Struct RegistryProvider 

pub struct RegistryProvider { /* private fields */ }

StatsProvider backed by an IndexRegistry.

Closes the loop between the index trait layer and the planner stats surface: storage components publish their indexes into an IndexRegistry, and this adapter surfaces those statistics to the cost estimator through the trait it already consumes.

Table-level statistics (row counts, page counts) still need an external source — the registry only knows about indexes, not base-table cardinality. Callers can chain a StaticProvider via RegistryProvider::with_table_fallback when they want both.

Implementations

impl RegistryProvider

pub fn new(registry: Arc<IndexRegistry>) -> RegistryProvider

Wrap an existing registry. Without a fallback, table_stats always returns None — only index-level stats are served.

pub fn with_table_fallback( self, fallback: Arc<dyn StatsProvider>, ) -> RegistryProvider

Attach a secondary provider consulted for table-level stats the registry cannot answer.

Trait Implementations

impl StatsProvider for RegistryProvider

fn table_stats(&self, table: &str) -> Option<TableStats>

Return row-count / page-count / column metadata for table, or None when stats are not available.

fn index_stats(&self, table: &str, column: &str) -> Option<IndexStats>

Return the IndexStats backing a secondary index on (table, column), if one exists. The planner uses IndexStats::point_selectivity to derive equality selectivity instead of the 0.01 constant.

fn column_stats(&self, table: &str, column: &str) -> Option<ColumnStats>

Return per-column statistics (distinct count, null count, min/max) when available. Default implementation derives from StatsProvider::table_stats when present.

fn has_index(&self, table: &str, column: &str) -> bool

Convenience: does a usable index exist on (table, column)?

fn distinct_values(&self, table: &str, column: &str) -> Option<u64>

Convenience: distinct-value count for a column, via column stats or an index on that column, whichever is available.

fn column_histogram(&self, _table: &str, _column: &str) -> Option<Histogram>

Optional equi-depth histogram for the column. Defaults to None, in which case the planner falls back to its uniform 0.3 range heuristic.

fn column_mcv(&self, _table: &str, _column: &str) -> Option<MostCommonValues>

Optional most-common-values list for the column. Defaults to None, in which case the planner falls back to its uniform 0.01 equality heuristic. Use for skewed columns where one or two values dominate the table.