Faceted Search & Aggregation Engines: Columnar Doc-Values & Distributed Merging

Faceted search enables users to filter search results dynamically by category, price, date, and attributes with instant count aggregations.

1. Columnar Doc-Values vs Inverted Index

While inverted indexes map terms to documents, faceted aggregations require the inverse mapping: given a set of matching documents, retrieve their field values. Columnar doc-values store un-inverted columnar arrays on disk with OS page caching for high-speed sequential iterations.

2. Global Ordinals & String Compaction

String facets are encoded into dense integer ordinals per index segment. Global ordinals unify segment-level mappings into a single cluster-wide dictionary, allowing bitset aggregations to execute via CPU cache-friendly array increments.

3. Distributed Scatter-Gather Aggregation

Queries broadcast across distributed shards compute localized histograms in parallel. The coordinating node merges shard results, executing final bucket reductions and cardinality estimates in sub-millisecond response windows.