Market-data per repo — GH Actions cron + GH Pages JSON serve, one repo per API

decision Mon Jun 22 2026 00:00:00 GMT+0000 (Coordinated Universal Time) decisionarchitecturemarket-datagithub-actionsgithub-pagesjsonindiafree-tierno-cf-workers

Market-data per repo

Decision

Each India-market data feed lives in its own GitHub repo. GitHub Actions cron runs the scraper, commits the resulting JSON back to that repo's data/ directory, and GitHub Pages publishes the same directory at a stable HTTPS URL. Apps fetch via Pages or raw.githubusercontent.com. Zero Cloudflare Workers, zero shared aggregator, zero recurring cost.

Scrapers are Node 22 native fetch + cheerio, ~55 LOC each, in scripts/scrape.mjs. FII/DII tries NSE official API first, falls back to Moneycontrol HTML scrape. MMI scrapes Tickertape's public /market-mood-index page (Next.js __NEXT_DATA__ regex with DOM-text fallback).

Each repo has two workflows:

• .github/workflows/scrape.yml — cron + workflow_dispatch, runs the scraper, commits + pushes data/.
• .github/workflows/pages.yml — on push to data/**, deploys the data/ directory to GitHub Pages (Actions source, not legacy branch source).

Why per-repo, not the aggregator we tried briefly

The aggregator fused two unrelated upstreams into one repo for no real benefit:

• Slug stays 1:1 with product. flow-fii-dii.api.oriz.in and mmi.api.oriz.in were already mapped to per-API slugs; mashing them under a third slug broke that mapping.
• No shared scraper code. NSE FII/DII and Tickertape MMI scrapers have nothing in common beyond fetch + cheerio. Co-locating them shaved no LOC.
• Independent cadence. Weekday post-close vs hourly cron are different ops shapes; a shared repo would have run two unrelated GH Actions back-to-back.
• Cleaner blast radius. If Tickertape changes shape and MMI scraper breaks, FII/DII commits stay clean (and vice-versa). With a shared repo, one bad scrape could pollute the git log of the other.

Why GitHub Pages + raw, not CF Workers

Same rationale as the original draft:

• Free forever. GH Actions: 2000 min/mo free on public repos. Pages + raw: unlimited reads via GitHub's CDN. No card, no quota worry.
• Audit trail built in. Every scrape is a commit. git log data/ is the time-series history.
• Lower blast radius. A bad scrape is a revertable commit. A bad KV write is opaque.
• One less moving piece. No wrangler deploy, no DNS CNAME, no CLOUDFLARE_API_TOKEN rotation.

Trade-off accepted: a small eventual-consistency window (apps see data up to 1h stale for MMI, up to 24h for FII/DII) and reliance on GH Actions uptime. Both acceptable for non-trading UI.

Implications

• Submodules restored. repos/oriz/own/svc/api/oriz-flow-fii-dii-activity-api + repos/oriz/own/svc/api/oriz-mmi-tickertape-mmi-api are wired back into master .gitmodules. The shared repos/oriz/own/content/data/oriz-market-data submodule and its upstream repo were deleted on 2026-06-22.
• CF Worker code removed. src/index.ts (Hono routes) + wrangler.toml + tsconfig.json + vitest.config.ts + test/ were deleted from both repos. Subdomains flow-fii-dii.api.oriz.in and mmi.api.oriz.in are unused; the canonical fetch URL is now the GH Pages URL above.
• Apps consume via URL constants. First user: oriz-paisa-finance-tools-app/src/data/market-data-urls.ts — both FII_DII_LATEST_URL and MMI_LATEST_URL point at raw.githubusercontent.com per-repo URLs.
• GH Pages enabled on both repos via REST API (build_type=workflow).
• License MIT per mit-license-all-repos.
• Linux-only CI per rules/linux-ci-only.

Cross-refs

• market-data-apis — the original CF-Worker design, now archived as a historical alternative
• — the rejected aggregator pattern (same-day revert)
• github-pages-as-json-api — the sibling pattern this generalises
• cf-worker-quota-mitigation — quota-pressure context that motivated moving off Workers