This walkthrough needs JavaScript. In short: LinkedIn's professional network, as this practice designs it, keeps the records (Espresso, MySQL, Ambry) apart from copies (Venice, LIquid, FollowFeed, Galene) that are rebuilt from a Kafka log. A feed page is built in 300 milliseconds with a deadline on every call. Posts reach readers by hybrid fan-out: pushed for authors under about 10,000 followers, pulled for bigger ones. Privacy is checked on every read, retries carry idempotency keys, and what each feature does when a dependency fails is written down in advance.