Convert Spark Structure Streaming DataFrames to Pandas DataFrame

Question

I have a Spark Streaming App set up that consumes from a Kafka topic and I need to use some APIs that takes in Pandas Dataframe but when I try to convert it I get this

: org.apache.spark.sql.AnalysisException: Queries with streaming sources must be executed with writeStream.start();;
kafka
        at org.apache.spark.sql.catalyst.analysis.UnsupportedOperationChecker$.org$apache$spark$sql$catalyst$analysis$UnsupportedOperationChecker$$throwError(UnsupportedOperationChecker.scala:297)
        at org.apache.spark.sql.catalyst.analysis.UnsupportedOperationChecker$$anonfun$checkForBatch$1.apply(UnsupportedOperationChecker.scala:36)
        at org.apache.spark.sql.catalyst.analysis.UnsupportedOperationChecker$$anonfun$checkForBatch$1.apply(UnsupportedOperationChecker.scala:34)
        at org.apache.spark.sql.catalyst.trees.TreeNode.foreachUp(TreeNode.scala:127)
        at org.apache.spark.sql.catalyst.analysis.UnsupportedOperationChecker$.checkForBatch(UnsupportedOperationChecker.scala:34)
        at org.apache.spark.sql.execution.QueryExecution.assertSupported(QueryExecution.scala:63)
        at org.apache.spark.sql.execution.QueryExecution.withCachedData$lzycompute(QueryExecution.scala:74)
        at org.apache.spark.sql.execution.QueryExecution.withCachedData(QueryExecution.scala:72)
        at org.apache.spark.sql.execution.QueryExecution.optimizedPlan$lzycompute(QueryExecution.scala:78)
        at org.apache.spark.sql.execution.QueryExecution.optimizedPlan(QueryExecution.scala:78)
        at org.apache.spark.sql.execution.QueryExecution.completeString(QueryExecution.scala:219)
        at org.apache.spark.sql.execution.QueryExecution.toString(QueryExecution.scala:202)
        at org.apache.spark.sql.execution.SQLExecution$.withNewExecutionId(SQLExecution.scala:62)
        at org.apache.spark.sql.Dataset.withNewExecutionId(Dataset.scala:2832)
        at org.apache.spark.sql.Dataset.collectToPython(Dataset.scala:2809)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:498)
        at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
        at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
        at py4j.Gateway.invoke(Gateway.java:282)
        at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
        at py4j.commands.CallCommand.execute(CallCommand.java:79)
        at py4j.GatewayConnection.run(GatewayConnection.java:238)
        at java.lang.Thread.run(Thread.java:745)

Here is my python code

spark = SparkSession\
    .builder\
    .appName("sparkDf to pandasDf")\
    .getOrCreate()

sparkDf = spark.readStream\
    .format("kafka")\
    .option("kafka.bootstrap.servers", "kafkahost:9092")\
    .option("subscribe", "mytopic")\
    .option("startingOffsets", "earliest")\
    .load()


pandas_df =  sparkDf.toPandas()

query = sparkDf.writeStream\
    .outputMode("append")\
    .format("console")\
    .option("truncate", "false")\
    .trigger(processingTime="5 seconds")\
    .start()\
    .awaitTermination()

Now I am aware I am creating another instance of a streaming Dataframe but no matter where I try to use start() and awaitTermination(), I get the same error.

Any ideas?

Answer 1

TL;DR Such operation just cannot work.

Now I am aware I am creating another instance of a streaming Dataframe

Well, the problem is that you really don't. toPandas, called on a DataFrame creates a simple, local, non-distributed Pandas DataFrame, in memory of the driver node.

It not only has nothing to do with Spark, but as an abstraction is inherently incompatible with Structured Streaming - Pandas DataFrame represents a fixed set of tuples, while structured stream represent a an infinite stream of tuples.

It is not exactly clear what you're trying to achieve here, and it might be the XY-problem, but if you really need to use Pandas with Structured Streaming, you can try using pandas_udf - SCALAR and GROUPED_MAP variants are compatible with at least basic time based triggers (other variants might be supported as well, though some combinations clearly don't make any sense, and I am not aware of any official compatibility matrix).