Distributed Systems

• Good for data that's read frequently but modified infrequently
• It's a good practice to add an expiration policy for when to remove data from the cache
• LRU - evict the least-recently-used data
• Databases frequently have some sort of caching. You can tweak this for more specific use cases

Caching Queries

• This approach is tough with complex queries
• If one piece of data changes, you need to find and modify all queries that might include the cached data

Caching Objects

• Cache web pages, user sessions
• Cache objects with cache.set(key, json.dumps(object))

Write-Through

• The cache always sits between the db and the client - for reads and writes
• Writes go to the cache, the cache writes to db

Disadvantages

• The db could lose data if the cache goes down prior to writes

Hotkey Issue

• When one key receives a disproportional amount of traffic
• To resolve this, we append a random number to the key that we hash

Redis

• Set a TTL expiry on data
• Or setup LRU eviction

Redis as a Rate-Limiter

• Ex. is a service can handle 5 requests in a minute, then we use Redis with a 60s TTL and increment requests as they come in

Message Queues in Redis

• Consumer group - pointer into a stream