Periodically collects and pushes VM metrics to arbitrary consumer processes, which, in-turn, can do whatever they want with the given data (such as serialize and forward to some time series storage). Includes, off by default, example implementations of consumers for:
- StatsD (
beam_stats_consumer_statsd
) - Graphite (
beam_stats_consumer_graphite
) - CSV file (
beam_stats_consumer_csv
)
Essentially like folsomite
, but different. Different in the following ways:
- More-general: consumers other than graphite can be defined
- More-focused: only concerned with VM metrics, while
folsomite
ships off everything fromfolsom
(in addition to VM metrics) - Easier-(for me!)-to-reason-about implementation:
- Well-defined metrics-to-binary conversions, as opposed to the
nearly-arbitrary term-to-string conversions used in
folsomite
- Spec'd, tested and Dialyzed
- Well-defined metrics-to-binary conversions, as opposed to the
nearly-arbitrary term-to-string conversions used in
- More detailed stats:
- per-process. As much process ancestry is collected as possible, then
anonymous processes are aggregated to their youngest-known, named
predecessor - this aggregation keeps the useful breadcrumbs, while
reducing the number of unique names from exploding, which
avoids the associated problems:
- not very useful when there're lots of short-lived processes
- exploading disk space usage in Whisper
- per-ETS-table
- and more ... see records defined in
include
directory
- per-process. As much process ancestry is collected as possible, then
anonymous processes are aggregated to their youngest-known, named
predecessor - this aggregation keeps the useful breadcrumbs, while
reducing the number of unique names from exploding, which
avoids the associated problems:
For an example of using pre-process stats to track-down memory leaks, here's a screenshot of the SSL connection process memory usage growth drop after upgrade from 17.5 to 18.1 (back in 2015):
{env,
[ {production_interval , 30000}
, {consumers,
[ {beam_stats_consumer_statsd,
[ {consumption_interval , 60000}
, {dst_host , "localhost"}
, {dst_port , 8125}
, {src_port , 8124}
, {num_msgs_per_packet , 10}
% If you want to name your node something other than what
% erlang:node() returns:
, {static_node_name , <<"unicorn_at_rainbow">>}
]}
, {beam_stats_consumer_graphite,
[ {consumption_interval , 60000}
, {host , "localhost"}
, {port , 2003}
, {timeout , 5000}
]}
, {beam_stats_consumer_csv,
[ {consumption_interval , 60000}
, {path , "beam_stats.csv"}
]}
, {some_custom_consumer_module,
[ {some_custom_option_a, "abc"}
, {some_custom_option_b, 123}
]}
]}
]}
beam_stats_consumer:add(consumer_module, ConsumerOptions).
Not yet implemented.