feat: add the minimal operational module lifecycle
CI / verify (pull_request) Successful in 35s
CI / infrastructure-smoke (pull_request) Successful in 13s
CI / migration-smoke (pull_request) Successful in 46s

Ekklesia had no notion of a module being on or off: the accepted module set
was exactly "whatever discovery found on disk", and `ModuleLifecycleStage`
named nine states nothing produced or consumed. This replaces that vocabulary
with two axes that real code produces — persistent, administrator-owned
configuration (`enabled`/`disabled`) and per-boot runtime observation
(`running`/`disabled`/`blocked`) — and wires them through startup.

- One Core table (`ekklesia_core_module_state`, migration 0004) holding only
  `module_id` and `enabled`. A module with no record is enabled by default, so
  deploying a module is still sufficient to run it and discovery never needs an
  initialization write. A record for an undeployed module is retained and
  ignored, so redeploying restores the decision.
- `resolveModuleEligibility` is pure: a disabled module transitively blocks
  every module that *requires* it (module and capability edges alike) with a
  structured reason naming the provider; optional dependencies never block; and
  a missing, incompatible, or cyclic dependency stays discovery's concern.
- `startApi` starts only the eligible set and builds the UI publication
  registry from that same list, so routes, artifacts, browser inventory, and
  Dashboard navigation all follow one authoritative set. A disabled module is
  never imported, so there is no per-request enablement check anywhere.
- Startup stays all-or-nothing, deliberately; there is consequently no `failed`
  runtime state. `pnpm modules:enable`/`modules:disable` is the out-of-band
  recovery path for a module that breaks startup.
- Changes take effect at the next start, and the read model reports
  configuration, runtime, and `restartRequired` separately rather than claiming
  a module stopped when it has not.
- Migration participation and permission composition still span the deployed
  and discovered inventories respectively, unchanged.
- New restricted platform permission `core.modules.lifecycle.manage` gates both
  reading and changing enablement, at a sibling prefix (`/api/v1/module-lifecycle`)
  that cannot collide with the module route namespace.
- `pnpm doctor` gains a `modules.lifecycle` check so "why is this module
  unavailable?" is answerable without reading logs.

See ADR 0028.

Co-Authored-By: Claude Opus 5 <[email protected]>
This commit is contained in:
ekklesia
2026-08-22 14:45:34 -05:00
co-authored by Claude Opus 5
parent db67cf7700
commit 7729c99e24
52 changed files with 5311 additions and 116 deletions
+331 -23
View File
@@ -195,6 +195,7 @@ async function applyCoreMigrations(url: string): Promise<void> {
"core:0001_authorization_core_tables",
"core:0002_authorization_integrity",
"core:0003_dashboard_navigation_tables",
"core:0004_module_lifecycle_state",
]);
expect(aggregate.migrations[0]?.checksum).toBe(
"f8f2086ef06f7a6943e8ffc4350bb010b37aed04c0a72030d2b153c986b1325f",
@@ -612,6 +613,32 @@ async function signIn(
return { cookiePair: setCookie ? firstCookiePair(setCookie) : undefined, response };
}
/**
* Signs up an account over real HTTP, grants it platform-owner authority through Core's own real
* bootstrap operation (exactly what `pnpm authorization:bootstrap-owner` performs), and signs it
* in — so a scenario needing an administrator exercises the genuine evaluator path rather than a
* stubbed decision.
*/
async function signUpPlatformOwner(
origin: string,
url: string,
): Promise<{ readonly cookie: string; readonly userId: string }> {
const email = `owner-${randomBytes(6).toString("hex")}@example.test`;
const password = "correct horse battery staple";
const created = await signUp(origin, email, password);
const { close, db } = createAuthorizationDrizzleDatabase(url);
try {
await bootstrapPlatformOwner(db, { userId: parseEkklesiaUserId(created.body.user.id) });
} finally {
await close();
}
const session = await signIn(origin, email, password);
if (!session.cookiePair) {
throw harnessError("The platform-owner session cookie was not issued.");
}
return { cookie: session.cookiePair, userId: created.body.user.id };
}
async function fetchPrincipal(
origin: string,
headers: Record<string, string> = {},
@@ -1055,7 +1082,7 @@ describe.sequential("live API — signup, sign-in, principal, sign-out", () => {
});
describe.sequential("live API — startup and migration guarantees", () => {
it("does not create any tables or migration history when started against an unmigrated database", async () => {
it("refuses to start against an unmigrated database, and creates no tables or migration history", async () => {
const database = await createDisposableDatabase("unmigrated");
try {
const config = createApiRuntimeConfig(
@@ -1071,33 +1098,36 @@ describe.sequential("live API — startup and migration guarantees", () => {
port: 0,
},
// Development validation, for the same reason `startConfiguredApi` documents: this scenario
// drives a real sign-up request, which production configuration refuses to enable at all.
// uses the disposable plaintext loopback database, which production configuration refuses.
{ environment: "development" },
);
const running = await startApi(config, { rateLimitOverride: { max: 1000, window: 1 } });
const origin = apiOrigin(running);
// Startup reads the stored module enablement configuration before importing any module
// entrypoint (ADR 0028). Against an unmigrated database that read cannot succeed, and the
// platform fails closed rather than starting every discovered module: it cannot know what an
// administrator decided, and starting a module they disabled is worse than not starting.
// The failure is Core's own sanitized store error — no connection string, host, SQL, driver
// message, or stack.
let captured: unknown;
try {
// A request that touches the (nonexistent) identity tables fails, but must not repair
// or create anything. The exact failure body is not asserted here (Better Auth's
// internal-error shape for this case is not a stable public contract); only that it is
// not a successful sign-up.
const response = await fetch(`${origin}/api/auth/sign-up/email`, {
body: JSON.stringify({
email: "[email protected]",
name: "Integration Test User",
password: "correct horse battery staple",
}),
headers: { "content-type": "application/json" },
method: "POST",
});
expect(response.status).not.toBe(200);
} finally {
await running.shutdown();
await startApi(config, { rateLimitOverride: { max: 1000, window: 1 } });
} catch (error) {
captured = error;
}
expect(captured).toMatchObject({
code: "MODULE_LIFECYCLE_READ_FAILED",
name: "ModuleLifecycleStoreError",
});
const rendered = `${String(captured)} ${JSON.stringify(captured)}`;
expect(rendered).not.toContain(database.name);
expect(rendered).not.toContain("127.0.0.1");
expect(rendered).not.toContain("select");
const admin = await openAdministrativeClient(database.url);
try {
// The invariant this scenario has always existed to protect, unchanged and now proven on a
// path that never even reached Better Auth: a failed start repairs nothing, creates
// nothing, and applies no migration.
const rows = await safeQuery<{ count: string }>(
admin,
"select count(*)::text as count from information_schema.tables where table_schema = 'public'",
@@ -2159,9 +2189,10 @@ describe.sequential("live API — installed module permission end-to-end", () =>
expect(moduleCheck).toMatchObject({ status: "pass" });
expect(moduleCheck?.details).toEqual(["calendar 0.1.0"]);
expect(permissionsCheck).toMatchObject({
// Core's own three (`core.authorization.grants.manage`, `core.modules.ui.read`,
// `core.dashboard.navigation.manage`) plus the fixture module's one.
details: ["4 permissions registered"],
// Core's own four (`core.authorization.grants.manage`, `core.modules.ui.read`,
// `core.dashboard.navigation.manage`, `core.modules.lifecycle.manage`) plus the fixture
// module's one.
details: ["5 permissions registered"],
status: "pass",
});
// Doctor and permission composition read only the manifest — the fixture's register/start
@@ -2979,6 +3010,283 @@ describe.sequential("live API — Node header/request timeouts and client errors
* disposable database is closed in `finally`, and each scenario carries its own outer deadline so
* a regression fails rather than hangs.
*/
describe.sequential("live API — module lifecycle governs what actually runs", () => {
/** Writes one explicit enablement decision through the real table, before the API starts —
* exactly what `pnpm modules:disable` or a prior `PUT` would have left behind. Nothing in this
* suite reaches into the store through a double. */
async function storeEnablement(url: string, moduleId: string, enabled: boolean): Promise<void> {
const client = await openAdministrativeClient(url);
try {
await safeQuery(
client,
`insert into ekklesia_core_module_state (module_id, enabled) values ($1, $2)
on conflict (module_id) do update set enabled = excluded.enabled`,
[moduleId, enabled],
);
} finally {
await safeEnd(client);
}
}
it("never imports, registers, starts, or mounts a disabled module, and still runs its neighbours", async () => {
const modulesDirectory = await createFixtureModulesDirectory();
const recorder = createEventRecorder();
await createFixtureModule(modulesDirectory, "calendar", recorder, {
permissions: [
{ baseline: "public", description: "Read calendar events.", id: "calendar.events.read" },
],
});
await createFixtureModule(modulesDirectory, "announcements", recorder, {
permissions: [
{
baseline: "public",
description: "Read announcements.",
id: "announcements.events.read",
},
],
});
const database = await createDisposableDatabase("lifecycle_off");
await applyCoreMigrations(database.url);
await storeEnablement(database.url, "announcements", false);
const running = await startConfiguredApi(database.url, {
modulesDirectoryOverride: modulesDirectory,
});
try {
const origin = apiOrigin(running);
// The enabled module really did run: its markers are present, and its route answers.
expect(recorder.events).toContain("calendar:register");
expect(recorder.events).toContain("calendar:start");
const calendarRoute = await fetch(`${origin}/api/v1/modules/calendar/events/abc`);
expect(calendarRoute.status).toBe(200);
// The disabled module ran nothing at all. Not "registered but skipped", not "started and
// gated per request" — its compiled entrypoint was never imported, so no hook could run.
expect(recorder.events).not.toContain("announcements:register");
expect(recorder.events).not.toContain("announcements:start");
expect(recorder.events.filter((event) => event.startsWith("announcements:"))).toEqual([]);
// No route was ever mounted, so the platform's ordinary 404 answers — there is no per-request
// "is this module enabled?" check anywhere, because there is nothing mounted to check.
const disabledRoute = await fetch(`${origin}/api/v1/modules/announcements/events/abc`);
expect(disabledRoute.status).toBe(404);
expect(await disabledRoute.json()).toEqual({ code: "NOT_FOUND" });
// Shutdown never invokes a stop hook for a module that never started.
await running.shutdown();
expect(recorder.events).toContain("calendar:stop");
expect(recorder.events).not.toContain("announcements:stop");
} finally {
await running.shutdown().catch(() => undefined);
await dropDatabase(database.name);
}
});
it("blocks a dependent of a disabled required dependency, and explains why over HTTP", async () => {
const modulesDirectory = await createFixtureModulesDirectory();
const recorder = createEventRecorder();
await createFixtureModule(modulesDirectory, "storage", recorder);
await createFixtureModule(modulesDirectory, "announcements", recorder, {
dependencies: { modules: { required: { storage: "^0.1.0" } } },
});
const database = await createDisposableDatabase("lifecycle_block");
await applyCoreMigrations(database.url);
await storeEnablement(database.url, "storage", false);
const running = await startConfiguredApi(database.url, {
modulesDirectoryOverride: modulesDirectory,
});
try {
const origin = apiOrigin(running);
// Neither ran: the disabled dependency by decision, the dependent because starting it would
// violate its own required-dependency contract.
expect(recorder.events).toEqual([]);
const owner = await signUpPlatformOwner(origin, database.url);
const lifecycle = await fetch(`${origin}/api/v1/module-lifecycle`, {
headers: { cookie: owner.cookie, origin: dashboardOrigin },
});
expect(lifecycle.status).toBe(200);
const body = (await lifecycle.json()) as {
readonly modules: readonly {
readonly id: string;
readonly configured: { readonly enabled: boolean; readonly explicit: boolean };
readonly runtime: {
readonly state: string;
readonly reason?: string;
readonly dependency?: { readonly kind: string; readonly moduleId: string };
};
readonly restartRequired: boolean;
}[];
};
expect(body.modules).toEqual([
{
configured: { enabled: false, explicit: true },
id: "storage",
restartRequired: false,
runtime: { state: "disabled" },
version: "0.1.0",
},
{
configured: { enabled: true, explicit: false },
id: "announcements",
restartRequired: false,
runtime: {
dependency: { kind: "module", moduleId: "storage" },
reason: "required_dependency_disabled",
state: "blocked",
},
version: "0.1.0",
},
]);
// The operator's question — "why is this module unavailable?" — answered without a log line,
// and without a filesystem path anywhere in the payload.
const raw = JSON.stringify(body);
expect(raw).not.toContain(modulesDirectory);
expect(raw).not.toContain("dist/");
} finally {
await running.shutdown().catch(() => undefined);
await dropDatabase(database.name);
}
});
it("applies a change made over HTTP only at the next start, and says so in the meantime", async () => {
const modulesDirectory = await createFixtureModulesDirectory();
const recorder = createEventRecorder();
await createFixtureModule(modulesDirectory, "calendar", recorder, {
permissions: [
{ baseline: "public", description: "Read calendar events.", id: "calendar.events.read" },
],
});
const database = await createDisposableDatabase("lifecycle_restart");
await applyCoreMigrations(database.url);
const first = await startConfiguredApi(database.url, {
modulesDirectoryOverride: modulesDirectory,
});
try {
const origin = apiOrigin(first);
const owner = await signUpPlatformOwner(origin, database.url);
const mutation = await fetch(`${origin}/api/v1/module-lifecycle/calendar/enabled`, {
body: JSON.stringify({ enabled: false }),
headers: {
"content-type": "application/json",
cookie: owner.cookie,
origin: dashboardOrigin,
},
method: "PUT",
});
expect(mutation.status).toBe(200);
expect(await mutation.json()).toEqual({
configured: { enabled: false, explicit: true },
id: "calendar",
restartRequired: true,
// Honest: this process really is still serving that module.
runtime: { state: "running" },
version: "0.1.0",
});
// And it demonstrably still is — nothing was hot-unloaded to make the change feel immediate.
expect((await fetch(`${origin}/api/v1/modules/calendar/events/abc`)).status).toBe(200);
} finally {
await first.shutdown().catch(() => undefined);
}
recorder.events.length = 0;
const second = await startConfiguredApi(database.url, {
modulesDirectoryOverride: modulesDirectory,
});
try {
const origin = apiOrigin(second);
// The next start applies the stored decision: nothing imported, nothing mounted.
expect(recorder.events).toEqual([]);
expect((await fetch(`${origin}/api/v1/modules/calendar/events/abc`)).status).toBe(404);
} finally {
await second.shutdown().catch(() => undefined);
await dropDatabase(database.name);
}
});
it("keeps a disabled module's UI out of the inventory and out of Dashboard navigation", async () => {
const modulesDirectory = await createFixtureModulesDirectory();
await createUiModule(
modulesDirectory,
"shown",
"1.0.0",
'export default { id: "shown", version: "1.0.0", mount() {} };\n',
);
await createUiModule(
modulesDirectory,
"hidden",
"1.0.0",
'export default { id: "hidden", version: "1.0.0", mount() {} };\n',
);
const database = await createDisposableDatabase("lifecycle_ui");
await applyCoreMigrations(database.url);
await storeEnablement(database.url, "hidden", false);
const running = await startConfiguredApi(database.url, {
modulesDirectoryOverride: modulesDirectory,
});
try {
const origin = apiOrigin(running);
const owner = await signUpPlatformOwner(origin, database.url);
const headers = { cookie: owner.cookie, origin: dashboardOrigin };
const inventory = (await (await fetch(`${origin}/api/v1/module-ui`, { headers })).json()) as {
readonly modules: readonly { readonly id: string }[];
};
const navigation = (await (
await fetch(`${origin}/api/v1/dashboard/navigation`, { headers })
).json()) as {
readonly ungrouped: readonly { readonly module: { readonly id: string } }[];
};
expect(inventory.modules.map((entry) => entry.id)).toEqual(["shown"]);
expect(navigation.ungrouped.map((entry) => entry.module.id)).toEqual(["shown"]);
// The artifact itself is unreachable too, even to a caller that knows both halves of its
// identity — publication is derived from the eligible set, not from the file's existence.
const artifact = await fetch(`${origin}/api/v1/module-ui/hidden/1.0.0`);
expect(artifact.status).toBe(404);
expect((await fetch(`${origin}/api/v1/module-ui/shown/1.0.0`)).status).toBe(200);
} finally {
await running.shutdown().catch(() => undefined);
await dropDatabase(database.name);
}
});
it("still discovers a disabled module's migrations: deployed schema ownership is not runtime state", async () => {
const modulesDirectory = await createFixtureModulesDirectory();
const recorder = createEventRecorder();
await createFixtureModule(modulesDirectory, "calendar", recorder);
const database = await createDisposableDatabase("lifecycle_migrations");
await applyCoreMigrations(database.url);
await storeEnablement(database.url, "calendar", false);
// Migration discovery is offline and reads the deployed inventory, never the enablement table.
// A disabled module's tables must still be migrated, or enabling it later would need a schema
// change in the middle of a restart.
const aggregate = await discoverAllMigrations({
coreMigrationsDirectory,
moduleDiscovery: {
moduleSdkVersion: "0.1.0",
modulesDirectory,
platformVersion: "0.1.0",
},
});
expect(aggregate.discoveredModules.map((module) => module.manifest.id)).toEqual(["calendar"]);
});
});
describe.sequential("live API — bounded HTTP drain and shutdown", () => {
async function withDeadline<T>(operation: Promise<T>, timeoutMs: number): Promise<T> {
let timer: ReturnType<typeof setTimeout> | undefined;