ds-chat/backend/README.md

KeepItTalking backend (Phase 1 + 2 + 4 + 5 + 6)

FastAPI + SQLAlchemy 2.0 (async) + PostgreSQL + Redis. Implements auth, room CRUD (open and private), room roles (owner/admin/member) and invites, a WebSocket chat endpoint that fans out across multiple app-server instances via Redis pub/sub, Web Push notifications for offline room members, and a site-admin portal (user/room management + an audit log). See ../ARCHITECTURE.md for the full system design and the phased build plan.

This is an invite-only site: there is no public registration endpoint. Accounts are created by an operator on the app server — see step 4 below.

Local dev setup

1. Postgres

Any local Postgres 14+ works. The quickest option is a container:

docker run -d --name chatapp-postgres \
  -e POSTGRES_USER=chatapp -e POSTGRES_PASSWORD=chatapp -e POSTGRES_DB=chatapp \
  -p 5432:5432 postgres:16-alpine

Then create the test database (used by the test suite, kept separate from dev data):

docker exec chatapp-postgres psql -U chatapp -d chatapp -c "CREATE DATABASE chatapp_test;"

(Docker here is purely a local-dev convenience for standing up Postgres quickly — the actual deployment target has no containers at all, see ARCHITECTURE.md §9.)

2. Redis

Used for cross-instance WebSocket fan-out and presence (see the section below). Required — there's no in-memory fallback.

docker run -d --name chatapp-redis -p 6379:6379 redis:7-alpine

3. Python environment

cd backend
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
cp .env.example .env
# edit .env: set SESSION_SECRET to a long random string, e.g.
#   python3 -c "import secrets; print(secrets.token_urlsafe(32))"

4. Migrations

.venv/bin/alembic upgrade head

5. Create a user

There's no public sign-up. Create accounts directly with the CLI (add --admin to grant is_site_admin, which unlocks the admin portal at /admin on the frontend and the /api/admin/* routes below):

.venv/bin/python -m app.cli create-user alice alice@example.com "some-password"

6. (Optional) Set up push notifications

Push works without any setup — VAPID_PUBLIC_KEY/VAPID_PRIVATE_KEY are unset by default and push delivery is silently skipped. To enable it:

.venv/bin/python -m app.cli generate-vapid-keys
# paste the three printed lines into backend/.env

7. Run the dev server

.venv/bin/uvicorn app.main:app --reload

API docs: http://localhost:8000/docs. WebSocket chat endpoint: ws://localhost:8000/ws/chat.

To try horizontal scaling locally, run a second instance on another port against the same Postgres + Redis (.venv/bin/uvicorn app.main:app --port 8001) — a message sent through one instance's WebSocket is delivered to clients connected to the other, purely via Redis.

8. Run tests

Tests run against a real Postgres database (chatapp_test by default — native ENUM/UUID types aren't faithfully reproduced by SQLite) and a real Redis (db 15 by default, kept separate from dev use of db 0), with each test wrapped in a transaction that's rolled back afterward:

DATABASE_URL=postgresql+asyncpg://chatapp:chatapp@localhost:5432/chatapp_test .venv/bin/pytest

Layout

app/
  main.py            create_app(), session middleware, router/WS mounting
  config.py           environment-driven settings (pydantic-settings)
  database.py          async engine/session, get_db() dependency
  dependencies.py       get_current_user, require_room_member, require_room_role,
                           require_site_admin
  security.py            argon2 password hashing
  cli.py                  `python -m app.cli create-user` / `generate-vapid-keys`
  models/                 SQLAlchemy models (users, rooms, room_memberships,
                             messages, room_invites, push_subscriptions,
                             admin_audit_log)
  schemas/                 Pydantic request/response models
  routers/                  auth, rooms, invites, push, admin, health
  services/                  business logic called by routers
  ws/                        connection_manager (local sockets), presence +
                               broadcaster (Redis), /ws/chat handler
alembic/                      migrations
tests/                         pytest + httpx/TestClient tests

Admin portal (Phase 6)

Every /api/admin/* route (app/routers/admin.py) requires current_user.is_site_admin (checked via require_site_admin, app/dependencies.py) and is backed by app/services/admin_service.py:

• Users: list, deactivate/reactivate (User.is_active), reset password, promote/demote is_site_admin. An admin can't deactivate or demote their own account (CannotActOnSelfError → 400) — the one guard against an admin locking themselves out. Deactivation takes effect immediately, even for an already-open session: get_current_user re-checks is_active on every request since it already loads the user row.
• Rooms: list every room including private ones (unlike the member-facing GET /api/rooms, which is open-rooms-only), archive/ unarchive (Room.is_archived — archived rooms drop out of the open-room browse list but stay readable for existing members, matching how Mattermost archive works), and force a transfer of ownership to any existing member without needing to already be the owner (the "admin override" of the member-initiated transfer in room_service.py, which otherwise requires exactly that).
• Audit log: every mutating admin action writes one AdminAuditLog row (actor, action, target type/id, JSON metadata) in the same transaction as the change, listed newest-first via GET /api/admin/audit-log.

Two items from the original phase scope are deliberately not here yet:

• Bot/integration management — nothing to manage until Phase 7 builds the actual bot data model (api_tokens, webhooks_incoming, event_subscriptions per ARCHITECTURE.md §4); it'll be built alongside that data model instead of as an empty panel now.
• System settings — no settings storage or concrete setting exists yet. The frontend has an empty "Settings" tab as a placeholder for when one does.

Cross-instance broadcast (Phase 5)

The WebSocket layer is split into three pieces so that running one app instance and running many behave identically:

• app/ws/connection_manager.py — purely local: which sockets on this process are in which room, used only to actually send_json to them.
• app/ws/broadcaster.py (RoomBroadcaster) — on a chat message, publish()s it to a Redis channel scoped to the room (room:{id}). Every app instance, including the publisher, runs a single background listen() task (started in app/main.py's lifespan) pattern-subscribed to room:*; each message it receives is handed to its own local ConnectionManager.broadcast(). One instance just talks to itself through Redis, so there's no separate single-instance code path.
• app/ws/presence.py (Presence) — a Redis hash per room (presence:{room_id}, field = user ID, value = connection refcount) is the cross-instance answer to "is this member connected anywhere right now," which is what the Phase 4 offline-push check uses instead of the local ConnectionManager. Refcounted so a user connected from two tabs (or two instances) isn't marked offline until every connection closes.

Known limitation: Presence has no heartbeat/TTL, so a hard process crash (not a clean disconnect) leaks that connection's increment forever — same category of simplification as the "no server-side session revocation" note below.

Push notifications (Phase 4)

POST /api/push/subscribe (upserts by endpoint) / DELETE /api/push/subscribe manage a user's push_subscriptions rows; GET /api/push/vapid-public-key gives the frontend the key it needs for PushManager.subscribe(). On every chat message, app/ws/chat.py computes room members - Presence. connected_user_ids(room_id) (who's actually connected to that room right now, across every app instance — see Phase 5 below) and sends each offline member a push via pywebpush, awaited inline against the same request-scoped session rather than fired as a background task — the broadcast to online members already happened by that point, so nothing online-facing is delayed, and it sidesteps asyncio.create_task()s outliving the session/event loop they were created on. An expired/invalid subscription (pywebpush 404/410) is deleted automatically.

Room roles and invites (Phase 2)

Rooms can be open (anyone can join via POST /api/rooms/{id}/join) or private (is_private: true at creation — joinable only via invite). Room roles are owner > admin > member:

• member: post messages, leave the room
• admin: edit room settings, create/list/revoke invites, remove plain members
• owner: everything admin can, plus delete the room, remove admins, change member roles, and transfer ownership

Invite flow: an admin+ calls POST /api/rooms/{id}/invites with an existing target_username; the invited user sees it via GET /api/invites/mine and calls POST /api/invites/{id}/accept (or /decline). GET /api/rooms/mine lists every room (open + private) the current user belongs to, alongside their role.

Notes / scope decisions

• Invite-only site registration: no POST /api/auth/register. Accounts are provisioned with python -m app.cli create-user (see step 4 above). This is separate from room invites above — site accounts vs. room membership.
• Room invites are by username only — room_invites.target_email exists in the schema (per ARCHITECTURE.md) but is unused, since there's no email-delivery mechanism anywhere in the stack yet.
• Sessions are signed cookies (Starlette SessionMiddleware), not a server-side session table — see ARCHITECTURE.md's rationale (simplest way to carry auth through a WebSocket handshake). This means there's currently no way to force- revoke a session server-side; that needs a real session table later.
• No CSRF token yet — SameSite=Lax cookies plus a same-origin frontend dev proxy (see ../frontend/vite.config.ts) is the accepted phase-1 mitigation.
• Deleting a room explicitly deletes its messages/memberships/invites first (room_service.delete_room) rather than relying on DB-level cascades.
• admin_audit_log has no admin UI for filtering/searching yet — it's a flat newest-first list with limit/offset pagination, no filter by actor/action/target. Fine at current scale; revisit if the log grows.