RedisReplicasDisconnected

The number of replicas connected to a Redis master has just dropped.


Severity	warning
Source	oliver006/redis_exporter 1.50+
Key metric	`redis_connected_slaves` (from `connected_slaves` in `INFO replication`)

What it means

The alert watches the master’s replica count and fires when it goes down within a short window. It stays active for a while afterwards so a brief drop is not missed between notifications. It does not fire if a replica was never connected; it detects losses.

Fewer replicas means less redundancy and fewer failover candidates. If you use min-replicas-to-write, the master may start rejecting writes when too many replicas go away.

Common causes

A replica pod or host was restarted, rescheduled or scaled down.
Replica disconnected for exceeding the replica output buffer limit during a large write burst or full sync.
Network interruption between master and a replica.
A failover reconfigured replicas to follow a different master.
The replica process was OOM killed while loading a large RDB.

First checks

See which master lost replicas and when:
```
redis_connected_slaves
```
List the replicas the master currently sees, with their offsets:
```
redis-cli -h <master> INFO replication
```
Lines like slave0:ip=...,port=...,state=online,offset=...,lag=... show each connected replica.

Compare with the replicas you expect, and check each missing one:

redis-cli -h <replica> INFO replication | grep -E '^(role|master_host|master_link_status)'
kubectl -n <ns> get pods -l <redis-selector> -o wide

Look in the master log for why the connection closed:

kubectl -n <ns> logs <master-pod> --since=30m | grep -iE 'replica|output buffer|connection lost'

Check whether an ongoing full sync is involved: INFO persistence on the master shows rdb_bgsave_in_progress.

Fixing it

If the replica is restarting, it usually reconnects on its own; confirm state=online on the master. For output buffer disconnects, raise client-output-buffer-limit replica and repl-backlog-size. If a failover moved the replica elsewhere, that is expected; confirm the topology matches what Sentinel reports. Replace or repair a replica that is repeatedly OOM killed.

RedisReplicaLinkDown: the replica-side alert for a lasting disconnect.
RedisMissingMaster: check the group still has a master.
RedisMemoryHigh: memory pressure during sync often kills replicas.