Storage
KV review
Score: 9/10Dead-simple global key-value storage. Design for its consistency model and it's superb.
The breakdown explains the score. It is not an average. How to read these.
Last updated
Workers KV is the storage product I recommend first: a global key-value store with an API you can learn in five minutes. Reads are fast everywhere (hot keys are cached at the edge), writes are cheap, and the free tier is generous enough for real projects.
You do need to understand its eventual consistency: a write can take up to a minute to appear globally. That is fine for configuration, feature flags, cached lookups, and other data where some staleness is acceptable. Problems start when people treat KV like a database.
The one missing point covers the per-key write limit of one write per second, the occasional surprise for someone who missed the consistency model, and a billing quirk worth knowing: cached reads are charged at the full read rate. A hot key served from the edge cache costs the same as a cold one, so the caching helps latency but never your bill. Within those limits, KV has been boringly reliable for years. That is high praise for infrastructure.
Status page incidents
Published under Workers KV, plus incidents whose title names it.
- Last 3 years
- 26
- 3 from the title
- Major or critical
- 2
- Median time open
- 1.3 h
- Since January 2019
- 50
Counts what Cloudflare published, not what broke: time open is how long the status entry stayed open, and nothing before January 2019 is tracked at all. Incidents Cloudflare tagged against nothing but named in the title are counted too. The caveats in full.
Consider instead
- R2
Use R2 when the value is a file or large blob, not a small read-heavy record.
- Hyperdrive
Use a real database behind Hyperdrive when writes must be immediately consistent or transactional.