用avro模式(avsc)在Java Spark作业中向s3写入avro数据

Question

{
  "type" : "record",
  "name" : "name1",
  "namespace" : "com.data"
  "fields" : [
  {
    "name" : "id",
    "type" : "string"
  },
  {
    "name" : "count",
    "type" : "int"
  },
  {
    "name" : "val_type",
    "type" : {
      "type" : "enum",
      "name" : "ValType"
      "symbols" : [ "s1", "s2" ]
    }
  }
  ]
}

当我尝试基于avro模式将avro数据写入s3时

DF数据类型：

root
 |-- id: string (nullable = true)
 |-- count: integer (nullable = true)
 |-- val_type: string (nullable = true)

finaldf.write（）.option（“avroschema”,string.valueof（inAvroSchema））.format（“com.databricks.spark.avro”）.mode（“overwrite”）.save（“target_s3_path”）；

我得到了错误：

User class threw exception: org.apache.spark.SparkException: Job aborted.
    ......
    Caused by: org.apache.avro.AvroRuntimeException: **Not a union: "string"**
        at org.apache.avro.Schema.getTypes(Schema.java:299)
        at 
org.apache.spark.sql.avro.AvroSerializer.org$apache$spark$sql$avro$AvroSerializer$$resolveNullableType(AvroSerializer.scala:229)

Answer 1

我在附近挖了一些东西，发现了一些有趣的东西，

case class Name1(id: String, count: Int, val_type: String)

val schema = """{
                   |  "type" : "record",
                   |  "name" : "name1",
                   |  "namespace" : "com.data",
                   |  "fields" : [
                   |  {
                   |    "name" : "id",
                   |    "type" : "string"
                   |  },
                   |  {
                   |    "name" : "count",
                   |    "type" : "int"
                   |  },
                   |  {
                   |    "name" : "val_type",
                   |    "type" : {
                   |      "type" : "enum",
                   |      "name" : "ValType",
                   |      "symbols" : [ "s1", "s2" ]
                   |    }
                   |  }
                   |  ]
                   |}""".stripMargin


val d = Seq(Name1("1",2,"s1"),Name1("1",3,"s2"),Name1("1",4,"s2"),Name1("11",2,"s1")).toDF()

d.write.mode(SaveMode.Overwrite).format("avro").option("avroSchema",schema).save("data/tes2/")

当我使用Spark2.4.x执行代码时，上面的代码失败了，但是当我使用新的Spark3.0.0运行相同的代码时，代码成功了，数据被成功写入。

val df = spark.read.format("avro").load("data/tes2/")
df.printSchema()
df.show(10)

root
 |-- id: string (nullable = true)
 |-- count: integer (nullable = true)
 |-- val_type: string (nullable = true)

+---+-----+--------+
| id|count|val_type|
+---+-----+--------+
| 11|    2|      s1|
|  1|    4|      s2|
|  1|    3|      s2|
|  1|    2|      s1|
+---+-----+--------+

我想最好的办法是升级spark版本或更改avro模式定义。