[
https://issues.apache.org/jira/browse/SPARK-21021?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16047815#comment-16047815
]
jin xing commented on SPARK-21021:
----------------------------------
I think the reason of the incompatibility is that the {{userSpecifiedSchema}}
is split to two parts -- {{dataSchema}}(f3) and {{partitionSchema}}(f1, f2) in
{{HadoopFsRelation}}. And the schema of {{HadoopFsRelation}} is {{dataSchema}}
++ {{partitionSchema}}.
I guess this is not a problem?
You can rectify the incompatibility by {{dfRead.select("f1", "f2", "f3")}}
> Reading partitioned parquet does not respect specified schema column order
> --------------------------------------------------------------------------
>
> Key: SPARK-21021
> URL: https://issues.apache.org/jira/browse/SPARK-21021
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.1.0
> Reporter: Michel Lemay
> Priority: Minor
>
> When reading back a partitioned parquet folder, column order gets messed up.
> Consider the following example:
> {code}
> case class Event(f1: String, f2: String, f3: String)
> val df = Seq(Event("v1", "v2", "v3")).toDF
> df.write.partitionBy("f1", "f2").parquet("out")
> val schema: StructType = StructType(StructField("f1", StringType, true) ::
> StructField("f2", StringType, true) :: StructField("f3", StringType, true) ::
> Nil)
> val dfRead = spark.read.schema(schema).parquet("out")
> dfRead.show
> +---+---+---+
> | f3| f1| f2|
> +---+---+---+
> | v3| v1| v2|
> +---+---+---+
> dfRead.columns
> Array[String] = Array(f3, f1, f2)
> schema.fields
> Array(StructField(f1,StringType,true), StructField(f2,StringType,true),
> StructField(f3,StringType,true))
> {code}
> This makes it really hard to have compatible schema when reading from
> multiple sources.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
