Opscraft
Nginx & reverse proxies · Lesson 4 of 6 · 3 min read

TLS, security and performance

TLS termination

server {
    listen 443 ssl;
    http2 on;
    server_name shop.example.com;
    ssl_certificate     /etc/letsencrypt/live/shop.example.com/fullchain.pem;
    ssl_certificate_key /etc/letsencrypt/live/shop.example.com/privkey.pem;
    ssl_protocols TLSv1.2 TLSv1.3;
    ssl_session_cache shared:SSL:10m;
    add_header Strict-Transport-Security "max-age=31536000" always;
}
server {
    listen 80;
    server_name shop.example.com;
    return 301 https://$host$request_uri;
}
  • Use the full chain (fullchain.pem). A missing intermediate certificate works in some browsers (they cache intermediates) and fails in curl and mobile apps.
  • ssl_session_cache lets returning clients resume sessions and skip a full handshake.
  • HSTS tells browsers to use HTTPS only. Start with a short max-age because it can't be undone quickly.
  • Let's Encrypt with certbot --nginx or a webroot challenge renews every 60–90 days; reload Nginx after renewal (certbot's deploy hook does it).
  • To re-encrypt to the backend, use proxy_pass https://... with proxy_ssl_verify on; and proxy_ssl_trusted_certificate, otherwise the upstream certificate isn't checked at all.

Rate limiting and access control

Nginx rate limiting is a leaky bucket per key:

http {
    limit_req_zone $binary_remote_addr zone=login:10m rate=5r/m;
    limit_req_status 429;
    server {
        location = /login {
            limit_req zone=login burst=10 nodelay;
            proxy_pass http://app;
        }
    }
}
  • rate=5r/m is the steady rate; Nginx tracks it in milliseconds, so 10r/s means one request every 100 ms, not ten at once.
  • burst=10 queues up to 10 extra requests; without nodelay they are spaced out at the rate, with nodelay they are served immediately but still use up the burst.
  • Requests beyond that are rejected with limit_req_status (default 503; set 429).
  • 10m of zone memory holds about 160,000 IP addresses.
  • limit_conn caps concurrent connections per key, useful for downloads.
Enlarge diagramrequestsfrom one IPburst queueup to 10 waitingrate5 per minute passoverflow429 Too ManyRequests
Figure 1. Rate limiting is a leaky bucket per client: requests drain at the set rate, the burst is the bucket size, and overflow gets 429.
Swipe sideways to see all of it, or tap to open it full screen.

Also: allow 10.0.0.0/8; deny all; for admin paths, auth_basic for quick protection, server_tokens off; to hide the version, and security headers (X-Content-Type-Options, Content-Security-Policy) with always so they're added to error responses too.

Caching and compression

proxy_cache_path /var/cache/nginx keys_zone=api:50m max_size=2g inactive=30m;
location /catalog/ {
    proxy_cache api;
    proxy_cache_valid 200 5m;
    proxy_cache_use_stale error timeout updating;
    proxy_cache_lock on;
    add_header X-Cache-Status $upstream_cache_status;
    proxy_pass http://catalog;
}
  • proxy_cache_use_stale serves an old copy when the backend is down or slow: a cheap resilience win.
  • proxy_cache_lock lets only one request refill an expired item, so a popular page expiring doesn't stampede the backend.
  • X-Cache-Status (HIT, MISS, EXPIRED, STALE) makes caching visible when debugging.
  • gzip on; gzip_types text/css application/json application/javascript; compresses text (HTML is included by default). Don't compress images or already-compressed formats.

The Caching & CDNs topic goes deeper into Cache-Control headers and edge caching.