Skip to content

Consumer apps at scale · 6 of 13

Step 1 · The pattern

One essential feature, many extras

Viewers start the video (the chips marked play) and the app also loads extras: recommendations, live chat, reactions. Extras make two in three requests. One shared backend serves both, 8 at a time. (All numbers are illustrative.)

  • request
  • served

In short

Graceful degradation: how it works and how it breaks

A consumer app does one essential thing (play the video, take the payment, open the chat) and many extras around it: recommendations, reactions, personalised rows, badges. At a big moment, such as the start of a match or the final over, everyone arrives together and the extras make most of the requests. On a shared backend they take the capacity the essential path needs, and the core feature fails along with everything else. Graceful degradation decides in advance which features matter, wires each extra to a switch, and turns the extras off when the backend is under pressure or just before a moment you know is coming. The app hides those widgets and stops calling them. People lose the extras for a few minutes and keep the thing they came for.

What this simulation simplifies

  • One shared backend serves every feature, and requests are either essential (video start) or extra. Real apps have many services with their own limits.
  • The switch turns on automatically at a threshold. Many teams also flip it by hand, or on a schedule just before a known moment.
  • The surge lasts six seconds. Real match moments last minutes, and several can come in one evening.

How this comes up in interviews

"Your streaming app fails at the start of every big match, though playback alone would fit the capacity. How do you keep the video playing?"

Then they ask

  1. Who decides which features can be switched off, and when is that decided?
  2. What does a viewer see when chat is switched off?
  3. Would you switch automatically or by hand, and why?

A strong answer covers

  • At the big moment everyone arrives together, and the extras outnumber the essential requests on a shared backend.
  • Rank features in advance and put every non-essential one behind a switch the app reads.
  • Switch extras off on a pressure threshold, or before a moment you know is coming, and back on after.
  • The app hides switched-off features rather than showing errors, so nobody retries them.
  • Where possible, give the essential path its own capacity so extras cannot take it.

Points to cover, not a script. Say them in your own words and tie them to a system you have worked on.

On one page: the track cheat sheet.

Last reviewed: Official documentation

Diagrams are our own simulations: the timings, rates and counts in them are illustrative, not measurements. Captions are original wording.