- New Go backend binary (backend + runner) replacing old scraper/ - Rename SCRAPER_API_URL → BACKEND_API_URL in UI env and docker-compose - Rename scraperFetch → backendFetch across all 19 UI server files - Remove SCRAPER_PROXY env var and proxy transport from browser.Config - Add Meilisearch, Valkey, Caddy to docker-compose - Add docs/: api-endpoints.md, request-flow.mermaid.md, data-flow.mermaid.md
3.7 KiB
3.7 KiB
Data Flow — Scrape & TTS Job Pipeline
How content moves from novelfire.net through the runner into storage, and how audio is generated on-demand via the backend.
Catalogue Scrape Pipeline
The runner performs a background catalogue walk on startup and then on a
configurable interval (RUNNER_CATALOGUE_REFRESH_INTERVAL, default 24 h).
flowchart TD
A([Runner starts / refresh tick]) --> B[Walk novelfire.net catalogue\npages 1…N]
B --> C{Book already\nin PocketBase?}
C -- no --> D[Scrape book metadata\ntitle · author · genres\ncover · summary · status]
C -- yes --> E[Check for new chapters\ncompare total_chapters]
D --> F[Write BookMeta\nto PocketBase books]
E --> G{New chapters\nfound?}
G -- no --> Z([Done — next book])
G -- yes --> H
F --> H[Scrape chapter list\n→ chapters_idx in PocketBase]
H --> I[Worker pool — N goroutines\nRUNNER_MAX_CONCURRENT_SCRAPE]
I --> J[For each missing chapter:\nGET chapter HTML from novelfire.net]
J --> K[Parse HTML → Markdown\nhtmlutil.NodeToMarkdown]
K --> L[PUT object to MinIO\nlibnovel-chapters/{slug}/{n}.md]
L --> M[Upsert book doc\nto Meilisearch index: books]
M --> Z
F --> M
On-Demand Single-Book Scrape
Triggered when a user visits /books/{slug} and the book is not in PocketBase.
The UI calls GET /api/book-preview/{slug} → backend enqueues a task.
sequenceDiagram
actor U as User
participant UI as SvelteKit UI
participant BE as Backend API
participant TQ as Task Queue (PocketBase)
participant RN as Runner
participant NF as novelfire.net
participant PB as PocketBase
participant MN as MinIO
participant MS as Meilisearch
U->>UI: Visit /books/{slug}
UI->>BE: GET /api/book-preview/{slug}
BE->>PB: getBook(slug) — not found
BE->>TQ: INSERT scrape_task (slug, status=pending)
BE-->>UI: 202 {task_id, message}
UI-->>U: "Scraping…" placeholder
RN->>TQ: Poll for pending tasks
TQ-->>RN: scrape_task (slug)
RN->>NF: GET novelfire.net/book/{slug}
NF-->>RN: HTML
RN->>PB: upsert book + chapters_idx
RN->>MN: PUT chapter objects
RN->>MS: UpsertBook doc
RN->>TQ: UPDATE task status=done
U->>UI: Poll GET /api/scrape/tasks/{task_id}
UI->>BE: GET /api/scrape/status
BE->>TQ: get task
TQ-->>BE: status=done
BE-->>UI: {status:"done"}
UI-->>U: Redirect to /books/{slug}
TTS Audio Generation Pipeline
Audio is generated lazily: on first request the job is enqueued; subsequent requests poll for completion and then stream from MinIO via presigned URL.
flowchart TD
A([POST /api/audio/{slug}/{n}\nbody: voice=af_bella]) --> B{Audio already\nin MinIO?}
B -- yes --> C[200 status: done]
B -- no --> D{Job already\nin queue?}
D -- yes pending/generating --> E[202 task_id + status]
D -- no --> F[INSERT audio_task\nstatus=pending\nin PocketBase]
F --> E
G([Runner polls task queue]) --> H[Claim audio_task\nstatus=generating]
H --> I[GET /api/chapter-text/{slug}/{n}\nfrom backend — plain text]
I --> J[POST /v1/audio/speech\nto Kokoro-FastAPI\nbody: text + voice]
J --> K[Stream MP3 response]
K --> L[PUT object to MinIO\nlibnovel-audio/{slug}/{n}/{voice}.mp3]
L --> M[UPDATE audio_task\nstatus=done]
N([Client polls\nGET /api/audio/status/{slug}/{n}]) --> O{status?}
O -- pending/generating --> N
O -- done --> P[GET /api/presign/audio/{slug}/{n}]
P --> Q{Valkey cache hit?}
Q -- yes --> R[302 → presigned URL]
Q -- no --> S[GeneratePresignedURL\nfrom MinIO — TTL 1h]
S --> T[Cache in Valkey\nTTL 3500s]
T --> R
R --> U([Client streams audio\ndirectly from MinIO])