[
https://issues.apache.org/jira/browse/HIVE-30177?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Denys Kuzmenko updated HIVE-30177:
----------------------------------
Description:
Lost pruning, not wrong results.
- Nested paths: for $.a.b, VariantParquetFilters builds the column path
v.typed_value.a.b.typed_value, but the
shredding spec layout is v.typed_value.a.typed_value.b.typed_value. No column
matches, so nothing is pruned.
TestVariantParquetFilters passes only because it uses a non-spec schema.
- Struct-nested variants (HIVE-29373): VariantFilterRewriter rewrites only
predicates whose first argument is a
plain column, so variant_get(s.v, ...) is never turned into an extract
predicate.
was:Lost pruning: $.a.b paths never prune.
> Iceberg: variant row-group pruning never applies to nested paths or
> struct-nested variants
> -------------------------------------------------------------------------------------------
>
> Key: HIVE-30177
> URL: https://issues.apache.org/jira/browse/HIVE-30177
> Project: Hive
> Issue Type: Sub-task
> Reporter: Denys Kuzmenko
> Priority: Major
>
> Lost pruning, not wrong results.
> - Nested paths: for $.a.b, VariantParquetFilters builds the column path
> v.typed_value.a.b.typed_value, but the
> shredding spec layout is v.typed_value.a.typed_value.b.typed_value. No
> column matches, so nothing is pruned.
> TestVariantParquetFilters passes only because it uses a non-spec schema.
> - Struct-nested variants (HIVE-29373): VariantFilterRewriter rewrites only
> predicates whose first argument is a
> plain column, so variant_get(s.v, ...) is never turned into an extract
> predicate.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
