RedisDown

redis_exporter has failed to talk to its Redis instance for a couple of minutes, so Redis is down or unreachable.


Severity	critical
Source	oliver006/redis_exporter 1.50+
Key metric	`redis_up` (1 = last scrape reached Redis, 0 = failed)

What it means

On each scrape the exporter connects to Redis and runs INFO. If that fails, redis_up is 0. The alert fires when this persists beyond a short grace period. The exporter process itself is running (its own up series is 1), so the problem is between the exporter and Redis, or Redis itself.

Applications using Redis as a cache usually degrade (slower, more database load); applications using it for sessions, queues or locks usually fail outright.

Common causes

Redis crashed or was OOM killed (container memory limit below actual usage).
Redis is blocked by a slow command (KEYS *, big DEL, long Lua script), so connections time out.
A requirepass/ACL change and the exporter’s password is now wrong.
The instance moved (failover, new pod IP) and the exporter points at a stale address.
maxclients reached, so new connections are refused.

First checks

Try to reach it from where the exporter runs:

redis-cli -h <host> -p 6379 -a '<password>' --no-auth-warning PING

Read the exporter log for the exact error (timeout, auth, connection refused):
```
kubectl -n <ns> logs deploy/<redis-exporter> --tail=50
```

Check the Redis process or pod and its last termination reason:

kubectl -n <ns> get pod <redis-pod> -o jsonpath='{.status.containerStatuses[*].lastState.terminated.reason}'
kubectl -n <ns> logs <redis-pod> --previous --tail=50

If Redis answers but slowly, look for blocking commands:

redis-cli -h <host> SLOWLOG GET 10
redis-cli -h <host> INFO clients

Check the scope:
```
count by (job) (redis_up == 0)
```

Fixing it

Restart a crashed instance and check the log for the reason before it happens again; for OOM, set maxmemory below the container limit with headroom for fork and buffers. Fix exporter credentials or address after a password rotation or failover (point it at the Sentinel-managed master or a stable Service). Ban KEYS in favour of SCAN if blocking commands caused it.

RedisMissingMaster: the down instance may have been the only master.
RedisReplicaLinkDown: replicas notice when their master disappears.
RedisRejectedConnections: a full client table can look like an outage.