Mongodb 如何通过在pyspark上插入子文档将两个文档合并为一个文档?
我有一个大问题,希望能清楚地解释我想做什么。 我正在尝试在pyspark(Spark Structured Streaming)上获取流结构,我想在从Kafka的抓取中获取新数据时更新同一文档 以下是在本地主机和MongoCompass上发送的JSON示例:Mongodb 如何通过在pyspark上插入子文档将两个文档合并为一个文档?,mongodb,apache-spark,pyspark,kafka-consumer-api,spark-structured-streaming,Mongodb,Apache Spark,Pyspark,Kafka Consumer Api,Spark Structured Streaming,我有一个大问题,希望能清楚地解释我想做什么。 我正在尝试在pyspark(Spark Structured Streaming)上获取流结构,我想在从Kafka的抓取中获取新数据时更新同一文档 以下是在本地主机和MongoCompass上发送的JSON示例: { _id: ObjectId("28276465847392747") id: reply Company: reply Value:{ Date: 20-05-2020 Last_Hour_Contract: 09.1
{
_id: ObjectId("28276465847392747")
id: reply
Company: reply
Value:{
Date: 20-05-2020
Last_Hour_Contract: 09.12.25
Last_Hour: 09.14.30
Price: 16.08
Quantity: 8000
Medium_Price: 8.98
Min_Price: 8.98
Max_Price: 20.33
News: { id_news: Reply_20-05-20
title_news: "titolo news"
text: "text"
date: 20-05-2020
hour: 09:13:00
subject: Reply
}
}
}
{
_id: ObjectId("28276465847392747")
id: reply
Company: reply
Value:{
Date: 20-05-2020
Last_Hour_Contract: 09.12.25
Last_Hour: 09.14.30
Price: 17.78
Quantity: 9000
Medium_Price: 67.98
Min_Price: 8.98
Max_Price: 20.33
News: { id_news: Reply_20-05-20
title_news: "title_news"
text: "text"
date: 20-05-2020
hour: 09:13:00
subject: Reply
}
}
}
我想要实现的是在新数据到达时将各种文档(基于Company_Name=“Name_Company”)合并到一个文档中
我想要的JSON文档的设置如下:
{
_id: ObjectId("3333884747656565"),
id: reply
Date: 21-05-2020
Company: Reply
Value:{
Date: 20-05-2020
Last_Hour_Contract: 09.12.25
Last_Hour: 09.14.30
Price: 16.08
Quantity: 8000
Medium_Price: 8.98
Min_Price: 8.98
Max_Price: 20.33
News: {id_news: Reply_20-05-20
title_news: "title news..."
text: "text..."
date: 20-05-2020
hour: 09:13:00
subject: Reply
}
Date: 21-05-2020
Last_Hour_Contract: 09.12.25
Last_Hour: 09.16.50
Price: 16.68
Quantity: 7000
Medium_Price: 8.98
Min_Price: 8.98
Max_Price: 20.33
News: {id_news: Reply_20-05-20
title_news: "title news..."
text: "text..."
date: 21-05-2020
hour: 09:14:00
subject: Reply
}
}
}
我还插入了一个图像,以便您更好地理解(我希望两个箭头可以理解):
如何使用Pyspark实现这一点?谢谢
这是我的代码:
def writeStreamer(sdf):
sdf.select("id_Borsa","NomeAzienda","Valori_Di_Borsa") \
.dropDuplicates(["id_Borsa","NomeAzienda","Valori_Di_Borsa"]) \
.writeStream \
.outputMode("append") \
.foreachBatch(foreach_batch_function) \
.start()
def foreach_batch_function(sdf, epoch_id):
sdf.write \
.format("mongo") \
.mode("append") \
.option("spark.mongodb.input.uri", "mongodb://127.0.0.1:27017/DataManagement.Data") \
.option("spark.mongodb.output.uri", "mongodb://127.0.0.1:27017/DataManagement.Data") \
.save() #"com.mongodb.spark.sql.DefaultSource"
df_borsa = spark.readStream.format("kafka") \
.option("kafka.bootstrap.servers", kafka_broker) \
.option("startingOffsets", "latest") \
.option("subscribe","Reply_borsa") \
.load() \
.selectExpr("CAST(value AS STRING)")
df_news = spark.readStream.format("kafka") \
.option("kafka.bootstrap.servers", kafka_broker) \
.option("startingOffsets", "latest") \
.option("subscribe","Reply_news") \
.load() \
.selectExpr("CAST(value AS STRING)")
df_borsa = df_borsa.withColumn("Valori_Di_Borsa",F.struct(F.col("Data"),F.col("PrezzoUltimoContratto"),F.col("Var%"),F.col("VarAssoluta"),F.col("OraUltimoContratto"),F.col("QuantitaUltimo"),F.col("QuantitaAcquisto"),F.col("QuantitaVendita"),F.col("QuantitaTotale"),F.col("NumeroContratti"),F.col("MaxOggi"),F.col("MinOggi")))
df_news = df_news.withColumn("News",F.struct(F.col("id_News"),F.col("TitoloNews"),F.col("TestoNews"),F.col("DataNews"),F.col("OraNews")))
# Apply watermarks on event-time columns
dfWithWatermark = df_borsa.select("id_Borsa","NomeAzienda","StartTime","Valori_Di_Borsa").withWatermark("StartTime", "2 hours") # maximal delay
df1WithWatermark = df_news.select("SoggettoNews","EndTime").withWatermark("EndTime", "3 hours") # maximal delay
# Join with event-time constraints
sdf = dfWithWatermark.join(df1WithWatermark,expr("""
SoggettoNews = NomeAzienda AND
EndTime >= StartTime AND
EndTime <= StartTime + interval 1 hours
"""),
"leftOuter").withColumn("Valori_Di_Borsa",F.struct(F.col("Valori_Di_Borsa.*"),F.col("News")))
query = writeStreamer(sdf)
spark.streams.awaitAnyTermination()
def writeStreamer(sdf):
sdf.选择(“id_Borsa”、“NomeAzienda”、“Valori_Di_Borsa”)\
.dropDuplicates([“id_Borsa”、“NomeAzienda”、“Valori_Di_Borsa”])\
.writeStream\
.outputMode(“追加”)\
.foreachBatch(foreach\u batch\u函数)\
.start()
def foreach_批处理_函数(sdf,历元id):
写\
.格式(“mongo”)\
.mode(“追加”)\
.option(“spark.mongodb.input.uri”mongodb://127.0.0.1:27017/DataManagement.Data") \
.option(“spark.mongodb.output.uri”mongodb://127.0.0.1:27017/DataManagement.Data") \
.save()#“com.mongodb.spark.sql.DefaultSource”
df_borsa=spark.readStream.format(“卡夫卡”)\
.option(“kafka.bootstrap.servers”,kafka\u代理)\
.选项(“起始偏移量”、“最新”)\
.期权(“认购”、“回复”)\
.load()\
.selectExpr(“转换(值为字符串)”)
df_news=spark.readStream.format(“卡夫卡”)\
.option(“kafka.bootstrap.servers”,kafka\u代理)\
.选项(“起始偏移量”、“最新”)\
.选项(“订阅”、“回复新闻”)\
.load()\
.selectExpr(“转换(值为字符串)”)
df_borsa=df_borsa.带列(“Valori_Di_borsa”)、F.struct(F.col(“数据”)、F.col(“Prezzoultimocontrato”)、F.col(“Varassolta”)、F.col(“Orultimocontrato”)、F.col(“QuantitaUltimo”)、F.col(“QuantitaAcquisto”)、F.col(“QuantitaVendita”)、F.col(“QuantitaTotale”)、F.col(“Numerocontratio”)、F.col(“MaxOggi”)、F.col(“Minogi”))
df_news=df_news.withColumn(“新闻”)、F.struct(F.col(“id_新闻”)、F.col(“TitoloNews”)、F.col(“TestoNews”)、F.col(“数据新闻”)、F.col(“OraNews”))
#在事件时间列上应用水印
dfWithWatermark=df_borsa。选择(“id_borsa”、“NomeAzienda”、“StartTime”、“Valori_Di_borsa”)。withWatermark(“StartTime”、“2小时”)#最大延迟
df1WithWatermark=df#U新闻。选择(“SoggettoNews”,“EndTime”)。withWatermark(“EndTime”,“3小时”)#最大延迟
#加入事件时间约束
sdf=dfWithWatermark.join(df1WithWatermark,expr
SoggettoNews=NomeAzienda和
EndTime>=开始时间和结束时间
EndTime您只需使用分组
操作符按公司
字段对文档进行分组,并使用$push
操作符将每个分组文档的值
对象添加到新形成的数组字段值
因此,上述mongo实现如下所示:
db.collection.aggregate([{
$group:{
_id:“$Company”,
id:{$first:'$id'},
日期:{$first:'$first'},
值:{$push:'$value'}
}
}])
您可以轻松地将上述聚合转换为PySpark实现
你需要做如下的事情:
pipeline=“{'$group':{'$id':'$Company','id':{'$first':'$id'},'日期:{'$first':'$first'},'值:{'$push':'$value'}”
df=spark.read.format(“mongo”).option(“pipeline”,pipeline.load())
df.show()
注意:我不是PySpark的专家,但您可以轻松地将其转换为所需的实现。在mongocompass上它工作得很好!但在PySpark上还没有……我试图在此处将您的管道插入foreach_batch_函数,但一旦我进入mongocompass,它就会返回我想要的两个或更多文档正是你写的,但我一进入MongoCompass就什么也没发生……我试图在我的代码的不同点插入你的管道,但什么都没有。我比你更不幸。我甚至不知道PySpark。如果你愿意,我会删除这个答案并对这个问题进行投票,这样你就能从专家那里得到答案?而且,这将是一个l也为我挣钱。(不过,我只是指导你)好吧!如果你能做到,你会帮我一个大忙的,谢谢