Elephant by the pool: PostgreSQL connection poolers overview
summary
Explains why connection creation costs matter with concrete numbers: auth drops TPS from 130 to 95 and TLS to 72 while persistent connections jump throughput to 658, and pgBouncer reaches ~6000 in a specific test. Compares PgPool as feature‑heavy HA router, PgBouncer as lightweight single‑threaded pooler with SSL scaling limits, and Odyssey as a new multi‑threaded drop‑in addressing CPU and SSL bottlenecks. Describes real tradeoffs: queued requests, lost session state (prepared statements and temp tables), logging/auditing blind spots, and single‑point‑of‑failure risks.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:06Why a connection pooler
- 8:46PostgreSQL connection poolers
- 15:56Takeaway
- 16:18Limitations
- 19:51References
related talks
59mComparing Connection Poolers for PostgreSQL
51mConnection pooling routing and queueing with pgBouncer
46mPgBouncer Connection Pooling and Routing
23mComparing Postgres connection pooler support for prepared statements
46mOdyssey — a scalable connection pooler for PostgreSQL
27mThe world's worst pool party: Connection management with Prisma
53mPgBouncer: Present and Future
13mA Deep Dive into the DB Connection Pool