Module: Tracelit::Metrics

Defined in:
lib/tracelit/metrics.rb

Class Method Summary collapse

Class Method Details

.counter(name, description: "", unit: "") ⇒ Object

Exposes a counter for manual instrumentation in user code:

Tracelit::Metrics.counter("orders.placed").add(1)



76
77
78
79
80
81
 
# File 'lib/tracelit/metrics.rb', line 76

def self.counter(name, description: "", unit: "")
  @meter&.create_counter(name,
    description: description,
    unit: unit
  )
end

.gauge(name, description: "", unit: "") ⇒ Object 



90
91
92
93
94
95
 
# File 'lib/tracelit/metrics.rb', line 90

def self.gauge(name, description: "", unit: "")
  @meter&.create_gauge(name,
    description: description,
    unit: unit
  )
end

.histogram(name, description: "", unit: "") ⇒ Object 



83
84
85
86
87
88
 
# File 'lib/tracelit/metrics.rb', line 83

def self.histogram(name, description: "", unit: "")
  @meter&.create_histogram(name,
    description: description,
    unit: unit
  )
end

.install_connection_pool_poller ⇒ Object

Polls ActiveRecord connection pool stats every 30 seconds on a daemon thread and records them as gauges. Does not require a live connection at install time — errors during polling are silently retried next cycle.

Fix 11: version-safe pool access that works on Rails 6.0–8.x.



235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
 
# File 'lib/tracelit/metrics.rb', line 235

def self.install_connection_pool_poller
  return if @connection_pool_poller_installed
  @connection_pool_poller_installed = true

  pool_size = @meter.create_gauge(
    "db.connection_pool.size",
    description: "Maximum connections in the pool",
    unit: "{connections}"
  )

  pool_busy = @meter.create_gauge(
    "db.connection_pool.busy",
    description: "Connections currently checked out",
    unit: "{connections}"
  )

  pool_idle = @meter.create_gauge(
    "db.connection_pool.idle",
    description: "Connections available for checkout",
    unit: "{connections}"
  )

  pool_waiting = @meter.create_gauge(
    "db.connection_pool.waiting",
    description: "Threads waiting for a connection",
    unit: "{threads}"
  )

  thread = Thread.new do
    Thread.current[:tracelit_pool_poller] = true
    loop do
      sleep 30
      begin
        # Fix 11a: Rails 7.2 soft-deprecated connection_pool on the base
        # class. Use the connection handler when available; fall back for
        # Rails 6.0 compatibility.
        pool = if ActiveRecord::Base.respond_to?(:connection_handler)
          ActiveRecord::Base.connection_handler
            .retrieve_connection_pool(ActiveRecord::Base.connection_specification_name) rescue
          ActiveRecord::Base.connection_pool
        else
          ActiveRecord::Base.connection_pool
        end

        next unless pool

        stat = pool.stat

        # Fix 11b: pool_config.db_config was added in Rails 6.1.
        # Fall back gracefully on older setups.
        adapter = if pool.respond_to?(:pool_config)
          pool.pool_config.db_config.adapter.to_s rescue "unknown"
        else
          "unknown"
        end

        attrs = { "db.system" => adapter }
        pool_size.record(stat[:size], attributes: attrs)
        pool_busy.record(stat[:busy], attributes: attrs)
        pool_idle.record(stat[:idle], attributes: attrs)
        pool_waiting.record(stat[:waiting], attributes: attrs)
      rescue StandardError
        # Pool may not be connected yet — retry next cycle
      end
    end
  end
  thread.abort_on_exception = false
  thread
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to install connection pool poller: #{e.message}")
end

.install_cpu_poller ⇒ Object

Polls process CPU utilisation every 30 seconds on a daemon thread. Computes a percentage by tracking the delta in CPU time (user + system) against wall-clock elapsed time — same approach as the Go and Node SDKs.

On Linux: reads /proc/self/stat (utime + stime in jiffies at 100 Hz). On macOS: reads ‘ps -o %cpu= -p <pid>` as a direct percentage.

Emits: process.runtime.cpu.usage (%) Attributes: process.pid, process.runtime



365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
 
# File 'lib/tracelit/metrics.rb', line 365

def self.install_cpu_poller
  return if @cpu_poller_installed
  @cpu_poller_installed = true

  cpu_gauge = @meter.create_gauge(
    "process.runtime.cpu.usage",
    description: "Process CPU utilisation percentage",
    unit: "%"
  )

  pid      = Process.pid
  linux    = File.exist?("/proc/self/stat")
  interval = 30 # seconds

  thread = Thread.new do
    Thread.current[:tracelit_cpu_poller] = true

    last_cpu_time  = read_cpu_time_s(pid, linux)
    last_wall_time = Process.clock_gettime(Process::CLOCK_MONOTONIC)

    loop do
      sleep interval
      begin
        now      = Process.clock_gettime(Process::CLOCK_MONOTONIC)
        elapsed  = now - last_wall_time
        cpu_time = read_cpu_time_s(pid, linux)

        next if elapsed <= 0 || cpu_time.nil? || last_cpu_time.nil?

        delta = cpu_time - last_cpu_time
        last_cpu_time  = cpu_time
        last_wall_time = now

        next if delta < 0

        pct = [[delta / elapsed * 100.0, 100.0].min, 0.0].max

        cpu_gauge.record(pct, attributes: {
          "process.pid"     => pid.to_s,
          "process.runtime" => "ruby",
        })
      rescue StandardError
        # Retry next cycle — never crash on a metric poll failure
      end
    end
  end
  thread.abort_on_exception = false
  thread
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to install CPU poller: #{e.message}")
end

.install_memory_poller ⇒ Object

Polls process RSS memory every 60 seconds on a daemon thread.

Fix 12: On Linux use /proc/self/status (always present, no subprocess). Fall back to ‘ps` on macOS/BSD. The previous implementation always used a shell backtick which spawns a child process and fails silently in minimal Docker containers that lack procps.



313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
 
# File 'lib/tracelit/metrics.rb', line 313

def self.install_memory_poller
  return if @memory_poller_installed
  @memory_poller_installed = true

  memory_gauge = @meter.create_gauge(
    "process.memory.rss",
    description: "Process resident set size (RSS)",
    unit: "MB"
  )

  pid = Process.pid

  thread = Thread.new do
    Thread.current[:tracelit_memory_poller] = true
    loop do
      sleep 60
      begin
        rss_kb = if File.exist?("/proc/self/status")
          # Linux: read VmRSS from /proc — no subprocess, always available
          File.read("/proc/self/status")[/VmRSS:\s+(\d+)/, 1].to_i
        else
          # macOS / BSD fallback
          `ps -o rss= -p #{Integer(pid)} 2>/dev/null`.strip.to_i
        end

        next if rss_kb == 0

        rss_mb = rss_kb / 1024.0
        memory_gauge.record(rss_mb, attributes: {
          "process.pid"     => pid.to_s,
          "process.runtime" => "ruby",
        })
      rescue StandardError
        # Ignore — environment may not support RSS polling
      end
    end
  end
  thread.abort_on_exception = false
  thread
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to install memory poller: #{e.message}")
end

.install_rails_subscriber ⇒ Object

Subscribes to Rails process_action.action_controller to emit:

http.server.request.count    — counter per request
http.server.request.duration — histogram in milliseconds
http.server.error.count      — counter for 5xx responses
db.query.duration            — histogram for ActiveRecord time per request

Fix 6: guarded against double-registration so reset! + re-setup in tests or Rails code-reloading scenarios does not duplicate metric counts.



105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
 
# File 'lib/tracelit/metrics.rb', line 105

def self.install_rails_subscriber
  return if @rails_subscriber_installed
  @rails_subscriber_installed = true

  request_counter = @meter.create_counter(
    "http.server.request.count",
    description: "Total HTTP requests processed",
    unit: "{requests}"
  )

  duration_histogram = @meter.create_histogram(
    "http.server.request.duration",
    description: "HTTP request duration",
    unit: "ms"
  )

  error_counter = @meter.create_counter(
    "http.server.error.count",
    description: "Total HTTP 5xx responses",
    unit: "{errors}"
  )

  db_duration_histogram = @meter.create_histogram(
    "db.query.duration",
    description: "Database query duration",
    unit: "ms"
  )

  ActiveSupport::Notifications.subscribe("process_action.action_controller") do |*args|
    event   = ActiveSupport::Notifications::Event.new(*args)
    payload = event.payload

    attrs = {
      # Fix 7: use controller#action (stable, low-cardinality route template)
      # instead of payload[:path] which contains raw IDs and causes metric
      # cardinality explosion on apps with resource IDs in URLs.
      "http.route"       => "#{payload[:controller]}##{payload[:action]}",
      "http.method"      => payload[:method].to_s,
      "http.status_code" => payload[:status].to_s,
      "controller"       => payload[:controller].to_s,
      "action"           => payload[:action].to_s,
    }

    request_counter.add(1, attributes: attrs)
    duration_histogram.record(event.duration, attributes: attrs)

    error_counter.add(1, attributes: attrs) if payload[:status].to_i >= 500

    if payload[:db_runtime]
      db_duration_histogram.record(
        payload[:db_runtime].to_f,
        attributes: { "controller" => payload[:controller].to_s }
      )
    end
  rescue StandardError
    # Never let metric errors surface to the application
  end
end

.install_sidekiq_middleware ⇒ Object

Installs a Sidekiq server middleware that emits per-job metrics. Uses a dynamically defined class so the instrument references are captured in the closure without global state.

Fix 6: guarded against double-registration.



169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
 
# File 'lib/tracelit/metrics.rb', line 169

def self.install_sidekiq_middleware
  return if @sidekiq_middleware_installed
  @sidekiq_middleware_installed = true

  job_counter = @meter.create_counter(
    "sidekiq.job.count",
    description: "Total Sidekiq jobs processed",
    unit: "{jobs}"
  )

  job_duration = @meter.create_histogram(
    "sidekiq.job.duration",
    description: "Sidekiq job execution duration",
    unit: "ms"
  )

  job_error_counter = @meter.create_counter(
    "sidekiq.job.error.count",
    description: "Total Sidekiq jobs that raised an error",
    unit: "{jobs}"
  )

  _job_counter       = job_counter
  _job_duration      = job_duration
  _job_error_counter = job_error_counter

  middleware_class = Class.new do
    define_method(:call) do |_worker, msg, queue, &block|
      start = Process.clock_gettime(Process::CLOCK_MONOTONIC)
      error_raised = false

      begin
        block.call
      rescue StandardError
        error_raised = true
        raise
      ensure
        elapsed_ms = (Process.clock_gettime(Process::CLOCK_MONOTONIC) - start) * 1000.0

        attrs = {
          "sidekiq.job.class" => msg["class"].to_s,
          "sidekiq.queue"     => queue.to_s,
          "sidekiq.status"    => error_raised ? "error" : "success",
        }

        _job_counter.add(1, attributes: attrs)
        _job_duration.record(elapsed_ms, attributes: attrs)
        _job_error_counter.add(1, attributes: attrs) if error_raised
      end
    end
  end

  Sidekiq.configure_server do |config|
    config.server_middleware do |chain|
      chain.add middleware_class
    end
  end
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to install Sidekiq middleware: #{e.message}")
end

.meter ⇒ Object 



56
57
58
 
# File 'lib/tracelit/metrics.rb', line 56

def self.meter
  @meter
end

.read_cpu_time_s(pid, linux) ⇒ Object

Returns cumulative CPU time (user + system) for this process in seconds. On Linux reads /proc/self/stat; on macOS/BSD falls back to ps %cpu which gives an instantaneous percentage instead (treated as fractional seconds over a 1-second window — good enough for a 30 s gauge).



421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
 
# File 'lib/tracelit/metrics.rb', line 421

def self.read_cpu_time_s(pid, linux)
  if linux
    stat = begin
      File.read("/proc/self/stat")
    rescue
      return nil
    end
    # Format: pid (comm) state ppid ... utime stime ...
    # comm can contain spaces — find last ')' and split from there.
    after_comm = stat[stat.rindex(")").to_i + 1..]
    return nil unless after_comm

    fields = after_comm.split
    # After ')': state(0) ppid(1) ... utime(11) stime(12)
    utime = fields[11]&.to_i
    stime = fields[12]&.to_i
    return nil unless utime && stime

    # Jiffies at 100 Hz → seconds
    (utime + stime) / 100.0
  else
    # macOS/BSD: `ps` gives current CPU % directly.
    # Return it as a fractional "seconds per second" proxy so the
    # delta calculation above yields the right percentage.
    out = `ps -o %cpu= -p #{Integer(pid)} 2>/dev/null`.strip
    return nil if out.empty?
    out.to_f / 100.0
  end
end

.restart_pollers(config) ⇒ Object

Fix 5 (support): Called from the Process._fork hook in Instrumentation to restart background polling threads inside each forked Puma/Unicorn worker. The parent-process threads are dead in the child; this revives them.



63
64
65
66
67
68
69
70
71
72
 
# File 'lib/tracelit/metrics.rb', line 63

def self.restart_pollers(config)
  @connection_pool_poller_installed = false
  @memory_poller_installed          = false
  @cpu_poller_installed             = false
  install_connection_pool_poller if defined?(::ActiveRecord)
  install_memory_poller
  install_cpu_poller
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to restart pollers after fork: #{e.message}")
end

.setup(config) ⇒ Object

Sets up the OpenTelemetry MeterProvider with OTLP exporter. Called once from Instrumentation.setup after trace setup.



16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
 
# File 'lib/tracelit/metrics.rb', line 16

def self.setup(config)
  exporter = OpenTelemetry::Exporter::OTLP::Metrics::MetricsExporter.new(
    endpoint: "#{config.endpoint}/v1/metrics",
    headers: {
      "Authorization"  => "Bearer #{config.api_key}",
      "X-Service-Name" => config.resolved_service_name,
      "X-Environment"  => config.environment,
    }
  )

  reader = OpenTelemetry::SDK::Metrics::Export::PeriodicMetricReader.new(
    exporter: exporter,
    export_interval_millis: 60_000,
    export_timeout_millis:  10_000
  )

  tp = OpenTelemetry.tracer_provider
  resource = tp.respond_to?(:resource) ? tp.resource : OpenTelemetry::SDK::Resources::Resource.create({})

  provider = OpenTelemetry::SDK::Metrics::MeterProvider.new(
    resource: resource
  )
  provider.add_metric_reader(reader)

  OpenTelemetry.meter_provider = provider

  @meter = provider.meter(
    config.resolved_service_name,
    version: Tracelit::VERSION
  )

  install_rails_subscriber       if defined?(::Rails)
  install_sidekiq_middleware      if defined?(::Sidekiq)
  install_connection_pool_poller  if defined?(::ActiveRecord)
  install_memory_poller
  install_cpu_poller
rescue StandardError => e
  OpenTelemetry.logger.warn("[Tracelit] failed to set up metrics: #{e.message}")
end