microsplit

Overview ¶

Command microsplit is the runnable demo for the deployable-modules preview. The same binary boots as a monolith (no env vars set) or as a single split unit when `nexus build --deployment X` produces it:

# monolith — both modules in one process
nexus build --deployment monolith
./bin/monolith

# split — users on :8081, checkout on :8080 talking to it
nexus build --deployment users-svc
nexus build --deployment checkout-svc
./bin/users-svc &
./bin/checkout-svc

Or run both at once with auto-wiring:

nexus dev --split

All deployment-specific config (port, peers, timeouts, listeners) lives in nexus.deploy.yaml — this file stays deployment-agnostic. The manifest's listeners block adds an admin listener at port+1000 so the dashboard sits off the public surface.

Verifying the production-grade features ¶

After `nexus dev --split` is up, each subprocess listens on two ports: the public one declared in nexus.deploy.yaml and an admin one at +1000. Curl probes:

# health: 200 once fx Start completes — used by k8s liveness
curl -i http://localhost:8081/__nexus/health

# ready: 200 when alive AND every declared peer is reachable;
#        body shows per-peer probe state. Stop users-svc and
#        watch checkout-svc's /ready flip to 503.
curl http://localhost:8080/__nexus/ready | jq

# scope filter: dashboard 404s on the public port, served on
#               the admin port (+1000).
curl -o /dev/null -w 'public:%{http_code}\n' http://localhost:8081/__nexus/config
curl -o /dev/null -w 'admin: %{http_code}\n' http://localhost:9081/__nexus/config

# routing: same userId pins to one users-svc replica when scaled.
#          (See scenario_test.go for the in-process verification.)
curl -X POST http://localhost:8080/checkout \
     -d '{"userId":"u1","orderId":"o1"}' -H 'Content-Type: application/json'