Fail Fast Instead of Spinning Forever

Systems & Architecture · errors, loading, status

Updated 2026-08-04

Show a failure the moment it is known rather than waiting for a timeout to expire. When a circuit breaker has already marked a dependency as down, a spinner is a lie that costs the user thirty seconds.
Give every request a deadline the user can feel. If a screen has no answer within a few seconds, it needs to say something other than the same spinner.
Let the page render without the parts that failed. A recommendations panel that is down should not take the product page with it, and a checkout that still works is worth more than a complete page.
Mark the missing section instead of hiding it silently. A small Unavailable right now in the slot preserves the layout and tells the user nothing is wrong with their action.
Separate the essential from the optional before the outage, not during it. The list of modules a screen can survive without is a design decision, and if nobody makes it, everything is essential by default.
Recover automatically and visibly. When the dependency returns, fill the section in rather than waiting for the user to reload a page they have already given up on.
Never let a slow dependency hold the whole response. One service taking ten seconds should degrade one panel, not delay first paint for everything.
Write the copy for a system that is down, not for a user who made a mistake. We cannot load reviews right now is accurate, and Something went wrong implies they did something.

A fast, honest failure keeps a product usable, and an endless spinner turns one broken dependency into a broken product.