Use the Replication Lag Seconds alert in Mini DBA to monitor MySQL and MariaDB instances and make this condition visible before it becomes a wider database incident.
Mini DBA describes this alert as: A MySQL replica is behind its source. Sustained lag can make failover unsafe, cause stale reads, or indicate that the replica cannot keep up with the primary workload. The default evaluation frequency is Minute, so the alert is intended to be close enough to operational reality for live triage. Because duration is used, Mini DBA can avoid treating a single short spike as a full incident when the condition clears quickly.
This alert protects high availability and disaster recovery by warning when replicas, replication slots, or apply processes fall behind. Lag can turn a planned failover into data loss risk or leave reporting users looking at stale data.
Enable it where replication, read replicas, failover, or reporting copies are part of the service design. Disable it on standalone systems that do not use replication so the alert list stays focused.
Threshold meaning: Seconds behind source. Major threshold: 300 seconds. Minor threshold: 10 seconds. Comparison direction: "over". Duration is used, so prefer requiring the condition to persist before paging people for transient spikes. Use higher thresholds on batch-heavy, development, or intentionally bursty systems where brief pressure is expected. Use lower thresholds on latency-sensitive production systems, small instances with little headroom, and services with strict recovery or availability commitments.
Check the sender, receiver, apply process, network path, and oldest retained log position. Restart failed replication workers only after capturing the error, then address the root cause such as disk pressure, long transactions, schema drift, network latency, or insufficient replica capacity.
If this alert is noisy, check whether the workload naturally has short bursts. Raise the duration or threshold for expected batch windows, but keep a lower route for production systems where user-facing latency matters. Avoid simply disabling the alert until you know whether the noise is caused by threshold choice or by a real capacity trend.