[
https://issues.apache.org/jira/browse/SPARK-18753?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Apache Spark reassigned SPARK-18753:
------------------------------------
Assignee: Apache Spark
> Inconsistent behavior after writing to parquet files
> ----------------------------------------------------
>
> Key: SPARK-18753
> URL: https://issues.apache.org/jira/browse/SPARK-18753
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.0.2, 2.1.0
> Reporter: Shixiong Zhu
> Assignee: Apache Spark
>
> Found an inconsistent behavior when using parquet.
> {code}
> scala> val ds = Seq[java.lang.Boolean](new java.lang.Boolean(true), null:
> java.lang.Boolean, new java.lang.Boolean(false)).toDS
> ds: org.apache.spark.sql.Dataset[Boolean] = [value: boolean]
> scala> ds.filter('value === "true").show
> +-----+
> |value|
> +-----+
> +-----+
> {code}
> In the above example, `ds.filter('value === "true")` returns nothing as
> "true" will be converted to null and the filter expression will be always
> null, so it drops all rows.
> However, if I store `ds` to a parquet file and read it back, `filter('value
> === "true")` will return non null values.
> {code}
> scala> ds.write.parquet("testfile")
> SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
> SLF4J: Defaulting to no-operation (NOP) logger implementation
> SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further
> details.
> scala> val ds2 = spark.read.parquet("testfile")
> ds2: org.apache.spark.sql.DataFrame = [value: boolean]
> scala> ds2.filter('value === "true").show
> +-----+
> |value|
> +-----+
> | true|
> |false|
> +-----+
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
