TestElixir

Phoenix-based JSON API for reminders, backed by a small in-memory OTP service.

Design

• TestElixir.Reminders contains pure domain logic.
• TestElixir.Reminders.Server owns mutable state in a supervised GenServer.
• TestElixirWeb is the Phoenix boundary for routing, controllers, and JSON contracts.
• Tests are split into domain tests and HTTP contract tests.

Setup

cd /Users/mz/sandbox/test-elixir
mix setup

With just:

just setup

Start the server

mix phx.server

Or with IEx:

iex -S mix phx.server

The server listens on http://localhost:4000.

Opening http://localhost:4000/ in a browser now shows a small landing page with the available API routes.

For the browser-based Connect Four UI, open:

http://localhost:4000/connect-four

Create a room in the lobby, then open the same room URL in a second browser window with another player_id.

A third browser can join the same room as a spectator. Spectators receive live updates but cannot drop tokens.

API

Create a Connect Four room:

curl -X POST http://127.0.0.1:4000/api/connect-four/rooms

Then connect players to the returned topic over Phoenix Channels:

ws://127.0.0.1:4000/socket/websocket?vsn=2.0.0&player_id=alice

Reconnect with the same player_id to reclaim the same seat after a disconnect.

Join topic:

["1","1","connect_four:ROOM_ID","phx_join",{}]

Drop a token:

["2","2","connect_four:ROOM_ID","drop_token",{"column":0}]

List reminders:

curl http://127.0.0.1:4000/api/reminders

Create a reminder:

curl -X POST http://127.0.0.1:4000/api/reminders \
  -H 'content-type: application/json' \
  -d '{"title":"Pay rent","due_on":"2026-03-15"}'

Complete a reminder:

curl -X PATCH http://127.0.0.1:4000/api/reminders/1/complete

Quality

mix quality

The alias runs:

• mix format --check-formatted
• mix compile --warnings-as-errors
• mix test

For static analysis with Dialyzer:

mix typecheck

The first run builds PLTs, so it is significantly slower than normal tests.

Task shortcuts via just:

just quality
just typecheck
just ci
just docker-build

Benchmark

Start the server in one shell:

just server

Then run k6 in another shell:

just bench-http
just bench-channel
just bench-game
just bench-spectator
just bench-soak
just bench-wasm

Both scripts accept k6-style environment overrides:

VUS=20 DURATION=30s just bench-http
VUS=50 DURATION=20s just bench-channel
VUS=10 DURATION=20s just bench-game
VUS=10 DURATION=20s just bench-spectator
just bench-soak
FIB_INPUT=20 BENCH_TIME_S=3 just bench-wasm

bench-http measures the landing page, LiveView lobby HTML, reminders API, and Connect Four room creation. bench-channel measures room creation plus Phoenix Channel join latency for Connect Four. bench-game creates a room, joins Alice and Bob over separate WebSockets, and plays a fixed 7-move match until red wins. bench-spectator adds Carol as a spectator, verifies that spectator actions are rejected, and measures how quickly match updates fan out to the spectator socket. bench-soak is a longer-running wrapper around bench-spectator with defaults tuned for local soak runs.

The spectator benchmarks also emit soak-friendly counters in the default k6 summary:

• rooms_created_total
• spectator_matches_completed_total
• spectator_state_updates_total
• spectator_rejections_total
• match_moves_sent_total

bench-wasm compares two Elixir-side Wasm execution paths against the same fixture module in priv/wasm/sample.wat:

• Wasmex, which embeds Wasmtime through a Rust NIF
• Port + native wasmtime host, which keeps Wasmtime in an external process

The first run builds the native host under native/wasmtime_host/ with Cargo. The benchmark prints two sections:

• cold_add: start runtime + one add/2 call
• hot_add / hot_fib: repeated calls against already-started runtimes

Deploy

This repository now includes both Fly.io and Linode deployment settings. Both paths use the same release-oriented Dockerfile.

Shared constraints

• GET /healthz is available for load balancer and platform health checks.
• The current reminders store and Connect Four rooms live in memory.
• Because of that, production should stay on a single app node unless you externalize state or add node-aware room routing.

Fly.io

• Config file: fly.toml
• Release env hooks: rel/env.sh.eex
• Default region: nrt (Tokyo)

Typical flow:

fly auth login
fly apps create mizchi-test-elixir
fly secrets set SECRET_KEY_BASE="$(mix phx.gen.secret)"
fly deploy

You will probably want to change app and PHX_HOST in fly.toml if mizchi-test-elixir is already taken.

ENABLE_DISTRIBUTED_ERLANG is intentionally false by default. Turn it on only after you externalize state or add node-aware room routing, and then set a shared RELEASE_COOKIE secret.

Linode

• Compose setup: deploy/linode/compose.yaml
• Env template: deploy/linode/.env.example
• Deployment notes: deploy/linode/README.md

Typical flow:

cp deploy/linode/.env.example deploy/linode/.env
$EDITOR deploy/linode/.env
docker compose -f deploy/linode/compose.yaml up -d --build

For Linode NodeBalancer, point HTTP health checks at /healthz. TLS termination can stay at the balancer while Phoenix serves plain HTTP on port 4000.

Keep ENABLE_DISTRIBUTED_ERLANG=false for the current single-node setup.