[
https://issues.apache.org/jira/browse/SPARK-42704?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ryan Johnson updated SPARK-42704:
---------------------------------
Description:
The `AddMetadataColumns` analyzer rule intends to make resolve available
metadata columns, even if the plan already contains projections that did not
explicitly mention the metadata column.
The `SubqueryAlias` plan node intentionally does not propagate metadata columns
automatically from a non-leaf/non-subquery child node, because the following
should _not_ work:
{code:java}
spark.read.table("t").select("a", "b").as("s").select("_metadata"){code}
However, today it is too strict in breaks the metadata chain, in case the child
node's output already includes the metadata column:
{code:java}
// expected to work (and does)
spark.read.table("t")
.select("a", "b").select("_metadata")
// by extension, should also work (but does not)
spark.read.table("t").select("a", "b", "_metadata").as("s")
.select("a", "b").select("_metadata"){code}
The solution is for `SubqueryAlias` to always propagate metadata columns that
are already in the child's output, thus preserving the `metadataOutput` chain
for that column.
was:
The `AddMetadataColumns` analyzer rule intends to make resolve available
metadata columns, even if the plan already contains projections that did not
explicitly mention the metadata column.
The `SubqueryAlias` plan node intentionally does not propagate metadata columns
automatically from a non-leaf/non-subquery child node, because the following
should _not_ work:
{code:java}
spark.read.table("t").select("a", "b").as("s").select("_metadata"){code}
However, today it is too strict in breaks the metadata chain, in case the child
node's output already includes the metadata column:
{code:java}
// expected to work
spark.read.table("t")
.select("a", "b").select("_metadata")
// by extension, should also work (but does not)
spark.read.table("t").select("a", "b", "_metadata").as("s")
.select("a", "b").select("_metadata"){code}
The solution is for `SubqueryAlias` to always propagate metadata columns that
are already in the child's output, thus preserving the `metadataOutput` chain
for that column.
> SubqueryAlias should propagate metadata columns its child already selects
> --------------------------------------------------------------------------
>
> Key: SPARK-42704
> URL: https://issues.apache.org/jira/browse/SPARK-42704
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 3.3.2, 3.4.0
> Reporter: Ryan Johnson
> Priority: Major
>
> The `AddMetadataColumns` analyzer rule intends to make resolve available
> metadata columns, even if the plan already contains projections that did not
> explicitly mention the metadata column.
> The `SubqueryAlias` plan node intentionally does not propagate metadata
> columns automatically from a non-leaf/non-subquery child node, because the
> following should _not_ work:
>
> {code:java}
> spark.read.table("t").select("a", "b").as("s").select("_metadata"){code}
> However, today it is too strict in breaks the metadata chain, in case the
> child node's output already includes the metadata column:
>
> {code:java}
> // expected to work (and does)
> spark.read.table("t")
> .select("a", "b").select("_metadata")
> // by extension, should also work (but does not)
> spark.read.table("t").select("a", "b", "_metadata").as("s")
> .select("a", "b").select("_metadata"){code}
> The solution is for `SubqueryAlias` to always propagate metadata columns that
> are already in the child's output, thus preserving the `metadataOutput` chain
> for that column.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
