auth-yes/ONBOARDING.md

299 lines
13 KiB
Markdown

# Auth-Yes — Application Integration & Architecture Guide
Auth-Yes is a standalone, zero-trust Identity and Access Management (IAM) fabric
and WebAuthn Passkey authority. It provides microsecond session validation,
cryptographic workload identity, and decoupled application-level Role-Based
Access Control (RBAC).
This guide provides a universal, technology-agnostic architectural reference for
developers and system administrators integrating services into the Auth-Yes
ecosystem.
---
## 1. System Architecture & Defense-in-Depth Model
Auth-Yes implements a **three-tier defense model**, allowing applications to
choose the integration depth that best fits their architecture:
```
[ Public Internet / User Browser ]
│
1. WebAuthn Passkey Login
2. Wildcard Cookie: *.yourdomain.org
│
▼
═════════════════════════════════════════
Ingress Reverse Proxy (Traefik, Nginx, etc.)
═════════════════════════════════════════
│
┌────────────────────┴────────────────────┐
▼ ▼
[ Tier 1 & 2: Edge ForwardAuth ] [ Tier 3: Native Zero-Trust ]
(Off-the-shelf UIs / 3rd-Party) (Microservices & APIs)
│ │
│ GET /api/forward-auth │ 1. Extract session token
│ (Valkey Cache Lookup) │ 2. Query Workload API Socket
│ ▼
│ ┌─────────────────────────────────┐
│ │ SPIRE Agent (Workload API) │
│ │ - Issues X.509 SVID │
│ │ - spiffe://<trust-domain>/<app> │
│ └────────────────┬────────────────┘
│ │
│ │ ConnectRPC over HTTP/2
▼ ▼
═════════════════════════════════════════════════════════════════════
Auth-Yes Central Gateway (auth-api:8000)
- Microsecond Session Validation (Valkey L1/L2 with RESP3 Tracking)
- Cryptographic Workload SVID Attestation (spire_ffi)
- Default-Deny PostgreSQL Application RBAC Grants (grants table)
═════════════════════════════════════════════════════════════════════
```
### The Three Integration Tiers:
| Tier | Target Use Case | Integration Mechanism | Protocol / Transport | Code Required |
| :--------------------------- | :---------------------------------------------------- | :------------------------------------------ | :------------------------------------------ | :------------------------- |
| **Tier 1: Global Edge** | Pre-release & unmapped internal services | Universal proxy perimeter fallback | HTTP 302 Redirect to `/login` | None |
| **Tier 2: Edge ForwardAuth** | Off-the-shelf apps, admin dashboards, legacy web UIs | Reverse proxy ForwardAuth route | HTTP `GET /api/forward-auth` | None |
| **Tier 3: Zero-Trust RPC** | Native microservices, custom APIs, streaming backends | Application SDK or direct ConnectRPC client | ConnectRPC / gRPC over HTTP/2 + SPIFFE mTLS | Minimal (Middleware / RPC) |
---
## 2. Core Identity & Authorization Primitives
### 2.1. User Session Identifiers
- **Browser Clients:** Auth-Yes issues an HTTP-only, secure session cookie
(`session_id`) scoped to the parent domain (e.g., `.yourdomain.org`). Browsers
automatically transmit this cookie across all subdomains.
- **API / Programmatic Clients:** Applications accept standard Bearer
authorization headers: `Authorization: Bearer <session_token>`.
### 2.2. Workload Identity (SPIFFE / SPIRE)
- Microservices obtain cryptographic identities (X.509 SVIDs) by connecting to
the local SPIRE Workload API UNIX domain socket at
`/var/run/spire/agent.sock`.
- SVIDs follow the standard SPIFFE ID URI format:
`spiffe://<trust-domain>/<workload-name>` (e.g.,
`spiffe://system.local/payment-service`).
- When an application calls Auth-Yes, Auth-Yes extracts and attests the caller's
SPIFFE ID to verify application authenticity.
### 2.3. Decoupled Default-Deny RBAC
Auth-Yes decouples **Authentication** (who the user is) from **Authorization**
(what application they are permitted to access):
1. **Default-Deny:** A valid user session has **zero application permissions by
default**.
2. **Grant Evaluation:** When a service validates a session, Auth-Yes evaluates
the internal grant matrix:
```sql
SELECT role FROM grants WHERE user_id = $1 AND app_id = $2;
```
3. **Scope Payloads:** If a grant exists, Auth-Yes returns
`{ valid: true, uuid: "...", scopes: ["<role>"] }`. If no grant exists, it
returns `{ valid: false, error: "Validation failed" }` resulting in an
immediate `403 Forbidden`.
---
## 3. Integration Patterns
### Pattern A: Edge ForwardAuth (Zero-Code / Reverse Proxy)
For off-the-shelf web applications (e.g., Grafana, Portainer, PgAdmin),
configure your ingress reverse proxy to query Auth-Yes before routing traffic to
the upstream container.
#### Traefik Example:
```yaml
# In Traefik dynamic configuration or container labels:
labels:
- "traefik.http.routers.myapp.middlewares=auth-forward@docker"
- "traefik.http.middlewares.auth-forward.forwardauth.address=http://auth-api:8000/api/forward-auth"
- "traefik.http.middlewares.auth-forward.forwardauth.trustForwardHeader=true"
- "traefik.http.middlewares.auth-forward.forwardauth.authResponseHeaders=X-Forwarded-User,X-Forwarded-User-Id"
```
- **Behavior:**
- If valid: Auth-Yes returns HTTP `200 OK` and injects upstream identity
headers (`X-Forwarded-User: <username>`, `X-Forwarded-User-Id: <uuid>`).
- If missing / invalid: Auth-Yes returns HTTP `401 Unauthorized` or redirects
unauthenticated browsers to `/login`.
---
### Pattern B: Native ConnectRPC / gRPC (Language-Agnostic)
For services written in Go, Rust, Python, Java, or C#, applications can
communicate directly with the Auth-Yes ConnectRPC endpoint over internal HTTP/2
multiplexing.
#### Protobuf Contract (`auth.proto`):
```protobuf
syntax = "proto3";
package auth.v1;
service AuthService {
rpc ValidateSession (ValidateSessionRequest) returns (ValidateSessionResponse);
}
message ValidateSessionRequest {
string token = 1;
}
message ValidateSessionResponse {
bool valid = 1;
string uuid = 2;
repeated string scopes = 3;
string error = 4;
}
```
- **Endpoint:** `http://auth-api:8000/auth.v1.AuthService/ValidateSession` (or
internal mesh DNS).
- **Workload Identity:** Pass client TLS certificates fetched from the SPIRE
Workload API socket to enable mTLS attestation.
---
### Pattern C: TypeScript & Deno Applications (`@auth-yes/sdk`)
For TypeScript/JavaScript runtimes, `@auth-yes/sdk` provides a zero-dependency
client with built-in **RESP3 client-side cache tracking** (L1 memory cache
invalidated in real time via Valkey broadcast events).
#### 1. Import Reference:
```json
{
"imports": {
"@auth-yes/sdk": "https://git.atyg.org/tylerg/auth-yes/raw/branch/main/sdk/mod.ts",
"@auth-yes/sdk/hono": "https://git.atyg.org/tylerg/auth-yes/raw/branch/main/sdk/hono.ts"
}
}
```
#### 2. Service Integration:
```typescript
import { createAuthSdk } from "@auth-yes/sdk";
// Initialize SDK with internal mesh endpoints
const authSdk = createAuthSdk({
authApiUrl: Deno.env.get("AUTH_API_URL") || "http://auth-api:8000",
valkeyUrl: Deno.env.get("VALKEY_URL"), // Optional: Enables real-time L1 cache tracking
});
// Validate any session token (cookie or header)
const result = await authSdk.validateSession(sessionToken);
if (result.valid) {
console.log(`User Authenticated: ${result.uuid}, Roles:`, result.scopes);
} else {
console.warn(`Access Denied: ${result.error}`);
}
```
---
## 4. Container & Infrastructure Topology
To connect any application container to Auth-Yes and SPIRE, ensure two
infrastructure prerequisites:
1. **Network Mesh:** Attach the container to the internal bridge network
(`auth-internal-net`) and ingress proxy network (`traefik-net`).
2. **Workload Socket:** Mount the pre-existing named volume `spire-socket` as
read-only (`:ro`).
### Generic Compose Template:
```yaml
version: "3.8"
services:
my-service:
image: ${REG}/library/my-service:latest
container_name: my-service
environment:
- AUTH_API_URL=http://auth-api:8000
- VALKEY_URL=redis://auth-valkey:6379
- SPIFFE_ENDPOINT_SOCKET=/var/run/spire/agent.sock
networks:
- default # auth-internal-net
- traefik-net # Ingress proxy
volumes:
# Read-only SPIRE Workload API UNIX domain socket
- spire-socket:/var/run/spire:ro
volumes:
spire-socket:
external: true
networks:
default:
name: auth-internal-net
external: true
traefik-net:
external: true
```
---
## 5. Administrative Lifecycle & Access Provisioning
Administrative operations are managed via the **Auth-Yes Admin Console**
(`https://auth.<domain>/admin`):
### 5.1. Application Registration (`/admin/apps`)
Every internal service connecting to Auth-Yes must be registered:
- **Application Name:** Human-readable label (e.g., `Analytics Engine`).
- **SPIFFE ID:** Cryptographic workload URI matching the service's container
attestation (e.g., `spiffe://system.local/analytics-engine`).
- **Description:** Functional description of the application.
### 5.2. User Onboarding & Invite Token Taxonomy (`/admin/invites`)
Auth-Yes uses a structured invite token model for user registration:
1. **Site-Scoped Token (Recommended for subsidiary applications):**
- Pre-binds an invite code to a specific application and role (`user`,
`operator`, `admin`).
- When a user registers their passkey using this token, their account is
automatically activated and granted immediate access to the designated
application.
2. **Global Admin Token:**
- Grants system-wide administrative access across Auth-Yes management views.
3. **Open / Pending Token:**
- Creates an account in `pending` state requiring explicit administrator
review and role assignment before login is permitted.
### 5.3. Dynamic Role Assignment & Instant Revocation (`/admin/users/:id`)
- **Grant Management:** Administrators can attach or adjust application roles
directly in the user profile matrix.
- **Instant Session Revocation:** Terminating a session or revoking an
application grant purges active tokens from Valkey in $< 1$ millisecond,
immediately blocking subsequent requests across all microservices.
---
## 6. Verification & Troubleshooting
| Check | Verification Command / Procedure | Expected Diagnostic |
| :---------------------------------- | :------------------------------------------------------------- | :--------------------------------------------------------------------- |
| **1. Workload Socket Availability** | `ls -la /var/run/spire/agent.sock` inside container | File exists with type `s` (UNIX domain socket). |
| **2. Auth Mesh Connectivity** | `curl -I http://auth-api:8000/health` inside container | Returns HTTP `200 OK`. |
| **3. Unregistered Service Check** | Calling `ValidateSession` from an unassigned application | Returns HTTP `403 Forbidden: Validation failed`. |
| **4. Grant Authorization Check** | Calling `ValidateSession` with an active user grant | Returns `{ valid: true, uuid: "...", scopes: ["..."] }`. |
| **5. Revocation Propagation** | Revoke session in Admin Console $\rightarrow$ Re-issue request | Instant transition to HTTP `401 Unauthorized` without server restarts. |