Automatic vs Manual Instrumentation for JDBC Connection Pool Exhaustion

Diagnosing intermittent connection pool exhaustion in high-throughput Java applications requires a balance between visibility and performance overhead. OpenTelemetry provides two primary paths for capturing this telemetry.

Automatic Instrumentation

The OpenTelemetry Java Agent provides out-of-the-box visibility into JDBC connection pools, capturing span events for acquire and release cycles without code changes. However, this approach may lack internal pool state details, such as the number of pending threads, depending on the specific pool implementation.

Manual Instrumentation

Manual instrumentation allows for the injection of custom attributes, such as specific pool identifiers or precise queue wait times, which are essential for pinpointing the root cause of saturation. The trade-off is the increased development effort and the risk of introducing CPU overhead if high-frequency acquire/release events are sampled too aggressively.

Given these constraints, it is unclear which approach is more sustainable for identifying transient spikes in pool usage without degrading application performance.

• Does automatic instrumentation provide sufficient granularity to distinguish between a slow database and a depleted connection pool?
• At what sampling frequency does manual instrumentation of connection acquire events become a performance bottleneck?