CrownChu opened a new pull request, #8618:
URL: https://github.com/apache/paimon/pull/8618
## Purpose
This is a follow-up to #8000. It fixes persisted full-text query
compatibility and vector metric handling, and changes `paimon-eslib` to a thin
JAR so that ESLib and Lucene classes are no longer bundled into the Apache
artifact.
## Changes
### Thin packaging
- Remove the module-specific shading configuration that bundled ESLib and
Lucene classes.
- Keep ESLib and Lucene as regular Maven dependencies.
- Resolve ESLib 1.0.6 artifacts from the public GitHub Maven repository.
- Document the runtime dependency requirements in `paimon-eslib/README.md`.
- Include `paimon-eslib` in the JDK 17 snapshot build and deployment
workflow.
This change only separates the artifact boundary. It does not relicense
ESLib or change its dependency license.
### Persisted full-text query compatibility
- Accept the serialized `column` field as a routing hint when reading
persisted SQL full-text predicates.
- Continue using the resolved physical index field for query execution.
- Add regression coverage for persisted predicates on complementary text
fields.
### Vector metric correctness
- Implement the `VectorGlobalIndexer` metric contract.
- Support canonical `l2`, `cosine`, and `inner_product` metric mappings.
- Use the metric persisted in each index shard instead of relying on the
current table configuration.
- Reject searches across shards with inconsistent vector metrics.
- Convert Lucene vector scores to Paimon's expected score semantics.
- Close the reader when vector metric registration fails to avoid leaking
resources.
## Tests
```shell
mvn -pl paimon-eslib clean verify
Result:
Tests run: 70, Failures: 0, Errors: 0, Skipped: 0
BUILD SUCCESS
The generated paimon-eslib JAR was also verified not to contain ESLib or
Lucene classes.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
