(I tried to refresh cached mapping in the index pattern page, it didn't help.)
I have two clusters with very similar architecture, one of them have this problem, but the other one is fine, differences between them are that problematic cluster is on version 7.10.2 and the working one is 7.12.0, not sure if it's related tho.
The problem is that in the monitoring index, some important fields shows "No cached mapping for this field." For instance, cluster_stats and cluster_state ones.
I tried to revert to built-in monitoring instead of metricbeat and use .monitoring-es-7-2021* instead of .monitoring-es-7-mb-2021* index pattern, but same fields are not cached there as well. Some of them are fine, for instance node_stats and index_stats are cached.
I need them because i want to make Grafana dashboard with overview of the cluster health. Otherwise, cluster monitoring inside kibana is not bothered by the fact that they are not cached, but in Grafana, that's make them unsearchable.
.monitoring-es index template is not edited in any way, it has same mappings as my working cluster have.