Prometheus.erl

Prometheus.io monitoring system and time series database client in Erlang.

Metrics

Standard

• [x] Counter
• [x] Gauge
• [x] Summary
• [x] Histogram

Custom Collectors

• erlang_vm_memory_collector - Collects information about Erlang VM memory usage mainly using erlang:memory/0
• erlang_vm_statistics_collector - Collects Erlang VM statistics using erlang:statistics/1

You can write custom collector/exporter for any library/app you’d like. For example here is Queue info collector from RabbitMQ Exporter.

Integrations / Collectors / Instrumenters

• Ecto collector
• Elixir client
• Plugs Instrumenter/Exporter
• Elli middleware
• Fuse plugin
• Phoenix instrumenter
• Process Info Collector
• RabbitMQ Exporter

Example Console Session

Run shell with compiled and loaded app:

$ rebar3 shell

Start prometheus app:

prometheus:start().

Register metrics:

prometheus_gauge:new([{name, pool_size}, {help, "MongoDB Connections pool size"}]),
prometheus_counter:new([{name, http_requests_total}, {help, "Http request count"}]).
prometheus_summary:new([{name, orders}, {help, "Track orders count/total sum"}]).
prometheus_histogram:new([{name, http_request_duration_milliseconds},
                               {labels, [method]},
                               {bounds, [100, 300, 500, 750, 1000]},
                               {help, "Http Request execution time"}]).

Use metrics:

prometheus_gauge:set(pool_size, 365),
prometheus_counter:inc(http_requests_total).
prometheus_summary:observe(orders, 10).
prometheus_summary:observe(orders, 15).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 95).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 100).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 102).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 150).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 250).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 75).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 350).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 550).
prometheus_histogram:observe(http_request_duration_milliseconds, [get], 950).
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 500),
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 150).
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 450).
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 850).
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 750).
prometheus_histogram:observe(http_request_duration_milliseconds, [post], 1650).

Export metrics as text:

io:format(prometheus_text_format:format()).

->

# TYPE http_requests_total counter
# HELP http_requests_total Http request count
http_requests_total 2
# TYPE pool_size gauge
# HELP pool_size MongoDB Connections pool size
pool_size 365
# TYPE orders summary
# HELP orders Track orders count/total sum
orders_count 4
orders_sum 50
# TYPE http_request_duration_milliseconds histogram
# HELP http_request_duration_milliseconds Http Request execution time
http_request_duration_milliseconds_bucket{method="post",le="100"} 0
http_request_duration_milliseconds_bucket{method="post",le="300"} 1
http_request_duration_milliseconds_bucket{method="post",le="500"} 3
http_request_duration_milliseconds_bucket{method="post",le="750"} 4
http_request_duration_milliseconds_bucket{method="post",le="1000"} 5
http_request_duration_milliseconds_bucket{method="post",le="+Inf"} 6
http_request_duration_milliseconds_count{method="post"} 6
http_request_duration_milliseconds_sum{method="post"} 4350
http_request_duration_milliseconds_bucket{method="get",le="100"} 3
http_request_duration_milliseconds_bucket{method="get",le="300"} 6
http_request_duration_milliseconds_bucket{method="get",le="500"} 7
http_request_duration_milliseconds_bucket{method="get",le="750"} 8
http_request_duration_milliseconds_bucket{method="get",le="1000"} 9
http_request_duration_milliseconds_bucket{method="get",le="+Inf"} 9
http_request_duration_milliseconds_count{method="get"} 9
http_request_duration_milliseconds_sum{method="get"} 2622

Bucket generators

prometheus_buckets:linear(-15, 5, 6) produces [-15, -10, -5, 0, 5, 10]

prometheus_buckets:exponential(100, 1.2, 3) produces [100, 120, 144]

Implementation Note

Prometheus.erl exports two API sets.

• For Integers: prometheus_coutner:inc, prometheus_summary:observe and prometheus_histogram:observe. Implementation is based on ets:update_counter. While this is expected to be much faster than using processes for synchronization it restricts us to integers-only while Prometheus expects series values to be double. ETS-based metrics are optimistic - for basic metrics such as counters/gauges it first tries to increment and iff series doesn’t exist it queries ETS to check if metric actually registered and if so it inserts series. For histograms at least one lookup is required - we need buckets to compute bucket counter position.
• For Floats: prometheus_coutner:dinc, prometheus_summary:dobserve and prometheus_histogram:dobserve. Implementation is based on gen_server which is used for synchronizations (ets doesn’t support float atomic increments).

NOTE: you can use float APIs after integer but not vice-versa.

Configuration

Prometheus.erl supports standard Erlang app configuration.

• default_collectors - List of custom collectors modules to be registered automatically. If undefined list of all modules implementing prometheus_collector behaviour will be used.
• default_metrics - List of metrics to be registered during app startup. Metric format: {Registry, Metric, Spec} where Registry is registry name, Metric is metric type (prometheus_counter, prometheus_gauge … etc), Spec is a list to be passed to Metric:register/2.

Collectors & Exporters Conventions

Configuration

All 3d-party libraries should be configured via prometheus app env.

Exproters are responsible for maintianing scrape endpoint. Exporters usually tightly coupled with web server and are singletons. They should understand these keys:

• path - url for scraping;
• format - scrape format as module name i.e. prometheus_text_format or prometheus_protobuf_format. Exporter-specific options should be under <exporter_name>_exporter for erlang or <Exporter_name>Exporter for Elixir i.e. PlugsExporter or elli_exporter

Collectors collect integration specific metrics i.e. ecto timings, process informations and so on. Their configuration should be under <collector_name>_collectorfor erlang or <Collector_name>Collector for Elixir i.e. process_collector, EctoCollector and so on.

Naming

For Erlang: prometheus_<name>_collector/prometheus_<name>_exporter.

For Elixir: Prometheus.Collectors.<name>/Prometheus.Exporters.<name>.

TODO

• [x] Floats support
• [x] Tests
• [x] Bucket generators
• [x] Protobuf format
• [ ] Full summary implementation
• [ ] Extend custom collectors collection?

Build

$ rebar3 compile

Contributing

Types -> Macros -> Callbacks -> Public API -> Deprecations -> Private Parts

install git precommit hook:

./bin/pre-commit.sh install

Pre-commit check can be skipped passing --no-verify option to git commit.

License

MIT