PostgreSQL 18: Generate Production Query Plans Without Real Data (2026)

A significant advancement in database management has emerged with PostgreSQL 18, allowing engineers to generate accurate production query plans without the need for the actual production data. This breakthrough, reported by database expert Radim Marek, centers on two new functions: pg_restore_relation_stats() and pg_restore_attribute_stats(). These tools let teams simulate the exact conditions that guide the PostgreSQL query planner’s decisions — all within a lightweight development or testing environment.

How PostgreSQL 18 Replicates Production Query Plans

The PostgreSQL query planner relies on internal statistics — like table sizes, data distribution, and column uniqueness — to determine the most efficient execution path. In production, these stats reflect terabytes of real data. But in development, sanitized subsets often lack this fidelity, leading to misleading query plans.

With PostgreSQL 18, you can now export these critical statistics from production and import them into dev environments using pg_restore_relation_stats(). For example, you can set a status column to 95% 'delivered' and 5% 'shipped', forcing the planner to choose index scans for rare values and full scans for common ones — just like in production.

This eliminates the guesswork in performance tuning. No more deploying code that works fine in dev but fails in prod because of a missed index.

Step-by-Step: Exporting and Applying Statistics

• Export: Run pg_dump --data-only --table=pg_statistic or use pg_restore_relation_stats() to generate a stats dump.
• Store: Save the output (often <1MB) in version control alongside your schema.
• Apply: In dev, use pg_restore_relation_stats() to inject the stats before running queries.
• Test: Run your queries — now the planner behaves identically to production.

Why This Beats SQLite’s Approach

SQLite has long supported writable statistics via sqlite_stat1 and sqlite_stat4 tables, and the .fullschema CLI command can output both schema and stats. While powerful, SQLite’s approach requires manual editing and lacks built-in export/import functions.

PostgreSQL 18 formalizes this concept with dedicated, safe, and scalable functions. Unlike SQLite’s ad-hoc method, PostgreSQL’s implementation integrates seamlessly with backup/restore workflows and supports complex schemas without risk of corruption.

Key Differences: PostgreSQL 18 vs SQLite

Feature	PostgreSQL 18	SQLite
Export Function	pg_restore_relation_stats()	.fullschema (manual)
Automation	Programmatic, scriptable	Manual editing required
Schema Scale	Handles 1000s of tables	Best for small-to-medium
Integration	Native to pg_dump/pg_restore	CLI-only

Transforming Dev-to-Prod Sync with Plan Reproducibility

By treating query planner statistics as a first-class artifact — separate from data — teams achieve true plan reproducibility. This transforms performance testing from a guessing game into a deterministic process.

CI/CD pipelines can now validate query performance before deployment. DevOps teams can detect regressions early. Security teams sleep easier knowing no real PII is copied into dev.

This innovation isn’t just about speed — it’s about trust. With statistics inheritance and dev environment simulation now standardized, PostgreSQL 18 bridges the last major gap between development and production.

PostgreSQL 18 marks a pivotal step toward more deterministic and efficient database operations. By enabling the creation of precise production query plans without the underlying data, it empowers teams to build more robust and performant applications from day one of development.