Edge functions now run in their own MicroVMs that boot in under a millisecond (2 ms at p99) using memory-mapped, compressed images to minimize startup costs, and these MicroVMs scale to zero when idle and are snapshotted for rapid restarts.

Compute nodes use rendezvous hashing to assign services, balancing load with controlled stickiness so destinations can scale to meet traffic without hotspots.

The new edge approach improves security and reliability and enables more sophisticated compute at the edge without changing how Edge Functions are authored or deployed.

Isolation is enforced at the MicroVM layer, ensuring deploys stay isolated even if a runtime vulnerability exists, addressing gaps left by earlier V8-only isolation.

A defined lifecycle—boot, snapshot, restore, and scale-to-zero—is managed by Unikraft tooling and a dedicated compute control plane that tracks healthy nodes and coordinates rolling deployments.

Requests follow a defined path: the edge node handles TLS, routes to a compute node, then selects or creates a MicroVM to execute the edge function, with specifications for runtime, platform, and edge function images.

Lessons from the rollout include deploying local DNS resolvers on compute nodes, expanding metrics (boot time, port-open time, user-code start time), and using circuit breakers to reroute and decommission nodes.

Netlify migrated Edge Functions to run on Firecracker MicroVMs within their own edge network, delivering roughly 5x faster median latency (about 5–6 ms) than prior infrastructure.

Live status confirms no changes needed for existing Edge Functions workflows while enabling greater capabilities like full npm package support and adjustable operation limits within current pricing.

Performance highlights show a 47.4% improvement in p99 latency, near-perfect availability at 99.998%, and fivefold faster edge function log delivery, with cold starts averaging around 9 ms for 1.2% of invocations when images are fetched.