Moving a legacy commerce platform to Cloudflare Workers

The situation

A small internal tool on a Singapore VPS. Read-mostly, forty thousand rows, no heavy writes. Paying for a server around the clock to serve traffic that peaked during a two-hour window each morning.

Why not just optimise the server

Honestly, they could have. The server was not wrong. But three things tipped it:

The split

Read-mostly and file-heavy, which is exactly where edge execution wins:

• Static assets → Pages, cached at the edge
• Read API → Workers + D1, served from the nearest PoP
• Files → R2, never through the database
• Writes → a single Worker with D1’s transaction model respected, not worked around

import { z } from 'astro:schema';

const OrderQuery = z.object({
  customerId: z.string().uuid(),
  since: z.coerce.date().optional(),
});

export async function GET({ request, env }: ExportedHandler<Env>) {
  const url = new URL(request.url);
  const parsed = OrderQuery.safeParse(Object.fromEntries(url.searchParams));
  if (!parsed.success) return Response.json({ error: 'bad_request' }, { status: 400 });

  const { customerId, since } = parsed.data;
  return Response.json(
    await env.DB.prepare(
      'SELECT id, placed_at, total FROM orders WHERE customer_id = ?1 AND placed_at >= ?2 ORDER BY placed_at DESC',
    )
      .bind(customerId, since ?? new Date(0))
      .all(),
  );
}

Result

• 68ms

  p95 after

• 3

  Servers removed

• £0

  Fixed monthly

• 14 days

  Elapsed

What I’d raise with a different client