Failure models

How this question breaks down

1. 1Crash failuresCore18 minName the ways a machine can fail, from stopping cleanly to lying, because each model decides what a design must detect, how many replicas it needs and what it can safely assume.
2. 2PartitionsCore20 minWhen the network splits a cluster into groups that cannot hear each other, each group must decide whether to keep working, and a majority rule plus fencing stops both sides from acting as the leader at once.
3. 3TimeoutsCore18 minGive every remote call a deadline set from its latency percentiles, retry only what is safe with capped, jittered backoff, and cap retries with a budget so a failing service is not buried by its own callers.

Work through the parts in order and you’ve covered the full design. Each part is one topic.