Support distributed key

[sundy-li]

Since databend is using cloud service as its storage infra, it'll be easier to have hash-based strategy partition.

create table ... Engine = Fuse
DISTRIBUTED BY HASH(k1)
max_hash_bucket = 32

This may help in performance. Such as hash join and aggregate(distinct), which may avoid local shuffle and benefit co-located join.

[xudong963]

Agree

[wubx]

Agree. The distributed key is a great performance improvement.

[sundy-li]

Refer to:

1. https://www.sisense.com/blog/double-your-redshift-performance-with-the-right-sortkeys-and-distkeys/
2. https://doris.apache.org/zh-CN/docs/advanced/join-optimization/doris-join-optimization/

[artorias1024]

Hi,Is there a corresponding pull request or RFC to track this idea?

[sundy-li]

Not yet, still investigating.