RedisKeyEvictions

Redis has been evicting keys for a sustained period because it keeps reaching its memory limit.


Severity	info
Source	oliver006/redis_exporter 1.50+
Key metric	`redis_evicted_keys_total` (from `evicted_keys` in `INFO stats`)

What it means

When used memory hits maxmemory and the policy is anything other than noeviction, Redis removes keys to make room for new writes. The alert fires when evictions continue over a longer window rather than a single burst.

It is informational because for a pure cache with allkeys-lru or allkeys-lfu, steady eviction is the design. It becomes a real problem when Redis also holds data that must not disappear (sessions, rate-limit counters, job queues, locks), or when eviction is so heavy that the cache hit rate collapses.

Common causes

The dataset has outgrown maxmemory.
Keys stored without TTLs crowding out short-lived ones.
A single large key or a new feature caching much more data.
Cache and durable data mixed on the same instance.
maxmemory set lower than intended (or lowered during an incident and never restored).

First checks

Confirm the policy and how full memory is:

redis-cli -h <host> CONFIG GET maxmemory-policy
redis-cli -h <host> INFO memory | grep -E '^(used_memory_human|maxmemory_human)'

Check evictions against cache effectiveness:
```
redis-cli -h <host> INFO stats | grep -E '^(evicted_keys|expired_keys|keyspace_hits|keyspace_misses)'
```
A falling hit ratio alongside evictions means the cache is too small for the working set.

Graph eviction rate per instance:

sum by (instance) (rate(redis_evicted_keys_total[5m]))

Find what is using the memory:

redis-cli -h <host> --bigkeys
redis-cli -h <host> INFO keyspace

Fixing it

If this is a cache and hit rates are fine, you can accept it or lower the alert’s priority for that instance. Otherwise give Redis more memory (CONFIG SET maxmemory, a larger node, or sharding), add TTLs to keys that lack them, and trim oversized keys. Move data that must not be evicted to a separate instance with noeviction and alert on its memory instead.

RedisMemoryHigh: the memory pressure that leads to evictions.
RedisRdbLastSaveFailed: memory pressure also breaks snapshots.
RedisTooManyConnections: client buffers add to used memory.