[
https://issues.apache.org/jira/browse/SPARK-24233?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon updated SPARK-24233:
---------------------------------
Description:
I know that I can use wild card * to read all subfolders. But, I am trying to
use .par and .schema to speed up the read process.
{code}
val absolutePath = "adl://datalakename.azuredatalakestore.net/testU/"
Seq((1, "one"), (2, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "1")
Seq((11, "one"), (22, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "2")
Seq((111, "one"), (222, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "3")
Seq((1111, "one"), (2222, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "4")
Seq((2, "one"), (2, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "5")
{code}
{code}
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs.{FileSystem, Path}
import java.net.URI
def readDir(path: String): DataFrame = {
val fs = FileSystem.get(new URI(path), new Configuration())
val subDir = fs.listStatus(new Path(path)).map(i => i.getPath.toString)
var df = spark.read.parquet(subDir.head)
val dfSchema = df.schema subDir.tail.par.foreach(p => df =
df.union(spark.read.schema(dfSchema).parquet(p)).select(df.columns.head,
df.columns.tail:_*))
df
}
val dfAll = readDir(absolutePath)
dfAll.count
{code}
The count of produced dfAll is 4, which in this example should be 10.
was:
I know that I can use wild card * to read all subfolders. But, I am trying to
use .par and .schema to speed up the read process.
{code}
val absolutePath = "adl://datalakename.azuredatalakestore.net/testU/"
Seq((1, "one"), (2, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "1")
Seq((11, "one"), (22, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "2")
Seq((111, "one"), (222, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "3")
Seq((1111, "one"), (2222, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "4")
Seq((2, "one"), (2, "two")).toDF("k",
"v").write.mode("overwrite").parquet(absolutePath + "5")
{code}
{code}
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs.\{FileSystem, Path}
import java.net.URI
def readDir(path: String): DataFrame =
{ val fs = FileSystem.get(new URI(path), new Configuration()) val subDir =
fs.listStatus(new Path(path)).map(i => i.getPath.toString) var df =
spark.read.parquet(subDir.head) val dfSchema = df.schema
subDir.tail.par.foreach(p => df =
df.union(spark.read.schema(dfSchema).parquet(p)).select(df.columns.head,
df.columns.tail:_*)) df }
val dfAll = readDir(absolutePath)
dfAll.count
{code}
The count of produced dfAll is 4, which in this example should be 10.
> Union Operation on Read of Dataframe does NOT produce correct result
> ---------------------------------------------------------------------
>
> Key: SPARK-24233
> URL: https://issues.apache.org/jira/browse/SPARK-24233
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.1.0
> Reporter: smohr003
> Priority: Major
>
> I know that I can use wild card * to read all subfolders. But, I am trying to
> use .par and .schema to speed up the read process.
> {code}
> val absolutePath = "adl://datalakename.azuredatalakestore.net/testU/"
> Seq((1, "one"), (2, "two")).toDF("k",
> "v").write.mode("overwrite").parquet(absolutePath + "1")
> Seq((11, "one"), (22, "two")).toDF("k",
> "v").write.mode("overwrite").parquet(absolutePath + "2")
> Seq((111, "one"), (222, "two")).toDF("k",
> "v").write.mode("overwrite").parquet(absolutePath + "3")
> Seq((1111, "one"), (2222, "two")).toDF("k",
> "v").write.mode("overwrite").parquet(absolutePath + "4")
> Seq((2, "one"), (2, "two")).toDF("k",
> "v").write.mode("overwrite").parquet(absolutePath + "5")
> {code}
>
> {code}
> import org.apache.hadoop.conf.Configuration
> import org.apache.hadoop.fs.{FileSystem, Path}
> import java.net.URI
> def readDir(path: String): DataFrame = {
> val fs = FileSystem.get(new URI(path), new Configuration())
> val subDir = fs.listStatus(new Path(path)).map(i => i.getPath.toString)
> var df = spark.read.parquet(subDir.head)
> val dfSchema = df.schema subDir.tail.par.foreach(p => df =
> df.union(spark.read.schema(dfSchema).parquet(p)).select(df.columns.head,
> df.columns.tail:_*))
> df
> }
> val dfAll = readDir(absolutePath)
> dfAll.count
> {code}
> The count of produced dfAll is 4, which in this example should be 10.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
