All the persistence of metrics goes first to the ring buffer. Then, the threads from the writer pool are responsible for grabbing data from the ring buffer and send it to Cassandra.

When the ring buffer is full, the Newts writer starts to drop data and an error message is logged. I think that besides logging an error is would be very useful to count how many metrics we are dropping, so we can estimate better if we should increase the thread pool, or the ring buffer, or both.

Of course, you might need an external tool like Hawtio to extract the data, as for obvious reason, it might be possible that it won't be available on Cassandra, if there are issues with the cluster, which is exactly the kind of things I'm interested on.