Redis Eviction Policies

16. Redis Eviction Policies

What is Eviction?

When Redis reaches its configured memory limit (maxmemory), it must evict (delete) some keys to make room for new writes. The eviction policy decides which keys to remove.

Analogy: Eviction is like your refrigerator being full — when you need to add something new, you must throw away old items. The policy decides whether to throw away the oldest, the least used, or whatever is expiring soon.

Eviction Decision Flow

flowchart TB
    App[App writes new data<br/>SET / LPUSH / SADD etc.] --> Check{Is Redis<br/>over maxmemory?}
    Check -->|No ✅| Accept[Store the data normally]
    Check -->|Yes ❌| Policy{Which eviction<br/>policy is set?}

    Policy -->|noeviction| Reject[Return error<br/>Writes rejected]

    subgraph Lru[LRU — Least Recently Used]
        LRU1[Remove keys that<br/>haven't been accessed<br/>the longest]
    end
    Policy -->|allkeys-lru| Lru

    subgraph Lfu[LFU — Least Frequently Used]
        LFU1[Remove keys that<br/>are accessed the<br/>least often]
    end
    Policy -->|allkeys-lfu| Lfu

    subgraph Ttl[TTL — Expiring Soonest]
        TTL1[Remove keys with<br/>the shortest TTL /<br/>nearest expiration]
    end
    Policy -->|volatile-ttl| Ttl

    subgraph Random[Random]
        RAND1[Remove a<br/>random key]
    end
    Policy -->|allkeys-random| Random

    Lru --> Evict[Evict selected key(s)]
    Lfu --> Evict
    Ttl --> Evict
    RAND1 --> Evict

    Evict --> Check

    style App fill:#3b82f6,color:#fff
    style Check fill:#f59e0b,color:#fff
    style Accept fill:#059669,color:#fff
    style Reject fill:#ef4444,color:#fff
    style Policy fill:#7c3aed,color:#fff
    style Lru fill:#7c3aed,color:#fff
    style Lfu fill:#3b82f6,color:#fff
    style Ttl fill:#f59e0b,color:#fff
    style Evict fill:#ec4899,color:#fff

All Eviction Policies

Policy	Description	Best For
`noeviction`	Return error on writes when full (default)	When every key matters
`allkeys-lru`	Evict least recently used keys from all keys	General-purpose caching ✅
`allkeys-lfu`	Evict least frequently used keys from all keys	When access frequency matters
`allkeys-random`	Evict a random key	Simple, predictable eviction
`volatile-lru`	Evict LRU from keys with TTL set only	Mixed caching + persistent data
`volatile-lfu`	Evict LFU from keys with TTL only	When TTL keys have varying frequency
`volatile-ttl`	Evict keys with the shortest remaining TTL	When expiry is the priority
`volatile-random`	Evict random key from TTL keys only	Simple eviction for TTL keys

Configure in redis.conf

# Set maximum memory (e.g., 256 MB)
maxmemory 256mb

# Set eviction policy
maxmemory-policy allkeys-lru

# For LRU/LFU: sample size for approximation (default 5)
maxmemory-samples 10  # Higher = more accurate, slower

LRU vs LFU vs TTL

# allkeys-lru — removes keys you haven't touched recently
# Good for: Most caching use cases
# Example: User session data — old sessions get evicted first

# allkeys-lfu — removes keys you rarely access
# Good for: Content caching where some items are "evergreen"
# Example: Product catalog — popular products stay cached

# volatile-ttl — removes keys about to expire
# Good for: When you only want to evict TTL keys
# Example: OTP codes — ones about to expire are removed first

Real-World Examples

Example 1: General Cache (Recommended)

maxmemory 512mb
maxmemory-policy allkeys-lru

Example 2: Session Store Only

maxmemory 1gb
maxmemory-policy volatile-ttl
# Only keys with TTL get evicted — persistent keys stay safe

Example 3: Mixed Data (Cache + Persistent)

maxmemory 2gb
maxmemory-policy volatile-lru
# Cached data (with TTL) gets evicted before persistent data

Common Mistakes

Mistake	Why	Fix
No `maxmemory` set	Redis uses all available RAM	Always set `maxmemory`
Using `noeviction` for caching	Writes fail when full	Use `allkeys-lru` instead
Setting `maxmemory-policy` without `maxmemory`	Policy has no effect without a limit	Always set both

In Simple Words

Eviction is Redis’s way of freeing memory when it’s full
allkeys-lru is the best default for most caching use cases
noeviction (default) just returns errors — only use if data loss is unacceptable
LFU tracks access frequency; LRU tracks access recency
Always set maxmemory and maxmemory-policy together in production