[SPARK-23889] DataSourceV2: Add interfaces to pass required sorting and clustering for writes

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Implemented
• Affects Version/s: 2.3.0
• Component/s: SQL
• Assignee: Unassigned
• Reporter: Ryan Blue
• Created: 06/Apr/18 16:42
• Updated: 08/Feb/22 22:14
• Resolved: 08/Feb/22 22:14

Description

From a discussion on the dev list, there is consensus around adding interfaces to pass required sorting and clustering to Spark. The proposal is to add:

interface RequiresClustering {
  Set<Expression> requiredClustering();
}

interface RequiresSort {
  List<SortOrder> requiredOrdering();
}

When only RequiresSort is present, the sort would produce a global sort. The partitioning introduced by that sort would be overridden by RequiresClustering, making the sort local to each partition.

Issue Links

• relates to: SPARK-33779 DataSource V2: API to request distribution and ordering on write
• relates to: SPARK-33808 DataSource V2: Build logical writes in the optimizer
• relates to: SPARK-34026 DataSource V2: Inject repartition and sort nodes to satisfy required distribution and ordering
• relates to: SPARK-34049 DataSource V2: Use Write abstraction in StreamExecution
• links to: [Github] Pull Request #29066 (aokolnychyi)
• links to: [Github] Pull Request #30706 (aokolnychyi)
• links to: [Github] Pull Request #30706 (aokolnychyi)