Cache & Redis Basic

Adding a cache is often the fastest way to take load off a database. In exchange, the system gains a second copy of the data.

Every hard question revolves around that copy: when it gets updated, who is responsible when it is wrong, and what happens when it disappears at peak load.

The articles and the questions they answer

Question	Article	Key conclusion
Which pattern does the application use to read/write the cache?	Cache-aside	Write the DB first, delete the key second; a race can still leave a stale value in cache until the TTL expires.
What happens when a hot key expires?	Cache stampede	Many requests miss at once and all hit the DB; coalesce requests, use a token lock, or refresh early.
Did the key vanish because it expired or because memory ran out?	Expiration and eviction	Compare expired_keys with evicted_keys before touching TTLs.
How does Redis execute commands?	Event loop and I/O threads	Commands run sequentially on one thread; one slow command makes other clients’ PING wait.
Which type should store the data?	Redis data structures	Choose by operation: sorted set for rankings, HyperLogLog for unique counts, hash for objects.
How much is lost on restart or kill -9?	Persistence: RDB and AOF	By default only RDB is on: every write after the last snapshot is lost; AOF everysec loses about 1 s when the machine crashes.
Is batching several commands atomic?	Pipelining, MULTI and Lua	A pipeline only cuts round trips; check-then-write needs Lua or WATCH.
Does a Redis lock guarantee only one party runs?	Distributed lock	A lock with a TTL can still have two owners at times; the target resource needs a fencing token.
The cache is there, yet the DB is still overloaded?	Cache penetration & avalanche	Nonexistent ids pierce the cache; many keys expire at once; use a bloom filter and TTL jitter.

These four problems are independent: correct cache-aside can still collapse from a stampede, and avoiding stampedes can still leave a bad p99 because a reporting job hogs the command thread.

Pick a reading path by symptom

Current situation	Read in this order
Just starting to use Redis as a cache	Cache-aside → Event loop
p99 jumps suddenly while request volume is unchanged	Event loop → check SLOWLOG and key sizes
The database gets slammed every time the cache expires	Cache stampede
Hit rate drops, keys vanish early, or Redis runs out of RAM	Expiration and eviction
Cached data drifts from the database	Cache-aside, the race and invalidation sections
Overselling stock when decrementing in Redis	Pipelining, MULTI and Lua
A cron job runs twice across pods despite a lock	Distributed lock
DB load spikes periodically or bots request garbage ids	Cache penetration & avalanche
Sessions are lost after a Redis restart	Persistence: RDB and AOF