Expiration & TTL

Introduction

TTL (time-to-live) is non-negotiable for cache keys — without EX/PX every SET lives forever until manual DELETE or OOM. Redis expires keys passively on access and actively via sampling (~10 keys/second per DB plus faster cycle when memory pressure hits).

Mass simultaneous expiry causes cache stampede — thousands of requests hit DB when hot keys expire together. Jitter TTL and single-flight refresh prevent avalanches.

volatile-* eviction policies only remove keys with TTL set — understand the difference from allkeys-*.

Understanding the topic

Key concepts

• EX seconds / PX milliseconds on SET — set with expiry.
• EXPIRE key seconds — add TTL to existing key.
• TTL key — remaining seconds; -1 no expiry; -2 missing.
• PERSIST — remove TTL.
• Passive expire on read; active random sampling.
• volatile-lru evicts expirable keys when maxmemory hit.

flowchart LR
  SET -->|EX| KeyTTL
  Access -->|passive| Delete
  Sample -->|active| Evict

Step-by-step explanation

1. Expiry stored in key metadata dict.
2. On access, Redis checks expiry and deletes if past.
3. Active cycle samples random keys with TTL.
4. Expired keys deleted or evicted per policy.
5. EXPIREAT/PEXPIREAT use absolute Unix timestamps.

Syntax reference

Common commands

• Always EX on cache writes.
• TTL -2 means key gone — treat as cache miss.
• Add random jitter in app: baseTtl + random(0, 60).

SET session:abc data EX 3600
TTL session:abc
EXPIRE cache:product:1 1800
SET cache:flag 1 EX 300 NX

Informative example

Cache write with jittered TTL to reduce stampede:

# Shell illustrates jitter concept
BASE=3600
JITTER=$(( RANDOM % 300 ))
redis-cli SET "product:42" "$JSON" EX $((BASE + JITTER))

In Java: Duration.ofSeconds(3600 + ThreadLocalRandom.current().nextInt(300)). Spring @Cacheable supports custom TTL per cache name.

Real-world use

Real-world use cases

• Session timeout — EXPIRE aligned with JWT.
• API cache — 5–60 min TTL by volatility.
• Rate limit windows — EXPIRE on counter key.
• OTP codes — EX 300 one-time.
• Distributed lock — SET NX EX auto-release.

Best practices

• TTL on every cache key — no exceptions.
• Jitter TTL on hot keys.
• Refresh TTL on session activity (sliding window).
• Monitor expired_keys_per_sec in INFO stats.
• Use volatile-lru when most keys have TTL.
• Plan stampede protection (sync=true, mutex).

Common mistakes

• Forever keys — memory leak.
• Identical TTL on millions of keys — thundering herd.
• Assuming EXPIRE extends on read without EXPIRE call.
• volatile-lru with keys missing TTL — they never evict.

Advanced interview questions

Summary

TTL prevents unbounded memory growth.