How do you optimize faceted search performance on large collections?

Faceting cost grows with both the number of unique values in a field and the number of matching documents, so optimization usually targets one of those two dimensions.

1. Enable docValues on faceted fields. Without it, Solr builds an in-memory uninverted index (fieldCache) on first use, which is slow to build and heavy on heap. DocValues moves this to an on-disk, memory-mapped columnar structure built at index time instead.
2. Choose the right facet.method - fc (field cache/docValues-based) generally wins for fields with many unique values, while enum can be faster for low-cardinality fields like a status flag.
3. Limit facet.limit and use facet.mincount=1 to avoid computing and transferring buckets nobody will use.
4. Prefer the JSON Facet API for multi-level or nested facets - it can compute several aggregations in a single pass instead of Solr's older, separate facet parameters requiring multiple internal computations.
5. Cache filter clauses shared with facets so the base document set doesn't get recomputed for every facet field.

On very large, high-cardinality fields (like a free-text tag field with millions of unique values), consider whether faceting is even the right tool - a dedicated aggregation or a pre-computed summary table sometimes serves the UI need better than live faceting.