[Bug] Vector partial updates can return stale values in Data Evolution tables

[zhoulii]

Search before asking

• I searched in the issues and found nothing similar.

Paimon version

2.0

Compute Engine

sdk

Minimal reproduce step

1. Create a table with row tracking, Data Evolution, and two dedicated vector columns: v1 and v2.
2. Insert rows with both vectors in the same file, with v2 initially NULL.
3. Update only v2 to a non-NULL vector, producing a separate vector file.
4. Read the table, including v2.

What doesn't meet your expectations?

Expected: v2 returns the updated vector, while v1 remains unchanged.

Actual: v2 still returns NULL from the older file. Existing non-NULL values can also remain stale after updates.

Anything else?

Column-group ordering can take precedence over file sequence numbers. The latest provider needs to be selected per field and row range; sorting entire groups by their maximum sequence is insufficient.

Are you willing to submit a PR?

• I'm willing to submit a PR!