# Services Architecture

Galaxy: Turn-based Strategy Game

## Purpose

This document defines the high-level architecture of the Galaxy Ga,e platform as a single source of truth for implementing all core microservices.

It describes:

* public and trusted service boundaries;
* ownership of main business entities and state;
* request routing and transport rules;
* interaction rules between services;
* runtime model for game containers;
* notification and event propagation model;
* recommended implementation order.

Detailed behavior of each concrete service belongs in its own README.
This document fixes the system-level structure and the architectural rules that must remain stable across service implementations.

## Scope

Galaxy Game is a multiplayer turn-based online strategy game platform.

Core product properties:

* many game sessions may exist simultaneously;
* one user may participate in multiple games at once;
* users authenticate by e-mail confirmation code;
* users have platform roles and tariff/entitlement state;
* games may be public or private;
* public games are managed by system administrators;
* private games are created and managed by eligible paid users;
* each running game is executed inside its own dedicated game engine container;
* each running game is bound to one concrete engine version;
* in-place upgrade of a running game is allowed only as a patch update within the same semver major/minor line;
* player commands are turn-bound and are accepted only before the next scheduled turn generation cutoff.

The platform stores durable business state in PostgreSQL (one shared database, schema per service) and uses Redis with Redis Streams for ephemeral state, caches, and the internal event bus. The backend split, library stack, and staged migration plan live in [`PG_PLAN.md`](PG_PLAN.md) and the [Persistence Backends](#persistence-backends) section below.

## Main Principles

* The platform exposes a single external entry point: **Edge Gateway**.
* Public unauthenticated flows use REST/JSON.
* Authenticated user edge traffic uses signed gRPC over HTTP/2 with protobuf control envelopes and FlatBuffers payload bytes.
* Trusted synchronous inter-service traffic uses REST/JSON unless a service-specific contract states otherwise.
* For the direct `Gateway -> User` self-service boundary, gateway keeps the external authenticated gRPC + FlatBuffers contract and performs REST/JSON transcoding toward `User Service` internally.
* The gateway handles only edge concerns: parsing, authentication, integrity checks, anti-replay, rate limiting, routing, and push delivery. Business authorization and domain rules remain in downstream services.
* `Auth / Session Service` is the source of truth for `device_session`, but it is not on the hot path of every authenticated request. Gateway authenticates steady-state traffic from session cache and lifecycle updates.
* `Game Lobby` owns platform-level metadata of game sessions.
* `Game Master` owns runtime and operational state of running games.
* `Runtime Manager` is the only service allowed to access Docker API directly.
* `Notification Service` is the platform-level delivery/orchestration layer for push and most non-auth email notifications.
* `Mail Service` sends email; auth-code mail is sent directly by `Auth / Session Service`, while all other platform mail is initiated through `Notification Service`.
* `Geo Profile Service` is auxiliary and fail-open relative to gameplay; it never blocks the currently processed request and may affect only later requests.
* If a user-facing request must complete with a deterministic result in the same flow, the critical internal chain must be synchronous. If the interaction is propagation, notification, cache update, runtime job completion, telemetry, or denormalized read-model update, it should be asynchronous.

## Security and Transport Model

The former standalone security model is part of the main architecture and is no longer treated as a separate subsystem.

### Public and authenticated transport classes

The gateway already distinguishes:

* public REST/JSON for unauthenticated traffic such as health checks and public auth;
* authenticated gRPC over HTTP/2 for verified commands and push delivery.

For downstream business services, the current default trusted transport is
strict REST/JSON. Gateway may therefore authenticate and verify one external
FlatBuffers command, then transcode it to one trusted downstream REST call.

When forwarding an authenticated command to a downstream service, `Edge Gateway`
enriches the REST call with the `X-User-ID` header carrying the verified platform
user identifier. Downstream services derive the acting user identity exclusively
from this header and must never accept identity claims from request body fields.

The public auth contract is:

* `send-email-code(email) -> challenge_id`
* `confirm-email-code(challenge_id, code, client_public_key, time_zone) -> device_session_id`

The authenticated request contract is based on:

* `device_session_id`
* `message_type`
* `timestamp_ms`
* `request_id`
* `payload_hash`
* Ed25519 client signature over canonical envelope fields.

Server responses and push events are signed by the gateway so clients can verify server-originated messages. Push streams are bound to authenticated `user_id` and `device_session_id`, and session revoke closes only streams bound to the revoked session.

### Verification boundary

Before routing an authenticated request, gateway must:

1. validate envelope presence and protocol version;
2. resolve session from session cache;
3. reject unknown or revoked sessions;
4. verify `payload_hash`;
5. verify client signature;
6. verify freshness window;
7. verify anti-replay by `device_session_id + request_id`;
8. apply edge rate limits and basic policy checks;
9. build an authenticated internal command context and only then route downstream.

Downstream services must never receive unauthenticated external traffic.

## High-Level System Diagram

```mermaid
flowchart LR
    Client["Game Client\n(native / browser)"]
    AdminUI["Admin UI"]
    Gateway["Edge Gateway\nPublic REST\nAuthenticated gRPC\nAdmin REST"]
    Auth["Auth / Session Service"]
    User["User Service"]
    Lobby["Game Lobby Service"]
    GM["Game Master"]
    Runtime["Runtime Manager"]
    Notify["Notification Service"]
    Mail["Mail Service"]
    Geo["Geo Profile Service"]
    Billing["Billing Service\nfuture"]
    Redis["Redis\nCache, Streams, Leases"]
    Postgres["PostgreSQL\nDurable Business State"]
    Telemetry["Telemetry"]

    Client --> Gateway
    AdminUI --> Gateway

    Gateway --> Auth
    Gateway --> User
    Gateway --> Lobby
    Gateway --> GM
    Gateway --> Geo

    Auth --> User
    Auth --> Mail
    Auth --> Redis

    User --> Redis

    Lobby --> User
    Lobby --> GM
    Lobby --> Runtime
    Lobby --> Redis

    User --> Lobby

    GM --> Lobby
    GM --> Runtime
    GM --> Redis

    Geo --> Auth
    Geo --> User
    Geo --> Redis

    Notify --> Gateway
    Notify --> Mail
    Notify --> Redis

    Runtime --> Redis

    Mail --> Redis
    User --> Postgres
    Mail --> Postgres
    Notify --> Postgres
    Lobby --> Postgres

    Billing --> User
    Telemetry --- Gateway
    Telemetry --- Auth
    Telemetry --- User
    Telemetry --- Lobby
    Telemetry --- GM
    Telemetry --- Runtime
    Telemetry --- Notify
    Telemetry --- Geo
```

The baseline gateway/auth/session/pub-sub model above is consistent with the existing architecture and service READMEs.

## Service List and Responsibility Boundaries

## 1. [Edge Gateway](gateway/README.md)

`Edge Gateway` is the only public entry point for all external traffic. It already owns transport parsing, session-cache-based authentication, signature verification, freshness/replay checks, edge rate limiting, routing, and push delivery. It must remain free of domain-specific business logic.

External surfaces:

* public REST:

  * health and readiness;
  * public auth commands;
  * browser/bootstrap and public route classes where needed.
* authenticated gRPC:

  * generic `ExecuteCommand`;
  * authenticated `SubscribeEvents`.
* admin REST:

  * separate public administrative surface for system administrators;
  * routed only for authenticated users with admin role.

The gateway does not directly access game engine containers.
For running games it routes to `Game Master`.
For pre-game platform flows it routes to `Game Lobby`.
For user-profile requests it routes to `User Service`.
For public auth it routes to `Auth / Session Service`.

## 2. [Auth / Session Service](authsession/README.md)

`Auth / Session Service` owns:

* challenge lifecycle;
* e-mail-code authentication;
* creation of `device_session`;
* registration of the client Ed25519 public key;
* revoke/logout/block state;
* trusted internal read/revoke/block API;
* projection of session lifecycle state into gateway-consumable Redis data.

It is the source of truth for:

* authentication challenges;
* `device_session`;
* revoke/block state.

Important architectural rules:

* public auth stays synchronous;
* `confirm-email-code` returns a ready `device_session_id`;
* no async “pending session provisioning” step exists;
* session source of truth and gateway-facing projection remain separate;
* active-session limits are configuration-driven;
* `send-email-code` stays success-shaped for existing, new, blocked, and throttled email flows.

When `confirm-email-code` reaches first successful completion for an e-mail
address that does not yet belong to a user, auth may pass create-only
registration context to `User Service` during the synchronous ensure/create
step.

Direct integrations:

* synchronous to `User Service` for user resolution/create/block decision;
* synchronous to `Mail Service` for auth-code delivery;
* asynchronous session lifecycle projection into Redis for gateway consumption.

## 3. [User Service](user/README.md)

`User Service` owns regular-user identity and profile as platform-level
business data.

It is the source of truth for:

* `user_id` of regular platform users;
* `user_name` — immutable auto-generated unique platform handle in
  `player-<suffix>` form; never used as foreign key in other models;
* `display_name` — mutable free-text user-editable label validated through
  `pkg/util/string.go:ValidateTypeName`; not required to be unique; default
  empty for new accounts;
* editable user settings (`preferred_language`, `time_zone`);
* current tariff/entitlement state including `max_registered_race_names`;
* user-specific limits and platform sanctions (including
  `permanent_block` and `max_registered_race_names` override limits);
* latest effective `declared_country`;
* soft-delete state via `DeleteUser`.

`User Service` does not own in-game `race_name` values; those live in
`Game Lobby` Race Name Directory.

System-administrator identity remains outside this service and belongs to the
later `Admin Service`. Trusted administrative reads and mutations against
regular-user state do not make `User Service` the owner of administrator
identity.

It is directly reachable through gateway for selected user-facing operations such as:

* reading and editing allowed profile fields;
* viewing tariff and entitlement state;
* viewing user settings;
* viewing current restrictions and sanctions.

Not every profile mutation goes directly here. For example:

* email change must use a code-confirm flow;
* `declared_country` change remains under admin approval flow via `Geo Profile Service`.

Architectural rules fixed for this service:

* `User Service` owns regular-user identity only; system-admin identity is out
  of scope.
* `User Service` stores only the current effective `declared_country`; review
  workflow and history belong to `Geo Profile Service`.
* `User Service` does not own in-game `race_name` values. All in-game name
  state (registered, reserved, pending registration) lives in the Game Lobby
  Race Name Directory. The only identity strings owned by `User Service` are
  `user_name` (immutable) and `display_name` (mutable, non-unique).
* `permanent_block` is a dedicated sanction code that collapses every
  `can_*` eligibility marker to false and triggers RND cascade release via
  the `user:lifecycle_events` stream.
* `DeleteUser` is a trusted internal endpoint that soft-deletes the account,
  rejects all subsequent operations with `subject_not_found`, and triggers
  the same RND cascade release.
* During the current auth-registration rollout, `Auth / Session Service`
  passes a preferred-language candidate derived from public
  `Accept-Language`, falling back to `en` when no supported value is
  available, plus the confirmed `time_zone` into `User Service`.

Future billing does not become a direct dependency of other services. `Billing Service` will feed entitlement/payment outcomes into `User Service`, and the rest of the platform will continue to use `User Service` as the source of truth for current entitlements.

## 4. [Mail Service](mail/README.md)

`Mail Service` is the internal email delivery service.

Split of responsibility:

* auth code emails: `Auth / Session Service -> Mail Service` directly;
* all other user/admin notification emails: `Notification Service -> Mail Service`.

Transport rules:

* `Auth / Session Service -> Mail Service` uses the dedicated synchronous
  trusted internal REST contract `POST /api/v1/internal/login-code-deliveries`;
* `Notification Service -> Mail Service` is an asynchronous internal command
  flow carried through dedicated queue-backed handoff after durable route
  acceptance inside `Notification Service`.

This split is covered by integration tests: auth-code delivery bypasses
`Notification Service`, while notification-generated mail uses template-mode
commands whose `template_id` equals `notification_type`.

`Mail Service` may internally queue both flows.
Its trusted operator read and resend APIs are part of the v1 service surface,
not a later add-on.
For auth callers, a successful result means the request was durably accepted
into the mail-delivery pipeline or intentionally suppressed; it does not
require that the external SMTP exchange already completed before the response
is returned.
Stable service-local delivery rules, retry semantics, and storage details
(PostgreSQL for the durable delivery record, attempt history, dead letters,
and audit; Redis for the inbound `mail:delivery_commands` stream and its
consumer offset) belong in [`mail/README.md`](mail/README.md), not in the
root architecture document.

## 5. [Geo Profile Service](geoprofile/README.md)

`Geo Profile Service` is an internal trusted auxiliary service for country-level connection signals of authenticated users.

It integrates with:

* gateway as asynchronous ingest producer;
* `User Service` for current effective `declared_country`;
* `Auth / Session Service` for suspicious session blocking;
* `Notification Service` for optional admin notifications.

It owns:

* observed country facts;
* per-session country aggregation;
* `usual_connection_country`;
* `country_review_recommended`;
* history of `declared_country` changes.

It does not block the request that triggered suspicion.
It can only request block of suspicious sessions for subsequent requests.
It does not call `Mail Service` directly; optional admin mail must flow
through `Notification Service`.

In this document, references to `Edge Service` in older geo documentation should be understood as `Edge Gateway`.

## 6. Admin Service

`Admin Service` is the external backend/orchestration layer for the administrative UI.

It is not a heavy domain owner.
Its job is to:

* expose administrator-facing workflows;
* call trusted internal APIs of other services;
* aggregate administrative views where needed;
* enforce system-admin role checks at the gateway/admin boundary.

System administrators can view and operate on all games, including private ones.

## 7. [Game Lobby Service](lobby/README.md)

`Game Lobby` owns platform-level metadata and lifecycle of game sessions as platform entities.

It is the source of truth for:

* game records before and after runtime existence;
* public/private game type;
* owner of a private game;
* user-bound invitations and invite lifecycle;
* applications and approvals;
* membership and roster;
* blocked/removed participants at platform level;
* turn schedule configuration;
* target engine version for launch;
* user-facing lists of games;
* denormalized runtime snapshot imported from `Game Master`.

`Game Lobby` is the source of truth for:

* party membership;
* invited / pending / active / finished / removed status of players relative to games;
* user-visible lists such as `active / finished / pending / invited games`.

It also stores a denormalized runtime snapshot for convenience, at least:

* `current_turn`;
* `runtime_status`;
* `engine_health_summary`.

Additionally, `Game Lobby` aggregates per-member game statistics from
`player_turn_stats` carried on each `runtime_snapshot_update` event: current
and running-max of `planets`, `population`, and `ships_built`. The aggregate
is retained from game start until capability evaluation at `game_finished`.

This prevents user-facing list/read flows from fan-out requests into `Game Master`.

### Lobby status model

Minimum platform-level status set:

* `draft`
* `enrollment_open`
* `ready_to_start`
* `starting`
* `start_failed`
* `running`
* `paused`
* `finished`
* `cancelled`

`Lobby.paused` is a business/platform pause, distinct from engine/runtime failure states.

`start_failed` indicates that the runtime container could not be started or that
metadata persistence failed after a successful container start.
From `start_failed` an admin or owner may retry (→ `ready_to_start`) or cancel (→ `cancelled`).

### Enrollment rules

Each game stores three enrollment configuration fields set at creation:

* `min_players` — minimum approved participants required before the game may start.
* `max_players` — target roster size that activates the gap admission window.
* `start_gap_hours` — hours to keep enrollment open after `max_players` is reached.
* `start_gap_players` — additional players admitted during the gap window.
* `enrollment_ends_at` — UTC Unix timestamp at which enrollment closes automatically.

Transition from `enrollment_open` to `ready_to_start` occurs via one of three paths:

1. **Manual**: an admin (public game) or owner (private game) issues a close-enrollment
   command when `approved_count >= min_players`.
2. **Deadline**: `enrollment_ends_at` is reached and `approved_count >= min_players`.
3. **Gap exhaustion**: `approved_count >= max_players` activates a gap window of
   `start_gap_hours` during which up to `start_gap_players` additional participants
   may join; the transition fires when the gap window expires or
   `approved_count >= max_players + start_gap_players`.

All pending invites transition to `expired` when the game moves to `ready_to_start`.

### Membership rules

* `User Service` owns users of the platform as identities.
* `Game Lobby` owns membership in concrete games.
* game engine does not own platform membership;
* `Game Master` may cache membership for runtime authorization, but `Game Lobby` remains the source of truth.

### Public vs private game rules

Public games:

* created and controlled by system administrators;
* visible in public list;
* joining is based on application and manual admin approval in v1.

Private games:

* can be created only by eligible paid users;
* visible only to their owner and to invited users whose invitation is bound
  to a concrete `user_id` and later accepted;
* joining uses a user-bound invite; accepting the invite immediately creates active
  membership without a separate owner-approval step;
* invite lifecycle belongs entirely to `Game Lobby`.

Private-party owners get a limited owner-admin capability set, not full system admin power.

### Race Name Directory

`Race Name Directory` (RND) is the platform source of truth for in-game player
names (`race_name`). It is owned by `Game Lobby` in v1 and is scheduled to move
to a dedicated `Race Name Service` later without changing the domain or
service-layer logic.

RND owns three levels of state per name:

* **registered** — platform-unique permanent names owned by one regular user.
  A registered name cannot be transferred, released, or renamed; the only path
  back to availability is `permanent_block` or `DeleteUser` on the owning
  account. The number of registered names a user can hold is bounded by the
  current tariff (`max_registered_race_names` in the `User Service` eligibility
  snapshot): `free=1`, `paid_monthly=2`, `paid_yearly=6`,
  `paid_lifetime=unlimited`. Tariff downgrade never revokes existing
  registrations; it only constrains new ones.
* **reservation** — per-game binding created when a participant joins a game
  through application approval or invite redeem. The reservation key is
  `(game_id, canonical_key)`. One user may hold the same name simultaneously
  across multiple active games. A reservation survives until the game
  finishes, then either becomes a `pending_registration` (see below) or is
  released.
* **pending_registration** — a reservation that survived a capable finish and
  is now waiting up to 30 days for the owner to upgrade it into a registered
  name via `lobby.race_name.register`. Expiration releases the binding.

**Canonical key** — RND uses a canonical key (lowercase + frozen
confusable-pair policy) to enforce uniqueness. A name is considered taken for
another user when any `registered`, active `reservation`, or
`pending_registration` with a different `user_id` exists under the same
canonical key. The confusable-pair policy lives in Lobby
(`lobby/internal/domain/racename/policy.go`).

**Capability gating** — at `game_finished` `Game Lobby` evaluates per-member
capability: `capable = max_planets > initial_planets AND max_population >
initial_population`, computed from the `player_turn_stats` stream published by
`Game Master`. Capable reservations transition to `pending_registration` with
`eligible_until = finished_at + 30 days`; non-capable reservations are
released immediately.

**Registration** — a user initiates registration via `lobby.race_name.register`
inside the 30-day window. Registration succeeds only when the user is still
eligible (no `permanent_block`, tariff slot available) and the pending entry
is still within its window. Expired pending entries are released by a
background worker.

**Cascade release** — `User Service` publishes
`user.lifecycle.permanent_blocked` and `user.lifecycle.deleted` events to
`user:lifecycle_events`. `Game Lobby` consumes this stream and calls
`RND.ReleaseAllByUser(user_id)` atomically with membership/application/invite
cancellations for the affected user.

## 8. Game Master

`Game Master` owns runtime and operational metadata of already running games.

It is the only trusted service allowed to communicate with game engine containers.

It owns:

* runtime mapping of running game to container endpoint/binding;
* current turn number;
* runtime status;
* generation status;
* engine health;
* patch state;
* engine version registry and version-specific engine options;
* runtime mapping `platform user_id -> engine player UUID` for each running game.

### Game Master status model

Minimum runtime-level status set:

* `starting`
* `running`
* `generation_in_progress`
* `generation_failed`
* `stopped`
* `engine_unreachable`

`running` here means `running_accepting_commands`.

### Game command routing

All game-related `message_type` include `game_id`.

Gateway enriches them with authenticated `user_id` and routes them to `Game Master`.
`Game Master` checks whether this user may access this running game, using membership data sourced from `Game Lobby`, then routes the command to the correct engine container.

The gateway never routes directly to game engine containers.

### Runtime admin operations

For already running games, `Game Master` handles:

* `stop game`
* `force next turn`
* `patch engine`
* admin/runtime status reads
* player deactivation/removal inside engine when required
* regular collection of game runtime metrics

System admin can use all of them.
Private-game owner can use the subset allowed for the owner of that game.

### Turn cutoff and scheduling

`Game Master` is the owner of authoritative platform time for turn cutoff decisions.

Commands arriving exactly on the boundary of a new turn are considered stale and must not reach the engine.

The scheduler is a subsystem inside `Game Master`.
It triggers turn generation according to the game schedule.

If a manual “force next turn” is executed, the next scheduled turn slot must be skipped so that players still get at least one full normal schedule interval before the following generated turn.

### Runtime snapshot publishing

`Game Master` publishes runtime updates to the `gm:lobby_events` Redis Stream
consumed by `Game Lobby`. Events include:

* `runtime_snapshot_update` — carries the current `current_turn`,
  `runtime_status`, `engine_health_summary`, and a `player_turn_stats` array
  with one entry per active member (`user_id`, `planets`, `population`,
  `ships_built`). `Game Lobby` maintains a per-game per-user stats aggregate
  from these events for capability evaluation at game finish.
* `game_finished` — carries the final snapshot values and triggers the
  platform status transition plus Race Name Directory capability evaluation
  inside `Game Lobby`.

`Game Master` does not retain the aggregate; it only publishes the per-turn
observation. `Game Lobby` is responsible for holding initial values and
running maxima across the lifetime of the game.

### Runtime/engine finish flow

When the engine determines that a game is finished:

1. engine reports finish to `Game Master`;
2. `Game Master` updates runtime state;
3. `Game Master` notifies `Game Lobby`;
4. `Game Lobby` updates the platform-level game record to `finished`.

### Player removal after start

After a game has started, two different actions exist:

* temporary removal/block at platform level:

  * the player cannot send commands through gateway/platform;
  * the engine still keeps the player slot;
* final removal or account-level block:

  * `Game Master` must additionally send an admin command to the engine to deactivate/remove the player inside the game.

This distinction is architectural and must remain explicit.

## 9. Runtime Manager

`Runtime Manager` is the only internal service allowed to access Docker API directly.

It owns:

* starting game engine containers;
* stopping containers;
* restarting containers where allowed;
* patching/replacing containers where allowed;
* technical runtime inspection/status;
* monitoring containers and publishing technical health events.

It does **not** own platform metadata of games.
It does **not** own runtime business state of games.
It executes runtime jobs for `Game Lobby` and `Game Master`.

### Container model

* one game = one container;
* one container = one game.

This is a hard invariant.

## 10. [Notification Service](notification/README.md)

`Notification Service` is the async delivery/orchestration layer for platform notifications.

It has a deliberately minimal role:

* consume normalized notification intents from services through dedicated
  Redis Stream `notification:intents`;
* validate idempotency and persist durable notification route state;
* enrich user-targeted routes with `email` and `preferred_language` from
  `User Service`;
* decide whether a given notification type results in `push`, `email`, or
  both;
* send user-targeted `push` events toward gateway by `user_id`;
* send non-auth email asynchronous commands toward `Mail Service`.

It is not a source of truth for user preferences in v1 unless a later feature requires it.

For user-targeted intents, upstream producers publish the concrete recipient
`user_id` values. `Notification Service` resolves user email and locale from
`User Service`, uses configured administrator email lists per
`notification_type` for admin-only notifications, keeps
`template_id == notification_type` for notification-generated email, and
treats private-game invite flows in v1 as user-bound by internal `user_id`.
Go producers use the shared `galaxy/notificationintent` module to build and
append compatible intents into `notification:intents`; a failed append is a
notification degradation signal and must not roll back already committed source
business state.
Acceptance of a user-targeted notification intent is complete only after every
published recipient `user_id` resolves through `User Service`; unresolved user
ids are treated as producer input defects and are recorded as malformed
notification intents rather than deferred publication failures.

User-facing notifications use `push+email` unless a type explicitly opts out of
one channel. Administrator-facing notifications are `email`-only in v1.

All platform notifications except auth-code delivery flow through this service, including:

* game lifecycle notifications;
* invite/application updates;
* new turn notifications;
* operational/admin notifications where appropriate.

The current process surface exposes only one private probe HTTP listener with
`GET /healthz` and `GET /readyz`; that probe surface is documented in
[`notification/openapi.yaml`](notification/openapi.yaml). The canonical
notification-intent stream contract remains
[`notification/api/intents-asyncapi.yaml`](notification/api/intents-asyncapi.yaml).
It does not expose an operator REST API.

## 11. Billing Service (future)

`Billing Service` is not part of the first implementation wave.

When introduced, it will:

* process payment/billing events;
* calculate or validate payment outcomes;
* feed resulting entitlement changes into `User Service`.

`User Service` remains the source of truth for current entitlement used by the rest of the platform.

Billing-driven tariff changes alter only the headroom for *new* registered
race names: tariff downgrade never revokes already registered names. The
affected ceiling is materialized as `max_registered_race_names` in the
eligibility snapshot consumed by `Game Lobby`.

## Data Ownership Summary

```mermaid
flowchart TD
    U["User Service"]
    A["Auth / Session Service"]
    L["Game Lobby"]
    G["Game Master"]
    R["Runtime Manager"]
    P["Geo Profile Service"]
    N["Notification Service"]
    M["Mail Service"]

    U -->|"regular users, user_name/display_name, settings, tariffs, limits, sanctions, declared_country, soft-delete"| X1["Platform user identity"]
    A -->|"challenges, device sessions, revoke/block state"| X2["Auth/session state"]
    L -->|"game metadata, invites, applications, membership, roster, race names (registered/reservations/pending)"| X3["Platform game records"]
    G -->|"runtime state, current turn, engine health, engine mapping, engine version registry"| X4["Running-game state"]
    R -->|"container execution and technical runtime control"| X5["Container runtime"]
    P -->|"observed country, usual_connection_country, review state, declared_country history"| X6["Geo state"]
    N -->|"notification routing only"| X7["Notification orchestration"]
    M -->|"email delivery only"| X8["Email transport"]
```

## Internal Transport Semantics

The platform uses one simple rule:

* if the user-facing request must complete with a deterministic result in the same flow, the critical internal chain is synchronous;
* if the interaction is propagation, notification, cache invalidation, runtime job completion, telemetry, or denormalized read-model update, it is asynchronous.

### Fixed synchronous interactions

* `Gateway -> Auth / Session Service`
* `Gateway -> Admin Service`
* `Gateway -> User Service`
* `Gateway -> Game Lobby`
* `Gateway -> Game Master`
* `Auth / Session Service -> User Service`
* `Auth / Session Service -> Mail Service`
* `Geo Profile Service -> Auth / Session Service`
* `Geo Profile Service -> User Service`
* `Game Lobby -> User Service`
* `Game Lobby -> Game Master` for critical registration/update calls

### Fixed asynchronous interactions

* session lifecycle projection toward gateway cache;
* revoke propagation;
* `Lobby -> Runtime Manager` runtime jobs;
* `Game Master -> Runtime Manager` runtime jobs;
* all event-bus propagation;
* `Game Master -> Game Lobby` runtime snapshot updates (including
  `player_turn_stats` for capability aggregation) and game-finish events
  through a dedicated Redis Stream consumed by `Game Lobby`;
* `User Service -> Game Lobby` user lifecycle events
  (`user.lifecycle.permanent_blocked`, `user.lifecycle.deleted`) through the
  `user:lifecycle_events` Redis Stream, consumed by `Game Lobby` to cascade
  RND release and membership/application/invite cancellation;
* `Game Master -> Notification Service` notification intents through
  `notification:intents`;
* `Game Lobby -> Notification Service` notification intents through
  `notification:intents`;
* `Geo Profile Service -> Notification Service` notification intents through
  `notification:intents`;
* `Notification Service -> Gateway`;
* `Notification Service -> Mail Service`;
* geo auxiliary ingest from gateway to geo service;
* runtime health events from `Runtime Manager`.

### Mixed interactions

Some service pairs may use both styles for different flows.
The main example is `Lobby -> Game Master`:

* synchronous for critical registration/update after successful start;
* asynchronous for secondary propagation and denormalized status fan-out.

## Persistence Backends

The platform splits durable state across two backends.

PostgreSQL is the source of truth for table-shaped business state:

* user identity, profile settings, tariffs/entitlements, sanctions, limits,
  and the blocked-email registry;
* mail deliveries, attempt history, dead letters, payloads, and
  malformed-command audit;
* notification records, route materialisations, dead letters, and
  malformed-intent audit;
* lobby games, applications, invites, memberships, and the race-name
  registry (registered/reservation/pending tiers);
* idempotency records, expressed as `UNIQUE` constraints on the durable
  table — not as a separate kv;
* retry scheduling state, expressed as a `next_attempt_at` column on the
  durable table and worked off via `SELECT ... FOR UPDATE SKIP LOCKED`.

Redis is the source of truth for ephemeral and runtime-coordination state:

* the platform event bus implemented as Redis Streams (`user:domain_events`,
  `user:lifecycle_events`, `gm:lobby_events`, `runtime:job_results`,
  `notification:intents`, `gateway:client-events`, `mail:delivery_commands`);
* stream consumer offsets;
* gateway session cache, replay reservations, rate-limit counters, and
  short-lived runtime locks/leases (e.g. notification `route_leases`);
* `Auth / Session Service` challenges and active session tokens, which are
  TTL-bounded and where loss is recoverable by re-authentication;
* lobby per-game runtime aggregates that are deleted at game finish
  (`game_turn_stats`, `gap_activated_at`, capability evaluation marker).

### Database topology

* Single PostgreSQL database `galaxy`.
* Schema per service: `user`, `mail`, `notification`, `lobby`. Reserved for
  future use: `geoprofile`. Not allocated unless needed: `gateway`,
  `authsession`.
* Each service connects with its own PostgreSQL role whose grants are
  restricted to its own schema (defense-in-depth).
* Authentication is username + password only. `sslmode=disable`. No client
  certificates and no SCRAM channel binding.
* Each service connects to one primary plus zero-or-more read-only
  replicas. Only the primary is used in this iteration; the replica pool
  is wired but receives no traffic. Future read-routing is a non-breaking
  change.

### Redis topology

* Each service connects to one master plus zero-or-more replicas.
* All connections require a password. `USERNAME`/ACL is not used. TLS is
  off.
* Only the master is used in this iteration; the replica list is wired but
  unused. Failover/read routing is added later without a config break.
* The legacy env vars `*_REDIS_TLS_ENABLED` and `*_REDIS_USERNAME` are
  removed without a backward-compat shim.

### Library stack and migration discipline

* Driver: `github.com/jackc/pgx/v5`, exposed as `*sql.DB` via
  `github.com/jackc/pgx/v5/stdlib` so it is consumable by query builders
  written against `database/sql`.
* Query layer: `github.com/go-jet/jet/v2` (PostgreSQL dialect). Generated
  code lives under each service `internal/adapters/postgres/jet/`,
  regenerated by a per-service `make jet` target (testcontainers + goose +
  jet) and committed to the repo so consumers don't need Docker just to
  build.
* Migrations: `github.com/pressly/goose/v3` library API. Migration files
  are embedded via `//go:embed *.sql`, applied at service startup before
  any listener opens; the service exits non-zero on failure. Files are
  forward-only, sequence-numbered, and use the standard `-- +goose Up` /
  `-- +goose Down` markers.
* Single-init policy during pre-launch development: each PG-backed
  service ships exactly one migration file, `00001_init.sql`, that
  represents the full current schema. New tables, columns, and indexes
  are added by editing that file directly rather than by appending
  `00002_*.sql`, `00003_*.sql`, etc. The trade-off is intentional —
  schema clarity beats migration-history granularity while no production
  database exists. Once the platform reaches its first production
  deploy, future schema evolution switches to additive sequence-numbered
  migrations.
* Test infrastructure: `github.com/testcontainers/testcontainers-go` plus
  the `modules/postgres` submodule for unit tests and for `make jet`.

Per-service decision records that capture schema and adapter choices live
at `galaxy/<service>/docs/postgres-migration.md`.

### Timestamp handling

Every time-valued column in every Galaxy schema is `timestamptz`. The
adapter layer is responsible for ensuring that all `time.Time` values
crossing the SQL boundary carry `time.UTC` as their location.

* **Writes.** Every `time.Time` parameter bound through `database/sql`
  (`ExecContext`, `QueryContext`, `QueryRowContext`) is normalised with
  `.UTC()` at the binding site. Optional `*time.Time` columns are bound
  through a shared helper (`nullableTime` or equivalent per adapter) that
  returns `value.UTC()` when non-nil and SQL `NULL` otherwise. Helper
  bindings of `cutoff`, `now`, etc. (retention, schedulers) follow the
  same rule even when the input was already produced via
  `clock.Now().UTC()` — defensive `.UTC()` calls are intentional and
  cheap.
* **Reads.** Every `time.Time` scanned out of PostgreSQL is re-wrapped
  with `.UTC()` (directly or via a small helper that mirrors
  `nullableTime` for the read path) before it leaves the adapter. The
  domain layer therefore never observes a `time.Time` whose location is
  anything other than `time.UTC`.
* **Why.** PostgreSQL stores `timestamptz` as UTC at rest, but the Go
  driver returns scanned values in `time.Local`. Mixing locations across
  the boundary produces inequalities in tests, drift in JSON output, and
  comparison bugs against pointer fields. The defensive `.UTC()` rule on
  both sides removes that class of bug entirely.

### Configuration

For each service `<S>` ∈ { `USERSERVICE`, `MAIL`, `NOTIFICATION`,
`LOBBY`, `GATEWAY`, `AUTHSESSION` }, the Redis connection accepts:

* `<S>_REDIS_MASTER_ADDR` (required)
* `<S>_REDIS_REPLICA_ADDRS` (optional, comma-separated)
* `<S>_REDIS_PASSWORD` (required)
* `<S>_REDIS_DB`, `<S>_REDIS_OPERATION_TIMEOUT`

For PG-backed services (`USERSERVICE`, `MAIL`, `NOTIFICATION`, `LOBBY`)
the Postgres connection accepts:

* `<S>_POSTGRES_PRIMARY_DSN` (required;
  `postgres://<role>:<pwd>@<host>:5432/galaxy?search_path=<schema>&sslmode=disable`)
* `<S>_POSTGRES_REPLICA_DSNS` (optional, comma-separated)
* `<S>_POSTGRES_OPERATION_TIMEOUT`, `<S>_POSTGRES_MAX_OPEN_CONNS`,
  `<S>_POSTGRES_MAX_IDLE_CONNS`, `<S>_POSTGRES_CONN_MAX_LIFETIME`

Stream- and key-shape env vars (`*_REDIS_DOMAIN_EVENTS_STREAM`,
`*_REDIS_LIFECYCLE_EVENTS_STREAM`, `*_REDIS_KEYSPACE_PREFIX`,
`MAIL_REDIS_COMMAND_STREAM`, `NOTIFICATION_INTENTS_STREAM`, etc.) keep
their current names and semantics — they describe stream/key shapes, not
connection topology.

## Main End-to-End Flows

## 1. Public authentication flow

```mermaid
sequenceDiagram
    participant Client
    participant Gateway
    participant Auth
    participant User
    participant Mail
    participant Redis

    Client->>Gateway: POST send-email-code
    Gateway->>Auth: send-email-code
    Auth->>User: resolve existing/creatable/blocked
    User-->>Auth: decision
    Auth->>Mail: send or suppress code
    Auth-->>Gateway: challenge_id
    Gateway-->>Client: challenge_id

    Client->>Gateway: POST confirm-email-code(time_zone)
    Gateway->>Auth: confirm-email-code(time_zone)
    Auth->>Auth: validate challenge/code/public key/time_zone
    Auth->>User: resolve/create/block with create-only registration context when needed
    User-->>Auth: user_id or deny
    Auth->>Auth: create device_session
    Auth->>Redis: write gateway session projection
    Auth->>Redis: publish session lifecycle update
    Auth-->>Gateway: device_session_id
    Gateway-->>Client: device_session_id
```

This preserves the existing gateway/auth contract and the rule that auth is not on the steady-state hot path.

## 2. Authenticated game/platform request flow

```mermaid
sequenceDiagram
    participant Client
    participant Gateway
    participant Lobby
    participant GM as Game Master

    Client->>Gateway: ExecuteCommand(message_type, payload, signature)
    Gateway->>Gateway: verify session, signature, freshness, replay
    alt platform-level command
        Gateway->>Lobby: verified authenticated command
        Lobby-->>Gateway: response
    else running-game command
        Gateway->>GM: verified authenticated command with game_id
        GM-->>Gateway: response
    end
    Gateway-->>Client: signed response
```

## 3. Game creation and pre-start lifecycle

```mermaid
sequenceDiagram
    participant Client
    participant Gateway
    participant Lobby
    participant User

    Client->>Gateway: create/apply/invite/approve/start-preparation commands
    Gateway->>Lobby: verified platform command
    Lobby->>User: entitlement/limit checks when needed
    User-->>Lobby: allow/deny and user metadata
    Lobby->>Lobby: update game metadata, roster, schedule, target engine version
    Lobby-->>Gateway: response
    Gateway-->>Client: signed response
```

## 4. Game start flow

```mermaid
sequenceDiagram
    participant Owner as Admin or Private Owner
    participant Gateway
    participant Lobby
    participant Runtime
    participant GM as Game Master
    participant Engine as Game Engine Container
    participant Redis

    Owner->>Gateway: start game
    Gateway->>Lobby: verified start command
    Lobby->>Lobby: validate ready_to_start and roster
    Lobby->>Runtime: async start job
    Runtime-->>Redis: runtime job result event

    alt start failed
        Lobby->>Lobby: keep failure / starting error state
        Lobby-->>Gateway: failure or accepted-then-observed failure path
    else container started
        Lobby->>Lobby: persist game metadata and runtime binding
        Lobby->>GM: sync running-game registration
        GM->>Engine: initial engine setup API
        GM->>GM: initialize runtime state
        GM-->>Lobby: registration result
        Lobby->>Lobby: mark game running or paused
    end
```

Critical rule:
if the container starts but `Lobby` cannot persist metadata, the launch is considered a full failure and the container must be removed.
If metadata is persisted but `Game Master` is unavailable, the game is placed into `paused` and administrators are notified.

## 5. Running-game command flow

```mermaid
sequenceDiagram
    participant Client
    participant Gateway
    participant GM as Game Master
    participant Lobby
    participant Engine

    Client->>Gateway: game-related ExecuteCommand(game_id,...)
    Gateway->>GM: verified authenticated command
    GM->>GM: check runtime status
    GM->>Lobby: resolve/cached-check membership if needed
    Lobby-->>GM: membership / permissions
    GM->>Engine: game or runtime-admin API call
    Engine-->>GM: result
    GM-->>Gateway: response payload
    Gateway-->>Client: signed response
```

## 6. Scheduled turn generation flow

```mermaid
sequenceDiagram
    participant Scheduler as Game Master Scheduler
    participant GM as Game Master
    participant Engine
    participant Lobby
    participant Notify as Notification Service
    participant Gateway

    Scheduler->>GM: due turn slot reached
    GM->>GM: switch runtime_status to generation_in_progress
    GM->>Engine: generate next turn
    alt generation success
        Engine-->>GM: new turn result / maybe finished
        GM->>GM: update current_turn and runtime state
        GM->>Lobby: sync runtime snapshot
        GM->>Notify: publish new-turn intent
        Notify->>Gateway: client-facing push events
    else generation failed
        Engine-->>GM: error / timeout
        GM->>GM: mark generation_failed
        GM->>Lobby: sync runtime snapshot
        GM->>Notify: notify administrators only
    end
```

Players receive only a lightweight push notification that a new turn exists.
They then request their own per-player game state separately.

If `force next turn` is used, the next scheduled slot is skipped so that the effective time between turns never becomes shorter than the schedule spacing.

## 7. Game finish flow

```mermaid
sequenceDiagram
    participant Engine
    participant GM as Game Master
    participant Lobby
    participant Notify as Notification Service
    participant Gateway

    Engine->>GM: game finished
    GM->>GM: update runtime state
    GM->>Lobby: mark platform game finished
    Lobby->>Lobby: finalize game record
    GM->>Notify: publish game-finished intent
    Notify->>Gateway: push user-facing/platform events
```

## 8. Geo profile auxiliary flow

```mermaid
sequenceDiagram
    participant Gateway
    participant Geo
    participant User
    participant Auth

    Gateway-->>Geo: async observation(user_id, device_session_id, ip_addr)
    Geo->>Geo: derive observed_country and aggregates
    alt suspicious multi-country pattern
        Geo->>Auth: sync block suspicious session(s)
    end
    alt declared_country admin change approved later
        Geo->>User: sync current declared_country update
    end
```

This flow is intentionally fail-open relative to gameplay.

## Separation of Platform Metadata and Engine State

This distinction is fundamental.

### Platform-level state

Owned by `Game Lobby`:

* who owns the game;
* who is invited;
* who applied;
* who was approved;
* who is currently a platform participant;
* what the schedule is;
* whether the game is public/private;
* whether the game is `draft`, `running`, `paused`, `finished`, etc. as a platform entity.

### Runtime/operational state

Owned by `Game Master`:

* current turn;
* runtime status;
* generation state;
* engine reachability;
* patch state;
* mapping to engine player UUIDs;
* engine version registry;
* operational metadata of the running game.

### Full game state

Owned only by the game engine container:

* actual per-player game state;
* internal mechanics and progression;
* player-visible game state snapshots;
* win/lose logic;
* domain truth of the game world.

The platform must not attempt to duplicate the full game state outside the engine.

## Versioning of Game Engines

Every game runs on one specific game engine version.

Rules:

* active games stay on the version with which they were started;
* upgrade during a running game is allowed only as a patch update within the same major/minor line;
* game-engine version management is manual in v1;
* each engine version may carry version-specific engine options;
* `Game Master` owns the engine version registry and its internal API.

## Administrative Access Model

Two distinct external admin modes exist.

### System administrator

Uses a separate admin-facing REST surface via gateway and `Admin Service`.

System administrator can:

* manage public games;
* see and operate on all private games;
* inspect platform operational state;
* launch, stop, patch, pause, and monitor games;
* approve/reject participation in public games;
* perform user/game administrative actions.

### Private-game owner

Uses the normal authenticated client protocol, not the separate system admin UI.

Allowed owner-admin actions are limited to the owner’s own private games and include at least:

* initiate enrollment;
* create and manage user-bound invites inside the system;
* approve/reject applicants;
* start game after enrollment;
* force next turn while running;
* stop game;
* temporarily or permanently remove/block players from that game according to allowed policy.

These operations use dedicated admin-related `message_type` values in the normal authenticated game/client protocol.

## Non-Goals

The architecture intentionally does not try to solve all future concerns now.

Current non-goals:

* a separate policy engine;
* automatic billing integration in v1;
* automatic match balancing in v1;
* direct external access to internal services;
* pushing full per-player game state over notification channels;
* allowing game engine containers to be called directly by clients or by services other than `Game Master`;
* using `Auth / Session Service` as a hot synchronous dependency for all authenticated traffic;
* making `Notification Service` the source of truth for notification preferences in v1.

## Recommended Order of Service Implementation

Recommended order for implementation is:

1. **Edge Gateway Service** (implemented)  
   First public ingress, transport boundary, authentication boundary, signed request/response model, push delivery, session cache, replay protection.

2. **Auth / Session Service** (implemented)  
   Public auth flow, `device_session`, revoke/block lifecycle, gateway session projection.

3. **User Service** (implemented)  
   Regular-user identity, profile/settings, tariffs/entitlements, user limits, sanctions, and current `declared_country`.

4. **Mail Service** (implemented)  
   Internal email delivery for auth codes and platform notification mail.

5. **Notification Service** (implemented)
   Unified async delivery of push and non-auth email notifications, with
   real Gateway and Mail Service boundary coverage.

6. **Game Lobby Service**  
   Platform game records, membership, invites, applications, approvals, schedules, user-facing lists, pre-start lifecycle.

7. **Runtime Manager**  
   Dedicated Docker-control service for container start/stop/patch/status and technical runtime monitoring.

8. **Game Master**  
   Running-game orchestration, engine version registry, runtime state, turn scheduler, engine API mediation, operational controls.

9. **Admin Service**  
   Admin UI backend that orchestrates trusted APIs of other services.

10. **Geo Profile Service** (planned)  
    Auxiliary geo aggregation, review recommendation, suspicious-session blocking, declared-country workflow.

11. **Billing Service**  
    Future payment and subscription source feeding entitlements into `User Service`.

This order gives the platform a usable public perimeter first, then identity/auth, then core gameplay lifecycle, then runtime orchestration, and only afterward secondary auxiliary services.