Guides
When hotcfg is down
What keeps serving your config if hotcfg's servers fail, for how long, and what stops working.
Config is read far more often than it's written, and an app has to start whether or not hotcfg answers. So reads don't depend on hotcfg's servers being up. Three layers sit between your app and them, and each one keeps working when the one behind it fails.
| Layer | Serves | If hotcfg's servers are down |
|---|---|---|
| 1. Cloudflare's edge | Public bins, from the data center nearest the user | Keeps answering from its copy for up to 7 days while our servers return errors or can't be reached. |
| 2. The app's saved copy | The last config the app received | Used when a read fails or takes longer than 2 seconds. Survives restarts. |
| 3. Bundled defaults | Config shipped inside the app | Used on the very first launch, before any read has worked. |
The edge cache
Public bins are served by Cloudflare. A read of the latest version is cached at the edge with:
Cache-Control: public, max-age=60, s-maxage=3600, stale-while-revalidate=3600, stale-if-error=604800
Cloudflare-CDN-Cache-Control: max-age=3600, stale-if-error=604800s-maxage=3600: the edge keeps a copy for an hour without asking our servers, then checks it with theETag(unchanged config is a bodyless304). Every publish, promote, rollback and unpublish also purges every cached copy of that environment, including?query=variants, so new config reaches the edge right away. If a purge ever fails, after several retries, the old copy still expires within the hour.stale-if-error=604800: once a copy has expired, if our servers fail or can't be reached, the edge serves it for up to a week rather than an error.max-age=60: browsers and HTTP clients may reuse a response for a minute, and then revalidate with theETag. Unchanged config is a bodyless304.
Pinned versions (?v=3) and queries (?query=) are cached the same way. Each Cloudflare data center keeps its own copy. One that hasn't seen a read since the last publish has nothing to serve yet, and Cloudflare may evict rarely read files early. That's what the two layers in your app are for.
In your app
The fetching recipes for every language do the same three things: send the ETag they saved, give up after two seconds, and on any failure return the last good copy or the bundled defaults. An outage costs your app at most two seconds at startup, and only when the edge can't answer either. We tested each recipe against a server that refuses connections, hangs and answers 503. To try your own code the same way, see Testing slow and failing networks.
To keep the bundled defaults recent, fetch production while you build the app: see Config from CI.
What stops working
While our servers are down:
- Writes fail: pushes, promotes, rollbacks and dashboard edits. The CLI exits with
1, so a CI job fails instead of reporting success. Run it again when we're back. - Private bins can't be read: they're never cached at the edge, since a shared cache would serve them to anyone. Apps fall back to their saved copy.
- The dashboard and sign-in are unavailable.
Tip
For config every user's app reads at startup, use a public bin. Then the edge cache covers it, with or without ?query=. Keep secrets out of it: public bins are readable by anyone.
Writes during an incident
If a bad config is live and our servers are slow, a rollback may take a few tries. The CLI waits up to 60 seconds per request, and a write that fails changes nothing, so you can safely run hotcfg rollback again. Once it succeeds, the edge purge usually completes within seconds.
Last updated 10 November 2026