apache-spark - 列数据到Spark结构化流中的嵌套json对象

Question

在我们的应用程序中，我们使用Spark sql获取字段值作为列。我正在尝试弄清楚如何将列值放入嵌套的json对象并推送到Elasticsearch。还有一种方法可以参数化selectExpr中的值以传递给正则表达式？

我们目前正在使用Spark Java API。

Dataset<Row> data = rowExtracted.selectExpr("split(value,\"[|]\")[0] as channelId",
                "split(value,\"[|]\")[1] as country",
                "split(value,\"[|]\")[2] as product",
                "split(value,\"[|]\")[3] as sourceId",
                "split(value,\"[|]\")[4] as systemId",
                "split(value,\"[|]\")[5] as destinationId",
                "split(value,\"[|]\")[6] as batchId",
                "split(value,\"[|]\")[7] as orgId",
                "split(value,\"[|]\")[8] as businessId",
                "split(value,\"[|]\")[9] as orgAccountId",
                "split(value,\"[|]\")[10] as orgBankCode",
                "split(value,\"[|]\")[11] as beneAccountId",
                "split(value,\"[|]\")[12] as beneBankId",
                "split(value,\"[|]\")[13] as currencyCode",
                "split(value,\"[|]\")[14] as amount",
                "split(value,\"[|]\")[15] as processingDate",
                "split(value,\"[|]\")[16] as status",
                "split(value,\"[|]\")[17] as rejectCode",
                "split(value,\"[|]\")[18] as stageId",
                "split(value,\"[|]\")[19] as stageStatus",
                "split(value,\"[|]\")[20] as stageUpdatedTime",
                "split(value,\"[|]\")[21] as receivedTime",
                "split(value,\"[|]\")[22] as sendTime"
        );

StreamingQuery query = data.writeStream()
                .outputMode(OutputMode.Append()).format("es").option("checkpointLocation", "C:\\checkpoint")
                .start("spark_index/doc")

实际输出:

{
  "_index": "spark_index",
  "_type": "doc",
  "_id": "test123",
  "_version": 1,
  "_score": 1,
  "_source": {
    "channelId": "test",
    "country": "SG",
    "product": "test",
    "sourceId": "",
    "systemId": "test123",
    "destinationId": "",
    "batchId": "",
    "orgId": "test",
    "businessId": "test",
    "orgAccountId": "test",
    "orgBankCode": "",
    "beneAccountId": "test",
    "beneBankId": "test",
    "currencyCode": "SGD",
    "amount": "53.0000",
    "processingDate": "",
    "status": "Pending",
    "rejectCode": "test",
    "stageId": "123",
    "stageStatus": "Comment",
    "stageUpdatedTime": "2019-08-05 18:11:05.999000",
    "receivedTime": "2019-08-05 18:10:12.701000",
    "sendTime": "2019-08-05 18:11:06.003000"
  }
}

我们需要在节点“txn_summary”下的上述列，例如以下json:

预期产量:

{
  "_index": "spark_index",
  "_type": "doc",
  "_id": "test123",
  "_version": 1,
  "_score": 1,
  "_source": {
    "txn_summary": {
      "channelId": "test",
      "country": "SG",
      "product": "test",
      "sourceId": "",
      "systemId": "test123",
      "destinationId": "",
      "batchId": "",
      "orgId": "test",
      "businessId": "test",
      "orgAccountId": "test",
      "orgBankCode": "",
      "beneAccountId": "test",
      "beneBankId": "test",
      "currencyCode": "SGD",
      "amount": "53.0000",
      "processingDate": "",
      "status": "Pending",
      "rejectCode": "test",
      "stageId": "123",
      "stageStatus": "Comment",
      "stageUpdatedTime": "2019-08-05 18:11:05.999000",
      "receivedTime": "2019-08-05 18:10:12.701000",
      "sendTime": "2019-08-05 18:11:06.003000"
    }
  }
}

Answer 1

将所有列添加到顶层结构应提供预期的输出。在Scala中:

import org.apache.spark.sql.functions._

data.select(struct(data.columns:_*).as("txn_summary"))

在Java中，我怀疑是这样的:

import org.apache.spark.sql.functions.struct;

data.select(struct(data.columns()).as("txn_summary"));