sparksql parquet 合并元数据

java

 1 public class ParquetMergeSchema {
 2     private static SparkConf conf = new SparkConf().setAppName("parquetmergeschema").setMaster("local");
 3     private static JavaSparkContext jsc = new JavaSparkContext(conf);
 4     private static SparkSession session = new SparkSession(jsc.sc());
 5
 6     public static void main(String[] args) {
 7         JavaRDD<Tuple2<String, Object>> rdd1 = jsc.parallelize(
 8                 Arrays.asList(new Tuple2<String, Object>("jack", 21), new Tuple2<String, Object>("lucy", 20)));
 9
10         JavaRDD<Row> row1 = rdd1.map(new Function<Tuple2<String, Object>, Row>() {
11
12             private static final long serialVersionUID = 1L;
13
14             @Override
15             public Row call(Tuple2<String, Object> v1) throws Exception {
16                 return RowFactory.create(v1._1, v1._2);
17             }
18         });
19
20         JavaRDD<Tuple2<String, Object>> rdd2 = jsc.parallelize(
21                 Arrays.asList(new Tuple2<String, Object>("jack", "A"), new Tuple2<String, Object>("yeye", "B")));
22
23         JavaRDD<Row> row2 = rdd2.map(new Function<Tuple2<String, Object>, Row>() {
24
25             private static final long serialVersionUID = 1L;
26
27             @Override
28             public Row call(Tuple2<String, Object> v1) throws Exception {
29                 return RowFactory.create(v1._1, v1._2);
30             }
31         });
32
33         StructType schema1 = DataTypes
34                 .createStructType(Arrays.asList(DataTypes.createStructField("name", DataTypes.StringType, false),
35                         DataTypes.createStructField("age", DataTypes.IntegerType, false)));
36
37         StructType schema2 = DataTypes
38                 .createStructType(Arrays.asList(DataTypes.createStructField("name", DataTypes.StringType, false),
39                         DataTypes.createStructField("grade", DataTypes.StringType, false)
40
41                 ));
42
43         // 将rdd转成dataset
44         Dataset<Row> ds1 = session.createDataFrame(row1, schema1);
45
46         Dataset<Row> ds2 = session.createDataFrame(row2, schema2);
47
48         // 保存为parquet文件
49         ds1.write().mode(SaveMode.Append).save("./src/main/java/cn/tele/spark_sql/parquet/mergetest");
50         ds2.write().mode(SaveMode.Append).save("./src/main/java/cn/tele/spark_sql/parquet/mergetest");
51
52         // 指定parquet文件的目录进行读取,设置mergeSchema为true进行合并
53         Dataset<Row> dataset = session.read().option("mergeSchema", true)
54                 .load("./src/main/java/cn/tele/spark_sql/parquet/mergetest");
55
56         dataset.printSchema();
57         dataset.show();
58
59         session.stop();
60         jsc.close();
61
62     }
63 }

scala

 1 object ParquetMergeSchema {
 2   def main(args: Array[String]): Unit = {
 3     val conf = new SparkConf().setAppName("parquetmergeschema").setMaster("local")
 4     val sc = new SparkContext(conf)
 5     val sqlContext = new SQLContext(sc)
 6
 7     val rdd1 = sc.parallelize(Array(("jack", 18), ("tele", 20)), 2).map(tuple => { Row(tuple._1, tuple._2) })
 8     val rdd2 = sc.parallelize(Array(("tele", "A"), ("wyc", "A"), ("yeye", "C")), 2).map(tuple => { Row(tuple._1, tuple._2) })
 9
10     //schema
11     val schema1 = DataTypes.createStructType(Array(
12       StructField("name", DataTypes.StringType, false),
13       StructField("age", DataTypes.IntegerType, false)))
14
15     val schema2 = DataTypes.createStructType(Array(
16       StructField("name", DataTypes.StringType, false),
17       StructField("grade", DataTypes.StringType, false)))
18
19     //转换
20     val df1 = sqlContext.createDataFrame(rdd1, schema1)
21     val df2 = sqlContext.createDataFrame(rdd2, schema2)
22
23     //写出
24     df1.write.mode(SaveMode.Append).save("./src/main/scala/cn/tele/spark_sql/parquet/mergetest")
25     df2.write.mode(SaveMode.Append).save("./src/main/scala/cn/tele/spark_sql/parquet/mergetest")
26
27     //读取进行合并
28     val df = sqlContext.read.option("mergeSchema", true).parquet("./src/main/scala/cn/tele/spark_sql/parquet/mergetest")
29     df.printSchema()
30     df.show()
31   }
32 }

原文地址：https://www.cnblogs.com/tele-share/p/10390972.html