데이터베이스 연결 풀링

Opening a database connection is expensive: a TCP handshake, authentication, and session setup can take tens of milliseconds. Doing this on every request adds latency and load you do not need.

A connection pool keeps a set of open connections ready to reuse. A request borrows one, runs its query, and returns it — no setup cost per request.

• Lower latency
• Less load on the database
• A natural cap on concurrent connections

The lifecycle is simple: acquire a connection, use it, release it back to the pool. The critical rule is to always release, even on error, or the pool leaks.

The key tuning knob is the maximum pool size. Too small and requests queue waiting for a connection; too large and you overwhelm the database with concurrent work and context switching.

Bigger is not better. A common starting heuristic is based on CPU cores, not request count, because the database can only do so much in parallel.

When all connections are busy, new borrowers wait. Set an acquisition timeout so a request fails fast instead of hanging forever when the pool is exhausted. Surfacing the error early is better than a silent stall.

A leak happens when a borrowed connection is never returned. The pool slowly drains until every request blocks. Always release in a finally block (or use a context manager / try-with-resources) and enable leak detection in production.

Tune connection longevity: an idle timeout closes connections the pool no longer needs, and a max lifetime recycles old connections to dodge stale sockets and server-side timeouts (e.g. a firewall dropping idle TCP).

Watch the math: 20 app servers each with a 50-connection pool means up to 1000 connections hitting one database — likely over its limit. Size per-instance pools with the total fleet in mind, or add a connection proxy.

Tools like PgBouncer sit between apps and the database, multiplexing many client connections onto a few database connections. They are essential when you have many app instances or serverless functions that each open connections.

Do not guess. Monitor pool metrics — active connections, wait time, timeouts — under realistic load, then adjust size and timeouts. The right pool size is the smallest one that keeps wait time near zero.

Test your understanding of connection pooling.

You learned to optimize database access with pooling:

• Pools reuse connections to avoid per-request setup cost
• Size pools by capacity (e.g. cores), not request volume
• Always release; guard against leaks with timeouts and detection
• Account for total connections across all instances; use poolers like PgBouncer at scale