Concurrent requests on warm functions

Fluid compute can multiplex multiple requests onto a warm function instance, so bursts reuse capacity instead of cold-starting a new isolate for every arrival.

How multiplexing works

While an isolate is warm, Easel can route additional concurrent requests to it within safe limits. Shared process caches stay available across those requests, which cuts cold starts during traffic spikes.

Each request still gets its own identity and auth context. Shared infrastructure must not leak tenant state between concurrent handlers.

What you should notice

Burst traffic that used to fan out into many cold starts now often lands on warm instances. Watch concurrency and duration in Observability after you enable Fluid on a project.

Design handlers to be safe under concurrency: avoid mutable globals for user identity, and keep per-request state on the request fork. Platform primitives favor process reuse with request-local binding.