import json
import os
import re
import sys
import time
import traceback
import zlib
import pandas as pd
import redis
from datetime import datetime
sys.path.append("/opt/module/spark-3.2.0-bin-hadoop3.2/demo/py_demo/")
sys.path.append(os.path.dirname(sys.path[0])) # 上级目录
from utils.templates import Templates
# from ..utils.templates import Templates
from utils.templates_mysql import TemplatesMysql
# from ..utils.templates_mysql import TemplatesMysql
from pyspark.sql.types import IntegerType
from pyspark.sql import functions as F
from pyspark.sql.types import *
from yswg_utils.common_udf import udf_rank_and_category
# from ..yswg_utils.common_udf import udf_rank_and_category
from yswg_utils.common_df import get_node_first_id_df
from kafka import KafkaConsumer, TopicPartition
from yswg_utils.common_udf import parse_weight_str
# from ..yswg_utils.common_udf import parse_weight_str
class DimStAsinInfo(Templates):
def __init__(self, site_name='us', date_type="day", date_info='2022-10-01', consumer_type='lastest', topic_name="us_asin_detail", batch_size=100000):
super().__init__()
self.site_name = site_name
self.date_type = date_type
self.date_info = date_info
self.consumer_type = consumer_type # 消费实时还是消费历史
self.topic_name = topic_name # 主题名字
self.batch_size = batch_size
self.batch_size_history = int(batch_size / 10)
# 连接到Redis服务器
self.redis_db = {
"us": 0,
"uk": 1,
"de": 2,
"es": 3,
"fr": 4,
"it": 5,
}
self.client = redis.Redis(host='192.168.10.224', port=6379, db=self.redis_db[self.site_name], password='yswg2023')
self.db_save = f'kafka_asin_detail'
self.spark = self.create_spark_object(
app_name=f"{self.db_save}: {self.site_name},{self.date_type}, {self.date_info}")
self.get_date_info_tuple()
self.df_save = self.spark.sql(f"select 1+1;")
self.df_st_asin = self.spark.sql(f"select 1+1;")
self.df_bs_report = self.spark.sql(f"select 1+1;")
self.df_asin_bs = self.spark.sql(f"select 1+1;")
self.df_self_asin = self.spark.sql(f"select 1+1;")
self.df_asin_sku = self.spark.sql(f"select 1+1;")
self.df_asin_templates = self.spark.sql("select asin_zr_counts, asin_sp_counts, asin_sb1_counts,asin_sb2_counts,asin_sb3_counts,asin_ac_counts,asin_bs_counts,asin_er_counts,asin_tr_counts from dwd_asin_measure limit 0")
self.df_asin_counts = self.spark.sql("select asin_zr_counts, asin_sp_counts, asin_sb1_counts,asin_sb2_counts,asin_sb3_counts,asin_ac_counts,asin_bs_counts,asin_er_counts,asin_tr_counts from dwd_asin_measure limit 0")
self.schema = self.init_schema()
# self.u_rank_and_category = self.spark.udf.register("u_rank_and_category", udf_rank_and_category, schema)
schema = StructType([
StructField('asin_bs_cate_1_rank', StringType(), True),
StructField('rank_and_category', StringType(), True),
])
self.u_rank_and_category = self.spark.udf.register("u_rank_and_category", self.udf_rank_and_category, schema)
self.u_cal_crc32 = self.spark.udf.register("u_cal_crc32", self.udf_cal_crc32, IntegerType())
self.u_cal_bkdr = self.spark.udf.register("u_cal_bkdr", self.udf_cal_bkdr, IntegerType())
self.u_extract_dimensions = self.spark.udf.register("u_cal_bkdr", self.udf_extract_dimensions, StringType())
self.u_extract_weight = self.spark.udf.register("u_cal_bkdr", self.udf_extract_weight, StringType())
self.pattern_1_rank_str = {
"us": "(\d+).*?See Top 100 in ",
"uk": "(\d+).*?See Top 100 in ",
"de": "(\d+).*?Siehe Top 100 in ",
"es": "(\d+).*?Ver el Top 100 en ",
"fr": "(\d+).*?Voir les 100 premiers en ",
"it": "(\d+).*?Visualizza i Top 100 nella categoria "
} # 匹配一级分类的排名
self.pattern_str = {
"us": "(\d+ in [\w&' ]+)",
"uk": "(\d+ in [\w&' ]+)",
"de": "Nr. (\d+ in [\w&' ]+)",
"es": "nº(\d+ en [\w&' ]+)",
"fr": "(\d+ en [\w&' ]+)",
"it": "n. (\d+ in [\w&' ]+)",
} # 匹配排名和分类
self.replace_str = {
"us": "See Top 100 in ",
"uk": "See Top 100 in ",
"de": "Siehe Top 100 in ",
"es": "Ver el Top 100 en ",
"fr": "Voir les 100 premiers en ",
"it": "Visualizza i Top 100 nella categoria ",
} # 去掉top100匹配
# 连接mysql
self.engine = self.get_connection()
def get_connection(self):
return TemplatesMysql(site_name="us").mysql_connect()
def judge_spider_asin_detail_is_finished(self):
while True:
try:
sql = f'SELECT * from workflow_progress WHERE page="ASIN详情" and site_name="{self.site_name}" and date_type="{self.date_type}" and date_info="{self.date_info}" and status_val=3'
df = pd.read_sql(sql, con=self.engine)
if df.shape[0] == 1:
print(f"ASIN详情状态为3, 抓取完成并终止程序, site_name:{self.site_name}, date_type:{self.date_type}, date_info:{self.date_info}")
self.spark.stop()
quit() # 退出程序
break
except Exception as e:
print(e, traceback.format_exc())
time.sleep(10)
self.engine = self.get_connection()
def fetch_self_asin(self):
while True:
try:
sql = f"""SELECT asin, 1 as isSelfAsin from {self.site_name}_self_asin"""
df_self_asin = pd.read_sql(sql, con=self.engine)
schema = StructType([
StructField("asin", StringType(), True),
StructField("isSelfAsin", IntegerType(), True),
])
self.df_self_asin = self.spark.createDataFrame(df_self_asin, schema=schema).cache()
self.df_self_asin.show(10, truncate=False)
break
except Exception as e:
print(e, traceback.format_exc())
time.sleep(10)
self.engine = self.get_connection()
@staticmethod
def udf_extract_dimensions(volume_str, asin_volume):
# 解析类型
# pattern = r'\b\w+\b'
pattern = r'[a-z]+'
matches = re.findall(pattern, asin_volume)
# 使用集合存储匹配的单词
type_set = set()
for word in matches:
if word in ['inches', 'inch']:
type_set.add('inches')
elif word in ['cm', 'centímetros', 'centimetres']:
type_set.add('cm')
elif word in ['milímetros', 'millimeter', 'mm']:
type_set.add('mm')
elif word in ['metros']:
type_set.add('m')
# 根据集合的长度返回结果
if len(type_set) == 1:
asin_volume_type = list(type_set)[0]
elif len(type_set) >= 2:
asin_volume_type = ','.join(type_set)
else:
asin_volume_type = 'none'
# 解析长宽高
length, width, height = None, None, None
if asin_volume_type == 'cm,inches':
num_inches = volume_str.find('inch')
num_cm = volume_str.find('cm')
volume_str = volume_str[:num_inches] if num_cm > num_inches else volume_str[num_cm:num_inches]
dimensions = re.findall(r"(\d+(\.\d+)?)", volume_str)
dimensions = [float(dim[0]) for dim in dimensions]
if len(dimensions) == 1:
length = dimensions[0]
elif len(dimensions) == 2:
if asin_volume_type == 'none':
if "l" in volume_str and "w" in volume_str:
length, width = dimensions
elif "w" in volume_str and "h" in volume_str:
width, height = dimensions
elif "l" in volume_str and "h" in volume_str:
length, height = dimensions
elif "d" in volume_str and "w" in volume_str:
length, width = dimensions
elif "d" in volume_str and "h" in volume_str:
length, height = dimensions
else:
length, width = dimensions
elif len(dimensions) == 3:
length, width, height = dimensions
elif len(dimensions) >= 4:
length, width, height = dimensions[:3]
return f"{length}*{width}*{height}{asin_volume_type}"
@staticmethod
def udf_extract_weight(weight_str: str):
"""
解析重量字符串获取重量和单位，逗号分隔
:param weight_str:
:param site_name:
:return:
"""
val = None
# weight_type = 'pounds' if site_name == 'us' else 'grams'
weight_type = 'g'
if weight_str is not None:
if 'pounds' in weight_str:
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,}pounds", weight_str)
val = round(float(match.group(1)) * 1000 * 0.454, 3) if match else None
elif 'ounces' in weight_str:
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,}ounces", weight_str)
val = round(float(match.group(1)) / 16 * 1000 * 0.454, 3) if match else None
elif any(substring in weight_str for substring in ['kilogram', ' kg']):
weight_str = weight_str.replace(' kg', ' kilogram')
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,}kilogram", weight_str)
val = round(float(match.group(1)) * 1000, 3) if match else None
elif any(substring in weight_str for substring in ['milligrams']):
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,}milligrams", weight_str)
val = round(float(match.group(1)) / 1000, 3) if match else None
elif ' gram' in weight_str:
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,} gram", weight_str)
val = round(float(match.group(1)), 3) if match else None
elif ' g' in weight_str:
match = re.search(r"(\d+\.{0,}\d{0,})\D{0,} g", weight_str)
val = round(float(match.group(1)), 3) if match else None
if val:
return f"{val}{weight_type}"
else:
return f"{val}"
def fetch_asin_sku_count(self):
while True:
try:
sql = f"""SELECT asin,count(id) as auctionsNum,count((case when sku!='' then sku else NULL end)) as skusNumCreat
from product_audit_asin_sku
-- where asin in ('B085WYH539')
GROUP BY asin
"""
df_asin_sku = pd.read_sql(sql, con=self.engine)
schema = StructType([
StructField("asin", StringType(), True),
StructField("auctionsNum", IntegerType(), True),
StructField("skusNumCreat", IntegerType(), True),
])
self.df_asin_sku = self.spark.createDataFrame(df_asin_sku, schema=schema).cache()
self.df_asin_sku.show(10, truncate=False)
break
except Exception as e:
print(e, traceback.format_exc())
time.sleep(10)
self.engine = self.get_connection()
@staticmethod
def init_schema():
schema = StructType([
StructField("asin", StringType(), True),
StructField("week", StringType(), True),
StructField("title", StringType(), True),
StructField("img_url", StringType(), True),
StructField("rating", StringType(), True),
StructField("total_comments", StringType(), True),
StructField("price", FloatType(), True),
StructField("rank", StringType(), True),
StructField("category", StringType(), True),
StructField("launch_time", StringType(), True),
StructField("volume", StringType(), True),
StructField("weight", StringType(), True),
StructField("page_inventory", IntegerType(), True),
StructField("buy_box_seller_type", IntegerType(), True),
StructField("asin_vartion_list", IntegerType(), True),
StructField("title_len", IntegerType(), True),
StructField("img_num", IntegerType(), True),
StructField("img_type", StringType(), True),
StructField("activity_type", StringType(), True),
StructField("one_two_val", StringType(), True),
StructField("three_four_val", StringType(), True),
StructField("eight_val", StringType(), True),
StructField("qa_num", IntegerType(), True),
StructField("five_star", IntegerType(), True),
StructField("four_star", IntegerType(), True),
StructField("three_star", IntegerType(), True),
StructField("two_star", IntegerType(), True),
StructField("one_star", IntegerType(), True),
StructField("low_star", IntegerType(), True),
StructField("together_asin", StringType(), True),
StructField("brand", StringType(), True),
StructField("ac_name", StringType(), True),
StructField("material", StringType(), True),
StructField("node_id", StringType(), True),
StructField("data_type", IntegerType(), True),
StructField("sp_num", StringType(), True),
StructField("describe", StringType(), True),
StructField("date_info", StringType(), True),
StructField("weight_str", StringType(), True),
StructField("package_quantity", StringType(), True),
StructField("pattern_name", StringType(), True),
StructField("seller_id", StringType(), True),
StructField("variat_num", IntegerType(), True),
StructField("site_name", StringType(), True),
StructField("best_sellers_rank", StringType(), True),
StructField("best_sellers_herf", StringType(), True),
StructField("account_url", StringType(), True),
StructField("account_name", StringType(), True),
StructField("parentAsin", StringType(), True),
StructField("asinUpdateTime", StringType(), True),
StructField("follow_sellers", StringType(), True),
])
return schema
@staticmethod
def udf_cal_crc32(asin, key_size):
# crc32算法 + 取余
# 获取asin字符串的字节表示形式
bytes_str = bytes(asin, 'utf-8')
# 使用zlib计算CRC-32校验和
checksum = zlib.crc32(bytes_str)
# 获取32位的二进制补码
checksum_signed = (checksum & 0xFFFFFFFF) - (1 << 32) if checksum & (1 << 31) else checksum
def java_mod(x, y):
# return x % y if x * y > 0 else x % y - y # 区分正负值
return abs(x) % y # 不区分正负值
# 取余
result = java_mod(checksum_signed, key_size)
return result
@staticmethod
def udf_cal_bkdr(asin):
# BKDR哈希算法
hash = 0
for c in asin:
hash = (hash * 33 + ord(c)) % 65535 # 对哈希值取模65535，以避免溢出
return hash
@staticmethod
def udf_rank_and_category(best_sellers_rank, pattern_1_rank_str, pattern_str, replace_str):
best_sellers_rank = str(best_sellers_rank).replace(",", "")
matches = re.findall(pattern_1_rank_str, best_sellers_rank)
asin_bs_cate_1_rank = matches[0] if matches else None
best_sellers_rank = best_sellers_rank.replace(replace_str, "")
matches = re.findall(pattern_str, best_sellers_rank)
rank_and_category = "&&&&".join([rank_cate.replace(",", "") for rank_cate in matches]) if matches else None
return asin_bs_cate_1_rank, rank_and_category
def df_read_data_by_kafka(self):
# .option("my_kafka.bootstrap.servers", "113.100.143.162:39092") \
# .option("startingOffsets", "lastest") # 偏移量， lastest, earliest
# .select(F.from_json("value", schema=self.schema).alias("data")) \
kafka_df = self.spark.readStream \
.format("my_kafka") \
.option("my_kafka.bootstrap.servers", "192.168.10.221:9092,192.168.10.220:9092,192.168.10.210:9092") \
.option("subscribe", f"{self.site_name}_asin_detail") \
.option("startingOffsets", "lastest") \
.load() \
.select(F.from_json(F.col("value").cast("string"), schema=self.schema).alias("data")) \
.select("data.*")
# assign_option = f"""{{"{self.site_name}_asin_detail": {{"7": 0}}}}"""
# # .option("subscribe", f""""{self.site_name}_asin_detail": {"7": 0}""")
# kafka_df = self.spark.readStream \
# .format("my_kafka") \
# .option("my_kafka.bootstrap.servers", "192.168.10.221:9092,192.168.10.220:9092,192.168.10.210:9092") \
# .option("subscribe", f"{self.site_name}_asin_detail")\
# .option("assign", assign_option) \
# .option("startingOffsets", "lastest") \
# .load() \
# .selectExpr("CAST(value AS STRING) AS value") \
# .select(F.from_json("value", schema=self.schema).alias("data")) \
# .select("data.*")
#"""{"your_topic_name": {"0": 100, "1": 200}}"""
# .option("my_kafka.fetch.max.bytes", "10485760") \
# .option("my_kafka.max.partition.fetch.bytes", "10485760") \
return kafka_df
def read_data(self):
print("1.1 读取dim_st_asin_info表, 计算ao值")
sql = f"select * from dim_st_asin_info where site_name='{self.site_name}' and date_type='{self.date_type}' and date_info='{self.date_info}'"
print("sql:", sql)
self.df_st_asin = self.spark.sql(sql)
self.df_st_asin = self.df_st_asin.drop_duplicates(['search_term', 'asin', 'data_type']).cache()
self.df_st_asin.show(10, truncate=False)
print("1.2 读取ods_one_category_report表")
if int(self.year) == 2022 and int(self.month) < 3:
sql = f"select category_id as asin_bs_cate_1_id, rank as asin_bs_cate_1_rank, orders as asin_bsr_orders from ods_one_category_report " \
f"where site_name='{self.site_name}' and date_type='month' and date_info='2022-12';"
else:
sql = f"select category_id as asin_bs_cate_1_id, rank as asin_bs_cate_1_rank, orders as asin_bsr_orders from ods_one_category_report " \
f"where site_name='{self.site_name}' and date_type='month' and date_info='{self.year}-{self.month}';"
print("sql:", sql)
self.df_bs_report = self.spark.sql(sqlQuery=sql).cache()
self.df_bs_report.show(10, truncate=False)
print("1.3 读取bsr一级分类表")
self.df_asin_bs = get_node_first_id_df(self.site_name, self.spark)
self.df_asin_bs = self.df_asin_bs.withColumnRenamed("category_first_id", "asin_bs_cate_1_id")
self.df_asin_bs.show(10, truncate=False)
print("1.4 读取内部asin表")
# sql = f"select asin, 1 as isSelfAsin from ods_self_asin where site_name='{self.site_name}';"
# print("sql:", sql)
# self.df_self_asin = self.spark.sql(sqlQuery=sql).cache()
# self.df_self_asin.show(10, truncate=False)
self.fetch_self_asin()
# 读取asin和sku计数关系
print("1.5 读取asin和sku计数关系")
self.fetch_asin_sku_count()
def handle_data(self):
# 计算asin的ao值
self.df_asin_counts = self.handle_st_asin_counts()
self.df_asin_counts = self.df_asin_counts.select("asin", "asin_ao").cache()
def handle_asin_bs_category_rank(self, df):
df = df.withColumn(
'bs_str', self.u_rank_and_category(
'best_sellers_rank',
F.lit(self.pattern_1_rank_str[self.site_name]),
F.lit(self.pattern_str[self.site_name]),
F.lit(self.replace_str[self.site_name])
)
)
df = df.withColumn('asin_bs_cate_1_rank', df.bs_str.getField('asin_bs_cate_1_rank')) \
.withColumn('rank_and_category', df.bs_str.getField('rank_and_category')) \
.drop('bs_str', 'best_sellers_rank')
df.show(10, truncate=False)
return df
def handle_st_asin_counts(self):
self.df_st_asin = self.df_st_asin.withColumn(
f"asin_data_type",
F.concat(F.lit(f"asin_"), self.df_st_asin.data_type, F.lit(f"_counts"))
)
df_asin_counts = self.df_st_asin.groupby([f'asin']). \
pivot(f"asin_data_type").count()
df_asin_counts = self.df_asin_templates.unionByName(df_asin_counts, allowMissingColumns=True) # 防止爬虫数据没有导致程序运行出错
df_asin_counts = df_asin_counts.fillna(0)
# df.show(10, truncate=False)
df_asin_counts = df_asin_counts.withColumn(
f"asin_sb_counts",
df_asin_counts[f"asin_sb1_counts"] + df_asin_counts[f"asin_sb2_counts"] + df_asin_counts[f"asin_sb3_counts"]
)
df_asin_counts = df_asin_counts.withColumn(
f"asin_adv_counts",
df_asin_counts[f"asin_sb_counts"] + df_asin_counts[f"asin_sp_counts"]
)
df_asin_counts = df_asin_counts.withColumn(
f"asin_ao",
df_asin_counts[f"asin_adv_counts"] / df_asin_counts[f"asin_zr_counts"]
) # 不要把null置为0, null值产生原因是zr类型没有搜到对应的搜索词
df_asin_counts = df_asin_counts.withColumn("asin_ao", F.round(df_asin_counts["asin_ao"], 4))
df_asin_counts.show(10, truncate=False)
return df_asin_counts
@staticmethod
def clean_kafka_df(df):
df = df.withColumnRenamed("seller_id", "account_id")
# cols_python = ["asin", "parentAsin", "variat_num", "best_sellers_rank", "best_sellers_herf", "price", "rating",
# "brand", "brand", "account_id", "account_name", "account_url", "buy_box_seller_type",
# "volume", "weight", "weight_str", "launchTime", "total_comments", "page_inventory"]
# oneCategoryRank, aoVal, bsrOrders, bsrOrdersSale
# siteName volumeFormat weightFormat asinUpdateTime
# java那边插件的字段名称
cols_java = ['asin', 'parentAsin', 'asinVarNum', 'oneCategoryRank', 'bestSellersRank', 'lastHerf', 'aoVal', 'price', 'rating',
'bsrOrders', 'bsrOrdersSale', 'brandName', 'accountId', 'accountName', 'accountUrl', 'siteName', 'buyBoxSellerType',
'volume', 'volumeFormat', 'weight', 'weightFormat', 'launchTime', 'totalComments', 'pageInventory', 'asinUpdateTime']
df = df.select("asin", "parentAsin", "variat_num", "best_sellers_rank", "best_sellers_herf", "price", "rating",
"brand", "account_id", "account_name", "account_url", "buy_box_seller_type",
"volume", "weight", "weight_str", "launch_time", "total_comments", "page_inventory", "asinUpdateTime", "site_name", "node_id")
return df
def rename_cols(self, df):
# 计算redis的key
df = df.withColumn(
'key_outer', self.u_cal_crc32('asin', F.lit(self.batch_size))
)
df = df.withColumn(
'key_inner', self.u_cal_bkdr('asin')
)
df.show(5, truncate=False)
df = df.withColumnRenamed("variat_num", "asinVarNum")
df = df.withColumnRenamed("asin_bs_cate_1_rank", "oneCategoryRank")
df = df.withColumnRenamed("rank_and_category", "bestSellersRank") # 解析后的
df = df.withColumnRenamed("best_sellers_herf", "lastHerf")
df = df.withColumnRenamed("asin_ao", "aoVal")
df = df.withColumnRenamed("asin_bsr_orders", "bsrOrders")
df = df.withColumnRenamed("asin_bsr_orders_sale", "bsrOrdersSale")
df = df.withColumnRenamed("brand", "brandName")
df = df.withColumnRenamed("account_id", "accountId")
df = df.withColumnRenamed("account_name", "accountName")
df = df.withColumnRenamed("account_url", "accountUrl")
df = df.withColumnRenamed("buy_box_seller_type", "buyBoxSellerType")
df = df.withColumnRenamed("launch_time", "launchTime")
df = df.withColumnRenamed("total_comments", "totalComments")
df = df.withColumnRenamed("page_inventory", "pageInventory")
df = df.select('asin', 'parentAsin', 'asinVarNum', 'oneCategoryRank', 'bestSellersRank', 'lastHerf', 'aoVal', 'price', 'rating',
'bsrOrders', 'bsrOrdersSale', 'brandName', 'accountId', 'accountName', 'accountUrl', 'buyBoxSellerType',
'volume', 'weight', 'launchTime', 'totalComments', 'pageInventory', 'asinUpdateTime',
"site_name", "key_outer", "key_inner")
return df
def process_batch(self, df, epoch_id):
try:
count = df.count()
print("当前批次传输的数据量为df.count():", count)
if count == 0:
self.judge_spider_asin_detail_is_finished()
# 确保schema非空以避免NoneType错误
if not self.schema:
raise ValueError("Schema is not defined")
# df.show(5, truncate=False)
print("df.columns:", df.columns)
# df = df.select("asin", "launch_time", "volume", "weight", "weight_str", "node_id", "variat_num", "best_sellers_rank", "best_sellers_herf", "seller_id", "account_url", "account_name", "site_name")
df = self.clean_kafka_df(df=df)
# df.show(5, truncate=False)
# # 提取排名和分类
df_bs = self.handle_asin_bs_category_rank(df=df.select("asin", "best_sellers_rank"))
# join
df_save = df.join(
df_bs, on='asin', how='left'
).join(
self.df_asin_counts, on='asin', how='left'
).join(
self.df_asin_bs, on='node_id', how='left'
).join(
self.df_self_asin, on='asin', how='left'
).join(
self.df_asin_sku, on='asin', how='left'
)
df_save = df_save.na.fill({"isSelfAsin": 0})
# 计算bsr效率
df_save = df_save.join(
self.df_bs_report, on=['asin_bs_cate_1_rank', 'asin_bs_cate_1_id'], how='left'
)
df_save = df_save.withColumn("asin_bsr_orders_sale", df_save.price * df_save.asin_bsr_orders)
df_save = self.rename_cols(df=df_save)
self.save_to_redis(df=df_save)
except Exception as e:
print(e, traceback.format_exc())
# # 与从Kafka读取的数据进行连接
# joined_df = df.join(self.df_asin_title, "asin", how='left')
# # 执行你的转换和聚合逻辑
# result_df = joined_df.groupBy("asin").count()
# result_df.show(10, truncate=False)
print("epoch_id:", epoch_id, datetime.now().strftime("%Y-%m-%d %H:%M:%S"))
def start_stream(self, processing_time=600):
# kafka_df = self.df_read_data_by_kafka()
if self.date_type == "month":
date_type = "_month"
else:
date_type = ""
topics = f"{self.site_name}_asin_detail{date_type}"
kafka_df = self.get_kafka_df_by_spark(schema=self.schema, consumption_type="latest", topics=topics)
query = kafka_df.writeStream \
.outputMode("append") \
.format("console") \
.option("checkpointLocation", "/root/tmp") \
.foreachBatch(self.process_batch) \
.trigger(processingTime=f'{processing_time} seconds').start()
query.awaitTermination()
def save_to_redis(self, df):
# 将Spark DataFrame转换为Pandas DataFrame
pdf = df.toPandas()
# 遍历Pandas DataFrame并将数据插入到Redis
for index, row in pdf.iterrows():
# 创建一个复合键，或者根据你的需要选择适当的键
# 1. 外层key为10197， 内层可以为10197:15931
# redis_key = f"{row['key_outer']}:{row['key_inner']}" #
# # 插入值到Redis - 在这里我仅仅存储了一个值，你可以存储一个字典来存储多个值
# self.client.set(redis_key, row['value'])
# row_json = row.to_json(orient='split')
# self.client.set(redis_key, row_json)
# 2. 外层key为10197， 内层可以为15931
# redis_key = row['key_outer']
# redis_field = row['key_inner']
# row_json = row.to_json(orient='split')
# self.client.hset(redis_key, redis_field, row_json)
# 3. hashmap + 外层key为10197， 内层可以为15931
redis_key = row['key_outer']
redis_field = row['key_inner']
row_dict = row.to_dict()
# row_dict = {k: str(v).lower().replace("none", "").replace("nan", "") for k, v in row_dict.items()} # 确保所有的值都是字符串
row_dict = {k: str(v).replace("None", "").replace("none", "").replace("NaN", "").replace("nan", "") for k, v in row_dict.items()} # 确保所有的值都是字符串
row_dict = {k: format(v, ".2f") if isinstance(v, (int, float)) else str(v).replace("None", "").replace(
"nan", "") for k, v in row_dict.items()}
del row_dict["key_outer"]
del row_dict["key_inner"]
row_json = json.dumps(row_dict)
self.client.hset(redis_key, redis_field, row_json)
def get_topic_name(self):
if self.site_name == "us" and self.date_type == "month":
self.topic_name = f"{site_name}_asin_detail_{self.date_type}"
else:
self.topic_name = f"{site_name}_asin_detail"
def handle_history(self):
self.get_topic_name()
consumer = self.get_kafka_object_by_python(topic_name=self.topic_name)
partition_offsets_dict = self.get_kafka_partitions_data(consumer=consumer, topic_name=self.topic_name)
partition_num = len(partition_offsets_dict)
beginning_offsets_dict = {}
end_offsets_dict = {}
while True:
num = 0
for key, value in partition_offsets_dict.items():
# 起始偏移量
beginning_offsets_dict[str(key)] = value['beginning_offsets']
# 结束偏移量
end_offsets = value['beginning_offsets'] + self.batch_size_history
end_offsets_partition = value['end_offsets']
end_offsets_dict[str(key)] = min(end_offsets, end_offsets_partition)
if end_offsets >= end_offsets_partition:
num += 1
else:
partition_offsets_dict[key]['beginning_offsets'] = end_offsets
starting_offsets_json = json.dumps({self.topic_name: beginning_offsets_dict})
ending_offsets_json = json.dumps({self.topic_name: end_offsets_dict})
print(f"starting_offsets_json: {starting_offsets_json}, ending_offsets_json:{ending_offsets_json}")
kafka_df = self.spark.read \
.format("kafka") \
.option("kafka.bootstrap.servers", self.kafka_servers) \
.option("subscribe", self.topic_name) \
.option("kafka.security.protocol", self.kafka_security_protocol) \
.option("kafka.sasl.mechanism", self.kafka_sasl_mechanism) \
.option("kafka.sasl.jaas.config",
f'org.apache.kafka.common.security.plain.PlainLoginModule required username="{self.kafka_username}" password="{self.kafka_password}";') \
.option("failOnDataLoss", "true") \
.option("startingOffsets", starting_offsets_json) \
.option("endingOffsets", ending_offsets_json) \
.load() \
.select(F.from_json(F.col("value").cast("string"), schema=self.schema).alias("data")) \
.select("data.*")
print(f"kafka_df.count():{kafka_df.count()}")
if num >= partition_num:
break
else:
continue
def handle_history_old(self):
self.get_topic_name()
consumer = self.get_kafka_object_by_python(topic_name=self.topic_name)
partition_data_count = self.get_kafka_partitions_data(consumer=consumer, topic_name=self.topic_name)
beginning_offsets_list = []
end_offsets_list = []
for values in partition_data_count.values():
beginning_offsets_list.append(values['beginning_offsets'])
end_offsets_list.append(values['end_offsets'])
min_offset = min(beginning_offsets_list)
# min_offset = max(beginning_offsets_list)
max_offset = max(end_offsets_list)
print(f"min_offset:{min_offset}, max_offset:{max_offset}")
# max_offset = max(partition_data_count.values())
# for start_offset in range(0, max_offset+1, self.batch_size_history):
# self.batch_size_history = 100
for start_offset in range(min_offset, max_offset+1, self.batch_size_history):
end_offset = max(start_offset + self.batch_size_history, max_offset)
starting_offsets_json = json.dumps({self.topic_name: {str(p): start_offset for p in partition_data_count.keys()}})
ending_offsets_json = json.dumps({self.topic_name: {str(p): end_offset for p in partition_data_count.keys()}})
# .option("failOnDataLoss", "true") # 设置 failOnDataLoss 为 true, 默认为False
kafka_df = self.spark.read \
.format("kafka") \
.option("kafka.bootstrap.servers", self.kafka_servers) \
.option("subscribe", self.topic_name) \
.option("kafka.security.protocol", self.kafka_security_protocol) \
.option("kafka.sasl.mechanism", self.kafka_sasl_mechanism) \
.option("kafka.sasl.jaas.config", f'org.apache.kafka.common.security.plain.PlainLoginModule required username="{self.kafka_username}" password="{self.kafka_password}";') \
.option("failOnDataLoss", "true") \
.option("startingOffsets", starting_offsets_json) \
.option("endingOffsets", ending_offsets_json) \
.load() \
.select(F.from_json(F.col("value").cast("string"), schema=self.schema).alias("data")) \
.select("data.*")
print(f"kafka_df.count():{kafka_df.count()}, start_offset:{start_offset}, end_offset:{end_offset}")
self.handle_batch_history(df=kafka_df)
# self.handle_batch_history(df=kafka_df)
# # current_offsets[partition] = end_offset
# .option("startingOffsets", starting_offsets_json) \
# .option("endingOffsets", ending_offsets_json) \
# \
# while not done:
# # for partition in partitions:
# # start_offset = current_offsets[partition]
# # end_offset = start_offset + self.batch_size
# # print(f"partition:{partition}, start_offset:{start_offset}, end_offset:{end_offset}")
# # 创建包含所有分区信息的JSON字符串
# start_offset, end_offset = 0, 0 + self.batch_size
# starting_offsets_json = json.dumps({self.topic_name: {str(p): start_offset for p in partitions}})
# # ending_offsets_json = json.dumps({self.topic_name: {str(p): (end_offset if p == partition else start_offset) for p in partitions}})
# ending_offsets_json = json.dumps({self.topic_name: {str(p): end_offset for p in partitions}})
# print(f"starting_offsets_json:{starting_offsets_json}, ending_offsets_json:{ending_offsets_json}")
# # 读取数据
# kafka_df = self.spark.read \
# .format("kafka") \
# .option("kafka.bootstrap.servers", self.kafka_servers) \
# .option("subscribe", self.topic_name) \
# .option("startingOffsets", starting_offsets_json) \
# .option("endingOffsets", ending_offsets_json) \
# .option("failOnDataLoss", "false") \
# .load()\
# .select(F.from_json(F.col("value").cast("string"), schema=self.schema).alias("data")) \
# .select("data.*")
#
# # TODO: 根据需要处理数据
# # kafka_df.show(10, truncate=False)
# print("kafka_df.count():", kafka_df.count())
# self.handle_batch_history(df=kafka_df)
# # current_offsets[partition] = end_offset
#
# done = all(offset >= partition_data_count[p] for p, offset in current_offsets.items())
#
# 关闭SparkSession
self.spark.stop()
def handle_batch_history(self, df):
try:
print("df.columns:", df.columns)
# df = df.select("asin", "launch_time", "volume", "weight", "weight_str", "node_id", "variat_num", "best_sellers_rank", "best_sellers_herf", "seller_id", "account_url", "account_name", "site_name")
df.show(10, truncate=False)
df = self.clean_kafka_df(df=df)
# df.show(5, truncate=False)
# # 提取排名和分类
df_bs = self.handle_asin_bs_category_rank(df=df.select("asin", "best_sellers_rank"))
# join
df_save = df.join(
df_bs, on='asin', how='left'
).join(
self.df_asin_counts, on='asin', how='left'
).join(
self.df_asin_bs, on='node_id', how='left'
).join(
self.df_self_asin, on='asin', how='left'
).join(
self.df_asin_sku, on='asin', how='left'
)
# 计算bsr效率
df_save = df_save.join(
self.df_bs_report, on=['asin_bs_cate_1_rank', 'asin_bs_cate_1_id'], how='left'
)
df_save = df_save.withColumn("asin_bsr_orders_sale", df_save.price * df_save.asin_bsr_orders)
df_save = self.rename_cols(df=df_save)
self.save_to_redis(df=df_save)
except Exception as e:
print(e, traceback.format_exc())
def run(self):
# self.read_data()
# self.handle_data()
if self.consumer_type == 'latest':
self.start_stream(processing_time=300)
else:
self.handle_history()
# # 将消息值转换为字符串，并创建一个临时视图
# stringifiedDF = self.my_kafka.selectExpr("CAST(value AS STRING)")
# stringifiedDF.createOrReplaceTempView("KafkaData")
# # 设置streaming查询，每5分钟触发一次
# query = stringifiedDF.writeStream.foreachBatch(self.process_batch).trigger(processingTime='600 seconds').start()
# # 等待查询终止
# query.awaitTermination()
if __name__ == '__main__':
site_name = sys.argv[1] # 参数1：站点
date_type = sys.argv[2] # 参数2：类型：week/4_week/month/quarter/day
date_info = sys.argv[3] # 参数3：年-周/年-月/年-季/年-月-日, 比如: 2022-1
consumer_type = sys.argv[4] # 参数3：年-周/年-月/年-季/年-月-日, 比如: 2022-1
handle_obj = DimStAsinInfo(site_name=site_name, date_type=date_type, date_info=date_info, consumer_type=consumer_type, batch_size=100000)
# handle_obj.run()
handle_obj.run_kafka()