How Google Search renders the web - a look behind the curtain
summary
Google Search uses a Web Rendering Service that runs headless evergreen Chrome in a managed pool, collects a render tree not pixels, and intentionally skips painting; cookies and service workers are cleared or rejected so renders mimic first-time visitors. Median queue time is about 5s, caching rates are around 60–70%, rendering cutoffs vary, and tests hit WRS directly without the cache. The system aims for deterministic outputs (fixed random seed) and per-render metrics for indexing.
AI-written summary · about
chapters (26)
- 0:00Intro
- 0:59What we'll talk about today...
- 2:26An example: Parsing into DOM tree DOM tree body
- 3:54An example: DOM tree to layout
- 6:24An example: Painting
- 7:52Hello, JavaScript
- 9:14Basically...
- 9:53A lay of the land
- 10:37Our focus today
- 11:10WRS - 10,000ft view
- 11:48Let's build a simple WRS w/ puppeteer const page = await browser.newPage()
- 15:24Actually that's more like it...
- 15:59By the way...
- 16:48But wait, there's more...
- 19:11WRS DOM tree (excerpt)
- 19:49WRS render tree (excerpt)
- 20:03A glimpse into the renderer
- 25:07Rendering time
- 25:58It's hard to measure this externally
- 28:35Use simple, clean caching
- 29:28Sample with Webpack
- 29:45Reduce resource requests
- 30:10Code bundling
- 30:56Code splitting, Vue.js example
- 31:15Code splitting, Webpack config module.exports
- 32:05Let's recap
