The pgvector production stack,
disassembled.
Most pgvector tutorials stop at CREATE INDEX USING hnsw. Here are the 5 pieces you actually need once you cross 10M embeddings.
The 5 things that break at scale
Each item: what breaks if you skip it, and how Rivestack handles it by default.
Pooler
A 10k-connection spike from embedder workers crashes Postgres directly. Connection storms take the DB down before your app layer notices.
PgBouncer in front, 10k pooled to 200 backend, transaction pooling mode by default, session pooling available per-database.
Point-in-Time Recovery (PITR)
"Daily backups" only recover you to yesterday. A dropped table at 4pm loses a day of writes. PITR with continuous WAL archiving recovers to any second in the retention window.
7-day PITR, continuous WAL archiving, sub-second RPO.
High availability
A single-node database is a single point of failure. Read traffic on the primary starves the write path. Vector queries with large ef values lock up connections analytics dashboards need.
HA cluster with streaming replication and automatic failover, separate read-only connection string for offloading analytics.
Index tuning
Default HNSW parameters (m=16, ef_construction=64) are a starting point, not a destination. Under-tuned = poor recall. Over-tuned = slow builds and high RAM.
pgvector 0.8 pre-installed, recall@10 ≈ 0.97 out of the box on a 50M-vector workload.
Monitoring
You can't tune what you can't see. The first time you hear about replica lag is when a user complains about stale data.
Built-in metrics dashboard per database — CPU, memory, connections, replica lag, query throughput, pg_stat_statements and slow query log enabled.
How the defaults hold up
Numbers from the default Rivestack configuration. No custom tuning required.
What "managed Postgres" usually ships
— and what Rivestack ships
| Feature | Typical managed Postgres | Rivestack |
|---|---|---|
| Pooler | Sold separately or premium add-on | Included, PgBouncer pre-configured |
| PITR retention | 1–3 days at base tier | 7 days default |
| HA + read replica | Higher tier only | HA cluster with automatic failover |
| pgvector version | Often lags upstream by months | Latest stable (0.8) |
| Per-database metrics | Account-wide only or add-on | Built-in dashboard per database |
| pg_stat_statements | Sometimes disabled | Enabled by default |
| Autovacuum tuning | Default knobs, no per-table tuning | Per-table tuning supported |
FAQ
Yes — logical replication from any Postgres 14+ source, or pg_dump + restore for smaller datasets. Migration window is typically 30–60 minutes for databases under 500 GB.
Yes. Free dev tier with 2 GB storage, no credit card. Start free.
Tune HNSW parameters, scale compute, add read replicas, switch pooling modes — all without downtime. Nothing is locked.
