sailfork / t3code-github / .repos/ alchemy-effect/ website/ src/ content/ docs/ testing/ test-harness.mdx

Test.make is the same Effect-aware harness on three runners: alchemy-test, bun test, and Vitest. For the end-to-end walkthrough, see Testing a Stack; for provider-lifecycle testing, see Testing Providers.

What Test.make returns

A single call returns a self-contained API for the file:

const { test, beforeAll, beforeEach, afterAll, afterEach, deploy, destroy } =
  Test.make({
    providers: Cloudflare.providers(),
    state: Cloudflare.state(),
  });
HelperPurpose
test(name, effect)Effect-aware test. HttpClient and your providers Layer are in scope.
test.skip / test.skipIf / test.only / test.todoSkip / focus / todo modifiers (same shape as bun.test).
test.provider(name, fn)Provider-lifecycle test against a scratch in-memory stack.
beforeAll(effect)Run an Effect once. Returns a lazy accessor (yield* result) usable inside tests.
beforeEach(effect)Run an Effect before every test.
afterAll(effect) / afterAll.skipIf(predicate)Cleanup hook with conditional teardown.
afterEach(effect)Run an Effect after every test.
deploy(Stack, opts?)Plan + apply a stack, resolve to its outputs.
destroy(Stack, opts?)Plan + apply against an empty desired state.

Test.getWhenReady / Test.executeWhenReady are module-level HTTP cold-start helpers, taught in Testing a Stack.

expect (and describe) come from the runner directly — @alchemy.run/test, bun:test, or @effect/vitest.

Test.make options

Test.make({
  providers,   // required
  state,       // optional
  profile,     // optional
  stage,       // optional
  dev,         // optional — run the suite against local emulators
  adopt,       // optional
  sidecar,     // optional — defaults to dev
});

providers (required)

The provider Layer that resolves resource implementations. Usually the same one your Stack uses:

providers: Cloudflare.providers(),
// or merge multiple:
providers: Layer.mergeAll(Cloudflare.providers(), Stripe.providers()),

Credentials resolve through the same AuthProviders registry as alchemy deploy, so tests pick up stored profiles or the env-var auth methods registered by each provider.

state

The state store used by top-level deploy(Stack) and destroy(Stack) (not by test.provider). Defaults to localState() — .alchemy/ on disk.

state: Cloudflare.state(),  // R2-backed, survives across CI runners
state: localState({ path: ".alchemy-test/" }),  // separate dir
state: undefined,  // omit → defaults to localState()

Persistent state lets deploy(Stack) skip recreating unchanged resources between runs. The run-against-an-existing-stack pattern built on it lives in Testing a Stack.

profile

Override ALCHEMY_PROFILE for this file only. Useful for pinning tests to a sandbox profile regardless of what's set in the environment:

Test.make({
  providers: AWS.providers(),
  profile: "test-sandbox",
});

When omitted, the harness reads ALCHEMY_PROFILE from env / .env the same way the CLI does.

stage

Default stage for deploy(Stack) / destroy(Stack). Defaults to test_$USER (e.g. test_sam) so two people running the same suite against one account don't collide. ALCHEMY_TEST_STAGE replaces the default for the whole run (from the command line), e.g. one stage per git worktree. Override per file, or per call:

Test.make({ providers, stage: "ci-pr-42" });

// or per-call:
beforeAll(deploy(Stack, { stage: "ci-pr-42" }));
afterAll.skipIf(!process.env.CI)(destroy(Stack, { stage: "ci-pr-42" }));

A unique stage per PR or test run lets multiple suites run in parallel against the same provider account without colliding.

dev

Off by default — tests deploy to the real cloud. dev: true runs the whole file in local-dev mode, the same wiring as alchemy dev: Cloudflare Workers, Durable Objects, KV, R2, D1, Queues, and Workflows run in workerd with local simulators, and AWS Lambda/ECS plus the emulated AWS surface run in a local Docker emulator. No cloud account is touched for emulated resources.

Test.make({
  providers: Cloudflare.providers(),
  dev: true,  // deploy(Stack) boots workerd / Docker simulators
});

Everything else in the file is unchanged — beforeAll(deploy(Stack)) boots the stack locally and the outputs (url, ids) point at the local instances, so the same HTTP assertions run against http://localhost:<port>. Local resource ids are dev:-prefixed, which doubles as proof no cloud call ran. Alchemy.remote() still pins individual resources live, exactly as it does under alchemy dev.

When omitted, the flag falls back to the ALCHEMY_DEV env var — leave it out of the file and set ALCHEMY_DEV=1 locally to run the same suite against emulators while CI runs it live. Separately, ALCHEMY_TEST_DEV=1 overrides the option in both directions — use it to force an entire existing suite through local providers without editing each Test.make.

What each cloud emulates is covered in Cloudflare local development and AWS local development; the step-by-step walkthrough is Cloudflare Tutorial Part 4.

adopt

Engine-level adoption policy for the run, matching the CLI's --adopt flag. When true, a resource with no prior state whose physical counterpart already exists in the cloud is adopted via provider.read instead of failing. Defaults to false.

sidecar

Only meaningful in dev mode, and defaults to the resolved dev flag: dev tests run local providers behind the same RPC sidecar process the real alchemy dev command uses. sidecar: false runs them in-process instead — useful when debugging provider code, since there's no child process between you and the breakpoint.

Hooks

beforeAll(effect) → Effect.Effect<A>

Runs the Effect once before any test in the file. Stores the result and returns a lazy accessor — yield* accessor inside any test or other hook returns the resolved value:

const stack = beforeAll(deploy(Stack));
const seed = beforeAll(Effect.gen(function* () {
  yield* DynamoDB.putItem({ /* ... */ });
  return Date.now();
}));

test(
  "uses both",
  Effect.gen(function* () {
    const { url } = yield* stack;
    const startedAt = yield* seed;
    /* ... */
  }),
);

Default timeout is 120s. Override with the second argument:

beforeAll(deploy(Stack), { timeout: 300_000 });

beforeEach(effect)

Runs the Effect before every test. No accessor returned — for side-effect setup only (truncate a table, reset a feature flag, …).

afterAll(effect) and afterAll.skipIf(predicate)

Cleanup hook with conditional teardown:

afterAll(destroy(Stack));                          // always destroy
afterAll.skipIf(!process.env.CI)(destroy(Stack));  // CI only
afterAll.skipIf(true)(destroy(Stack));             // never (debugging)

afterAll.skipIf(true) short-circuits without registering a hook at all — there's no risk of an Effect being constructed and dropped.

afterEach(effect)

Runs after every test. Combine with beforeEach for test-isolated fixtures.

Test variants

test.skip("not ready yet", Effect.gen(function* () { /* ... */ }));

test.skipIf(process.env.CI)(
  "local-only smoke test",
  Effect.gen(function* () { /* ... */ }),
);

test.only(
  "the one I'm debugging",
  Effect.gen(function* () { /* ... */ }),
);

test.todo("backfill once R2 has multipart helper");

test.provider mirrors the same shape (semantics in Testing Providers):

test.provider.skip(name, fn);
test.provider.skipIf(condition)(name, fn);

:::note[Vitest differences] On Vitest, test.only is supported but test.todo is a stub (maps to it.todo). test.skip / test.skipIf work identically. :::

HttpClient is in scope

HttpClient is wired into every test Effect, so you can call it directly:

import * as HttpClient from "effect/http/HttpClient";

test(
  "health check",
  Effect.gen(function* () {
    const { url } = yield* stack;
    const res = yield* HttpClient.get(`${url}/health`);
    expect(res.status).toBe(200);
  }),
);

The implementation comes from effect/http/FetchHttpClient — same client the CLI uses. For a full PUT/GET round-trip against a deployed stack, see Testing a Stack → Drive the live URL.

Runners

@alchemy.run/test holds Test.make for every runner. Install it once, whichever runner you use:

bun add -d @alchemy.run/test

Each runner has its own adapter in the package. The Test.make API is identical on all three; only the import and the command change.

alchemy-test

alchemy-test is the runner Alchemy's own suite uses, and it ships in the same package. It runs every test file in one bun process with files and tests in flight at once, which suits slow tests that wait on real clouds.

Import the harness from @alchemy.run/test/Alchemy and expect from @alchemy.run/test.

import * as Test from "@alchemy.run/test/Alchemy";
import { expect } from "@alchemy.run/test";

Run it with the alchemy-test binary. It picks up every *.test.ts under ./test, or the paths you pass.

bunx alchemy-test
bunx alchemy-test test/integ.test.ts -t "serves the url"

Flags set the environment before any test file is imported: --profile <name> sets ALCHEMY_PROFILE, --stage <name> sets ALCHEMY_TEST_STAGE, and --fast sets FAST=1.

bunx alchemy-test --profile testing --stage test_sam_feat-x

Passing tests stay quiet; a failing test prints its error and captured output inline. Every test's output goes to a per-run log under .alchemy/log/test/, whose path is printed at the end. Add --tui for a live view in an interactive terminal.

bun test

Import the harness from @alchemy.run/test/Bun and expect from bun:test.

import * as Test from "@alchemy.run/test/Bun";
import { expect } from "bun:test";

Run it with bun's built-in runner. Every test(...) becomes a bun:test test that runs its Effect.

bun test test/integ.test.ts

Vitest

Install vitest and @effect/vitest next to @alchemy.run/test.

npm add -D @alchemy.run/test vitest @effect/vitest

Import the harness from @alchemy.run/test/Vitest and expect from @effect/vitest.

import * as Test from "@alchemy.run/test/Vitest";
import { expect } from "@effect/vitest";

Every test(...) runs as an @effect/vitest it.live test, with the same 120s default hook timeout.

npx vitest run test/integ.test.ts

Setting the stage from the command line

The default stage is test_$USER on every runner. ALCHEMY_TEST_STAGE replaces it for the whole run. alchemy-test sets it with --stage; with bun test or Vitest, set the variable yourself.

bunx alchemy-test --stage ci-pr-42
ALCHEMY_TEST_STAGE=ci-pr-42 bun test
ALCHEMY_TEST_STAGE=ci-pr-42 npx vitest run

A stage passed to Test.make or to a single deploy still wins over the variable.

Where next

  • Testing a Stack — deploy once, drive the live URL, tear down.
  • Testing Providers — exercise create / update / replace / delete with test.provider.
  • Testing — the testing overview.
  • State Store — choosing between localState(), Cloudflare.state(), and friends.
  • Profiles — how ALCHEMY_PROFILE and the profile factory option resolve credentials.