This is an attempt to revive the concept introduced in
CASSANDRA-3527. Instead of specifically targeting JMX metrics, simplify by just exposing metrics into CQL tables as a dynamic view.
Exposing these metrics will give operators vital insight without resorting to anything more than what is installed with Apache Cassandra.
The scope of these tickets is only for reading metrics with a SELECT command. The actual data model can be discussed in the comments below.
Some basic requirements I would propose:
- SELECT syntax should allow for single node or cluster wide statistics
- Aggregations should be supported (min, max, avg)
- Inside metric tables, future column additions should be automatically created when new metrics are added to a family
- Large schema changes such as deletes should be postponed to a major release