Struct SubqueryUnnestingOptimizer 

pub struct SubqueryUnnestingOptimizer { /* private fields */ }

Subquery decorrelation optimizer

Based on batch plan analysis and statistical information, a decision is made as to whether to convert PatternApply to SemiJoin/AntiJoin.

Implementations

impl SubqueryUnnestingOptimizer

pub fn new() -> Self

Create a new optimizer.

pub fn with_max_rows(self, max_rows: u64) -> Self

Set a threshold for the maximum number of rows in subqueries

pub fn with_max_complexity(self, max_complexity: u32) -> Self

Set a threshold for the maximum complexity.

pub fn should_unnest( &self, pattern_apply: &PatternApplyNode, analysis: &BatchPlanAnalysis, stats: &StatsView<'_>, selectivity: &SelectivityEstimator, cardinality: &CardinalityFeedbackManager, advice: &DecorrelationAdvice, ) -> UnnestDecision

Determine whether decorrelation should be performed.

Parameters

pattern_apply: The PatternApply node analysis: The batch plan analysis result stats: The space-scoped statistics for the query selectivity: The selectivity estimator used for filter row estimates cardinality: The learned per-shape cardinality corrections (may be consulted to correct the row estimates of both inputs) advice: Measured execution feedback of both decorrelation paths

Decision

1. Empirical feedback: when past executions of the same shape have measured one path clearly faster, follow the measurement.
2. Otherwise compare corrected cost estimates.

pub fn is_simple_subquery_shape(node: &PlanNodeEnum) -> bool

Check whether the subquery has a decorrelatable shape: a scan of a single table (vertex/edge/index) — optionally a simple two-table inner-join tree — wrapped in equality / disjunctive-equality filters and projections, capped by a constant Limit within MAX_BOUNDED_SUBQUERY_ROWS.

Shared with the heuristic decorrelation rule (stat-free shape gate).

pub fn is_simple_subquery_shape_with_row_bound( node: &PlanNodeEnum, max_rows: i64, ) -> bool

Shape gate with a configurable constant row bound.

The stat-free heuristic rule passes MAX_BOUNDED_SUBQUERY_ROWS; the cost-based decision (should_unnest) passes i64::MAX and lets the CBO row estimate (step 4 of the decision) bound the subquery size instead of the static limit cap.

pub fn is_mark_join_shape(node: &PlanNodeEnum) -> bool

Check whether the subquery has a Mark-Join decorrelatable shape.

The Mark-Join path keeps the subquery plan as the join’s right side and moves the correlated residual into the join condition, so the right side only needs to be a single-table scan wrapped in arbitrary (subquery-local) filters and projections, optionally capped by a constant Limit. Unlike Self::is_simple_subquery_shape the inner filter conditions may be non-equality: they stay as residual filters on the materialized right side rather than becoming equi keys.

pub fn has_bounded_input(node: &PlanNodeEnum) -> bool

Check whether the subquery input is provably bounded: a plain leaf scan or a constant Limit wrapper with a literal count within MAX_BOUNDED_SUBQUERY_ROWS.

pub fn contains_aggregation(node: &PlanNodeEnum) -> bool

Whether the subquery contains an aggregation anywhere in its shape.

Shared with the heuristic decorrelation rule (stat-free shape gate).

pub fn unnest( &self, pattern_apply: PatternApplyNode, ) -> Result<PlanNodeEnum, PlannerError>

Perform the de-association transformation.

Parameters

• pattern_apply: PatternApply node

Returns

The transformed SemiJoin (EXISTS) / AntiJoin (NOT EXISTS) node

pub fn build_semi_join_from_pattern_apply( pattern_apply: PatternApplyNode, ) -> Result<PlanNodeEnum, PlannerError>

Shared transformation: PatternApply → SemiJoin / AntiJoin.

PatternApply has semi-join semantics: it keeps the left rows that have (EXISTS) / do not have (NOT EXISTS) a matching right row. The old InnerJoin conversion was unsound for both cases (duplicated left rows and leaked right columns).

The keys are a direct passthrough: PatternApplyNode already carries per-side key expressions (hash_keys for the left/outer layout, probe_keys for the right/subquery layout), matching the SemiJoinNode key convention.

Trait Implementations

impl Clone for SubqueryUnnestingOptimizer

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SubqueryUnnestingOptimizer

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for SubqueryUnnestingOptimizer

fn default() -> Self

Returns the “default value” for a type.