Module feedback 

Cardinality Feedback for Query Optimization

This module implements a learning system that improves cardinality estimates by tracking the difference between estimated and actual row counts during query execution. When similar predicates are seen again, the correction factors are applied to produce more accurate estimates.

How It Works

1. During EXPLAIN ANALYZE, we record estimated vs actual row counts
2. A fingerprint is computed for each predicate pattern (structure, not values)
3. Correction factors are stored: correction = actual / estimated
4. Future queries with similar patterns use the correction factor

Example

-- First query: estimated 100, actual 1000 → correction = 10.0
EXPLAIN ANALYZE SELECT * FROM users WHERE status = 'active';

-- Later query: base estimate 50, corrected estimate 500
SELECT * FROM users WHERE status = 'pending';

Structs

CardinalityFeedback

Cardinality feedback entry for a predicate pattern

FeedbackCache

Cache for cardinality feedback entries

Constants

DEFAULT_DECAY_FACTOR

Default decay factor for exponential moving average

MAX_CORRECTION_FACTOR

Maximum correction factor to prevent extreme adjustments

MIN_CORRECTION_FACTOR

Minimum correction factor

MIN_SAMPLE_COUNT

Minimum sample count before applying feedback

Functions

extract_column_from_predicate

Extract the column name from a simple predicate (col = value)

fingerprint_predicate

Compute a fingerprint for a predicate expression

global_feedback_cache

Get the global feedback cache