Instagram -

(not a guide for this question; only for how this question is different from all others)

Instagram

Main differences

There is a lot of overlap with Twitter design:

But Instagram/Tiktok is image/video based, so:

Bandwidth can be an issue in some localities, so on upload, producing videos at different bitrates is important. How about consistent loudness between videos?
Access pattern (a lot in the first few days, then fade away) becomes important for cost saving. Need a “cached timelines” storage on Redis/Memcached, and a “archived timelines” on Cassandra/HBase.
“The algorithm” is very important, so analytics feeding back to timelines MUST be mentioned.

Users need to be profiled thoroughly to best recommend them posts that will boost engagement & retention.
Their social graph & activity must be analyzed through collaborative filtering & other data science algorithms, and output some profiling artifact (e.g. tags) to be kept in some User Service’s cache.
If elaborating on “The algorithm”: need User Service, User cache, Trends service, Profiling Service, etc.
Then, use the feedback Post-process service in the diagram to explain how posts are sent to users timelines.

Data stored in Postgres cluster. IDs generated with Postgres clusters as well (Sharding IDs at Instagram).
Django/Python stack for site.
From here: “All of our Redis deployments run in master-slave, with the slave set to save to disk about every minute.”