Philippines staffing research · Published: September 11, 2026

When database connection wait time points to the wrong bottleneck — September 11 study

A bounded study design for separating pool pressure, slow transactions, and application concurrency.

A request waiting for a database connection may coincide with a full pool, but the pool is only the waiting room. Long transactions, leaked connections, traffic bursts, or a low configured limit can produce the same symptom.

PostgreSQL documents activity and wait-event views. OpenTelemetry defines database client metrics. These sources identify useful observations; they do not establish one correct pool size for every application.

Method: observe one service and pool during a defined load test. Record request concurrency, pool size, checked-out connections, acquisition wait, transaction duration, database wait events, timeout, and application revision. Change one variable per run.

Scope and inference limits: the comparison can locate a correlation in the test environment and show which constraint appears first. It cannot predict production capacity when queries, data volume, hardware, or traffic shape differ.

Limitations: instrumentation overhead, sampling intervals, warm caches, and pooled proxies can obscure short events. Keep failed runs. Database owners approve production observation, limits, and query or infrastructure changes.