PostgreSQL Best Practices (OpenAI Golden Knowledgebase)

Canonical Source

• Read {baseDir}/references/openai-scaling-postgresql.md first.
• Source URL: https://openai.com/index/scaling-postgresql/
• Published: January 22, 2026

Non-Negotiable Rules

1. Preserve all named technologies, numbers, and dates exactly as written in the source.
2. Preserve section order and intent exactly:
   • Use Postgres for everything
   • Partitioning and indexing strategy
   • Query optimization
   • Schema management
   • Replication and load balancing
   • Results
3. Keep technology and project names unchanged: Citus, Citus columnar, PgBouncer, PgAnalyze, Patroni, Haste, pgroll, pgai.
4. If a detail is not in the source file, explicitly state: Not stated in the source.
5. Do not add outside recommendations, interpretations, or alternative patterns that are not present in the source.

Required Coverage

When summarizing or operationalizing, cover all of the following source facts:

• Scale baseline and growth metrics:
  • Over 4 billion queries every day
  • Over 2,000 database instances
  • Growth from one Postgres instance in 2022
  • Over 1 billion monthly users and over 800 million weekly users
  • Traffic growth of about 2.5x in just over a year
• Platform strategy:
  • Keep PostgreSQL as the core datastore
  • Use Citus and Citus columnar for distributed and analytical/high-throughput use cases
• Partitioning/indexing decisions:
  • Hash partitioning for large event/log tables
  • Time-range partitioning for append-only datasets
  • Composite index strategy
  • Custom index migration tooling to avoid blocking writes
  • Bottlenecks: table bloat, autovacuum inefficiencies, poor index selectivity
  • Spike drivers: full table scans, huge B-tree indexes, write amplification from hot rows
• Query optimization operations:
  • Query fingerprints and execution stats
  • Lock contention and long-running transaction identification
  • Rewriting expensive joins/subqueries
  • Statement-level timeout and retry strategy
  • Query budget alerts and EXPLAIN-based review workflows
  • Reported impact: 43% DB load reduction, 38% p95 latency reduction
• Schema management workflows:
  • Expand -> migrate -> contract
  • Async backfills with throttling and checkpointing
  • Canary migrations with progressive rollout
  • Automatic rollback on failing migration health checks
  • Open-sourced tools: Haste and pgroll
• Replication and load balancing:
  • Region-aware read replicas
  • Replica lag monitoring and query shedding
  • Dedicated analytical replicas
  • Failover orchestration with Patroni and custom routing logic
  • Deep integration of PgBouncer and PgAnalyze
• Results and forward direction:
  • Low latency under exponential growth
  • High migration velocity without sacrificing uptime
  • Operating thousands of instances with lean infrastructure teams
  • Ongoing iteration and open-source contributions including haste, pgroll, pgai

Output Format

When producing a complete answer, use these exact headings in this order:

1. Use Postgres for everything
2. Partitioning and indexing strategy
3. Query optimization
4. Schema management
5. Replication and load balancing
6. Results

Always include the source URL when attribution is requested.