Connection Pooling

Tune database/sql pools, pgxpool, max open connections, idle timeouts, and PostgreSQL saturation.

7 min readBuilding Backend ApplicationsBuild

Why Connection Pools

Opening a database connection is expensive. It involves a TCP handshake, SSL negotiation, and authentication. Connection pools reuse established connections across requests, eliminating this overhead. Every production Go application needs a connection pool.

database/sql Pool

Goconfiguring-database/sql-pool.go

17 linesLn 1, Col 1Go

MaxOpenConns limits concurrent database work. MaxIdleConns keeps connections warm. ConnMaxLifetime prevents stale connections. ConnMaxIdleTime closes unused connections.

pgxpool

pgxpool is the native connection pool for pgx, the most popular pure Go PostgreSQL driver. It avoids the database/sql abstraction overhead and provides better performance for PostgreSQL-specific features.

Gopgxpool-configuration.go

22 linesLn 1, Col 1Go

pgxpool provides a simpler configuration model. MinConns keeps connections warm. HealthCheckPeriod verifies idle connections are still alive.

Pool Sizing

Pool size depends on your workload. Too few connections and queries wait. Too many and PostgreSQL saturates. A good starting point is 2x CPU cores for MaxOpenConns.

Workload	MaxOpenConns	MaxIdleConns
Low traffic API	10	5
Medium traffic API	25	10
High traffic API	50	25
Background worker	10	5
Mixed (API + worker)	30	15

Start conservative, tune with metrics

Start with MaxOpenConns=25 and MaxIdleConns=5. Monitor PostgreSQL connections with pg_stat_activity. If queries wait for connections, increase MaxOpenConns. If connections sit idle, decrease.

Monitoring Pools

Gopool-stats-for-monitoring.go

26 linesLn 1, Col 1Go

OpenConnections shows total connections. InUse shows actively running queries. WaitCount shows how many queries waited for a connection. High WaitCount means you need more connections.

Common Pitfalls

Not closing rows: Rows not closed hold connections indefinitely. Always defer rows.Close().
Unbounded queries: Large result sets hold connections longer. Use LIMIT and pagination.
Transaction timeouts: Long transactions hold connections and locks. Keep them short.
MaxOpenConns too high: More connections than PostgreSQL can handle causes context deadline exceeded errors.
Not monitoring: Without pool metrics, you cannot diagnose connection exhaustion.

MaxOpenConns limits are per-database

If your application connects to multiple databases, each connection pool is independent. MaxOpenConns=25 on each pool means up to 50 total connections. Account for this when sizing PostgreSQL max_connections.

Mark this lesson complete to store local progress and unlock a cleaner resume path the next time you visit.

Stage 3 · Build

Databases & Persistence