Pyspark 在spark流媒体/结构化流媒体中阅读来自卡夫卡的avro消息

我第一次使用pyspark。 Spark版本：2.3.0 卡夫卡版本：2.2.0

我有一个kafka制作人，它以avro格式发送嵌套数据，我正在尝试用pyspark编写spark流/结构化流的代码，它将来自kafka的avro反序列化为dataframe，并将其以拼花格式写入s3。 我能够在spark/scala中找到avro转换器，但pyspark中的支持尚未添加。如何在pyspark中转换相同的值。

---

谢谢。

正如您所提到的，阅读卡夫卡的Avro消息并通过pyspark进行解析，没有直接的库。但我们可以通过编写小包装器来读取/解析Avro消息，并在pyspark流代码中将该函数作为UDF调用，如下所示

参考资料：

注意：Avro是自Spark 2.4以来的内置但外部数据源模块。请按照“ApacheAvro数据源指南”的部署部分部署应用程序

参考：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

Spark提交：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

[调整软件包版本以匹配基于spark/avro版本的安装]

/usr/hdp/2.6.1.0-129/spark2/bin/pyspark --packages org.apache.spark:spark-avro_2.11:2.4.3 --conf spark.ui.port=4064

Pyspark流媒体代码：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

---

正如您所提到的，从卡夫卡读取Avro消息并通过pyspark进行解析，并没有直接的库。但我们可以通过编写小包装器来读取/解析Avro消息，并在pyspark流代码中将该函数作为UDF调用，如下所示

参考资料：

注意：Avro是自Spark 2.4以来的内置但外部数据源模块。请按照“ApacheAvro数据源指南”的部署部分部署应用程序

参考：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

Spark提交：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

[调整软件包版本以匹配基于spark/avro版本的安装]

/usr/hdp/2.6.1.0-129/spark2/bin/pyspark --packages org.apache.spark:spark-avro_2.11:2.4.3 --conf spark.ui.port=4064

Pyspark流媒体代码：

from pyspark.sql import SparkSession
from pyspark.sql.functions import *
from pyspark.sql.types import *
from pyspark.streaming import StreamingContext
from pyspark.sql.column import Column, _to_java_column
from pyspark.sql.functions import col, struct
from pyspark.sql.functions import udf
import json
import csv
import time
import os

#  Spark Streaming context :

spark = SparkSession.builder.appName('streamingdata').getOrCreate()
sc = spark.sparkContext
ssc = StreamingContext(sc, 20)

#  Kafka Topic Details :

KAFKA_TOPIC_NAME_CONS = "topicname"
KAFKA_OUTPUT_TOPIC_NAME_CONS = "topic_to_hdfs"
KAFKA_BOOTSTRAP_SERVERS_CONS = 'localhost.com:9093'

#  Creating  readstream DataFrame :

df = spark.readStream \
     .format("kafka") \
     .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS_CONS) \
     .option("subscribe", KAFKA_TOPIC_NAME_CONS) \
     .option("startingOffsets", "latest") \
     .option("failOnDataLoss" ,"false")\
     .option("kafka.security.protocol","SASL_SSL")\
     .option("kafka.client.id" ,"MCI-CIL")\
     .option("kafka.sasl.kerberos.service.name","kafka")\
     .option("kafka.ssl.truststore.location", "/path/kafka_trust.jks") \
     .option("kafka.ssl.truststore.password", "changeit") \
     .option("kafka.sasl.kerberos.keytab","/path/bdpda.headless.keytab") \
     .option("kafka.sasl.kerberos.principal","bdpda") \
     .load()


df1 = df.selectExpr( "CAST(value AS STRING)")

df1.registerTempTable("test")


# Deserilzing the Avro code function

from pyspark.sql.column import Column, _to_java_column 
def from_avro(col): 
     jsonFormatSchema = """
                    {
                     "type": "record",
                     "name": "struct",
                     "fields": [
                       {"name": "col1", "type": "long"},
                       {"name": "col2", "type": "string"}
                                ]
                     }"""
    sc = SparkContext._active_spark_context 
    avro = sc._jvm.org.apache.spark.sql.avro
    f = getattr(getattr(avro, "package$"), "MODULE$").from_avro
    return Column(f(_to_java_column(col), jsonFormatSchema))


spark.udf.register("JsonformatterWithPython", from_avro)

squared_udf = udf(from_avro)
df1 = spark.table("test")
df2 = df1.select(squared_udf("value"))

#  Declaring the Readstream Schema DataFrame :

df2.coalesce(1).writeStream \
   .format("parquet") \
   .option("checkpointLocation","/path/chk31") \
   .outputMode("append") \
   .start("/path/stream/tgt31")


ssc.awaitTermination()

---

为什么您需要df5和df1是相同的东西？Alsop，正如它对您链接的答案所做的评论，这对Confluent Schema Registry Avro Database不起作用。感谢您分享您的评论。你是对的！！df5不是必需的，并更正了该代码。正如在问题中没有提到的，关于confluent registry给定了通用Avro格式处理的codeRight，但这很少是Kafka中的内容，根据我的经验，请注意，这段代码，我已经在Cloudra环境中用经典的Apache KafkaOkay实现了，很好，那么您没有使用模式注册表？Confluent只使用ApacheKafka…为什么需要df5和df1是相同的东西？Alsop，正如它对您链接的答案所做的评论，这对Confluent Schema Registry Avro Database不起作用。感谢您分享您的评论。你是对的！！df5不是必需的，并更正了该代码。正如在问题中没有提到的，关于confluent registry给定了通用Avro格式处理的codeRight，但这很少是Kafka中的内容，根据我的经验，请注意，这段代码，我已经在Cloudra环境中用经典的Apache KafkaOkay实现了，很好，那么您没有使用模式注册表？Confluent只使用ApacheKafka。。。