At a glance
Expired Keys / minute is the rate at which Redis removes keys that have reached their TTL, derived from theexpired_keyscounter inINFO stats. Unlike eviction, expiry is intentional and healthy: it is the cache doing exactly what it was told to do. The line is most useful as a baseline; a sudden change in the expiry rate usually means a change in how the application is setting TTLs, not a problem with Redis itself.
What it tracks
The card plotsExpired Keys / minute over time, computed as the per-minute delta of the cumulative expired_keys counter from INFO stats. A key is counted here when it reaches its TTL and is removed, either lazily (touched after expiry) or by the active expiry cycle Redis runs in the background. This is distinct from Evicted Keys / minute: expired keys were always meant to die at their TTL, whereas evicted keys were still wanted and were dropped under memory pressure. A steady expiry rate that tracks your write volume is normal. A sudden jump can mean a batch of short-TTL keys was just written (for example a flush-and-reload of a session or rate-limit set); a drop to near zero can mean TTLs were accidentally removed or lengthened in a recent deploy, which will quietly grow memory use over time. The window is 1h (a rolling per-minute rate) and there is no alert threshold on this card (alert: -); it is a context and baseline metric rather than a pager. Read it next to memory and eviction to tell healthy TTL churn apart from memory pressure.
Reconciling against the source
Confirm against Redis directly withredis-cli INFO stats | grep -E 'expired_keys|evicted_keys', differencing two readings 60 seconds apart for the per-minute rate; for ElastiCache or MemoryDB the CloudWatch Reclaimed metric is the managed-service equivalent.