{"id":1130,"date":"2026-09-14T22:36:22","date_gmt":"2026-09-14T22:36:22","guid":{"rendered":"https:\/\/virtualserversvps.com\/blog\/?p=1130"},"modified":"2026-10-13T22:04:08","modified_gmt":"2026-10-13T22:04:08","slug":"nginx-upstream-keepalive-connection-pooling-vps","status":"publish","type":"post","link":"https:\/\/virtualserversvps.com\/blog\/nginx-upstream-keepalive-connection-pooling-vps\/","title":{"rendered":"Nginx Upstream Keepalive on a VPS: Sizing Connection Pools With Measured Evidence"},"content":{"rendered":"

Without keepalive, every proxied request makes Nginx open a fresh TCP connection to your app server, run the handshake, serve, and tear it down. Under load this shows up as TIME_WAIT accumulation, ephemeral port pressure, and latency you cannot explain from app timings alone. Upstream keepalive fixes it \u2014 but the defaults and the copy-pasted keepalive 32<\/code> line are frequently wrong for a small VPS. This article sizes the pool from measurement.<\/p>

The two-zone configuration<\/h2>

Keepalive needs a directive in the upstream<\/code> block and a protocol version in proxy_http_version<\/code>. Miss either and you get connection reuse silently disabled.<\/p>

upstream app_backend {\n    server 127.0.0.1:9000 max_fails=3 fail_timeout=10s;\n    keepalive 16;                 # idle keepalive conns cached per worker\n    keepalive_requests 1000;      # requests per conn before recycle\n    keepalive_timeout 60s;        # idle timeout of cached conns\n}\n\nserver {\n    location \/ {\n        proxy_pass http:\/\/app_backend;\n        proxy_http_version 1.1;   # REQUIRED for keepalive\n        proxy_set_header Connection \"\";\n        proxy_next_upstream error timeout http_502;\n    }\n}<\/code><\/pre>
proxy_set_header Connection \"\"<\/code> clears the Connection: close<\/code> that Nginx otherwise forwards, which would defeat reuse.<\/p>
Sizing keepalive<\/code> from concurrency, not vibes<\/h2>
The pool is per worker process. A reasonable starting point is the number of upstream connections a single worker needs concurrently, plus headroom:<\/p>
Metric<\/th>How to get it<\/th>Use<\/th><\/tr><\/thead>
Active upstream conns<\/td>ss -tn state established ‘( dport = :9000 )’ | wc -l<\/td>Peak concurrency split across workers<\/td><\/tr>
Worker count<\/td>worker_processes auto<\/td>Divide concurrency by this<\/td><\/tr>
Requests\/sec<\/td>access log rate<\/td>Feeds keepalive_requests<\/td><\/tr><\/tbody><\/table><\/figure>
# observe real upstream connection churn\nss -tn state established '( dport = :9000 )' | wc -l\nss -s | grep -i timewait\n\n# count new backend connections per second\nawk '{print $4}' \/var\/log\/nginx\/access.log | \\\n  cut -d: -f1 | uniq -c | sort -rn | head<\/code><\/pre>
If TIME_WAIT on the upstream side drops to near zero after enabling keepalive, reuse is working. If it does not, verify proxy_http_version 1.1<\/code> and the cleared Connection<\/code> header before touching pool size. The same principle \u2014 verify the mechanism, then tune the number \u2014 applies across the <\/p>
Memory and correctness on a small box<\/h2>