1. registerTempTable 与createOrReplaceTempView
以下面数据JSON为案例people.json：
{
"name": "John",
"age": "28",
"country":
"UK"
}
{
"name": "Cathy",
"age": "30",
"country": "AUS"
}
{
"name": "Mark",
"age": "50",
"country": "USA"
}
|
输入:%spark2
val peopleJsonDF = sqlContext.read.format("json").load("/tests/people.json");
peopleJsonDF.createOrReplaceTempView("jsonPeople");
sqlContext.sql("select * from jsonPeople where age > 30").show();
|
Zeppelin output:
peopleJsonDF: org.apache.spark.sql.DataFrame = [age: string, country: string ... 1 more field]
+---+-------+----+
|age|country|name|
+---+-------+----+
| 50| USA|Mark|
+---+-------+----+
|
结果：registerTempTable是Sparn 1.x语法，createOrReplaceTempView是Spark 2.x语法。两者都做同样的事情。
2. Apache Spark 2中的Spark-csv
任务：从JSON文件中获取数据，进行一些排序并将结果保存为CSV文件。Apache Spark 2在这里非常方便！
%spark2
val peopleJsonDF = sqlContext.read.option("multiline", true).format("json").load("/tests/people.json");
peopleJsonDF.createOrReplaceTempView("jsonPeople");
val orderedJsonDF = sqlContext.sql(" select * from jsonPeople ORDER BY age DESC ");
orderedJsonDF.show();
orderedJsonDF.write.mode("overwrite").option("sep", "|").option("header", true).format("csv").save("/tests/csvresult");
|
Zeppelin output:
peopleJsonDF: org.apache.spark.sql.DataFrame = [age: string, country: string ... 1 more field]
orderedJsonDF: org.apache.spark.sql.DataFrame = [age: string, country: string ... 1 more field]
+---+-------+-----+
|age|country| name|
+---+-------+-----+
| 50| USA| Mark|
| 30| AUS|Cathy|
| 28| UK| John|
+---+-------+-----+
|
HDFS output:
[root@sandbox ~]# hdfs dfs -ls /tests/csvresult
Found 5 items
-rw-r--r-- 1 zeppelin hdfs 0 2018-09-04 17:14 /tests/csvresult/_SUCCESS
-rw-r--r-- 1 zeppelin hdfs 29 2018-09-04 17:14 /tests/csvresult/part-00000-4f4204f7-636b-4f7c-a8b7-a9a8a4c395cf.csv
-rw-r--r-- 1 zeppelin hdfs 30 2018-09-04 17:14 /tests/csvresult/part-00001-4f4204f7-636b-4f7c-a8b7-a9a8a4c395cf.csv
-rw-r--r-- 1 zeppelin hdfs 28 2018-09-04 17:14 /tests/csvresult/part-00002-4f4204f7-636b-4f7c-a8b7-a9a8a4c395cf.csv
-rw-r--r-- 1 zeppelin hdfs 0 2018-09-04 17:14 /tests/csvresult/part-00003-4f4204f7-636b-4f7c-a8b7-a9a8a4c395cf.csv
|
3.从CSV中推断模式并将结果保存到ORC中并通过Hive将其恢复
让我们将之前的结果'/ tests / csvresult'作为此任务的输入。
%spark2
import org.apache.spark.sql.hive.HiveContext;
import org.apache.spark.sql.types._;
var hiveContext = new HiveContext(sc);
var csvDF = sqlContext.read
.option("header", true)
.option("sep", "|")
.option("inferSchema", true)
.format("csv").load("/tests/csvresult");
csvDF.createOrReplaceTempView("csvTable");
var sortedDF = sqlContext.sql("select age,country,name from csvTable order by age desc");
sortedDF.write.mode("overwrite").format("orc").save("/tests/orcresult");
hiveContext.sql(" DROP TABLE IF EXISTS people ");
hiveContext.sql("CREATE EXTERNAL TABLE people (age INT, country String, name String) "+
" STORED AS ORC "+
" LOCATION '/tests/orcresult'"
);
hiveContext.sql("select * from people").show();
|
Zeppelin output:
import org.apache.spark.sql.hive.HiveContext
import org.apache.spark.sql.types._
warning: there was one deprecation warning; re-run with -deprecation for details
hiveContext: org.apache.spark.sql.hive.HiveContext = org.apache.spark.sql.hive.HiveContext@4fc938b8
csvDF: org.apache.spark.sql.DataFrame = [age: int, country: string ... 1 more field]
sortedDF: org.apache.spark.sql.DataFrame = [age: int, country: string ... 1 more field]
res135: org.apache.spark.sql.DataFrame = []
res136: org.apache.spark.sql.DataFrame = []
+---+-------+-----+
|age|country| name|
+---+-------+-----+
| 50| USA| Mark|
| 30| AUS|Cathy|
| 28| UK| John|
+---+-------+-----+
|
4.删除格式错误的记录
假设有文件malformed.csv：
50|USA|Mark
30|AUS|Cathy
28|UK
|
并且任务是删除格式错误的记录并创建新的DF：
%spark2
val schema = StructType(Array(
StructField("age", IntegerType, false),
StructField("country", StringType, false),
StructField("name", StringType, false)
));
val malformedDF = sqlContext.read.format("csv")
.schema(schema)
.option("mode", "DROPMALFORMED")
.option("header", false)
.option("sep", "|")
.load("/tests/malformed.csv");
malformedDF.show();
|
Zeppelin output:
schema: org.apache.spark.sql.types.StructType = StructType(StructField(age,IntegerType,false), StructField(country,StringType,false), StructField(name,StringType,false))
malformedDF: org.apache.spark.sql.DataFrame = [age: int, country: string ... 1 more field]
+---+-------+-----+
|age|country| name|
+---+-------+-----+
| 50| USA| Mark|
| 30| AUS|Cathy|
+---+-------+-----+
|