Part 3 of 8 · 2 min read
The two flows
Publishing and building.
Feed publishing. A user sends a post. The book gives the API as POST /v1/me/feed with the post text and an auth token. The path through the system:
- A load balancer sends the request to a web server.
- The web server checks the token and applies a limit on posts per user per period. This stops spam.
- The post service saves the post in the database and the cache.
- The fanout service delivers the new post to friends' feeds.
- The notification service tells friends there is something new.
Feed building. The reader asks GET /v1/me/feed. A load balancer sends it to a web server, which calls the news feed service. That service reads a list of post ids from the news feed cache, then builds the full feed.
publish: user -> LB -> web server -> post service -> db + post cache
-> fanout service -> feed cache of each friend
-> notification service
read: user -> LB -> web server -> feed service -> feed cache (ids)
-> post cache + user cache (full objects)
-> CDN (images, video)
Why two stages on the read side? The feed cache holds only ids, to stay small. A feed needs names, pictures and text, so the service looks those up in a post cache and a user cache. This step is called hydration, filling a bare id with its full data. Media files come straight from a CDN (Level 1), not through the feed service.
Hydration
Hydration means replacing a list of ids with the full objects they point to, just before the answer is sent.
Example: a feed of 20 post ids becomes 20 posts with author name, picture, text and like count.
Check yourself: 1. Why does the web server check a post limit? 2. What does the feed cache hold? 3. Where do images come from?