Struct Optimizer 

pub struct Optimizer { /* private fields */ }

Query optimizer using statistics for cost-based decisions.

The optimizer transforms query plans to improve execution efficiency by applying rewrites like predicate pushdown and using statistics to estimate costs for different plan alternatives.

Implementations

impl Optimizer

pub fn new(stats: Arc<StatsManager>) -> Self

Creates a new optimizer with default configuration.

Arguments

• stats - Shared statistics manager for cardinality and selectivity estimates

pub fn with_config(stats: Arc<StatsManager>, config: OptimizerConfig) -> Self

Creates a new optimizer with custom configuration.

Arguments

• stats - Shared statistics manager
• config - Custom optimizer configuration

pub fn set_schemas(&mut self, schemas: HashMap<RelId, Schema>)

Sets the schemas for relations.

This information is used by the optimizer to accurately determine column widths during predicate pushdown.

pub fn config(&self) -> &OptimizerConfig

Returns a reference to the current configuration.

pub fn stats(&self) -> &Arc<StatsManager>

Returns a reference to the statistics manager.

pub fn optimize(&self, node: RirNode) -> RirNode

Optimizes an execution plan by applying transformation rules.

Currently applies:

• Predicate pushdown (if enabled)

Future optimizations may include:

• Join reordering based on cardinality estimates
• Projection pushdown
• Common subexpression elimination

Arguments

• node - The plan to optimize

Returns

An optimized plan that is semantically equivalent to the input

pub fn estimate_cost(&self, node: &RirNode) -> PlanCost

Estimates the cost of executing a plan node.

Recursively computes cost estimates for the entire plan tree, using statistics when available and falling back to heuristics.

Arguments

• node - The plan node to estimate

Returns

A PlanCost with estimated rows, CPU cost, GPU memory, and transfers

pub fn recommend_indexes(&self) -> Vec<RelId>

Returns relations that should have indexes built based on access heat.

This is useful for adaptive query processing where frequently accessed relations benefit from index structures.

pub fn should_use_greedy(&self, node: &RirNode) -> bool

Returns true if the query involves more relations than the DP threshold.

Used to decide between exhaustive and greedy join ordering algorithms.