Last active
November 7, 2015 14:29
-
-
Save schon/b49a616eeb0b78fb5f47 to your computer and use it in GitHub Desktop.
Workaround a equality bug of data frame in spark 1.5.1
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import org.apache.spark.sql.SQLContext | |
import org.apache.spark.sql.functions._ | |
import org.apache.spark.{SparkConf, SparkContext} | |
object SparkDFTest { | |
def main(args: Array[String]) { | |
val conf = new SparkConf().setAppName("SparkDFTest") | |
val sc = new SparkContext(conf) | |
val sqlContext = new SQLContext(sc) | |
import sqlContext.implicits._ | |
val eventTableColumns = Seq[String]( | |
"entityType" | |
, "entityId" | |
, "targetEntityType" | |
, "targetEntityId" | |
, "properties" | |
, "eventTime") | |
// insert event | |
/** | |
* input csv format without header | |
* | |
* 1223140803222504478701,361804026,buy,1,2015-11-01T05:40:20+09:00,ib_user | |
* 1223140803222504478701,361804026,buy,1,2015-11-01T05:40:20+09:00,ib_user | |
* 2611150613134148167296,431634010,buy,1,2015-11-01T16:13:18+09:00,user | |
* 2611151029200004176515,349617013,buy,1,2015-11-01T16:51:09+09:00,user | |
*/ | |
val eventDF = sc.textFile("events_s.csv").map(_.split(",")).filter(_.size >= 6) | |
.map { e => | |
( | |
e(5), e(0), "item", e(1), s"""{"rating": ${e(3).trim.toDouble}}""", e(3) | |
) | |
}.toDF(eventTableColumns:_*).cache() | |
// print schema | |
eventDF.printSchema() | |
/** | |
* root | |
* |-- entityType: string (nullable = true) | |
* |-- entityId: string (nullable = true) | |
* |-- targetEntityType: string (nullable = true) | |
* |-- targetEntityId: string (nullable = true) | |
* |-- properties: string (nullable = true) | |
* |-- eventTime: string (nullable = true) | |
*/ | |
// SELECT entityType, count(distinct entityId) as count FROM dataframe GROUP BY entityType | |
eventDF.groupBy("entityType").agg(countDistinct("entityId").alias("count")).show | |
/** | |
* +----------+-----+ | |
* |entityType|count| | |
* +----------+-----+ | |
* | ib_user| 4751| | |
* | user| 2091| | |
* +----------+-----+ | |
*/ | |
/** | |
* eventDF.filter($"entityType" === "user") does NOT work | |
* comparison '===' has bug in spark 1.5.1. | |
* It would be fixed on next versions. | |
* | |
* @see [[https://issues.apache.org/jira/browse/SPARK-10859]] | |
*/ | |
eventDF.filter($"entityType" === "user").select("entityId").distinct.count | |
/** | |
* Wrong Result | |
* | |
* scala> eventDF.filter($"entityType" === "user").select("entityId").distinct.count | |
* res56: Long = 1219 | |
*/ | |
/** | |
* two workarounds this bug. | |
* | |
* First is to use `isin` function. | |
*/ | |
eventDF.filter($"entityType" isin lit("user")).select("entityId").distinct.count | |
/** | |
* scala> eventDF.filter($"entityType" isin lit("user")).select("entityId").distinct.count | |
* res57: Long = 2091 | |
*/ | |
/** | |
* Second is to use case class instead of toDF("col1", "col2", ...) | |
*/ | |
val eventDF2 = sc.textFile("events_s.csv").map(_.split(",")).filter(_.size >= 6) | |
.map { e => | |
Event( | |
e(5), e(0), "item", e(1), s"""{"rating": ${e(3).trim.toDouble}}""", e(3) | |
) | |
}.toDF() | |
eventDF2.filter($"entityType" === "user").select("entityId").distinct.count | |
/** | |
* scala> eventDF2.filter($"entityType" === "user").select("entityId").distinct.count | |
* res58: Long = 2091 | |
*/ | |
/** | |
* Third is to turn off "spark.sql.inMemoryColumnarStorage.partitionPruning" | |
*/ | |
sqlContext.sql("SET spark.sql.inMemoryColumnarStorage.partitionPruning=false") | |
eventDF.filter($"entityType" === "user").select("entityId").distinct.count | |
/** | |
* scala> sqlContext.sql("SET spark.sql.inMemoryColumnarStorage.partitionPruning=false") | |
* res59: org.apache.spark.sql.DataFrame = [key: string, value: string] | |
* | |
* scala> eventDF.filter($"entityType" === "user").select("entityId").distinct.count | |
* res60: Long = 2091 | |
*/ | |
sc.stop() | |
} | |
case class Event(entityType: String, entityId: String, | |
targetEntityType: String, targetEntityId: String, | |
properties: String, eventTime: String) | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment