Query Plan Prediction Accuracy

[ringo380]

Goal

Integrate with actual database execution plans and train models on production query performance data.

Proposed Improvements

• Add EXPLAIN plan parsing for MySQL, PostgreSQL, Oracle
• Collect execution statistics from production databases
• Train cost estimation models on actual query plans
• Implement cardinality estimation integration

Success Metrics

• Execution time prediction within 20% error margin
• Cost estimation accuracy >85%
• Plan structure prediction accuracy >80%

Related Components

• analyzer/ml/optimization/plan_predictor.py (was query_plan_predictor.py)
• analyzer/ml/integration/performance_predictor.py (was performance_impact_predictor.py)

[ringo380]

Reconciliation: predictor modules exist and are wired (authenticated path)

The predictor code this issue proposes already exists and is reachable; the missing piece is real EXPLAIN-plan training data, not the modules.

Stale paths -> real locations:

• analyzer/ml/query_plan_predictor.py -> analyzer/ml/optimization/plan_predictor.py (QueryPlanPredictor)
• analyzer/ml/performance_impact_predictor.py -> analyzer/ml/integration/performance_predictor.py

Reachability: both are imported by analyzer/ml/analysis/unified_analyzer.py (:25 plan_predictor, :21 performance_predictor), which runs via UnifiedQueryAnalyzer.analyze_query() at analyzer/views/query_grading_views.py:216 - authenticated path only (query_grading_views.py:199). With no authenticated traffic, the predictors are built and connected but unexercised.

Suggested reframe: this is a data + validation issue, not a build issue. The concrete open work is EXPLAIN-plan parsing/collection for MySQL/PostgreSQL/Oracle and training the cost model against real plans, then measuring the success metrics (within-20% execution-time error, >85% cost accuracy) against the existing QueryPlanPredictor.