[
https://issues.apache.org/jira/browse/ARROW-11094?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrew Lamb closed ARROW-11094.
-------------------------------
Resolution: Invalid
> [Rust] [DataFusion] Implement Sort-Merge Join
> ---------------------------------------------
>
> Key: ARROW-11094
> URL: https://issues.apache.org/jira/browse/ARROW-11094
> Project: Apache Arrow
> Issue Type: New Feature
> Components: Rust - DataFusion
> Reporter: Andy Grove
> Priority: Major
>
> The current hash join works well when one side of the join can be loaded into
> memory but cannot scale beyond the available RAM.
> The advantage of implementing SMJ (Sort-Merge Join) is that we can sort the
> left and right partitions, and write the intermediate results to disk, and
> then stream both sides of the join by merging these sorted partitions and we
> do not need to load one side into memory. At most, we need to load all
> batches from both sides that contain the current join key values.
> In order to reduce memory pressure we will want to limit the concurrency of
> these sort operations.
> We would still want to default to hash join when we know that the build-side
> can fit into memory since it is more efficient than using a sort-merge join.
> [https://en.wikipedia.org/wiki/Sort-merge_join]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
