[jira] [Commented] (SPARK-32342) Kafka events are missing magic byte

["Jakub Waller (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-32342?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17213886#comment-17213886 ] 

Jakub Waller commented on SPARK-32342:
--------------------------------------

Here are the details about the confluent format [https://docs.confluent.io/current/schema-registry/serdes-develop/index.html#wire-format]

Would be nice if the *from_avro* function also supported that.

I understand it's still experimental but the combination of Spark Streaming + Avro + Kafka Producers/Consumers is unusable at the moment.

I tested with Scala 2.12.10, Spark 3.0.1, Confluent 5.4.1, Kafka 2.4.1, Avro 1.9.1.

 

P.S. Seems to work with this [https://github.com/AbsaOSS/ABRiS]

> Kafka events are missing magic byte
> -----------------------------------
>
>                 Key: SPARK-32342
>                 URL: https://issues.apache.org/jira/browse/SPARK-32342
>             Project: Spark
>          Issue Type: Bug
>          Components: Structured Streaming
>    Affects Versions: 3.0.0
>         Environment: Pyspark 3.0.0, Python 3.7 Confluent cloud Kafka with Schema registry 5.5 
>            Reporter: Sridhar Baddela
>            Priority: Major
>
> Please refer to the documentation link for to_avro and from_avro.[ http://spark.apache.org/docs/latest/sql-data-sources-avro.html|http://spark.apache.org/docs/latest/sql-data-sources-avro.html]
> Tested the to_avro function by making sure that data is sent to Kafka topic. But when a Confluent Avro consumer is used to read data from the same topic, the consumer fails with an error message that event is missing the magic byte. 
> Used another topic to simulate reads from Kafka and further deserialization using from_avro. Use case is, use a Confluent Avro producer to produce a few events. And when I attempt to read this topic using structured streaming and applying the function from_avro, it fails with a message indicating that malformed records are present. 
> Using from_avro (deserialization) and to_avro (serialization) from Spark, only works with Spark. And other consumers outside of Spark which do not use this approach are failing.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org