Creating a conflicting global index over a primary column succeeds at DDL and then breaks index-backed reads

[LuciferYang]

Search before asking

• I searched in the issues and found no similar issues.

Paimon version

master (1.5-SNAPSHOT)

Compute Engine

Flink and Spark (the create_global_index procedure).

Minimal reproduce step

1. Build a global index over a primary-key column, e.g. a single-column index on vec, via sys.create_global_index.
2. Build a second global index over vec plus another column (vec, txt) — same primary column, different column set.
3. Run any index-backed read.

What doesn't meet your expectations?

The second create succeeds and commits index files, but any later index-backed read builds DataEvolutionGlobalIndexScanner, whose groupIndexFiles throws Primary field %s owns multiple indexes with different columns .... So an accepted DDL leaves the table in a state where the index-backed queries the index exists to serve are broken until the extra index is dropped by hand. The conflicting create should be rejected at DDL time instead.

Anything else?

A second create with the same column set is the refresh flow and must stay allowed. The rejection has to be type-agnostic, matching the read-side grouping by indexFieldId, so a conflicting index of a different type over the same primary column is caught too.

Are you willing to submit a PR?

• I'm willing to submit a PR!