Docker 容器中的 Spark 不读取 Kafka 输入 - 结构化流

Question

When the Spark job is run locally without Docker via spark-submit everything works fine. However, running on a docker container results in no output being generated.

To see if Kafka itself was working, I extracted Kafka on to the Spark worker container, and make a console consumer listen to the same host, port and topic, (kafka:9092, crypto_topic) which was working correctly and showing output. (There's a producer constantly pushing data to the topic in another container)

Expected -

20/09/11 17:35:27 INFO BlockManagerMasterEndpoint: Registering block manager 192.168.29.10:42565 with 366.3 MB RAM, BlockManagerId(driver, 192.168.29.10, 42565, None)
20/09/11 17:35:27 INFO BlockManagerMaster: Registered BlockManager BlockManagerId(driver, 192.168.29.10, 42565, None)
20/09/11 17:35:27 INFO BlockManager: Initialized BlockManager: BlockManagerId(driver, 192.168.29.10, 42565, None)
-------------------------------------------
Batch: 0
-------------------------------------------
+---------+-----------+-----------------+------+----------+------------+-----+-------------------+---------+
|name_coin|symbol_coin|number_of_markets|volume|market_cap|total_supply|price|percent_change_24hr|timestamp|
+---------+-----------+-----------------+------+----------+------------+-----+-------------------+---------+
+---------+-----------+-----------------+------+----------+------------+-----+-------------------+---------+
...
...
...
followed by more output

Actual

20/09/11 14:49:44 INFO BlockManagerMasterEndpoint: Registering block manager d7443d94165c:46203 with 366.3 MB RAM, BlockManagerId(driver, d7443d94165c, 46203, None)
20/09/11 14:49:44 INFO BlockManagerMaster: Registered BlockManager BlockManagerId(driver, d7443d94165c, 46203, None)
20/09/11 14:49:44 INFO BlockManager: Initialized BlockManager: BlockManagerId(driver, d7443d94165c, 46203, None)
20/09/11 14:49:44 INFO StandaloneSchedulerBackend: SchedulerBackend is ready for scheduling beginning after reached minRegisteredResourcesRatio: 0.0

no more output, stuck here

docker-compose.yml file

version: "3"

services:

    zookeeper:
        image: zookeeper:3.6.1
        container_name: zookeeper
        hostname: zookeeper
        ports:
            - "2181:2181"
        networks:
            - crypto-network
      
    kafka:
        image: wurstmeister/kafka:2.13-2.6.0
        container_name: kafka
        hostname: kafka
        ports:
            - "9092:9092"
        environment:
            - KAFKA_ADVERTISED_HOST_NAME=kafka
            - KAFKA_ZOOKEEPER_CONNECT=zookeeper:2181 
            - KAFKA_ADVERTISED_PORT=9092
            # topic-name:partitions:in-sync-replicas:cleanup-policy
            - KAFKA_CREATE_TOPICS="crypto_topic:1:1:compact"
        networks:
            - crypto-network

    kafka-producer:
        image: python:3-alpine
        container_name: kafka-producer
        command: >
                sh -c "pip install -r /usr/src/producer/requirements.txt
                && python3 /usr/src/producer/kafkaProducerService.py"
        volumes:
            - ./kafkaProducer:/usr/src/producer
        networks: 
            - crypto-network
      
            
    cassandra:
        image: cassandra:3.11.8
        container_name: cassandra
        hostname: cassandra
        ports:
            - "9042:9042"
        #command:
        #    cqlsh -f /var/lib/cassandra/cql-queries.cql
        volumes:
            - ./cassandraData:/var/lib/cassandra

        networks:
            - crypto-network
            
    spark-master:
        image: bde2020/spark-master:2.4.5-hadoop2.7
        container_name: spark-master
        hostname: spark-master
        ports:
            - "8080:8080"
            - "7077:7077"
            - "6066:6066"
        networks:
            - crypto-network
            
    spark-consumer-worker:
        image: bde2020/spark-worker:2.4.5-hadoop2.7
        container_name: spark-consumer-worker
        environment:
            - SPARK_MASTER=spark://spark-master:7077
        ports:
            - "8081:8081"
        volumes:
            - ./sparkConsumer:/sparkConsumer
        networks:
            - crypto-network
    
            
networks:
  crypto-network:
    driver: bridge

spark-submit is run by

docker exec -it spark-consumer-worker bash

/spark/bin/spark-submit --master $SPARK_MASTER --class processing.SparkRealTimePriceUpdates \
--packages com.datastax.spark:spark-cassandra-connector_2.11:2.4.3,org.apache.spark:spark-sql-kafka-0-10_2.11:2.4.5 \
/sparkConsumer/sparkconsumer_2.11-1.0-RELEASE.jar 

Relevant parts of Spark code

  val inputDF: DataFrame = spark
    .readStream
    .format("kafka")
    .option("kafka.bootstrap.servers", "kafka:9092")
    .option("subscribe", "crypto_topic")
    .load()

...
...
...

  val queryPrice: StreamingQuery = castedDF
    .writeStream
    .outputMode("update")
    .format("console")
    .option("truncate", "false")
    .start()

    queryPrice.awaitTermination()

Answer 1

  val inputDF: DataFrame = spark
    .readStream
    .format("kafka")
    .option("kafka.bootstrap.servers", "kafka:9092")
    .option("subscribe", "crypto_topic")
    .load()

This part of the code was actually

  val inputDF: DataFrame = spark
    .readStream
    .format("kafka")
    .option("kafka.bootstrap.servers", KAFKA_BOOTSTRAP_SERVERS)
    .option("subscribe", KAFKA_TOPIC)
    .load()

Where KAFKA_BOOTSTRAP_SERVERS and KAFKA_TOPIC are read in from a config file while packaging the jar locally.

The best way to debug for me was to set the logs to be more verbose.

Locally, the value of KAFKA_BOOTSTRAP_SERVERS was localhost:9092 , but in the Docker container it was changed to kafka:9092 in the config file there. This however, didn't reflect as the JAR was packaged already. So changing the value to kafka:9092 while packaging locally fixed it.

I would appreciate any help about how to have a JAR pick up configurations dynamically though. I don't want to package via SBT on the Docker container.