Postgres

Activate When

• /godmode:postgres, "postgres performance"
• "tune postgres", "replication", "partition table"
• "pgbouncer", "pgvector", "full-text search"
• "VACUUM", "ANALYZE", "bloat", "pg_stat"
• "postgres migration", "write a migration", "postgres schema"

Workflow

1. Environment Assessment

SELECT version();
SHOW shared_buffers;
SHOW work_mem;
SHOW max_connections;

Version: PostgreSQL 14|15|16|17
Hosting: self-managed|RDS|Aurora|Supabase|Neon
Workload: OLTP|OLAP|mixed
Size: <total GB>, Tables: <N>, Largest: <N rows>

2. Advanced Features

CTEs: recursive for hierarchies, materialized for repeated subqueries (PG 12+ auto-optimizes).

Window Functions: ROW_NUMBER, RANK, LAG/LEAD, running totals with SUM() OVER().

JSONB: GIN index for containment (@>), use jsonb_path_query for complex extraction. Keep structured data in columns, metadata in JSONB.

Full-Text Search: tsvector + GIN index. to_tsvector('english', col) with @@ operator. IF FTS sufficient: skip Elasticsearch.

3. Extensions

• pgvector: vector similarity (RAG, embeddings). Use ivfflat (fast, approximate) or hnsw (accurate). IF <1M vectors: pgvector over Pinecone.
• PostGIS: geospatial. ST_DWithin for proximity.
• TimescaleDB: time-series. Hypertables with automatic partitioning by time.
• pg_stat_statements: ALWAYS install. Non-negotiable for understanding query performance.

4. Replication

Streaming (physical): byte-for-byte copy. Use for HA failover + read replicas. wal_level=replica.

Logical: table-level, selective. Use for cross- version upgrades, specific table replication, CDC. Requires wal_level=logical.

IF HA failover needed: streaming replication. IF selective table sync: logical replication.

5. Partitioning

CREATE TABLE events (
  id bigserial, created_at timestamptz, data jsonb
) PARTITION BY RANGE (created_at);
CREATE TABLE events_2024_q1 PARTITION OF events
  FOR VALUES FROM ('2024-01-01') TO ('2024-04-01');

• Range: time-series, logs (partition by month/quarter)
• List: known categories (partition by region/status)
• Hash: uniform distribution (partition by user_id)

IF table >10M rows: consider partitioning. IF queries always filter by time: range partition. ALWAYS include partition key in WHERE clause.

6. VACUUM & Autovacuum

SELECT relname, n_dead_tup,
  last_autovacuum, last_autoanalyze
FROM pg_stat_user_tables
WHERE n_dead_tup > 1000 ORDER BY n_dead_tup DESC;

VACUUM: reclaims dead tuples. VACUUM FULL: rewrites table (EXCLUSIVE LOCK -- maintenance window only). ANALYZE: updates statistics for query planner.

IF dead_tuple_ratio > 10%: tune autovacuum. IF bloat > 50%: consider pg_repack (no lock).

7. Connection Pooling

PgBouncer: transaction mode for web apps. Pool: (cores * 2) + 1 per instance. PG degrades above ~100 active connections.

IF multi-tenant SaaS on Supabase: use Supavisor. IF web app: PgBouncer in transaction mode.

Coordinate: pool_size * instances < max_conn * 0.8.

8. Performance Tuning

shared_buffers: 25% of RAM (start point)
work_mem: RAM / (max_connections * 2)
effective_cache_size: 75% of RAM
random_page_cost: 1.1 (SSD) or 4.0 (HDD)

-- Cache hit ratio (target >99%)
SELECT sum(heap_blks_hit) /
  nullif(sum(heap_blks_hit + heap_blks_read), 0)
FROM pg_statio_user_tables;

-- Index hit ratio (target >95%)
SELECT sum(idx_blks_hit) /
  nullif(sum(idx_blks_hit + idx_blks_read), 0)
FROM pg_statio_user_indexes;

9. EXPLAIN ANALYZE

EXPLAIN (ANALYZE, BUFFERS, FORMAT JSON) <query>;

Red flags: Seq Scan on >10K rows, Nested Loop on large tables, estimates off >10x (run ANALYZE). ALWAYS use CONCURRENTLY for production index creation.

# PostgreSQL diagnostics
psql -c "SELECT * FROM pg_stat_activity WHERE state = 'active'"
psql -c "EXPLAIN ANALYZE SELECT 1"
pg_dump --schema-only > schema.sql

# PostgreSQL diagnostics
psql -c "SELECT count(*) FROM pg_stat_activity WHERE state = 'active'"
pg_dump --schema-only > schema.sql

# PostgreSQL diagnostics
curl -s http://localhost:8080/health/db
grep -r "CREATE INDEX" migrations/ | wc -l

Hard Rules

1. NEVER VACUUM FULL without maintenance window.
2. NEVER disable autovacuum.
3. ALWAYS CREATE INDEX CONCURRENTLY in production.
4. ALWAYS install pg_stat_statements.
5. NEVER exceed ~100 direct connections (use pooler).
6. ALWAYS include partition key in WHERE clauses.
7. ALWAYS EXPLAIN ANALYZE before claiming optimized.
8. NEVER tune without measuring baseline first.

TSV Logging

Append .godmode/postgres-ops.tsv:

timestamp	operation	table	metric_before	metric_after	improvement_pct	verdict

Keep/Discard

KEEP if: execution time improved AND plan uses
  new index/config AND tests pass.
DISCARD if: no improvement OR regression OR
  correctness changed.

Stop Conditions

STOP when FIRST of:
  - Cache hit ratio > 99%
  - Dead tuple ratio < 10% all tables
  - Top-5 queries under target latency

Autonomous Operation

On failure: git reset --hard HEAD~1. Never pause.

Error Recovery