[CARBONDATA-3852] CCD Merge with Partition Table is giving different results in different spark deploy modes

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.0
• Fix Version/s: 2.1.0
• Component/s: spark-integration
• Assignee: Unassigned
• Reporter: Sachin Ramachandra Setty
• Created: 11/Jun/20 14:18
• Updated: 29/Oct/20 13:08
• Resolved: 13/Jul/20 18:01

Description

The result sets are different when run the sql queries in spark-shell --master local and spark-shell --master yarn (Two Different Spark Deploy Modes)

import scala.collection.JavaConverters._
import java.sql.Date
import org.apache.spark.sql._
import org.apache.spark.sql.CarbonSession._
import org.apache.spark.sql.catalyst.TableIdentifier
import org.apache.spark.sql.execution.command.mutation.merge.{CarbonMergeDataSetCommand, DeleteAction, InsertAction, InsertInHistoryTableAction, MergeDataSetMatches, MergeMatch, UpdateAction, WhenMatched, WhenNotMatched, WhenNotMatchedAndExistsOnlyOnTarget}

import org.apache.spark.sql.functions._
import org.apache.spark.sql.test.util.QueryTest
import org.apache.spark.sql.types.{BooleanType, DateType, IntegerType, StringType, StructField, StructType}
import spark.implicits._

val df1 = sc.parallelize(1 to 10, 4).map{ x => ("id"+x, s"order$x",s"customer$x", x*10, x*75, 1)}.toDF("id", "name", "c_name", "quantity", "price", "state")


df1.write.format("carbondata").option("tableName", "order").mode(SaveMode.Overwrite).save()
val dwframe = spark.read.format("carbondata").option("tableName", "order").load()
val dwSelframe = dwframe.as("A")


val ds1 = sc.parallelize(3 to 10, 4)
      .map {x =>
        if (x <= 4) {
          ("id"+x, s"order$x",s"customer$x", x*10, x*75, 2)
        } else {
          ("id"+x, s"order$x",s"customer$x", x*10, x*75, 1)
        }
      }.toDF("id", "name", "c_name", "quantity", "price", "state")
	  

val ds2 = sc.parallelize(1 to 2, 4).map {x => ("newid"+x, s"order$x",s"customer$x", x*10, x*75, 1)}.toDS().toDF()
val ds3 = ds1.union(ds2)	
val odsframe = ds3.as("B")
	
sql("drop table if exists target").show()

val initframe = spark.createDataFrame(Seq(
  Row("a", "0"),
  Row("b", "1"),
  Row("c", "2"),
  Row("d", "3")
).asJava, StructType(Seq(StructField("key", StringType), StructField("value", StringType))))

initframe.write
  .format("carbondata")
  .option("tableName", "target")
  .option("partitionColumns", "value")
  .mode(SaveMode.Overwrite)
  .save()
  
val target = spark.read.format("carbondata").option("tableName", "target").load()

var ccd =
  spark.createDataFrame(Seq(
    Row("a", "10", false,  0),
    Row("a", null, true, 1),   
    Row("b", null, true, 2),   
    Row("c", null, true, 3),   
    Row("c", "20", false, 4),
    Row("c", "200", false, 5),
    Row("e", "100", false, 6) 
  ).asJava,
    StructType(Seq(StructField("key", StringType),
      StructField("newValue", StringType),
      StructField("deleted", BooleanType), StructField("time", IntegerType))))
	  
ccd.createOrReplaceTempView("changes")

ccd = sql("SELECT key, latest.newValue as newValue, latest.deleted as deleted FROM ( SELECT key, max(struct(time, newValue, deleted)) as latest FROM changes GROUP BY key)")

val updateMap = Map("key" -> "B.key", "value" -> "B.newValue").asInstanceOf[Map[Any, Any]]

val insertMap = Map("key" -> "B.key", "value" -> "B.newValue").asInstanceOf[Map[Any, Any]]

target.as("A").merge(ccd.as("B"), "A.key=B.key").
  whenMatched("B.deleted=false").
  updateExpr(updateMap).
  whenNotMatched("B.deleted=false").
  insertExpr(insertMap).
  whenMatched("B.deleted=true").
  delete().execute()
  

SQL Queries to run :

 
sql("select count(*) from target").show()
sql("select * from target order by key").show()

Results in spark-shell --master yarn

scala> sql("select count(*) from target").show()
+--------+
|count(1)|
+--------+
|       4|
+--------+


scala> sql("select * from target order by key").show()
+---+-----+
|key|value|
+---+-----+
|  a|    0|
|  b|    1|
|  c|    2|
|  d|    3|
+---+-----+

Results in spark-shell --master local

scala> sql("select count(*) from target").show()
+--------+
|count(1)|
+--------+
|       3|
+--------+


scala> sql("select * from target order by key").show()
+---+-----+
|key|value|
+---+-----+
|  c|  200|
|  d|    3|
|  e|  100|
+---+-----+