Sharding by hash(key) % N works until N changes: then almost every key moves to a different server, and your caches go cold all at once. Consistent hashing places servers and keys on a ring; each key belongs to the next server clockwise. Adding or removing a server only moves the keys in one arc.
Virtual nodes give each server several points on the ring, which evens out the load.
The bucket holds up to capacity tokens and refills at a steady rate. Each request spends one token; with no token left it gets HTTP 429. Capacity allows short bursts, the refill rate caps the long-run average.
Alternatives: leaky bucket (smooths output to a constant rate), fixed window counters (simple, but allow double bursts at window edges) and sliding windows (accurate, more memory).
A cache is small, so when it's full something must go. LRU evicts the least recently used item, LFU the least frequently used, FIFO the oldest inserted. Real traffic is skewed (a few keys are hot), which is why caches work at all.
Data lives on N replicas. A write succeeds once W replicas acknowledge it; a read asks R replicas and keeps the newest version. When R + W > N every read set overlaps every write set, so reads see the latest write.
Click a replica to cut it off with a network partition. CAP says that during a partition you must choose: refuse requests (consistency) or answer with possibly stale data (availability).
Every topic that comes up in system design work and interviews, in two lines each. Ones you can practise link to a level or lab.