core practice · coding problem
Object key layout
Keys userId/timestamp/uuid vs flat uuid. Name one hot-partition risk and one listing benefit.
The mechanism
Object storage (S3/R2) is infinite, cheap, durable bytes — but with ~tens-of-ms latency and no in-place edits. Modern systems like Turbopuffer put cold data here and keep a hot cache in front. Storage and compute separate.
Expected evidence
Prefix listing per user; hot prefix if time-ordered without sharding.
Try this drill in the practice workspace →
Hints when you need them
- S3 listing is prefix-based.
- Shard prefixes for hot tenants.
Watch for these mistakes
- Treating object storage like a low-latency disk
- Many tiny objects instead of batched segment files
- Ignoring request cost (GETs/PUTs are billed)