Should you raise the file‑descriptor limit or enable TCP_TW_REUSE to prevent connection‑pool exhaustion on NAT‑exposed hosts?

Goal

Select a mitigation that reliably prevents intermittent connection‑pool exhaustion on Unix hosts handling high‑churn outbound traffic.

Constraints

• TCP_TW_REUSE is unsafe on systems that act as NAT routers or that rely on TIME_WAIT for duplicate‑packet suppression.
• Raising the per‑process file‑descriptor limit can hide application‑level leaks and requires the kernel global fs.file-max to be increased accordingly.

Open questions

1. Under what workload patterns does a modest limit increase alone eliminate exhaustion?
2. When is it safe to enable TCP_TW_REUSE together with a higher limit on mixed NAT and non‑NAT services?
3. What verification steps confirm that the chosen approach does not mask underlying connection leaks?