The current puppet logic for replication lag monitoring monitor_replication.pp#L47-L52:
# check the lag towards the mw_primary datacenter's master $mw_primary = mediawiki::state('primary_dc') nrpe::monitor_service { "mariadb_replica_sql_lag_${name}": description => "MariaDB Replica Lag: ${name}", nrpe_command => "${check_mariadb} --check=slave_sql_lag \ --shard=${name} --datacenter=${mw_primary} \ --sql-lag-warn=${lag_warn} \ --sql-lag-crit=${lag_crit}",
This causes the master in the primary DC to monitor lag from itself. In the case of unidirectional replication, this is a no-op as check-mariadb.pl notices that there's no slave thread running, and skips the check. With circular replication (a la x2, or core sections in the lead-up to a DC switchover), monitoring should be looking at the lag from the other DC, but currently does not.