Reference · Cheat sheet

TTL vs eviction

Time kills one key; memory pressure kills many. Print it, pin it.

From lesson 0003See also glossary

Two mechanisms, one outcome

	TTL expiry	Eviction
Trigger	Time — per key	Memory pressure — global
Scope	The one key you marked	Whatever the policy selects
Configured by	`EXPIRE` / `SET EX`	`maxmemory` + `maxmemory-policy`
Meaning	"This datum is stale"	"We're out of room"

TTL commands

SET k v EX 300 — set value + 300s TTL atomically (PX = ms; EXAT/PXAT = absolute).
EXPIRE k 300 — (re)set TTL, O(1). Options: NX/XX/GT/LT.
TTL k — seconds left · -1 exists, no TTL · -2 no such key.
PERSIST k — remove the TTL (volatile → persistent).
Bare SET wipes the TTL. Use SET … KEEPTTL to keep it.

How a TTL deletes

Passive: checked on access → invisible to reads the instant it expires.
Active: samples TTL'd keys periodically; if >25% of a sample is expired, repeat.
So: logically dead ≠ memory freed. Up to ~25% of expired keys may linger in RAM.

Eviction policies

Candidate set	lru	lfu	random	ttl
allkeys-* (any key)	✓	✓	✓	—
volatile-* (only TTL'd keys)	✓	✓	✓	✓

noeviction = default; refuses writes with an OOM error when full.
LRU/LFU are approximated by sampling (maxmemory-samples, default 5).
volatile-ttl evicts the shortest remaining TTL first.

Two production traps

Default is noeviction — a cache under memory pressure starts erroring, not evicting. Set allkeys-lru/allkeys-lfu.
volatile-* with no TTL'd keys = nothing to evict = behaves like noeviction.

Replica note

Only the primary expires keys; replicas wait for its synthesized DEL. A replica read can return a logically-expired value until then.

← lesson 0003 · Source: Redis Docs — Key eviction · expiration FAQ