BookestoReliability
No uptime number on this page.
We do not publish a percentage, because we have not run long enough for one to mean anything. What follows is what we run, what we watch, and what we do when it stops.
No card needed. We set up each new venue with you, and pricing is agreed before anything is charged.
What runs where
The actual architecture, out loud
- Where
- One server in Falkenstein, Germany
- In front
- Caddy
- Data
- PostgreSQL 17 and Redis
- Application
- The web application and a background worker
- How
- All of it as an unprivileged user in rootless containers
One server means one region and one machine. If Falkenstein has a bad day, so do you. We know that is the weak point and we are not going to describe it as anything else. A second region is a cost decision we have not made yet.
For scale, the provider’s own service agreement pays out in hosting credits and explicitly excludes backups from its scope. Put plainly, if the machine is lost, the loss is ours to fix, and nobody is compensating you for a lost trading day. Which is why the restore drill on the security page matters more than the hosting contract does.
What we measure
In buildWhat we can see, and what we cannot
Today the product exposes two health checks. One says the application is up. The other tries the database and Redis together and answers ready or not ready, so it tells us something is wrong without saying which.
What we do not yet measure is bookings per hour, which is the part that would tell us something was wrong before you did. The metric is defined and nothing feeds it yet, so it reports nothing. Payment attempts, job-queue depth and job failures are recorded.
A status page built on top of what we currently measure would be green during an outage that only affected bookings, which is worse than having no status page. So bookings per hour is wired up first, then the page.
During an incident
What we do when it breaks
Three things, and they are cheap to promise because they cost time rather than money.
- We say what is broken
- We tell affected businesses as soon as we can, in English today
- What to do meanwhile
- Which in the Gulf usually means take the bookings on WhatsApp and enter them once we are back
- Afterwards
- A written note explaining what happened and what changed
One thing we will not do during an outage
We do not take deposits in degraded mode. A charge we cannot match to a booking becomes a refund argument between you and your customer, and you are the one standing in front of them.
Status
WaitingThere is no public status page yet
When there is one it will be hosted somewhere we do not operate, because a status page that goes down with the thing it reports on is decoration. It will show booking pages, checkout, webhook receipt and message delivery as four separate components, each one something you would notice, rather than the names of our internal services.
Contractual commitments
WaitingNone yet, and we would rather say that
We would rather say that than write a number into a contract we cannot defend.
The backup is a nightly archive, so if the machine were lost, up to a day of changes could be lost with it. Closing that gap means more frequent archiving, and it is not done.
The first figure we publish is the restore time, as a fact: 26 seconds in the drill on 5 November 2026, with the database as small as it is today.