HTTP Interface Connection Pooling: Transitioning to Distributed Load Balancing

ClickHouse manages server-side connection limits through the max_connections setting, while the HTTP interface relies on the Connection: keep-alive header and keep_alive_timeout to maintain persistent sessions. In single-node environments, this mechanism handles basic connection reuse effectively.

As deployments scale to distributed clusters, relying on internal server-side pooling becomes problematic. High traffic spikes can lead to connection exhaustion or memory overhead due to the thread-per-connection model, necessitating a transition toward external load balancers like Nginx or HAProxy to manage traffic distribution and socket persistence.

There is uncertainty regarding the optimal alignment of keep_alive_timeout between the ClickHouse server and an external load balancer to prevent premature socket closure or "broken pipe" errors.

• What is the recommended configuration for keep_alive_timeout when using an external load balancer to ensure connection stability?
• How does the server-side max_connections limit interact with the concurrent connection pools maintained by a distributed load balancer?