Module parallel 

Parallel Query Execution

This module provides parallel execution strategies for CPU-intensive query operations:

• Parallel Scan + Filter: Process table rows in parallel chunks with WHERE evaluation
• Parallel Aggregation: Already implemented in aggregation.rs
• Parallel Join: Parallel hash join build/probe phases

Architecture

The parallel execution model works by:

1. Collecting rows from storage (sequential - storage layer limitation)
2. Splitting rows into chunks for parallel processing
3. Processing each chunk independently using Rayon’s work-stealing scheduler
4. Merging results back together

Thresholds

Parallelization has overhead, so we only use it when beneficial:

• Table scan + filter: 10,000+ rows
• Aggregation: 100,000+ rows (already in aggregation.rs)
• Hash join: 10,000+ build rows

Re-exports

pub use super::operators::hash_join::JoinType;

Structs

ParallelConfig

Configuration for parallel execution

ParallelHashJoinOperator

Pull-based parallel hash join.

ParallelJoinResult

Parallel hash join result

Enums

DistanceMetric

Distance metric for vector search

Constants

DEFAULT_PARALLEL_CHUNK_SIZE

DEFAULT_PARALLEL_FILTER_THRESHOLD

DEFAULT_PARALLEL_JOIN_OUTPUT_BATCH_BYTES

DEFAULT_PARALLEL_JOIN_OUTPUT_BATCH_ROWS

DEFAULT_PARALLEL_JOIN_THRESHOLD

Functions

parallel_filter

Parallel filter execution for WHERE clause evaluation

parallel_hash_join_cancellable

parallel_join_state_retained_bytes

parallel_topn_vector_search

Parallel brute-force k-NN vector search