darafarm/risk-controlled-postgresql-planner-steering

How darafarm/risk-controlled-postgresql-planner-steering works

RiskSteer is a PostgreSQL query-planner steering research system: it inspects a query plan, predicts a join-planner configuration, and applies steering only when predicted savings clear a risk threshold. The README documents this intended pre-execution workflow and its four planner arms. The sampled code primarily implements benchmark collection, model evaluation, analysis, and visualization; it does not show a deployed pre-execution policy or risk gate, so those relationships are documented rather than source-verified.

Main entry points: src Plan Inspection · job_queries_scaled JOB Workload · job_runner_10queries.py Arm Benchmark · train_model.py Model Evaluation

Components 14

Query Steering

Plan Inspection src
Plan Features src/untitled11.py
Arm Classifier src/train_model.py
Risk Threshold src/train_model.py
Planner Arms src/untitled11.py

Benchmark Experiments

JOB Workload queries/job_queries_scaled
Arm Benchmark src/job_runner_10queries.py
Run Dataset src/ML_dataset.py

Modeling and Analysis

Feature Dataset src/train_model.py
Model Evaluation src/train_model.py
Oracle Analysis src/analysis_904.py

Results and Reporting

Result Plots src/visualization.py

Other components

Query User
PostgreSQL

Connections 16

Query User → JOB Workload: submits query README.md
JOB Workload → Plan Inspection: supplies query README.md
Plan Inspection → PostgreSQL: requests plan README.md
Plan Inspection → Plan Features: returns plan README.md
Plan Features → Arm Classifier: provides features README.md
Arm Classifier → Risk Threshold: predicts arm README.md
Risk Threshold → Planner Arms: selects or defaults README.md
Planner Arms → PostgreSQL: sets planner options README.md
PostgreSQL → Run Dataset: records outcomes README.md
JOB Workload → Arm Benchmark: runs workload src/job_runner_10queries.py
Arm Benchmark → PostgreSQL: executes EXPLAIN ANALYZE src/job_runner_10queries.py
Arm Benchmark → Run Dataset: inserts run records src/ML_dataset.py
Run Dataset → Feature Dataset: provides samples README.md
Feature Dataset → Model Evaluation: trains and evaluates src/train_model.py
Model Evaluation → Oracle Analysis: compares outcomes README.md
Model Evaluation → Result Plots: summarizes results README.md

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·