prom_multi_proc_rb

Ruby client library for collecting Prometheus metrics in multi-process applications. Designed for forking servers (Unicorn, Puma). Writes metrics as JSON to a Unix socket listened to by prom_multi_proc.

Add this line to your application's Gemfile:

gem 'prom_multi_proc_rb', require: 'prom_multi_proc'

And then execute:

$ bundle

Or install it yourself as:

$ gem install prom_multi_proc_rb

Create a JSON file to define the Prometheus metrics your application will track:

[
    {
        "type": "counter",
        "name": "app_requests_total",
        "help": "Total HTTP requests",
        "labels": ["method", "status"]
    },
    {
        "type": "gauge",
        "name": "app_workers_active",
        "help": "Number of active workers"
    },
    {
        "type": "histogram",
        "name": "app_request_duration_seconds",
        "help": "Request duration in seconds",
        "labels": ["method"]
    },
    {
        "type": "summary",
        "name": "app_response_size_bytes",
        "help": "Response size in bytes",
        "labels": ["method"]
    }
]

This file is shared by both the aggregator process and the Ruby client. Metric names in the JSON file may or may not include the prefix — the client applies it automatically if it is missing.

Download, install, and start prom_multi_proc using the metrics JSON definition file. Note the socket path.

The Ruby client functions normally if no aggregator is listening on the socket — failed writes are logged and silently dropped so the host application is never interrupted.

Create a PromMultiProc::Base object and begin recording metrics:

metrics = PromMultiProc::Base.new(
  prefix:     "app_",
  socket:     "/var/run/prom_multi_proc/metrics.sock",
  metrics:    "/etc/myapp/metrics.json",
  batch_size: 10,
  validate:   true
)

metrics.requests_total.inc(method: "GET", status: "200")
metrics.request_duration_seconds.observe(0.042, method: "GET")

The prefix: option works the same way as the -metric-prefix flag on the aggregator server: if a metric name in the JSON spec already starts with the prefix, it is used as-is; otherwise the prefix is prepended. This means you can use either full names or short names in your spec file.

# Both of these result in wire name "app_requests_total":
# spec: {"name": "app_requests_total", ...}  → already has prefix, used as-is
# spec: {"name": "requests_total", ...}       → prefix applied → "app_requests_total"

Use #multi to record multiple metrics in a single atomic socket write:

metrics.multi do |m|
  m.requests_total.inc(method: "POST", status: "201")
  m.workers_active.set(8)
  m.request_duration_seconds.observe(0.015, method: "POST")
end

A Rails initializer helper is provided at PromMultiProc::Rails.init. Create config/initializers/prom_multi_proc.rb:

$prom = PromMultiProc::Rails.init

$prom.requests_total.inc(method: "GET", status: "200")
$prom.request_duration_seconds.observe(0.042, method: "GET")

Rails.init derives sensible defaults from the Rails environment (batch size, validate, socket path, etc.) and accepts keyword arguments to override any of them:

# Override prefix and bump batch size in production
$prom = PromMultiProc::Rails.init(prefix: "myapp_", batch_size: 50)

# Point at a non-default socket
$prom = PromMultiProc::Rails.init(socket: "/run/metrics/custom.sock")

The PROM_MULTI_PROC_SOCKET and PROM_MULTI_PROC_DEFINITION_FILE environment variables are still respected as fallback defaults for socket: and metrics:, but any value passed directly as a keyword argument takes precedence.

Each flush sends a JSON array over the Unix socket and then closes the connection:

[
  {"name":"app_requests_total","method":"inc","value":1.0,"label_values":["GET","200"]},
  {"name":"app_workers_active","method":"set","value":8.0,"label_values":[]}
]

Bug reports and pull requests are welcome on GitHub at https://github.com/atongen/prom_multi_proc_rb.

Available as open source under the MIT License.