How much server downtime during the cutover?

Zero, with the blue-green cutover I default to. Here’s the actual playbook:

1. Pre-warm. The Hyvä storefront runs on a parallel pre-warmed instance (same database, same media, same backend, just a different frontend theme). Cache pre-warmed by replaying recent traffic. Lighthouse re-verified.
2. Content freeze. 24 hours before cutover, content / catalog edits pause so the two instances stay in sync. Order/customer data keeps flowing, both instances write to the same database.
3. Final sync + smoke test. Sync media + URL rewrites + cache warmer. Run a smoke-test suite on the pre-warmed Hyvä instance.
4. Flip. Change the load-balancer / DNS / Cloudflare to route to the Hyvä instance. DNS propagation tail is <30 seconds for sites on managed DNS (Cloudflare / Route 53 / Fastly).
5. Watch. Dashboards monitored for 4 hours: error rate, conversion rate, AOV, support volume. Rollback button stays armed for 72 hours, if anything tanks, route back to Luma in one command.

Result: customers experience zero downtime, no maintenance page, no broken sessions. The only people who notice the cutover are admins who paused content edits.

The alternative (in-place cutover) is cheaper but has a 30-120 minute maintenance window. Only worth it for sub-$500k GMV stores where the blue-green hosting cost overhead isn’t justified.