RedisReplicaLinkDown

criticalRedis (redis_exporter)

A Redis replica has been disconnected from its master for several minutes and is no longer receiving updates.
Severity critical
Source oliver006/redis_exporter 1.50+
Key metric redis_master_link_up (1 = master_link_status:up)

What it means

On a replica, INFO replication reports whether the link to the master is up. The alert fires when it has stayed down for several minutes, longer than a normal reconnect or short resync.

While the link is down, the replica serves increasingly stale reads, and it is not a safe failover target: promoting it would lose everything written since the link dropped. If this replica is your only one, you have no redundancy.

Common causes

  • The master is down or was replaced, and the replica still points at the old address.
  • A full resync loops: the replica requests a full sync, the RDB transfer takes too long, and the master drops it because client-output-buffer-limit replica is exceeded.
  • The replication backlog (repl-backlog-size) is too small, so every short disconnect forces a full sync.
  • Wrong masterauth after a password or ACL change.
  • Network policy, firewall or DNS change between replica and master.

First checks

  1. On the replica, look at link state and how long it has been down:
    redis-cli -h <replica> INFO replication | grep -E '^(role|master_host|master_port|master_link_status|master_link_down_since_seconds|master_sync_in_progress)'
  2. On the master, check whether it sees the replica:
    redis-cli -h <master> INFO replication
  3. Read the replica log for the reason sync fails (auth errors, timeouts, "MASTER aborted replication"):
    kubectl -n <ns> logs <replica-pod> --tail=100 | grep -iE 'master|sync|replica'
  4. Check the master log for output buffer disconnects:
    kubectl -n <ns> logs <master-pod> --tail=200 | grep -i 'scheduled to be closed ASAP for overcoming of output buffer limits'
  5. Find all affected replicas:
    redis_master_link_up == 0

Fixing it

If the master moved, run REPLICAOF <new-master> <port> (or let Sentinel reconfigure it). If full syncs keep failing, raise the replica output buffer limit and repl-backlog-size on the master with CONFIG SET, and persist the change to the config. Fix masterauth after credential changes, and check connectivity with redis-cli -h <master> PING from the replica host.

Related alerts

← all alert runbooks