ngrok Agent and Origin Connection Pool: Which Boundary Saturates First?

An ngrok tunnel places the agent between public traffic and a locally hosted origin service. The agent dials outbound to the ngrok edge, which accepts inbound requests and forwards them over the established tunnel. Because the agent multiplexes many inbound requests across fewer outbound sockets, socket counts at the tunnel boundary do not directly reflect the concurrency reaching the origin.

Intermittent pool exhaustion can therefore appear on either side. A small or leaking origin-side pool — database connections, HTTP clients, worker threads — may surface only under bursty public traffic. Ngrok-side concurrency limits vary by account tier and agent version, so remembered figures should not be treated as current.

The unresolved decision is where to remediate: raise tunnel-boundary capacity or fix origin-side pooling, since both present as intermittent timeouts and 5xx responses.

Which side saturates first under controlled load? Does the symptom track public traffic volume or internal job bursts? Which agent version and account plan limits apply before any configuration change?