autodial

Beeminder autodialer. A single Cloudflare Worker serves the React SPA (as static assets), the user API, and the dialing cron, backed by Workers KV.

Local Development

• Copy .env.example to .env and set variables
• npm start — runs the frontend at http://localhost:3000/
• cd functions && npm run dev — runs the Worker at http://localhost:8787/

Creating your Beeminder test client

• Go to your account settings
• At the bottom of the page, click "Register a new app"
• Name it something like bm_autodial_dev
• Use http://localhost:3000 as the redirect and post-deauthorize callback urls
• Copy the client ID into your .env file

Worker (functions/)

The Worker serves the built SPA from ../build as static assets, exposes POST /update and POST /remove (called by the frontend), and runs a scheduled() cron handler that dials every stored user's goals. In production the SPA and API share an origin, so the frontend calls /update relatively (no CORS); local dev is cross-origin (:3000 → :8787), which the Worker's CORS headers cover.

Run tests:

cd functions/
npm run test

Run locally (set DRY_RUN=true in .dev.vars to avoid writing to Beeminder):

npm run dev

First-time Cloudflare setup

npm run build                            # build the SPA into ./build first
cd functions/
npx wrangler kv namespace create USERS   # paste the id into wrangler.toml
npx wrangler deploy                       # deploys the Worker + SPA assets + cron

Migrating existing users (Firestore → KV)

The cron only dials users present in KV, so before cutover copy the existing Firestore users collection across (each user's token goes in KV metadata, the shape getUsers reads):

functions/scripts/migrate-firestore-to-kv.sh

Needs gcloud read access to the old Firestore and an authed wrangler. Run it at cutover — it's a point-in-time snapshot; anyone who authorizes on the old system afterward is missed until you re-run it.

Deployment

Pushes to master deploy via GitHub Actions (.github/workflows/deploy.yaml): one job builds the SPA and runs wrangler deploy, which uploads the Worker, the static assets, and the cron trigger together. Requires CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID repo secrets, plus the REACT_APP_APP_URL and REACT_APP_BM_CLIENT_ID build secrets (REACT_APP_API_URL is left empty in prod — same origin). The cron cadence is set in functions/wrangler.toml.

Error tracking (Sentry)

Both halves are optional and no-op until their DSN is set:

• Frontend — REACT_APP_SENTRY_DSN repo secret (baked into the build).
• Worker — a SENTRY_DSN Worker secret, set once with cd functions && wrangler secret put SENTRY_DSN (not a build var; the deploy doesn't provision it).

Beeminder tokens are scrubbed from URLs before events are sent (redactToken).