caching

Cache-aside checks Redis first, and on a miss fetches from the database then writes the result back with setex and a TTL. This lazy-loading approach keeps the cache populated only with data actually requested.

Use a Redis lock key set with nx=True and an expiry so only one request fetches from the database while others wait and retry. This prevents a thundering herd when a hot key expires.

Match TTL to data volatility: rarely changing data like user profiles can use one hour, sessions around fifteen minutes, and frequently changing feeds about one minute. Avoid infinite TTLs.

Use tag-based invalidation: store each key in a Redis set per tag, then delete all keys in the set when that tag is invalidated. This clears groups like user and profile entries in one operation.

Skip caching for simple in-memory dictionaries used temporarily, database query optimization problems better handled by profiling, and static assets served with HTTP-level caching. Caching adds complexity that only pays off for repeated expensive reads.