The Sluggish Feed
Home feed p95 latency increased from 320ms to 8.4s.
- Users report the feed takes about eight seconds to load.
- Error rate is normal.
- The feed shipped a new feature this morning: live reaction counts per post.
Graded on 5 review questions (70%), 3 key clues pinned to the evidence board (15%), and finishing within 9 min (15%).
Services
The app calls frontend, which calls feed-api, which checks auth-service and then calls user-service, posts-service, ranking-service and reaction-counts-service, which reads reactions-db.
- frontend
Edge service for the mobile and web apps. Receives the home feed request and calls feed-api.
Calls: feed-api
- feed-api
Builds one home feed page by calling the services below. The service that paged.
Calls: auth-service, user-service, posts-service, ranking-service, reaction-counts-service
- auth-service
Validates the session token on every feed request.
No downstream calls
- user-service
Returns the viewer's profile and the accounts they follow.
No downstream calls
- posts-service
Returns the posts for one feed page, up to 25 per page.
No downstream calls
- ranking-service
Orders the posts on the page for this viewer.
No downstream calls
- reaction-counts-service
Returns the reaction count shown under each post on the page.
Calls: reactions-db
- reactions-db
Postgres database that stores one row per reaction.
No downstream calls
Evidence board
0 / 8Collect outputs from tools to build a defensible RCA.