Uptime and availability

What we commit to

We commit to 99.5% availability of the application, per calendar month. We measure it as 100% minus the average error rate across every five-minute interval in the month. An interval’s error rate is the share of requests that got a server error (a 5xx response). An interval with no requests counts as error-free only if our external health check passed throughout it, and as fully failed otherwise; any interval during an outage posted on our status page counts as fully failed.

What that number excludes

Two things:

• A broad failure at our cloud provider, the kind that takes out a whole region and many companies at once. If the provider acts on our account specifically, that counts against the number, even where we can’t lift it ourselves.
• Planned maintenance that we announce at least 72 hours ahead on our status page, up to four hours in any calendar month.

Time in either counts as error-free.

Emergency maintenance still counts against the number. When we have to act without notice, we say so on the status page as soon as we can, and that time counts like any other outage.

Planned maintenance is rare

We have never taken a planned maintenance outage. The product runs on managed cloud services, so there are no servers to patch and no databases to upgrade in place, and ordinary releases go out without interrupting anyone.

When we do need a window, we prefer 00:00 to 04:00 UTC, when the fewest people are working.

Where to check

You can see live system status and a history of past incidents at status.thisandthat.chat, and subscribe there to hear about issues as they happen.