The SparkStreaming operator fails to execute

[zenvzenv]

kafka as SparkStreaming input and output

1. Use spark.readStream.format("kafka") read kafka data and decode binary data to string
2. Use df.map(_.Seq.foldLeft(""))(_ + separtor + _).writeStream("kafka") output data to kafka
3. If I fail to output to kafka, then no matter how I change the kafka topic later, the stream computing will fail,ArrayIndexOutOfBoundsException: 1 exception will report.If I only output to the console there will be no error
4. If I run the same code snippet directly in spark-shell without using livy, the effect is the same as 3