[Feature] Support nearest-previous partition fallback for chain table when queried partition does not exist

[blackflash997997]

Search before asking

• I searched in the issues and found nothing similar.

Motivation

Currently, when querying a Paimon chain table table (e.g., time-series data partitioned by dt), if the queried partition does not exist, the query returns empty results. This is the default behavior.

In some scenarios , users expect the query to transparently fall back to the nearest previous existing partition rather than returning empty data.

Solution

1.If the queried partition does not exist in the table
2.The system automatically finds the nearest previous partition (the largest partition value that is smaller than the queried one)
3.Returns data from that partition instead of empty results

example

Table partitions: [2024-07-01, 2024-07-02, 2024-07-05]
snapshot paritions : [2024-07-01]
delta partitionss : [2024-07-02, 2024-07-05]

Query1 :
SELECT * FROM table WHERE dt = '2024-07-04'
Current behavior: returns empty
With feature enabled: returns data from dt = '2024-07-02' , and the value of dt column replaced with 2024-07-04

Query2 :
SELECT * FROM table WHERE dt = '2024-07-10'
Current behavior: returns empty
With feature enabled: returns data from dt = '2024-07-10' , and the value of dt column replaced with 2024-07-10

only work for query like : partition column = 'some value'
not work for query like : partition column <='some value' or partition column in ('some value')

Anything else?

I would like to ask whether this feature is considered reasonable from the project's design perspective. If the maintainers agree that this is a valuable addition and the proposed approach is acceptable, I have already implemented the core logic locally and am ready to clean up the code and submit a PR.

Are you willing to submit a PR?

• I'm willing to submit a PR!