Cheatsheet: Redis

EXPIRE key seconds sets a key to auto-delete after that many seconds (PEXPIRE takes milliseconds; EXPIREAT/PEXPIREAT take an absolute Unix timestamp instead of a duration). TTL key returns the remaining seconds until expiration, -1 if the key exists but has no expiration set, or -2 if the key doesn't exist at all (PTTL returns the same in milliseconds). PERSIST key removes any expiration from a key, making it live forever until explicitly deleted -- useful when a value that started as a cache entry turns out to need permanent storage.

Plain SET key value always writes, overwriting any existing value. SETNX key value (now largely superseded) only sets if the key doesn't already exist, historically used to implement locks. The modern equivalent is SET key value NX (only set if absent) or SET key value XX (only set if present) as options on regular SET, which can be combined with EX seconds / PX milliseconds to set an expiration atomically in the same command -- e.g. SET lock:foo 1 NX EX 30 acquires a 30-second lock only if nobody else holds it, all in one atomic round-trip.

Use a Hash (HSET user:1 name Alice age 30) when you have several related fields for one logical entity -- it groups them under a single key, lets you fetch/update one field without touching the others (HGET user:1 name), and is more memory-efficient than one String key per field (user:1:name, user:1:age) because Redis stores small hashes in a compact encoding. Reach for plain String keys when there's genuinely just one value per key, or when you need Redis's string-specific commands (INCR, APPEND, GETRANGE) on that value.

A List (LPUSH/RPUSH/LRANGE) preserves insertion order and allows duplicates -- good for queues, recent-activity feeds, or anything order-dependent. A Set (SADD/SMEMBERS/SISMEMBER) has no order and rejects duplicates -- good for membership tests (is X in this set?) and set algebra (SINTER, SUNION, SDIFF) like tag overlaps or mutual-friends lookups. A Sorted Set (ZADD/ZRANGE/ZSCORE) is like a Set but every member also has a floating-point score that keeps it ordered -- good for leaderboards, rate-limiting windows, or anything ranked by a numeric value where you need range queries by score or rank.

RDB takes periodic point-in-time snapshots of the whole dataset to disk (compact, fast to restore, but can lose the last few minutes of writes on a crash). AOF (Append Only File) logs every write command as it happens and replays them on restart (safer, configurable durability down to every write via fsync always, but a larger file and slower restore). Many deployments enable both. Replication (REPLICAOF host port, formerly SLAVEOF) is opt-in, not automatic -- a fresh Redis instance is standalone until you explicitly point a replica at a primary; Redis doesn't auto-discover or auto-replicate peers on its own.