Skip to content

Build Pipeline

Jump to bottom
github-actions[bot] edited this page Jul 9, 2026 · 7 revisions

Build Pipeline

360Ghar's npm run build is a production-aware pipeline with 11 stages. On non-production builds, API-heavy stages are skipped to keep previews fast; production runs include all stages plus IndexNow submission.

Key Files

File Role
package.json build script orchestrator + build:* sub-scripts
netlify.toml Netlify build command, env vars, redirects, headers
scripts/ingest-gurgaon-entities.mjs Stage 1a: fetch external sitemaps
scripts/merge-entities.mjs Stage 1b: merge + dedupe
scripts/build-localities-json.mjs Stage 1c: localities.json
scripts/build-localities-index.mjs Stage 1d: localities-index.json
scripts/generate-locality-sitemap.mjs Stage 1e + 3: locality sitemap
scripts/write-ai-discovery.mjs Stage 2: AI discovery artifacts
scripts/generate-sitemaps.mjs Stage 3: static + landing sitemaps
scripts/generate-datahub-sitemap.mjs Stage 3: data hub sitemap
scripts/generate-dynamic-sitemaps.mjs Stage 3: property + dynamic sitemaps
scripts/generate-rss.mjs Stage 4: RSS feeds
scripts/optimize-images.mjs Stage 5: AVIF/WebP image variants
scripts/generate-og-image.mjs Stage 6: OG image
vite.config.js Stage 7: Vite production build
scripts/purge-main-css.mjs Stage 8: PurgeCSS on main entry
scripts/generate-prerender-routes.mjs Stage 9a: prerender route manifest
scripts/fetch-prerender-data.mjs Stage 9b: bulk API data bundle
scripts/prerender-pages.mjs Stage 9c: Puppeteer prerendering (concurrent + cached)
scripts/lib/prerenderCache.mjs Stage 9 cache: content-hash HTML cache
scripts/purge-bootstrap.mjs Stage 10: PurgeCSS on Bootstrap
scripts/indexnow-submit.mjs Stage 11: Submit same-host URLs to IndexNow

The build Script

"build": "node scripts/build.mjs"

Sub-scripts:

"build:entities": "node scripts/ingest-gurgaon-entities.mjs && node scripts/merge-entities.mjs && node scripts/build-localities-json.mjs && node scripts/build-localities-index.mjs && node scripts/generate-locality-sitemap.mjs",
"build:ai-discovery": "node scripts/write-ai-discovery.mjs",
"build:sitemaps": "node scripts/generate-sitemaps.mjs && node scripts/generate-locality-sitemap.mjs && node scripts/generate-datahub-sitemap.mjs && node scripts/generate-dynamic-sitemaps.mjs",
"build:rss": "node scripts/generate-rss.mjs",
"build:images": "node scripts/optimize-images.mjs --quiet",
"build:prerender": "npm run build:prerender-routes && node scripts/fetch-prerender-data.mjs && node scripts/prerender-pages.mjs",
"build:prerender-routes": "node scripts/generate-prerender-routes.mjs",
"build:prerender-data": "node scripts/fetch-prerender-data.mjs",
"build:indexnow": "node scripts/indexnow-submit.mjs",

Stages in Order

1. Entity Ingestion (build:entities)

  1. ingest-gurgaon-entities.mjs - fetches sitemaps from Magicbricks, SquareYards, NoBroker, CommonFloor; extracts <loc> URLs matching gurgaon|gurugram; decodes locality/society/project names. Output: scripts/reports/entity-raw.json.
  2. merge-entities.mjs - dedupes + merges raw entities.
  3. build-localities-json.mjs - writes src/data/localities.json.
  4. build-localities-index.mjs - writes src/data/localities-index.json (sorted, slugged).
  5. generate-locality-sitemap.mjs - writes public/sitemap-localities.xml.

isPlaceholderName() filters junk tokens; each entity carries confidence and sourceCoverage. See SEO & Programmatic.

2. AI Discovery (build:ai-discovery)

write-ai-discovery.mjs calls buildAiDiscoveryArtifacts() from src/seo/aiDiscovery.js and enriches the feed with the top 50 localities + 20 societies from localities-index.json. Outputs:

  • public/.well-known/ai.txt
  • public/.well-known/api-catalog
  • public/llms.txt
  • public/llms-full.txt
  • public/data/llm-feed.json

3. Sitemap Generation (build:sitemaps)

  1. generate-sitemaps.mjs - sitemap.xml (index), sitemap-static.xml, sitemap-landing.xml. Emits hreflang alternates (en, hi, x-default) for every URL, prunes via src/data/pseo-prune-list.json, and supports SITEMAP_MAX_LANDING_PER_CITY / SITEMAP_BATCH for phased releases.
  2. generate-locality-sitemap.mjs - sitemap-localities.xml (re-run; idempotent).
  3. generate-datahub-sitemap.mjs - sitemap-datahub.xml.
  4. generate-dynamic-sitemaps.mjs - sitemap-properties.xml + any dynamic batches (fetches live properties from the API).

4. RSS Generation (build:rss)

generate-rss.mjs fetches blog posts and properties via scripts/lib/paginatedApi.mjs, plus localities from localities-index.json. Outputs:

  • public/rss.xml - main feed (blog + properties)
  • public/rss/localities.xml - locality feed

Each item is XML-escaped and dated with toRfc2822().

5. Image Optimization (build:images)

optimize-images.mjs uses sharp to generate next-gen variants for every PNG/JPG under public/assets/images (above 50KB):

  • <name>.webp (quality 80)
  • <name>.avif (quality 50)
  • For "responsive" hero/banner art, also width-constrained variants at 320 / 640 / 768 / 1024 px in both WebP + AVIF.

Idempotent: outputs are skipped if newer than the source. Flags: --force (regenerate all), --quiet (summary only).

6. OG Image Generation

generate-og-image.mjs resizes public/assets/images/thumbs/banner-three.webp to 1200x630 JPG (quality 85, progressive) at public/og-image-home.jpg for Open Graph / Twitter cards.

7. Vite Production Build

vite build runs the standard Vite production build (React plugin, SCSS, code-splitting, compression plugin, PWA plugin, bundle visualizer). Output: dist/.

8. Main CSS Purge

purge-main-css.mjs runs PurgeCSS on the main entry CSS chunk (dist/assets/index-[hash].css, ~179KB from main.scss + component styles). Scans built HTML/JS + source JSX for used selectors, with a safelist for dynamic classes (show, collapsing, collapse, modal, etc.). Reduces the entry chunk significantly.

9. Prerendering (build:prerender)

  1. generate-prerender-routes.mjs - builds scripts/prerender-routes.json from indexableStaticRoutes, seedLandingPrerenderRoutes, seedLocalityPrerenderRoutes, and locality data. Each route has waitForSelector / waitForText / waitForTitle.

  2. fetch-prerender-data.mjs - pre-fetches a single bulk bundle of the high-value API payloads (recommendations, default + per-route property searches, blog posts) keyed by the same buildRequestKey the SPA adapter uses. Writes dist/prerender-data.json = { meta, entries: { [requestKey]: payload } }. During capture the SPA reads from this bundle instead of firing live calls per route, so production prerender no longer hammers the backend with 244 × live requests. In bulk mode, unknown keys fall through to live requests; non-production empty mode returns empty payloads. Never fails the build on a network error.

  3. prerender-pages.mjs - spawns vite preview on 127.0.0.1:4317, launches headless Puppeteer (--no-sandbox on Linux), and renders each route. Optimizations:

    • Concurrency (PRERENDER_CONCURRENCY, default 5, clamped 1-8): a bounded worker pool renders multiple routes in parallel against one shared browser (one page per route).
    • Content-hash cache (PRERENDER_CACHE_DIR, default node_modules/.cache/prerender-html): Netlify persists this across builds. Each route's cache key is sha256 over the vite build hash (dist/.vite-build-hash), the bulk-data bundle hash, the route config, and the algorithm source hash. Unchanged routes skip the Puppeteer render and copy cached HTML to dist/. Disable with PRERENDER_CACHE_DISABLED=1.
    • Request interception: aborts maps/analytics/fonts/service-worker during capture (defense-in-depth alongside the in-app isPrerendering() short-circuits).
    • Lower timeouts: per-route wait lowered from 60s to 20s (waitForText 10s) so flaky routes fail fast.

    Per route: sets window.__PRERENDER_INJECTED = { isPrerendering: true }, navigates, waits for the configured signal, serializes page.content() to dist/<route>.html, and rewrites stylesheets to be non-blocking.

The isPrerendering flag is checked by authStore.initializeAuth, locationStore.initializeLocation, posthogService.init, and main.jsx to skip network calls. The bulk-data adapter in src/services/http.js (__PRERENDER_DATA_SOURCE__ define) serves from dist/prerender-data.json during production capture; non-production builds use the 'empty' short-circuit.

10. Bootstrap Purge

purge-bootstrap.mjs runs PurgeCSS on public/assets/css/bootstrap.min.css (~190KB) against built HTML/JS + source JSX, reducing it to ~40-50KB. Safelist preserves dynamically-added classes (show, collapsing, collapse, modal, tooltip, popover, fade, active, disabled, open, btn-, form-, and col- responsive variants).

11. IndexNow Submission (Production only)

indexnow-submit.mjs scans generated sitemap*.xml files from dist/ (fallback: public/), extracts same-host URLs, and submits them in batches to https://api.indexnow.org/IndexNow after build completion. It deduplicates URLs across all sitemap files and sends the payload fields expected by Bing IndexNow (host, key, keyLocation, urlList).

Production builds use the static key ba96fa507cb7447aa74f5ddd2f516a6d with its companion file at public/ba96fa507cb7447aa74f5ddd2f516a6d.txt.

  • Batch size defaults to 1000 and can be overridden via INDEXNOW_BATCH_SIZE.
  • Optional tuning env vars: INDEXNOW_ENDPOINT, INDEXNOW_DRY_RUN, INDEXNOW_ENFORCE_SUCCESS.

Build Pipeline Diagram

flowchart TD
    S1[1. Entity Ingestion] --> S2[2. AI Discovery]
    S2 --> S3[3. Sitemaps]
    S3 --> S4[4. RSS]
    S4 --> S5[5. Image Optimization]
    S5 --> S6[6. OG Image]
    S6 --> S7[7. Vite Build]
    S7 --> S8[8. Main CSS Purge]
    S8 --> S9[9. Prerender]
    S9 --> S10[10. Bootstrap Purge]
    S10 --> S11[11. IndexNow Submission]
    S11 --> Dist[dist/ deployed to Netlify]
Loading

Netlify Deployment

netlify.toml config:

[build]
  publish = "dist"
  command = "npx puppeteer browsers install chrome && npm run build"

[build.environment]
  PUPPETEER_CACHE_DIR = "./node_modules/.cache/puppeteer"
  NODE_VERSION = "20"
  VITE_API_SERVER = "https://api.360ghar.com"
  VITE_API_BASE_URL = "https://api.360ghar.com/api/v1"
  PRERENDER_CONCURRENCY = "5"
  PRERENDER_CACHE_DIR = "node_modules/.cache/prerender-html"

Puppeteer's Chrome is installed explicitly before the build so prerendering works in Netlify's CI. Netlify persists node_modules/.cache across builds, so the content-hash prerender cache survives deploys and skips unchanged renders. Prerender tuning env vars:

Var Default Effect
PRERENDER_CONCURRENCY 5 Parallel Puppeteer pages (clamped 1-8)
PRERENDER_CACHE_DIR node_modules/.cache/prerender-html Where cached HTML + manifest live
PRERENDER_CACHE_DISABLED unset 1 forces every route to render
PRERENDER_DATA_DISABLED unset 1 writes an empty bulk bundle
VITE_PRERENDER_DATA_SOURCE bulk (prod) / empty (other) Overrides the SPA adapter data source

IndexNow tuning env vars:

Var Default Effect
INDEXNOW_API_KEY ba96fa507cb7447aa74f5ddd2f516a6d API key used in request payload
INDEXNOW_ENDPOINT https://api.indexnow.org/IndexNow Submission endpoint
INDEXNOW_KEY_LOCATION https://360ghar.com/ba96fa507cb7447aa74f5ddd2f516a6d.txt Public key file URL
INDEXNOW_SOURCE_DIR dist (fallback public) Directory to parse sitemap files from
INDEXNOW_BATCH_SIZE 1000 URLs per API payload
INDEXNOW_DRY_RUN 0 1 logs submission attempts without network calls
INDEXNOW_ENFORCE_SUCCESS 0 1 fails the build on any failed batch

Key redirects:

  • www.360ghar.com -> 360ghar.com (301)
  • trailing-slash normalization (301)
  • /gurugram/* -> /gurgaon/* (301)
  • /apartments -> /flats (301, all intents)
  • Hindi equivalents under /hi/*
  • SPA catch-all /* -> /index.html (200)

Cache headers: hashed JS/CSS/fonts are immutable (1 year); images 30 days + stale-while-revalidate; HTML always revalidates (browser max-age=60 + a longer CDN-Cache-Control edge window, since Netlify purges the edge on every deploy); service worker never cached. The root / advertises AI-discovery Link headers (RFC 8288) for api-catalog, service-doc, service-meta, mcp-server, agent-skills, llms-txt, openid-configuration.

Edge Functions

Netlify auto-discovers Deno edge functions under netlify/edge-functions/ (in-file export const config = { path, excludedPath }, no netlify.toml block needed):

  • markdown-negotiation.js — content negotiation for /for-ai.
  • soft-404-guard.js — runs on /* before the SPA catch-all. Paths that match NONE of the SPA's valid route grammar (static pages, /property/:id, /blog/:slug, /project/:slug, /policies/:slug, tools, /:city/:intent/:type + facet variants, /localities, with a tolerated /hi locale prefix) are rewritten to /404.html and returned with HTTP 404 + X-Robots-Tag: noindex. Static assets, the /api/* proxy, and well-known files always pass through. This converts soft-404s (the /* -> /index.html 200 catch-all otherwise serves a 200 SPA shell for every path) into hard 404s. Verify post-deploy: a junk URL returns 404, valid dynamic routes still return 200.

Head-tag dedup gotcha: react-helmet-async only adds head tags — it never removes static tags authored in index.html. So index.html must NOT ship static <meta name="description">, <link rel="canonical">, or hreflang alternates, or they will appear first and shadow the per-page tags the SEO component renders via Helmet on every route. index.html keeps only OG/Twitter tags as a fallback for non-JS social crawlers.

Dependency Constraints

  • ESLint must stay on major version 9 (@eslint/js ^9.39.4). eslint-plugin-react@7.x declares peerDependency: eslint ^2 || ... || ^9 and does not support eslint 10+. Upgrading to eslint 10 causes Netlify npm install to fail with ERESOLVE.
  • Node >= 18 (Netlify pins 20).
  • Puppeteer ^24.40.0 (devDependency) for prerendering.
  • sharp ^0.34.5 for image optimization.
  • Before upgrading any devDependency to a new major, check peer ranges with npm info <pkg> peerDependencies, run npm install locally, and run npm run build && npm run lint.

Cross-References

360Ghar Frontend Wiki Footer

360Ghar Frontend Wiki | 360ghar.com | info@360ghar.com | +91 9999900876

360Ghar Frontend

Getting Started

Contributing

Pages

Components & Common

Architecture

Features

Tools

Reference

Clone this wiki locally