Add post-collection facets support to the new faceting API

[epotyom]

Add post-collection facets support to the new faceting API.

Normally, users can collect facets directly during search without needing FacetsCollector to store doc IDs. However, this class implements the second step of two-step collection: iterating over doc IDs already collected in FacetsCollector to compute facet results. This approach is useful when all matches must be known before computing facets, or when reusing the same matching documents to run faceting multiple times.

Source: Facets during collection development plan

[mikemccand]

+1, thanks @epotyom -- this adds the capabilities of the first faceting impl (two-phased) to the new sandbox faceting engine?

I'm curious about:

when all matches must be known before computing facets

What's an example use-case?

[epotyom]

+1, thanks @epotyom — this adds the capabilities of the first faceting impl (two-phase) to the new sandbox faceting engine?

You are correct.

when all matches must be known before computing facets

What's an example use-case?

In Amazon product search, we want to add support for rank-dependent associative facets, e.g. sorting facets by max doc relevance, where the relevance expression depends on the ranks of the top-N matching documents. Other potential use cases that come to mind are dynamic ranges (which require numeric values for all matches before they can be computed), and percentile aggregates for facets. Note that this change isn’t the only thing we need to integrate these features with the new faceting API, but it enables them in the future.