Some metrics are not filtered

[podile]

The following metrics are not filtered even if mentioned in --exclude list. These metrics generating too many records in case of larger clusters. For example, a cluster of 250 nodes with 150 keyspaces resulting 37500 metrics.

cassandra_keyspace_effective_ownership_ratio
cassandra_token_ownership_ratio
cassandra_jvm_gc_estimated_collection_duration_seconds_total
cassandra_jvm_nio_buffer_pool_estimated_buffers

StorageServiceMBeanMetricFamilyCollector is the source of the first two metrics, is there any reason for not filtering it? If not, I will do code change to filter the metrics and raise a PR.

[zegelin]

Try adding an exclusion for org.apache.cassandra.db:type=StorageService. That will disable all StorageService metrics.

Exclusions are implemented at registration time and can either filter on MBean names/patterns or collector names.

For most metrics (anything that uses FunctionalMetricFamilyCollector) the collector name is equivalent to the prometheus metric family name, as there is one-to-one mapping between MBeans, collectors and metric families. Hence you can create an exclusion based on a metric family name, since its actually matching the collector name.

But some MBeans, such as the StorageService, have a custom collector -- in this case StorageServiceMBeanMetricFamilyCollector. These collectors output multiple metric families, hence there is no longer a one-to-one mapping. As a result, the exclusion must be created against the MBean name (in this case org.apache.cassandra.db:type=StorageService). This does have the side effect of filtering all metrics families generated by the collector.