Render at build time and serve from memory

How to do it

On start, or on first request, render each page and work out its CSP hashes and ETag. Compress it only in the encoding a client asks for, and cache those bytes. Key the cache by path or a short hash, never by the page text itself. Keep page sources on disk, not in memory, so a large site fits a small server. Insert content into the template literally, as the literal replacement rule describes.

Example

import { brotliCompressSync, gzipSync } from 'node:zlib';
import { createHash } from 'node:crypto';
const ENCODERS = { br: brotliCompressSync, gzip: gzipSync };
const cache = new Map();
function encoded(page, encoding) {
  const entry = cache.get(page.path) ?? {};
  entry[encoding] ??= ENCODERS[encoding](Buffer.from(render(page)));
  cache.set(page.path, entry);
  return entry[encoding];
}
// A function replacer inserts the content exactly as written
const fill = (template, html) => template.replace('<!--page-->', () => html);

import gzip
ENCODERS = {"gzip": lambda body: gzip.compress(body, 9)}  # add "br" with the brotli package
cache = {}
def encoded(page, encoding):
    entry = cache.setdefault(page.path, {})
    if encoding not in entry:
        entry[encoding] = ENCODERS[encoding](render(page).encode())
    return entry[encoding]
# str.replace inserts the content exactly as written
fill = lambda template, html: template.replace("<!--page-->", html)

How to check

With a Server-Timing header on page responses, a warm page shows a dur under 1 ms. After warming every page, the server's memory stays within the budget the load test sets.