Understanding HTTP/1.1 Keep‑Alive, TCP Keepalive, and Nginx Configuration

HTTP/1.1 differs from HTTP/1.0 by enabling keep‑alive by default, allowing persistent TCP connections; the article explains the TCP keepalive mechanism, its three kernel parameters (keepalive_probes, keepalive_time, keepalive_intvl) and how to view and modify them on Linux.

It then shows how applications can set these options via setsockopt(rs, SOL_SOCKET, SO_KEEPALIVE, (void *)&keepalive, sizeof(keepalive)); and related calls, providing example C code for configuring keepalive, idle time, interval, and probe count.

Next, the article distinguishes TCP‑level keepalive from the HTTP “Connection: keep‑alive” header and describes how Nginx implements and configures keepalive on both client‑side and upstream connections, including so_keepalive , keepalive_timeout , keepalive_requests , and the keepalive directive in upstream blocks.

Configuration snippets illustrate enabling keepalive, adjusting idle, interval, and probe counts, and forcing HTTP/1.1 for upstream communication, e.g., listen 127.0.0.1:3306 so_keepalive=7m:75s:9; and keepalive_timeout 120s 120s; .

The internal Nginx workflow is detailed: parsing listen directives, inheriting socket options, managing connection pools, handling client requests, and cleaning up or reusing connections based on timeout and request limits, with code such as if (ls[i].keepalive) { setsockopt(ls[i].fd, SOL_SOCKET, SO_KEEPALIVE, (const void *)&value, sizeof(int)); } .

Finally, the article warns of potential issues such as “connection reset by peer” when upstream servers close idle connections, and concludes with a brief summary of the keepalive concepts.