apache-spark - 结构化流错误 py4j.protocol.Py4JNetworkError : Answer from Java side is empty

Question

我正在尝试使用 PySpark 和 Structured Streaming (Spark 2.3) 在两个 Kafka Stream 之间进行左外连接。

import os
import time

from pyspark.sql.types import *
from pyspark.sql.functions import from_json, col, struct, explode, get_json_object
from ast import literal_eval
from pyspark.sql import SparkSession
from pyspark.sql.functions import expr

os.environ['PYSPARK_SUBMIT_ARGS'] = '--packages org.apache.spark:spark-sql-kafka-0-10_2.11:2.3.0 pyspark-shell'

spark = SparkSession \
    .builder \
    .appName("Spark Kafka Structured Streaming") \
    .getOrCreate()

schema_impressions = StructType() \
    .add("id_req", StringType()) \
    .add("ts_imp_request", TimestampType()) \
    .add("country", StringType()) \
    .add("TS_IMPRESSION", TimestampType()) 

schema_requests = StructType() \
    .add("id_req", StringType()) \
    .add("page", StringType()) \
    .add("conntype", StringType()) \
    .add("TS_REQUEST", TimestampType()) 

impressions = spark.readStream \
  .format("kafka") \
  .option("kafka.bootstrap.servers", "ip-ec2.internal:9092") \
  .option("subscribe", "ssp.datascience_impressions") \
  .load()

requests = spark \
  .readStream \
  .format("kafka") \
  .option("kafka.bootstrap.servers", "ip-ec2.internal:9092") \
  .option("subscribe", "ssp.datascience_requests") \
  .option("startingOffsets", "latest") \
  .load()

query_requests = requests \
        .select(col("timestamp"), col("key").cast("string"), from_json(col("value").cast("string"), schema_requests).alias("parsed")) \
        .select(col("timestamp").alias("timestamp_req"), "parsed.id_req", "parsed.page", "parsed.conntype", "parsed.TS_REQUEST") \
        .withWatermark("timestamp_req", "120 seconds") 

query_impressions = impressions \
        .select(col("timestamp"), col("key").cast("string"), from_json(col("value").cast("string"), schema_impressions).alias("parsed")) \
        .select(col("timestamp").alias("timestamp_imp"), col("parsed.id_req").alias("id_imp"), "parsed.ts_imp_request", "parsed.country", "parsed.TS_IMPRESSION") \
        .withWatermark("timestamp_imp", "120 seconds") 

query_requests.printSchema()        
query_impressions.printSchema()

> root  
|-- timestamp_req: timestamp (nullable = true)  
|-- id_req: string (nullable = true)  
|-- page: string (nullable = true)  
|-- conntype: string (nullable = true)  
|-- TS_REQUEST: timestamp (nullable = true)
> 
> root  |-- timestamp_imp: timestamp (nullable = true)  
|-- id_imp: string (nullable = true)  
|-- ts_imp_request: timestamp (nullable = true)  
|-- country: string (nullable = true)  
|-- TS_IMPRESSION: timestamp (nullable = true)

在简历中，我将从两个 Kafka Streams 中获取数据，在接下来的几行中，我将尝试使用 ID 进行连接。

rawQuery = query_requests.join(query_impressions,  expr(""" 
    (id_req = id_imp AND 
    timestamp_imp >= timestamp_req AND 
    timestamp_imp <= timestamp_req + interval 5 minutes) 
    """), 
  "leftOuter")

rawQuery = rawQuery \
        .writeStream \
        .format("parquet") \
        .option("checkpointLocation", "/home/jovyan/streaming/applicationHistory") \
        .option("path", "/home/jovyan/streaming").start()
print(rawQuery.status)

{'message': 'Processing new data', 'isDataAvailable': True, 'isTriggerActive': True} ERROR:root:Exception while sending command. Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 1062, in send_command raise Py4JNetworkError("Answer from Java side is empty") py4j.protocol.Py4JNetworkError: Answer from Java side is empty

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 908, in send_command response = connection.send_command(command) File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 1067, in send_command "Error while receiving", e, proto.ERROR_ON_RECEIVE) py4j.protocol.Py4JNetworkError: Error while receiving ERROR:py4j.java_gateway:An error occurred while trying to connect to the Java server (127.0.0.1:33968) Traceback (most recent call last):
File "/opt/conda/lib/python3.6/site-packages/IPython/core/interactiveshell.py", line 2910, in run_code exec(code_obj, self.user_global_ns, self.user_ns) File "", line 3, in print(rawQuery.status) File "/opt/conda/lib/python3.6/site-packages/pyspark/sql/streaming.py", line 114, in status return json.loads(self._jsq.status().json()) File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 1160, in call answer, self.gateway_client, self.target_id, self.name) File "/opt/conda/lib/python3.6/site-packages/pyspark/sql/utils.py", line 63, in deco return f(*a, **kw) File "/opt/conda/lib/python3.6/site-packages/py4j/protocol.py", line 328, in get_return_value format(target_id, ".", name)) py4j.protocol.Py4JError: An error occurred while calling o92.status

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/IPython/core/interactiveshell.py", line 1828, in showtraceback stb = value._render_traceback_() AttributeError: 'Py4JError' object has no attribute '_render_traceback_'

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 852, in _get_connection connection = self.deque.pop() IndexError: pop from an empty deque

我正在使用 Jupyter Notebook 在本地运行 Spark。在 spark/conf/ spark-defaults.conf 我有:

# Example:
# spark.master                     spark://master:7077
# spark.eventLog.enabled           true
# spark.eventLog.dir               hdfs://namenode:8021/directory
# spark.serializer                 org.apache.spark.serializer.KryoSerializer
spark.driver.memory             15g
# spark.executor.extraJavaOptions  -XX:+PrintGCDetails -Dkey=value -Dnumbers="one two three"

如果我在上一个错误后尝试使用 Spark，我会收到该错误:

ERROR:root:Exception while sending command. Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 1062, in send_command raise Py4JNetworkError("Answer from Java side is empty") py4j.protocol.Py4JNetworkError: Answer from Java side is empty

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 908, in send_command response = connection.send_command(command) File "/opt/conda/lib/python3.6/site-packages/py4j/java_gateway.py", line 1067, in send_command "Error while receiving", e, proto.ERROR_ON_RECEIVE) py4j.protocol.Py4JNetworkError: Error while receiving

Answer 1

我解决了问题!基本上，该问题出于某种原因与 Jupyter Notebook 有关。我删除了前面代码的下一行:

os.environ['PYSPARK_SUBMIT_ARGS'] = '--packages org.apache.spark:spark-sql-kafka-0-10_2.11:2.3.0 pyspark-shell'

我使用控制台运行了代码:

> spark-submit --packages org.apache.spark:spark-sql-kafka-0-10_2.11:2.3.0 spark_structured.py

这样，我就可以毫无问题地运行所有代码。

如果你有同样的问题，你也可以改变
spark-default.conf 并增加 spark.driver.memory 和 spark.executor.memory