Admin process — current implementation
Source review: 2026-10-21, HEAD f38ee3c6ced7fcab6b34955a65cfb955ec097567, root package 1.6.9. This supersedes the October 14 diagrams for current behavior. Arrows describe source-visible calls; deployment is configuration, not a live-environment check.
2026-11-05 CDN update: The admin sync now persists per-environment progress under data/admin-cdn-sync, exposes an authenticated progress read, and resumes pending files after a failed request or process restart when the staged source is unchanged. This targeted update reflects the current task branch; it is not a full rereview of the other diagrams. Sources: sync runner, sync routes, admin Compose.
Components
flowchart LR
Browser["Admin browser"] --> UI["web-layer-admin pages and core/admin media/CDN pages"]
UI --> FETCH["useAdminFetch and media-client"]
FETCH --> API["core/admin Nitro routes"]
API --> AUTH["requireAdminToken"]
API --> SELECT["resolveAdminDatabase"]
SELECT --> BLOG["BlogPostStore"]
SELECT --> MSG["MessageStore"]
BLOG --> SQL["SQLite or PostgreSQL"]
BLOG --> DDB["DynamoDB"]
MSG --> SQL
MSG --> DDB
API --> CDN["MinIO client and output sync"]
API --> MEDIA["admin media-service"]
MEDIA --> LOCAL["local MediaLibrary / SQLite / filesystem"]
MEDIA -->|MEDIA_API_URL| WEB["web media API"]
The selected database applies to blog posts, messages and CDN source-output mapping; media has its own local or forwarded configuration. Sources: admin routes, database selector, auth gate, media service, CDN client.
Main types and responsibilities
classDiagram
class BlogPost {
+id
+slug
+status
+title
}
class BlogPostStore {
<<interface>>
}
class MessageStore {
<<interface>>
}
class AdminMediaAsset {
+id
+path
+publicUrl
+previewUrl
}
class MediaLibrary {
+ingest()
+renameAsset()
+deleteAsset()
}
BlogPostStore --> BlogPost : persists
These are selected implemented types, not an inheritance hierarchy. AdminMediaAsset is a type intersection created by the admin service, while BlogPostStore and MessageStore are store contracts. Sources: blog types, blog store, message store, media service, MediaLibrary.
Authentication and database selection
sequenceDiagram
actor Admin
participant UI as Admin page
participant Fetch as useAdminFetch
participant Route as Admin API route
participant Gate as requireAdminToken
participant Select as resolveAdminDatabase
participant Store as Blog/Message store
Admin->>UI: Enter bearer token and select database
UI->>Fetch: Request with token and database header
Fetch->>Route: HTTP request
Route->>Gate: Validate bearer against configured token
alt missing or invalid token
Gate-->>Fetch: 401
Fetch-->>UI: Expired-session redirect
else valid token
Route->>Select: Resolve query db, header, or default
Select->>Store: Select backend-specific store
Store-->>Route: Result
Route-->>UI: Response
end
The route middleware is a client navigation guard; server routes enforce the bearer token. Query db wins over X-Admin-Database. Sources: route middleware, fetch wrapper, auth gate, database selector.
Blog edit and deletion
sequenceDiagram
actor Admin
participant Editor as Blog editor
participant Form as useForm / dirty guard
participant List as Blog list / useCrudResource
participant API as Admin post API
participant Store as BlogPostStore
Admin->>Editor: Edit and save
Editor->>Form: Validate shared schema
Form->>API: POST new or PUT existing
API->>Store: Persist selected backend
Store-->>Editor: Saved post
Admin->>Editor: Confirm delete
Editor->>List: In-memory handoff and navigation
List->>API: GET list (abortable)
API-->>List: Posts
alt handoff is fresh and post is listed
List-->>Admin: Undo window
opt Undo not selected
List->>API: DELETE post after delay
API->>Store: Delete selected backend
end
else load fails or post absent
List-->>Admin: Not deleted
end
The list filters status in the route query and loads all posts in one response. The editor uses useUnsavedChangesGuard for navigation, but its database watcher calls loadPost() and resets form values without a separate dirty-edit check; a slow response can also replace a later selection. These are source-level risks, not observed production failures. Sources: blog adapter and handoff, blog list, blog editor, CRUD composable, dirty guard, post API.
Media administration
sequenceDiagram
actor Admin
participant Page as Admin media page / drawer
participant Client as media-client
participant Route as Admin media API
participant Service as media-service
participant Library as MediaLibrary or web media API
Admin->>Page: Upload asset
Page->>Client: Ingest request
Client->>Route: POST ingest
Route->>Service: ingestMediaUpload
alt local media
Service->>Library: Ingest with sanitization and processing
else MEDIA_API_URL configured
Service->>Service: Sanitize upload bytes
Service->>Library: SVG request or presign, PUT, complete
end
Library-->>Page: Asset and URLs
Admin->>Page: Edit metadata or rename
Page->>Client: PATCH metadata or POST rename
Client->>Route: Authorized request
Route->>Service: Update or rename
Service->>Library: Local call or forwarded request
Admin->>Page: Confirm delete and optionally Undo
Page->>Client: DELETE after undo window
Client->>Route: DELETE asset
Route->>Library: Cascade delete
The media page now has metadata and rename forms plus a processing drawer. Its delete uses the shared undo composable, whereas the API performs hard deletion when the timer commits. The media store is independent of the blog/message database selector. Sources: media page, editing composable, drawer, media routes, media service. See Media Library current architecture for processing and storage details.
Messages and CDN
flowchart LR
MUI["Messages page"] --> MGET["GET /api/admin/messages"]
MUI --> MDEL["DELETE /api/admin/messages/:id"]
MGET --> MS["selected MessageStore"]
MDEL --> MS
CUI["CDN page"] --> CLIST["list objects"]
CUI --> CPUT["upload object"]
CUI --> CDEL["delete object"]
CUI --> CSYNC["sync staged output"]
CUI --> CPOLL["read sync progress"]
CLIST --> MINIO["MinIO bucket"]
CPUT --> MINIO
CDEL --> MINIO
CSYNC --> OUTPUT[".output/<environment>/public"]
CSYNC --> STATE["persistent sync manifest"]
CPOLL --> STATE
OUTPUT --> MINIO
Messages use the selected database; CDN operations use one configured bucket. CDN sync reads staged build output, writes objects sequentially, and checkpoints a per-environment JSON manifest after each successful PUT. The selected database only chooses the staged source directory. The manifest lives in the admin data volume (NUXT_CDN_SYNC_STATE_DIR can select another persistent path); a repeat POST resumes pending files only when the source snapshot and normalized MinIO endpoint and bucket still match. Older manifests without an endpoint identity start a fresh sync. A repeat after completion returns the same operation without another PUT; the UI reads progress for the captured database before sending an explicit restart when the operator chooses to resync. There is no cross-process lease or conditional destination write, so this is not multi-replica reconciliation. Sources: messages page, message routes, CDN page, CDN routes, sync runner, CDN client, database selector.
Deployment
flowchart TB
Host["Local admin Nitro node server"] --> SQLite["SQLite blog/message database"]
Host --> Files["local media filesystem"]
Docker["Docker admin Nitro service"] --> PG["PostgreSQL option"]
Docker --> DD["DynamoDB Local option"]
Docker --> MinIO["MinIO CDN"]
Docker --> Forward["optional MEDIA_API_URL"]
Forward --> Web["Web media API"]
Web --> AWS["SAM web Lambda + media AWS resources"]
This diagram describes configured paths. There is no separate admin-app SAM deployment in the inspected configuration; AWS media belongs to the web stack when forwarding is enabled. Sources: admin Nuxt config, admin Compose, admin Dockerfile, SAM template, media service.
Current issues and recommendations
Verified source behavior: editor database changes can reset unsaved values; CDN sync now has a local durable manifest but no cross-process lock or destination overwrite guard; post list is one unpaged response; Dynamo post updates remain unconditional after the slug precheck; media selection is independent of the admin database control. Recommended work: guard/reconcile editor database switches, add CDN pagination if object counts grow, add conditional destination writes before multi-replica sync, introduce conditional Dynamo revisions and slug claims where concurrent editing matters, and label the media storage target in the UI. These are recommendations, not diagrammed services. Sources: editor, blog adapter, sync runner, Dynamo blog store, media service.