Skip to content

web-production.md

Latest commit

 

History

History
61 lines (49 loc) · 3.21 KB

web-production.md

File metadata and controls

61 lines (49 loc) · 3.21 KB
Raw

Web deployment and integrations

The web application includes a PDF reader with annotations, knowledge libraries, source-grounded organization and questions, and persistent notebook notes. Model jobs run on the worker; the browser displays organization progress immediately and follows the server's queued, running, succeeded, and failure states.

Account authentication

Local accounts use invite registration, password recovery, and workspace-bound sessions. FastResearch SSO and optional FastCAS linking retain the same account and workspace. Configure the Research connection on the backend:

FASTRESEARCH_API_URL=http://127.0.0.1:8787
FASTRESEARCH_PANEL_URL=https://research.example.org
FASTREAD_COOKIE_SECURE=true

The API consumes a one-time ticket at POST /api/sso/consume with audience fast-read. Each research identity owns a separate workspace. Re-entering with a new ticket invalidates that identity's previous sessions. The browser exchanges the ticket through /api/auth/research, removes it from the URL, and obtains its CSRF token from /api/auth/me. Cookie-authenticated mutations require X-CSRF-Token. Keep API credentials and production configuration outside Git.

Both backend document routing and the frontend bootstrap handle SSO because deployments may serve HTML through a separate static server. The bootstrap is part of the normal Vite build; a patched production bundle is no longer needed.

FastNews metadata intake

The web entry accepts source=fastnews with title, optional authors, abstract, url, venue, year, and external_id query parameters. After login, the bootstrap sends them to POST /api/integrations/fastnews/intake. This records metadata in the authenticated workspace without downloading the paper. Repeated intake is deduplicated within that workspace. Users attach their PDF through POST /api/imports/pdf with multipart file and paper_id. Reading endpoints return a request to upload the PDF until a document version exists.

Build, delivery, and rollback

Build the frontend using pnpm install --frozen-lockfile and pnpm run build inside fastread-frontend. Publish the complete dist directory, including vendored PDF.js files and their license notices. Keep previous assets available for clients with an older page open. Give HTML revalidation/no-cache semantics; fingerprinted assets may be cached. Retain authentication on PDF requests, byte range/206 and HEAD support, and ETag/304 revalidation. Do not gzip PDF responses or apply API-wide no-store rules to them. Other private API responses remain non-cacheable.

Use new release directories, preserve the previous release, and atomically swap the active symlink. Check service health and queued/running jobs before changing the backend or worker. A frontend-only release does not require a worker restart. Keep runtime data, uploaded papers, login files, provider secrets, and host-specific deployment helpers out of release source commits.

Validation: run go test ./... and go vet ./... in backend-go; run frontend pnpm run lint, pnpm run typecheck, pnpm test, and pnpm run build. The native worker uses Poppler for PDF extraction. FastDeploy includes its executables and shared libraries in the release.