Database connection timeout with Django CONN_MAX_AGE

Goal

Determine the optimal CONN_MAX_AGE setting that minimizes latency spikes when Django serves a high volume of concurrent requests.

Constraints & Uncertainty

• Persistent connections reuse a shared pool; the pool size is bounded by the database server’s maximum connections and the web server’s worker count.
• When the pool is saturated, new requests block waiting for a free connection, producing noticeable latency spikes.
• Django’s documentation recommends a positive CONN_MAX_AGE for production but offers no specific value, leaving the choice to the deployment environment.
• It is unclear whether a fixed age (e.g., 60 s) or a dynamic strategy (e.g., external pooler) yields lower latency under varying concurrency levels.

Unresolved Questions

• What CONN_MAX_AGE value produces the lowest maximum latency when handling 500 concurrent requests on PostgreSQL?
• Is a fixed CONN_MAX_AGE superior to a dynamic connection‑pooling approach for mitigating latency spikes?
• How does Django’s default connection‑close behavior compare to persistent connections in terms of latency under high concurrency?