F.monotonically_increasing_id() + self.all_counts)
final_df.show()
# 此处更改数据类型在pictures_dim_features_slice已经做了 -- 由于读取的是lzo文件，而不是直接读表，因此还需要转换类型
# 定义一个将字符串转换为列表的UDF
str_to_list_udf = F.udf(lambda s: ast.literal_eval(s), ArrayType(FloatType()))
# 对DataFrame中的列应用这个UDF
final_df = final_df.withColumn("embedding", str_to_list_udf(final_df["embedding"]))
# final_df.write.mode('overwrite').parquet("hdfs://hadoop5:8020/home/ffman/parquet")
# final_df = final_df.withColumn("block", F.lit(self.hdfs_block_name))
self.df_save = final_df.withColumn("site_name", F.lit(self.site_name))
self.df_save_local = self.df_save.select("embedding")
def update_state_after_save(self):
with self.engine_pg.begin() as conn:
sql_update = text(
f"UPDATE {self.tn_pics_hdfs_index} SET state=3 WHERE state=2 and id={self.id};")
print("sql_update:", sql_update)
conn.execute(sql_update)
def run(self):
self.read_data()
self.handle_data()
self.save_data()
# 存储到本地
self.save_data_local()
self.update_state_after_save()
def main():
handle_obj = PicturesIdIndex()
handle_obj.run()
if __name__ == "__main__":
main()
quit()
processes = []
for _ in range(5): # 用于设定进程数量
handle_obj = PicturesIdIndex()
process = multiprocessing.Process(target=handle_obj.run)
process.start()
processes.append(process)
# 等待所有进程完成
for process in processes:
process.join()
# if __name__ == '__main__':
# # 设置进程数
# num_processes = 4 # 设置为你需要的进程数
# # 创建进程池对象
# pool = Pool(processes=num_processes)
# # 使用进程池中的进程执行任务
# pool.apply(main)
# # 关闭进程池
# pool.close()
# # 等待所有进程完成
# pool.join()
# if __name__ == '__main__':
# # 创建进程对象
# process = Process(target=main)
# # 启动进程
# process.start()
# # 等待进程结束
# process.join()