Sonu Sharma created SOLR-18290:
----------------------------------
Summary: Support configurable fusion candidate pool in Combined
Query
Key: SOLR-18290
URL: https://issues.apache.org/jira/browse/SOLR-18290
Project: Solr
Issue Type: Improvement
Components: vector-search
Reporter: Sonu Sharma
The Combined Query / RRF flow has no explicit control over the fusion candidate
pool and its size is implicitly tied to \{{start}} + \{{rows}}. As a result,
the pool shifts with each page, causing the fused ranking to drift across
pagination even for the same underlying query (e.g., page 2 with \{{rows=10}}
only fetches ~10–20 candidates per subquery per shard, a different RRF input
than page 1).
The root cause is that the candidate pool depth is not a first-class,
independently configurable concern.
It requires a control by introducing a parameter like `combiner.queryDepth`.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
