[
https://issues.apache.org/jira/browse/SPARK-60203?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18126491#comment-18126491
]
Ammar Chalifah commented on SPARK-60203:
----------------------------------------
Aware of this patch: [https://github.com/apache/spark/pull/58013]
Tried repro with 4.1.4-rc2, the issue persists.
> SPJ with allowJoinKeysSubsetOfPartitionKeys fails in AQE skew-join validation
> when a non-join partition key comes first
> -----------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-60203
> URL: https://issues.apache.org/jira/browse/SPARK-60203
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 4.1.1, 4.1.2, 4.1.3, 4.0.4
> Reporter: Ammar Chalifah
> Priority: Minor
>
> In copy-on-write MERGE operations with Iceberg, for tables that are
> partitioned on superset of the join keys and join keys doesn't include the
> first partition key, the operations fails on query planning side.
> For example, a table partitioned on `date, bucket(12, id)`, and MERGE with
> `id` as the only join key, raised an error:
> {{```}}
> {{{}java.lang.IllegalStateException: Unknown type for int field. Type name:
> java.lang.String{}}},
> ```
> The plan fails at {{{}KeyGroupedPartitioning.createShuffleSpec{}}}, then
> {{{}KeyGroupedPartitioning.apply{}}}, then {{project}} over
> {{{}originalPartitionValues{}}}. The scan has already been cut down to the
> join keys, but {{originalPartitionValues}} were never projected, so Spark
> reads the full (date, id) row with the join-key types.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
