jerry-024 commented on PR #67: URL: https://github.com/apache/paimon-vector-index/pull/67#issuecomment-5164035350
Thanks for the review. I updated the PR description to match the intended semantics and removed the inconsistent “existing 64 MiB” and “Residual L2 paths are unchanged” statements. The code changes prepared for the next branch update: 1. Enforce `ivfpq_batch_table_reuse_max_bytes` before resident residual-L2 query-table allocation, with fallback to direct table construction when the budget is insufficient. 2. Expose the reuse mode and max-byte budget consistently in C, C++, and Python through ABI-compatible additive V2 C batch entry points; existing C symbols remain unchanged. 3. Keep the default at 512 MiB intentionally. For example, `nq=16384`, `M=16`, and 8-bit PQ requires 256 MiB of reusable query tables. The same configurable cap now governs non-residual and residual-L2 reuse, and callers can lower it per search when memory is constrained. These implementation changes are still local and will be included in the next push; the current PR head is `9fe98fa`. -- 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]
