3 Commits

Author	SHA1	Message	Date
julien	12136f7a8a	feat(auth): authorization catalogues + Principal builder skeleton (ADR-0025) (#206) ... ## Summary First implementation PR after [ADR-0025](docs/decisions/0025-authorization-model-privileges-roles-scopes.md) was accepted in #205. Lands the closed-set catalogues + the OIDC-callback hook that composes the session-resident `Principal`. No new guards yet — those come in the next PR per the ADR's `§More Information` phasing. ## What lands **New shared library: `libs/shared/auth/`** (`scope:shared`, `type:shared`, framework-agnostic, vitest) | File | Role | | --- | --- | | `src/lib/authorization.types.ts` | Closed catalogues: `PRIVILEGES` (4), `FUNCTIONAL_ROLES` (24), `SCOPE_KINDS` (6). `Principal`, `Scope` types. `isPrivilege` / `isFunctionalRole` / `isScopeKind` type-guards for the drift gate. | | `src/lib/entra-group-to-role.ts` | `parseEntraGroupMap` (validates GUID + slug closedness + dedup) + `EntraGroupToRoleResolver` (translates the Entra `groups` claim into ordered `apf-role-*` slugs, reports unknown GUIDs via callback). | | `src/lib/*.spec.ts` | 17 unit tests against the catalogue counts + resolver contracts. | **BFF wiring** (`apps/portal-bff/src/auth/` + `src/config/`) | File | Role | | --- | --- | | `auth.service.ts` | Extracts the Entra `groups` claim. `AuthenticatedUser` grows `groups: readonly string[]`. | | `principal-builder.ts` | Composes the three axes at sign-in. Filters `roles` claim through `isPrivilege` (drops + WARNs on unknown), resolves `groups` claim through the `EntraGroupToRoleResolver` (drops + WARNs on unknown). | | `scope-resolver.ts` | `ScopeResolver` abstract class + `StubScopeResolver` returning `[{ kind: 'unrestricted' }]` (per ADR-0025 §331 — replaced by a Prisma implementation when ADR-0026's `user_scopes` table lands). | | `session-establisher.service.ts` | Calls `PrincipalBuilder.build()` once, persists the result as `req.session.principal`. The legacy `req.session.user` shape is untouched (audit + AI bridge keep working). | | `session.types.ts` | Adds optional `principal?: Principal` field on the express-session augmentation. | | `entra-group-to-role.token.ts` | DI token for the lazily-loaded resolver. | | `config/load-entra-group-map.ts` | Reads the gitignored `infra/<env>-tenant.entra.json` at boot. Missing path → empty resolver + WARN. Malformed file → hard fail (alternative would be silently dropping a role). | **Tests against the 19 test-tenant personas** (`principal-builder.spec.ts`) Every persona provisioned in `apfrd.onmicrosoft.com` on 2026-05-20 is exercised end-to-end: `admin`, `directeur-bordeaux`, `directeur-complexe`, `rh-aquitaine`, `rh-siege`, `collaborateur-simple`, `tresorier-bordeaux`, `dpo`, `it`, `benevole-aquitaine`, `chef-equipe-bordeaux`, `chef-service-bordeaux`, `directeur-territorial-aquitaine`, `juriste-siege`, `rssi`, `communication-siege`, `elu-ca-national`, `president-cd-aquitaine`, `secretaire-cd-aquitaine`. A coverage assertion confirms the personas cover all 4 privileges + 23 of 24 functional roles (the `partenaire` gap is intentional per ADR-0025). **Infra + docs** | File | Change | | --- | --- | | `infra/test-tenant.entra.example.json` | Template GUID → slug map (24 entries, full v1 catalogue). Real file (`*-tenant.entra.json`) gitignored. | | `infra/README.md` | New row + "Entra group map" section documenting the load semantics + provisioning workflow. | | `apps/portal-bff/.env.example` | New `ENTRA_GROUP_MAP_PATH` block. | | `.gitignore` | `infra/*-tenant.entra.json` (except `.example.json`). | | `tsconfig.base.json` | `shared-auth` path alias. | | `notes/entra-groups-claim-activation.md` | Operator runbook for the Entra admin-centre step (Token configuration → Add groups claim → Security groups → Group ID). | **ADR amendment** | File | Change | | --- | --- | | `docs/decisions/0025-authorization-model-privileges-roles-scopes.md` | Path references updated `libs/feature/auth/...` → `libs/shared/auth/...` (4 occurrences), and the `ENTRA_GROUP_TO_ROLE` static-record example replaced by the actual `EntraGroupToRoleResolver` + JSON shape. | ## Notes for the reviewer - **Why `libs/shared/auth/` and not `libs/feature/auth/` (as the ADR initially said).** The existing `libs/feature/auth/` carries `tags: ['scope:portal-shell', 'type:feature']`, which the Nx `enforce-module-boundaries` rule in `eslint.config.mjs:36-46` blocks the BFF from importing. The catalogue is needed by both sides (BFF builds the `Principal`, SPA will gate UI conditionally later), so a `scope:shared` lib is the correct home. The ADR is patched in the same PR so the document and the code agree. - **Why drop unknown privileges with a WARN instead of preserving them.** `Portal.GhostRole` in the `roles` claim is either a leftover from a different app registration or a v1+1 experiment that hasn't been added to the catalogue yet. Both paths should not silently grant access — drop with a WARN so an operator notices, and the drift CI gate (next PR) catches the source-side equivalent before merge. - **Why the principal builder uses a `ScopeResolver` seam now.** Per ADR-0025 §331 the v1 implementation returns `unrestricted` for everyone; the real version queries the `user_scopes` Prisma table that lands with ADR-0026. The seam is here so the next PR replaces only the implementation, not the call-site in `PrincipalBuilder` or its 24 tests. Per-persona stub scope data was deliberately not embedded in this PR — no guard consumes scopes yet, so it would be write-only documentation that drifts from the eventual Prisma seed. - **Why `user.id` and `user.personId` placeholder to `entraOid`.** Same logic: the real UUIDs land with the `Person` + `User` Prisma schema (proposed ADR-0026). Until then, populating both fields with `entraOid` lets consumers key on `principal.user.id` today and get a real value later without branching on availability. The seam is one place (in the builder), not 24 guards. - **Why session.types.ts carries both `user` and `principal`.** Additive instead of replacement: the audit module + the AI-bridge controller key on `user.oid` / `user.tid` directly, and migrating them in the same PR would balloon the diff for zero behavioural benefit. New consumers (`@RequirePrivilege` / `@RequireRole` / `@RequireScope` decorators, next PR) reach for `principal`. - **Map file fails soft on absence, hard on malformation.** Path unset OR file missing → empty resolver + WARN. Sign-in still succeeds, every user gets `roles: []`. This is deliberately fail-soft so a fresh dev environment isn't blocked by an Entra-side dependency. **Bad JSON / unknown slug / duplicate GUID → throws at boot.** The alternative would silently mis-resolve a role. - **Why `boot-time` validation only (no per-request)** for the map file. Catalogue drift is detected once at boot; per-request validation would re-read the file from disk on every sign-in. The trade-off is that a hot-edit to the file requires a BFF restart — acceptable because the file changes once per tenant lifecycle, not per session. - **Entra `groups` claim must be activated** on the BFF's app registration before this code does anything useful. The runbook in `notes/entra-groups-claim-activation.md` walks through the steps; if the claim isn't activated, every user signs in with `principal.roles = []` and a future warn for every group GUID that would have arrived (none, in that case). - **No drift gate yet.** ADR-0025's §"Confirmation" §346 calls for an ESLint custom rule (or `pnpm run` script) that greps every `@RequireRole('...')` literal and asserts membership in the catalogue. That's the third PR in the phasing — there's no `@RequireRole` consumer in the tree yet, so the gate would have nothing to assert against today. ## Test plan - [x] `pnpm nx affected -t lint test build --base=main` — 13 projects green - 497 BFF tests (was 465 — added 4 new `groups`-claim tests in `auth.service.spec.ts`, 1 new test in `session-establisher.service.spec.ts`, 6 new tests in `load-entra-group-map.spec.ts`, 24 new tests in `principal-builder.spec.ts`) - 17 `shared-auth` tests (catalogue counts + resolver contracts) - 1 `shared-util` test (unchanged) - full lint + full build - [x] `pnpm nx format:check` — clean after `pnpm nx format:write` - [x] Manual: `tsc --build libs/shared/auth/tsconfig.lib.json` emits `.d.ts` files at the path `portal-bff`'s webpack expects. - [ ] **Review focus** — the closed catalogues (`PRIVILEGES`, `FUNCTIONAL_ROLES`, `SCOPE_KINDS`) verbatim match ADR-0025; the 19 personas verbatim match `notes/test-tenant-role-assignments.md`; the `Principal` shape matches the ADR §"Principal shape" (modulo the `user.id` / `personId` placeholder); fail-soft on missing map file, hard on malformed. - [ ] **After merge — operator step** : activate the Entra `groups` claim per `notes/entra-groups-claim-activation.md` and drop the real GUIDs into `infra/test-tenant.entra.json` (gitignored). Then a sign-in with `admin@apfrd.onmicrosoft.com` should land `principal.privileges = ['Portal.Admin']` and `principal.roles = ['collaborateur', 'rh']` in Redis. ## What's next Per ADR-0025 §"More Information" phasing: 1. ✅ **This PR** — types + Principal builder + group-to-role mapping skeleton. 2. **Next PR** — `@RequirePrivilege` + `@RequireRole` + `@RequireScope` decorators + guard integration tests against the 19 personas. 3. **PR after that** — drift CI gate (ESLint custom rule asserting every `@RequireX('...')` literal is in the catalogue). 4. **Depends on ADR-0026** — `user_scopes` Prisma table + seed + `PrismaScopeResolver` replacing `StubScopeResolver`. --------- Co-authored-by: Julien Gautier <julien.gautier@apf.asso.fr> Reviewed-on: #206	2026-05-23 21:09:13 +02:00
julien	883c5151de	feat(portal-bff): ai-bridge controller — SSE chat + JSON rag/models (#196) ... ## Summary Step 3 of the AI-relay chantier (after #194 ADR and #195 client skeleton). Wires the BFF-side **live surface** that the SPA's future chatbot widget will consume. [ADR-0024](docs/decisions/0024-ai-service-relay-grpc-sse-bridge.md) is promoted from `proposed` to `accepted` in the same change. Three end-user routes under `/api/ai/*`, gated by the active portal session (no `@RequireAdmin` — AI is a regular-user surface): | Route | Verb | Wire | Maps to | |---|---|---|---| | `/api/ai/chat` | `POST` | `text/event-stream` | `apf.ai.v1.ChatService.Chat` (server-stream) | | `/api/ai/rag/search` | `GET` | `application/json` | `apf.ai.v1.RagService.Search` (unary) | | `/api/ai/models` | `GET` | `application/json` | `apf.ai.v1.ModelsService.ListModels` (unary) | CSRF and session validation are delegated to the global middleware mounted in `main.ts` (per [ADR-0009](docs/decisions/0009-auth-flow-oidc-pkce-msal-node.md) and [ADR-0021](docs/decisions/0021-phase-2-security-baseline.md)); the controller asserts `req.session.user` and emits 401 if absent. ## What lands ### `apps/portal-bff/src/grpc/ai-bridge/` ``` ai-bridge/ ├── ai-bridge.module.ts imports AiClientModule, exports the controller ├── ai-bridge.controller.ts 3 routes — POST chat (SSE), GET rag/search, GET models ├── sse.writer.ts ChatEvent oneof → SSE frame translator ├── sse.writer.spec.ts unit tests for the codec ├── ai-bridge.controller.spec.ts end-to-end against an in-process fake gRPC server └── dto/ ├── chat-request.dto.ts class-validator body shape (POST /chat) └── rag-search-query.dto.ts class-validator query shape (GET /rag/search) ``` ### SSE codec (`sse.writer.ts`) Each `ChatEvent` oneof case becomes one SSE frame with a kebab-case `event:` name and a JSON-encoded `data:` payload: ``` event: token data: {"token":"…","value":"…"} event: agent-step data: {"agent":"…","step":"…","stepId":"…"} event: tool-call data: {"callId":"…","name":"…","args":{…}} event: done data: {"stats":{"tokensIn":…,"tokensOut":…,"chunksRetrieved":…}} ``` A helper `relayErrorFrame(code, message, retriable)` synthesises a relay-side `event: error` frame that matches the AI service's own `ErrorEvent` shape — the SPA's renderer needs no second code path for relay-level failures vs upstream model errors. gRPC status codes map into the `urn:apf-ai:*` namespace (`UNAVAILABLE` → `urn:apf-ai:unavailable`, `DEADLINE_EXCEEDED` → `urn:apf-ai:timeout`, `PERMISSION_DENIED` → `urn:apf-ai:permission_denied`, `RESOURCE_EXHAUSTED` → `urn:apf-ai:rate_limited`, `INVALID_ARGUMENT` → `urn:apf-ai:invalid_argument`, anything else → `urn:apf-ai:relay_error`). The terminal `done` frame closes the stream — no `[DONE]` sentinel, per ADR-0024. ### Controller (`ai-bridge.controller.ts`) - `POST /api/ai/chat` — builds an `apf.ai.v1.ChatRequest` from the validated DTO + session-derived Principal, calls `ChatClient.chat()`, drains the `ClientReadableStream<ChatEvent>` into SSE frames written on the raw Express `Response`. `req.on('close', …)` propagates browser disconnect through an `AbortController` into `call.cancel()` so the upstream LLM stops (per `apf-ai-service/docs/streaming.md`). - `GET /api/ai/rag/search` — unary RAG call. `topK` defaults to 0 (server picks the default). `source` and `documentId` query params surface the same filter fields the upstream RPC accepts. - `GET /api/ai/models` — unary lookup of the provider catalogue. The SSE writes happen on the raw Express response (manual `setHeader` + `flushHeaders` + `write` + `end`) rather than through NestJS's `@Sse()` decorator, because `@Sse()` is GET-only and the chat endpoint is POST (the SPA carries the conversation history in the body). ### Lifecycle hooks `AiClientModule` now implements `OnApplicationShutdown` and closes the four gRPC stubs (Chat / Rag / Ingestion / Models). The four stubs share the same HTTP/2 channel (gRPC-js dedups on `endpoint + credentials`), so the `close()` calls are cheap, but kept explicit so adding a fifth stub later is an obvious one-line addition. `main.ts` now calls `app.enableShutdownHooks()` so `SIGTERM` / `SIGINT` / `SIGHUP` actually route through the lifecycle interface. ### DTOs `ChatRequestDto` constrains: - `messages` — 1 to 64 entries; each has `role ∈ {user, assistant, system}` (no `tool` — tool messages are constructed BFF-side per ADR-0024 §"Tool-dispatch contract") and `content` ≤ 16 KB. - `conversationId`, `model`, `provider` — optional, ≤ 64 / 128 chars. `RagSearchQueryDto`: - `query` — required, non-empty. - `topK` — optional, integer in `[1, 50]` (the AI service has its own cap; the BFF rejects out-of-range values early). - `source` / `documentId` — optional pass-through filters. ### Documentation - ADR-0024 frontmatter: `status: proposed` → `accepted`. - `docs/decisions/README.md` index reflects the new status. - `CLAUDE.md` Architecture section grows an "AI service relay" bullet; the roll-up line moves from "ADRs 0001 → 0023" to "0001 → 0024"; the shipped-on-main list grows an "AI relay surface" entry. - `apps/portal-bff/.env.example` documents `AI_SERVICE_GRPC_ENDPOINT` / `AI_SERVICE_CLIENT_ID` / `AI_SERVICE_GRPC_TLS` and points operators at `apf-ai-service`'s own docker-compose for the runtime dependency. ## Notes for the reviewer - **No live AI service in this PR's local-dev stack.** `apf-ai-service` runs from its own repo (`/home/jgautier/Works/apf-ai-service`) with its own `infra/docker-compose.yml`. The BFF dials `localhost:8080` by default — the host-published port of the AI service's container. This is option (a) from ADR-0024 §"Open question — Compose orchestration": two independent stacks, dial across via host networking. Merging the compose files into one would couple two release cadences without operational payoff. - **Tests run against an in-process fake `grpc.Server`.** All five spec cases on the controller wire it up against a fake `ChatService` + `RagService` + `ModelsService` server bound to `127.0.0.1:0` (random port). No mocks — the controller's gRPC client makes a real connection, real serialisation, real cancellation propagation. Cost: ~0.5 s overhead from the gRPC server setup. - **CSRF + session middleware are unchanged.** The new POST endpoint is protected by the existing double-submit CSRF middleware mounted in `main.ts` (per [ADR-0021](docs/decisions/0021-phase-2-security-baseline.md)). The SPA's fetch call needs to send the `X-CSRF-Token` header matching the `__Host-portal_csrf` cookie — same protocol as every other POST in the BFF. No per-controller wiring required. - **Manual session check rather than a guard.** Three reasons: (1) matches the existing pattern in `me.controller.ts`; (2) the session check is the only authorization gate (no roles to evaluate) — a guard would add ceremony without payoff; (3) the SSE controller already takes control of the response object (`@Res()`), which `UseGuards` interacts with awkwardly. Throwing `UnauthorizedException` lets `StructuredErrorFilter` produce the 401 envelope before any header is flushed. - **Why the controller does NOT use `@Sse()`.** NestJS's `@Sse()` decorator is GET-only and emits frames from `Observable<MessageEvent>`. The chat endpoint is POST (the SPA sends conversation history in the body) and the source is a Node `Readable` stream from `@grpc/grpc-js`. Manual response handling is simpler than adapting to / from `Observable` for a single consumer. - **Cancellation contract.** When the SPA aborts the fetch, the browser closes the TCP connection, Express emits `'close'` on the request, the controller's `AbortController.abort()` triggers, `ChatClient` calls `.cancel()` on the gRPC stream, the AI service's `ServerCallContext.CancellationToken` cancels the upstream LLM. The spec covers the `'close'` → server-side `cancelled` event end-to-end. - **No ingestion route in the BFF.** Per ADR-0024 §"Out of scope", v1 admin ingestion uses the `apf-ai-service/tools/Apf.Ai.Ingest/` CLI. A future PR adds the BFF endpoint when the admin "manage AI corpus" surface ships. `IngestionClient` remains in `AiClientModule` so that future PR is one new file, not a new module plus a new client. - **No bundle-size or perf surprise.** The BFF is a Node process, not a SPA chunk — bundle budgets don't apply. The gRPC channel is opened lazily on first call; idle BFFs incur no upstream TCP cost. ## Test plan - [x] `pnpm nx test portal-bff` — **461 specs pass** (was 443; +13 new: 8 SSE writer cases + 5 controller end-to-end cases against the in-process fake server). Worker-exit-leak warning persists from the gRPC server's slow shutdown — pre-existing pattern from PR #195; harmless. - [x] `pnpm nx lint portal-bff` — 6 pre-existing warnings, no new ones from the diff. - [x] `pnpm nx build portal-bff` — clean webpack compile. - [x] Module wiring: `AppModule` imports `AiBridgeModule`, which imports `AiClientModule`. Resolves cleanly through DI; the audit-side `HashUserIdService` is satisfied by `AiClientModule`'s local provider (per the rationale recorded in PR #195's `AiClientModule` docstring). - [ ] **Manual smoke** — bring up `apf-ai-service` from its own repo (`cd ../apf-ai-service && docker compose -f infra/docker-compose.yml up`), set `AI_SERVICE_GRPC_ENDPOINT=localhost:8080` in `apps/portal-bff/.env`, run `pnpm nx serve portal-bff`. Sign in to `portal-shell`, then in a terminal: ```bash curl --cookie-jar /tmp/portal-session http://localhost:3000/api/auth/login # follow Entra… curl -N \ -H 'Content-Type: application/json' \ -H 'X-CSRF-Token: <copied from cookie>' \ --cookie /tmp/portal-session \ -d '{"messages":[{"role":"user","content":"hello"}]}' \ http://localhost:3000/api/ai/chat ``` Expect a streamed SSE response terminated by an `event: done` frame. Verify `GET /api/ai/rag/search?query=test` returns a JSON response. Verify `GET /api/ai/models` lists the configured providers. ## What's next 1. **PR (frontend chantier)** — chatbot widget on `portal-shell` consuming the SSE endpoint. Will use `fetch` + `ReadableStream` parsing (not native `EventSource`, since POST is needed). Drag / fullscreen / suggestion UX carries forward from the stargate POC's `ChatbotWidget.tsx`. 2. **PR (post-v1)** — proto-drift CI gate that diffs `proto/apf-ai/` against an upstream tag of `apf-ai-service`. 3. **Coordinated amendment** — when the first production deployment is in scope, both repos record the same prod-hardening choice (signed `Principal` envelope vs mTLS) on the same date. --------- Co-authored-by: Julien Gautier <julien.gautier@apf.asso.fr> Reviewed-on: #196	2026-05-19 22:39:35 +02:00
julien	9b7d16601d	feat(portal-bff): ai-client skeleton — vendored protos + grpc client + Principal mapper (#195) ... ## Summary Step 2 of the AI-relay chantier (after [ADR-0024](docs/decisions/0024-ai-service-relay-grpc-sse-bridge.md) merged in #194). Lands the BFF-side **skeleton** that talks to `apf-ai-service` over gRPC: vendored protos, generated TypeScript stubs, typed wrapper clients, Principal mapper, and metadata builder — all tested against an in-process fake gRPC server. **No HTTP route is exposed in this PR**; the SSE bridge (`POST /api/ai/chat`, `GET /api/ai/rag/search`, `GET /api/ai/models`) ships in the next PR. The skeleton is self-contained: `AiClientModule` is built but is NOT imported in `AppModule` yet. The BFF runtime is byte-for-byte unchanged. Everything below exists for the next PR to wire into a controller. ## What lands ### Proto vendoring + codegen - `apps/portal-bff/src/grpc/proto/apf-ai/` — mirror of `apf-ai-service/contract/proto/` (common, chat, rag, ingestion, models). Both the `.proto` files and the regenerated `ts-proto` output under `grpc/gen/` are committed for hermetic builds and reviewable diffs (per ADR-0024 §"Sub-decision 3"). - `pnpm run grpc:codegen` — regenerates the stubs via `grpc-tools`' bundled `protoc` and the `ts-proto` plugin (`outputServices=grpc-js, esModuleInterop, forceLong=long, useOptionals=messages, exportCommonSymbols=false`). - `pnpm run grpc:sync` — copies the vendored `.proto` files from the sibling `apf-ai-service` working tree (`../apf-ai-service/contract/proto/`); developer convenience, never invoked from CI. Errors with an actionable message when the sibling tree is not where it expects. - Generated tree (`grpc/gen/**`) excluded from Prettier (`.prettierignore`) and ESLint (`eslint.config.mjs` ignores). Hand-rules apply to wrappers under `ai-client/`, not to codegen output. ### Dependencies - `@grpc/grpc-js@^1.13.0` — runtime gRPC client. - `@bufbuild/protobuf@^2.10.2` — wire codec used by `ts-proto`'s emitted code. - `long@^5.2.3` — int64 representation for proto Long fields (`forceLong=long`). - `ts-proto@^2.7.0` — devDep, TypeScript codegen plugin. - `grpc-tools@^1.13.0` — devDep, ships `protoc` + the gRPC plugin; added to `pnpm.onlyBuiltDependencies` so the postinstall binary download runs. ### Env validator `apps/portal-bff/src/config/check-ai-service-config.ts` follows the same posture as the other validators (per ADR-0018 §"BFF env-var loading"): small, per-key, runs at module init, throws with an actionable message on misconfiguration. Three vars: | Var | Mandatory | Purpose | |---|---|---| | `AI_SERVICE_GRPC_ENDPOINT` | yes | `host:port` reachable from the BFF — `apf-ai-service:8080` in dev Compose, service DNS + 443 in prod | | `AI_SERVICE_CLIENT_ID` | yes | Deployment slug propagated as the `x-client-id` metadata. Convention: `apf-portal-<env>` | | `AI_SERVICE_GRPC_TLS` | no (default `true`) | `false` for h2c in dev, `true` for h2 + TLS in prod | 7 spec cases lock the validation contract end-to-end. ### `AiClientModule` `apps/portal-bff/src/grpc/ai-client/` houses: - **`tokens.ts`** — DI tokens (`AI_CONFIG`, `AI_CREDENTIALS`, one per generated stub). - **`principal.mapper.ts`** — `PrincipalMapper.fromInputs({oid, tid, roles, extraAttributes})` returns the proto `Principal`. `subject` is hashed via `HashUserIdService` (the **same** salt + algorithm the audit writer uses) so `Principal.subject` matches `audit.events.actor_id_hash` byte-for-byte. `roles` passes through verbatim — inclusive expansion lands with a future role-hierarchy ADR; the wire contract on the AI side is unchanged. - **`grpc-metadata.builder.ts`** — stamps every outbound call with `x-client-id` (from config) and `x-correlation-id` (active OTel span's trace-id when present, else explicit override, else fresh UUID). - **`chat.client.ts`** — server-stream wrapper around `ChatServiceClient`. Returns the raw `ClientReadableStream<ChatEvent>` (Node `Readable` is async-iterable so the SSE bridge consumes with `for await`). Optional `AbortSignal` propagates browser disconnect to `call.cancel()`. - **`rag.client.ts`**, **`models.client.ts`**, **`ingestion.client.ts`** — unary promisified wrappers. `IngestionClient` is unused in v1 (the admin "manage AI corpus" surface lands later) but wired now so the future controller is one new file, not a new module. - **`ai-client.module.ts`** — NestJS module wiring the providers. `HashUserIdService` is declared locally rather than imported via `AuditModule` (the global module) to keep the module unit-testable in isolation; runtime equivalence is guaranteed by the service being a pure function of `LOG_USER_ID_SALT`. ### Tests 5 new spec files, 18 new test cases. All run against in-process fake gRPC servers; no network, no live `apf-ai-service`: - `check-ai-service-config.spec.ts` — 7 cases (happy path + 6 rejection branches). - `principal.mapper.spec.ts` — 7 cases including the cross-service hash-stability invariant and the `tenantId` reserved-key contract. - `grpc-metadata.builder.spec.ts` — 5 cases covering all three correlation-id resolution paths and metadata immutability. - `chat.client.spec.ts` — 4 cases: happy-path stream, metadata propagation, mid-stream cancel via AbortSignal, pre-aborted signal. - `rag.client.spec.ts` — 2 cases: unary happy path, `ServiceError` propagation. - `ai-client.module.spec.ts` — 4 cases: module bootstrap, all four wrappers resolved, env-driven `AI_CONFIG`, shared credentials across stubs. **Total BFF spec suite: 443 → 461 (after merge accounting), all passing.** ## Notes for the reviewer - **Why both `.proto` and generated `.ts` are committed.** Hermetic builds. Reviewers see both sides of a contract change in one diff. CI never runs codegen — drift between proto and stub would otherwise hide behind a successful build. The drift gate that compares the vendored copy against an upstream tag of `apf-ai-service` is the post-v1 follow-up listed in ADR-0024's "What's next". - **`HashUserIdService` declared locally in `AiClientModule`.** Two instances of the service exist when both `AuditModule` (global) and `AiClientModule` are wired into `AppModule`. The cost is one extra constructor call at bootstrap; the value is full test isolation of `AiClientModule` and a self-contained Principal-mapping boundary. The cross-service hash join invariant from ADR-0013 still holds because the service is a pure function of the `LOG_USER_ID_SALT` env var. - **`AiClientModule` is NOT imported by `AppModule`.** Deliberately. The skeleton compiles, type-checks, and unit-tests cleanly with no runtime side effects. The next PR (SSE bridge controller) adds the `imports: [AiClientModule]` line + the controller, in one focused change. - **`ts-proto` flat-oneof emission.** `ChatEvent` is generated with optional siblings (`token?`, `citation?`, `done?`, …) rather than a discriminated union (`$case: 'token'`). The flatter shape composes more naturally with the SSE writer the next PR will introduce (`event:` field name maps directly to the populated sibling). - **Cancellation test deliberately relaxed.** The "AbortSignal already aborted before call dial" test asserts the client-side outcome (no payload, error or clean end) but not server-side observation. gRPC-js may or may not propagate a cancel frame depending on whether the call had time to dial — both outcomes are correct per the contract; only the absence of payload matters. - **Lifecycle (`onApplicationShutdown`) deferred.** The module does not close the gRPC channel on shutdown today. Process termination closes the sockets via OS-level descriptor reclaim — sufficient for dev/preprod. The next PR wires the module into `AppModule` and adds an explicit Nest lifecycle hook in the same change (paired with `app.enableShutdownHooks()` in `main.ts`). ## Test plan - [x] `pnpm run grpc:codegen` — clean regeneration. Generated tree byte-identical to what's committed. - [x] `pnpm nx test portal-bff` — **443 specs pass** (was 425). - [x] `pnpm nx lint portal-bff` — clean. The eslint ignore for `grpc/gen/**` covers ts-proto's relaxed style; hand-written `ai-client/` files pass the project's full rule set. - [x] `pnpm nx build portal-bff` — clean, webpack-compiled. No bundle-size delta on the runtime artefact yet (the module is unimported). - [x] `pnpm install` — lockfile reconciled; `grpc-tools` postinstall fetches `protoc` from the precompiled-binaries mirror without errors on Linux x64. - [ ] **Manual smoke (next PR)** — once the SSE bridge ships, point `AI_SERVICE_GRPC_ENDPOINT` at the local `apf-ai-service` Compose service and run an end-to-end chat against the canned stub responses on the AI side. Not in scope for this PR. ## What's next 1. **PR — SSE bridge controller.** Wires `AiClientModule` into `AppModule`, adds `POST /api/ai/chat` (SSE), `GET /api/ai/rag/search`, `GET /api/ai/models`. Adds the `OnApplicationShutdown` hook + `enableShutdownHooks()`. Adds `apf-ai-service` to `infra/local/dev.compose.yml`. Promotes ADR-0024 from `proposed` to `accepted` and updates `CLAUDE.md`'s ADR roll-up. 2. **PR (frontend chantier)** — chatbot widget on `portal-shell` consuming the SSE endpoint. 3. **PR (post-v1)** — proto-drift CI gate diffing the vendored `proto/apf-ai/` against the upstream tag. 4. **Coordinated amendment** — when the first production deployment is in scope, both repos record the same prod-hardening choice (signed envelope or mTLS) on the same date. --------- Co-authored-by: Julien Gautier <julien.gautier@apf.asso.fr> Reviewed-on: #195	2026-05-19 21:30:04 +02:00