metrics_reader

An application that exposes Folsom metrics in common standardised formats.

This is important for several reasons:

• Applications provide a common interface for the same purpose
• Exposing application metrics in standardised formats allows existing tools to scrape these metrics e.g. Prometheus
• Knowledge is transferable when the same problem is solved in a consistent manner.

The metrics reader exposes the application metrics via an rpc call to the nodetool:

./bin/metrics_reader metrics

The metrics reader server

Metrics should be registered with the metrics reader server if they are to be reported via the console:

metrics_reader:register(subscriber_acks).

It is also possible to de-register a registered metric:

metrics_reader:deregister(subscriber_acks).

Finally, in order to examine the registered metrics:

metrics_reader:metrics().

The counter histogram

A stateful wrapper around a Folsom histogram is provided for use by clients. This accumulates scalar values over a configured interval, and creates a sample resevoir over these accumulated values. This allows richer statistics to be derived for this structure. It is possible to extend this histogram beyond counters, to other scalar types. Also, it is not necessary to use Folsom as a backend, and future versions may allow this as a configuration option.

At a conceptual level, the abstraction is leaky, because normally counters provide monotonic guarantees. This counter may appear to oscillate between high and low values as the accumulator clears previous registers.

counter_histogram:start().
counter_histogram:new(subscriber_acks_per_second).
counter_histogram:inc(subscriber_acks_per_second).

Configuration

{format, prometheus_format}                         # defaults to prometheus exposition format
{histogram_acc_interval_sec, 10}                    # defaults to 1 second
{histogram_slide_interval_sec, 60}                  # defaults to 60 second

Export formats

An export_format behaviour is defined that consists of the following callbacks:

callback histogram(Name :: binary(), Histogram :: #{}) -> 
   [binary()].

An default prometheus_format formatter is already provided that will convert the metric values to the prometheus textual exposition format.

Build

$ rebar3 compile

In order to run tests…{to do}

Release

$ rebar3 release