[
https://issues.apache.org/jira/browse/SPARK-60347?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-60347:
-----------------------------------
Labels: pull-request-available (was: )
> SCD1 Ignore-null support; Refactor SCD1 target merge into strategies
> --------------------------------------------------------------------
>
> Key: SPARK-60347
> URL: https://issues.apache.org/jira/browse/SPARK-60347
> Project: Spark
> Issue Type: Sub-task
> Components: Declarative Pipelines
> Affects Versions: 4.4.0
> Reporter: Anish Mahto
> Priority: Major
> Labels: pull-request-available
>
> With ignore-null (leaf-level reconciliation), merging the reconciled
> microbatch onto the target table requires additional logic.
> With row-level reconciliation, a microbatch row replaces a target row
> wholesale when its upsert sequence is at least as new as the target row's.
> With leaf-level reconciliation, a microbatch row can update only some of a
> target row's fields, because each leaf is versioned individually. The merge
> must also write each row's updated version map, whereas row level always
> writes a null version map.
> In preparation for these differences, we should extract the target merge into
> its own strategy class, which Scd1BatchProcessor creates for each merge, so
> that a leaf-level strategy can be added beside it. Common expressions and
> references move into a shared abstract base class, and each strategy
> implements its own merge. This is a pure refactor: the row-level strategy is
> the existing merge, moved without logic changes.
> Keeping row level as a separate strategy keeps the default path identical to
> today's merge and free of any per-leaf work.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
