你可以这样做：

import org.apache.spark.sql.functions.udf

// extract the schema id (bytes 1-4) and convert it to integer
val extractSchemaId = udf((message: Array[Byte]) => {
  ByteBuffer.wrap(message.slice(1, 5)).getInt()
})

// assuming the "value" column holds the Avro message
df.withColumn("schema_id", extractSchemaId(df("value")))