apurtell commented on PR #2624: URL: https://github.com/apache/phoenix/pull/2624#issuecomment-6074662340
The copilot findings and additional findings from internal review were applied to the commits where the problems were introduced. **Correctness** - Distance `ORDER BY ... ASC` puts NULLs last. Exact plans and index plans now give the same rows. - A query whose primary key range is narrow uses the data table, not the vector index. - Tenant queries probe more posting lists if the result is too small. - `LIMIT 0` and `ORDER BY <ordinal>` work correctly with distance queries. - During a migration, the client returns each row one time only. - `DELETE` on an immutable table works when the vector index has no centroids. - Writes reject NaN and infinite vector elements. - Casts from a vector to an array or `VARCHAR` fail at compile time. - Views with a `WHERE` clause can write to a table that has a vector index, and can query it. **Rebuild and maintenance** - `ALTER INDEX ... REBUILD` trains an index that has no centroids, and makes the index active. - Rebuilds run on a separate executor, so other system tasks do not stop. - The reconcile task moves to a new row before the 10-day TTL can delete it. - The rebuild claim has a short lease. Only the holder can renew it. - `CREATE` and `IndexTool` train the first centroids only while they hold the claim. **Tools and packaging** - `indexfsck` accepts case-sensitive names. It retries correctly and keeps a record of each repair action. - The shaded jars have `Multi-Release: true`, so JDK 21 can load the SIMD kernel. - We removed dead code, one unused gate, and tests that cannot fail. **Comments and docs** - All new code comments are now in ASD-STE100 Simplified Technical English. - The design documents agree with the code. **Known limits** - A background `IndexTool` run holds the claim only while it trains, not while its MapReduce job builds the index. - An ordinal after a wildcard can put NULLs first on a functional index plan. All unit tests pass. -- 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]
