Compute
Workers review
Score: 10/10The best thing Cloudflare has ever shipped.
The breakdown explains the score. It is not an average. How to read these.
Last updated
Workers changed how I build software. The isolate model means effectively zero cold starts, code runs in hundreds of cities without you thinking about regions, and the pricing is far cheaper than traditional serverless. The reason is simple: you are billed for CPU time, not wall-clock time. Most serverless platforms charge for every millisecond your function exists, including the time it spends idle waiting on a database or an API call. Workers only charges while your code is actually doing work, so all that waiting costs nothing.
The developer experience is generally good rather than excellent. Wrangler is fine, local development with workerd has come a long way and now closely matches production, and static asset hosting lets a Worker serve a complete website rather than only an API. There are rough edges. They just aren’t enough to cost Workers the 10.
A decade in, entire production platforms run on Workers: routing, APIs, websites, and background jobs. Limits on CPU time and bundle size have risen repeatedly. My enthusiasm has grown with scale. The caveats belong to products around Workers, not Workers itself.
Status page incidents
Published under Workers, plus incidents whose title names it.
- Last 3 years
- 178
- 37 from the title
- Major or critical
- 11
- Median time open
- 1.6 h
- Since January 2019
- 303
Workers is tagged broadly. Plenty of these are really Pages build failures, Workers Builds problems or dashboard and API outages that happened to touch deployments, rather than the runtime dropping requests. Read it as "how often something in the Workers platform was published as broken", not as runtime reliability.
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.