Scaling & Traffic
Balance uneven traffic across uneven machines, then lose one of them mid-round.
Grow a cluster from 3 nodes to 30, one at a time, minimizing the keys that move.
Hold a global limit across twelve regions when the counter itself is the bottleneck.
Kill a random instance every thirty seconds. Keep every logged-in user logged in.
Serve typeahead over 200 million documents in under 50 ms at the ninety-ninth percentile.
The other tracks
The machine and the network under every diagram · 5 modules · 23h
Where the state lives, and what it costs to keep it · 6 modules · 30h
Making the second request cheap · 4 modules · 16h
Decoupling two services without losing a message · 5 modules · 21h
What happens on the worst day · 5 modules · 18h
The six systems you will actually be asked to draw · 6 modules · 22h
Running the forty-five minutes on purpose · 5 modules · 12h
