Conversation
| // TODO(timvisee): enable indexed_only handling with proxies | ||
| // // If the proxy segment has no deleted points, direct call to wrapped segment | ||
| // if self.deleted_points.is_empty() { | ||
| // return self.wrapped_segment.get().read().in_indexed_only_search( | ||
| // vector_name, | ||
| // search_optimized_threshold_kb, | ||
| // filter, | ||
| // hw_counter, | ||
| // ); | ||
| // } | ||
|
|
||
| // // Incorporate deleted point count in decision | ||
| // let wrapped_filter = Self::add_deleted_points_condition_to_filter( | ||
| // filter, | ||
| // self.deleted_points.keys().copied(), | ||
| // ); | ||
|
|
||
| // self.wrapped_segment.get().read().in_indexed_only_search( | ||
| // vector_name, | ||
| // search_optimized_threshold_kb, | ||
| // Some(&wrapped_filter), | ||
| // hw_counter, | ||
| // ) |
There was a problem hiding this comment.
The current behavior is actually the above. But it might be more correct to use this commented-out section instead.
I'd have to think about it for a bit.
If we do enable this, we must also adjust how we use the parameter in search.
indexed_only decision
af7ac72 to
3d55bce
Compare
938b102 to
2f25d7d
Compare
e1ddc22 to
a0fbeb2
Compare
|
We can postpone merging this until after the Qdrant 1.16 release. That gives us a bit more time to decide upon #7463 (review). |
6a6386c to
e7cbba6
Compare
There was a problem hiding this comment.
This name seems complicated and depends on a specific query parameter.
| pub unindexed_vectors: Option<HashMap<String, usize>>, |
Do we even need to expose this?
Idea: do not expose this new metric, and instead, consider those unoptimized segments with less than full_scan_threshold as optimized, and include them in the total_optimized_points count.
|
This is stale, and I'm not sure it's worth the effort to revive it |
Improves #7307
Add a
in_indexed_only_search()function to segments and their indices. Allow us to use just this function to decide howindexed_onlybehaves.This specifically improves the way we list excluded points for #7307 by keeping the
indexed_onlydecision within segments.All Submissions:
devbranch. Did you create your branch fromdev?