• Sources: post, HN discussion
  • Summary: Netlify reports a warm invocation at 5 to 6 ms at the median, down from 25 to 40 ms, p99 invocations 47.4 percent faster, 99.998 percent availability, and cold invocations on about 1.2 percent of requests at about 9 ms. Each function runs in its own Firecracker MicroVM started from a snapshot in about 2 ms at p99, with the bundle mounted as an uncompressed EROFS image and memory-mapped so only the used parts are read, and compute nodes are chosen by rendezvous hashing with the stickiness relaxed above a threshold to avoid hot spots. The lifecycle work is attributed to Unikraft, nothing changes for existing code, and pricing is unchanged.
  • Why it matters: Netlify states the 50 ms CPU per request, 512 MB memory, and 20 MB compressed code limits are artifacts of the isolate model and now open to revision, and that V8 isolates do not give deploy-level isolation despite the name, which is the reusable claim for anyone choosing between the two models.
  • Follow-up: Track whether the stated limits are actually raised and whether the npm package caveats are removed.