Query Optimization Reference

This page is the reference for pg_ripple's SPARQL query optimizer.

Overview

pg_ripple applies multiple optimization passes before generating SQL from SPARQL algebra:

1. sparopt algebra optimizer: first-pass algebra rewriting (variable substitution, filter pushdown, DISTINCT elimination with SHACL hints).
2. Self-join elimination: star patterns on the same subject are collapsed into single-scan plans with multiple joins.
3. Filter pushdown: FILTER constants are encoded to BIGINT at translation time so comparisons happen on integer values, not strings.
4. SHACL hints: sh:maxCount 1 predicates omit DISTINCT; sh:minCount 1 predicates use INNER JOIN instead of LEFT JOIN.
5. Plan cache: compiled SQL is stored in _pg_ripple.plan_cache and reused across identical queries (keyed by query text + current role + relevant GUCs).
6. TopN push-down: ORDER BY ... LIMIT N patterns are pushed into subqueries to avoid sorting full result sets.
7. Leapfrog TrieJoin (WCOJ): worst-case optimal join planning for cyclic SPARQL graph patterns.

Status

SELECT feature_name, status FROM pg_ripple.feature_status()
WHERE feature_name LIKE '%plan%' OR feature_name LIKE '%cache%' OR feature_name LIKE '%wcoj%';

Plan Cache

The plan cache avoids re-compiling SPARQL→SQL for repeated queries. Key details:

• Cache key: SHA-256 of (query text, current_role, GUC snapshot)
• Cache invalidated on: VP promotion (schema change), extension upgrade, plan_cache_reset()
• Maximum entries: pg_ripple.plan_cache_size (default: 512)
• Eviction policy: LRU

-- Inspect cache
SELECT * FROM _pg_ripple.plan_cache;

-- Manual invalidation
SELECT pg_ripple.plan_cache_reset();

Property Path Optimization

Property paths compile to WITH RECURSIVE … CYCLE queries using PostgreSQL 18's native CYCLE clause for hash-based cycle detection. Bounded depth paths ({n,m}) use iterative CTEs limited to pg_ripple.max_property_path_depth hops (default: 20). Early fixpoint termination avoids iterating past convergence for bounded hierarchies.

Magic Sets (Goal-Directed Inference)

When using pg_ripple.query_goal() for Datalog queries, magic sets transformation rewrites rules to focus inference on the bindings needed to answer the goal, avoiding full forward-chaining materialization.

SQL Functions

Related Pages

• Plan Cache
• SPARQL Reference
• Architecture Internals
• Feature Status Taxonomy