You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Johan Lasperas (Jira)" <ji...@apache.org> on 2023/11/24 14:53:00 UTC
[jira] [Updated] (SPARK-46092) Overflow in Parquet row group filter creation causes incorrect results
[ https://issues.apache.org/jira/browse/SPARK-46092?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Johan Lasperas updated SPARK-46092:
-----------------------------------
Description:
While the parquet readers don't support reading parquet values into larger Spark types, it's possible to trigger an overflow when creating a Parquet row group filter that will then incorrectly skip row groups and bypass the exception in the reader,
Repro:
{code:java}
Seq(0).toDF("a").write.parquet(path)
spark.read.schema("a LONG").parquet(path).where(s"a < ${Long.MaxValue}").collect(){code}
This succeeds and returns no results. This should either fail if the Parquet reader doesn't support the upcast from int to long or produce result `[0]` if it does.
was:
While the parquet readers don't support reading parquet values into larger Spark types, it's possible to trigger an overflow when creating a Parquet row group filter that will then incorrectly skip row groups and bypass the exception in the reader,
Repro:
```
Seq(0).toDF("a").write.parquet(path)
spark.read.schema("a LONG").parquet(path).where(s"a < ${Long.MaxValue}").collect()
```
This succeeds and returns no results. This should either fail if the Parquet reader doesn't support the upcast from int to long or produce result `[0]` if it does.
> Overflow in Parquet row group filter creation causes incorrect results
> ----------------------------------------------------------------------
>
> Key: SPARK-46092
> URL: https://issues.apache.org/jira/browse/SPARK-46092
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 3.5.0
> Reporter: Johan Lasperas
> Priority: Major
>
> While the parquet readers don't support reading parquet values into larger Spark types, it's possible to trigger an overflow when creating a Parquet row group filter that will then incorrectly skip row groups and bypass the exception in the reader,
> Repro:
> {code:java}
> Seq(0).toDF("a").write.parquet(path)
> spark.read.schema("a LONG").parquet(path).where(s"a < ${Long.MaxValue}").collect(){code}
> This succeeds and returns no results. This should either fail if the Parquet reader doesn't support the upcast from int to long or produce result `[0]` if it does.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org