Part 1 of 8 · 1 min read
Why this page comes fourth
In this module you will
- Explain why a server that accepts everything gets slower and then stops.
- Use load shedding, load levelling and rate limiting, and say how they differ.
- Describe the token bucket, leaking bucket and window algorithms for rate limiting, with their weak points.
- Decide what to drop first when a system cannot do everything.
- Pass a time budget down a chain of calls so that no one works on a request nobody waits for.
Pages 2 and 3 protected a caller from a failing dependency. This page protects a service from the callers. Load is the other half of the failure story from page 1: a spike, a greedy client or a retry storm. A service that cannot say no will fail for everyone. The books used are "Understanding Distributed Systems" (UDS), chapter 17, and "System Design Interview" (Xu), chapter 4 on the rate limiter. The numbers are made up to show the idea, unless the text names a book.