Commit d453bb86 authored by Rémy Coutable's avatar Rémy Coutable

Merge branch '33410-unicorn_metrics' into 'master'

Instrument Unicorn with Ruby exporter

Closes #33410

See merge request !12115
parents 53c626bc 26ac691a
......@@ -285,6 +285,7 @@ group :metrics do
# Prometheus
gem 'prometheus-client-mmap', '~>0.7.0.beta5'
gem 'raindrops', '~> 0.18'
end
group :development do
......
......@@ -599,7 +599,7 @@ GEM
premailer-rails (1.9.7)
actionmailer (>= 3, < 6)
premailer (~> 1.7, >= 1.7.9)
prometheus-client-mmap (0.7.0.beta5)
prometheus-client-mmap (0.7.0.beta7)
mmap2 (~> 2.2.6)
pry (0.10.4)
coderay (~> 1.1.0)
......@@ -658,7 +658,7 @@ GEM
thor (>= 0.18.1, < 2.0)
rainbow (2.2.2)
rake
raindrops (0.17.0)
raindrops (0.18.0)
rake (10.5.0)
rblineprof (0.3.6)
debugger-ruby_core_source (~> 1.3)
......@@ -1062,6 +1062,7 @@ DEPENDENCIES
rails-deprecated_sanitizer (~> 1.0.3)
rails-i18n (~> 4.0.9)
rainbow (~> 2.2)
raindrops (~> 0.18)
rblineprof (~> 0.3.6)
rdoc (~> 4.2)
recaptcha (~> 3.0)
......
......@@ -543,6 +543,10 @@ production: &base
# enabled: true
# host: localhost
# port: 3808
prometheus:
# Time between sampling of unicorn socket metrics, in seconds
# unicorn_sampler_interval: 10
#
# 5. Extra customization
......
......@@ -494,6 +494,12 @@ Settings.webpack.dev_server['enabled'] ||= false
Settings.webpack.dev_server['host'] ||= 'localhost'
Settings.webpack.dev_server['port'] ||= 3808
#
# Prometheus metrics settings
#
Settings['prometheus'] ||= Settingslogic.new({})
Settings.prometheus['unicorn_sampler_interval'] ||= 10
#
# Testing settings
#
......
......@@ -119,6 +119,13 @@ def instrument_classes(instrumentation)
end
# rubocop:enable Metrics/AbcSize
Gitlab::Metrics::UnicornSampler.initialize_instance(Settings.prometheus.unicorn_sampler_interval).start
Gitlab::Application.configure do |config|
# 0 should be Sentry to catch errors in this middleware
config.middleware.insert(1, Gitlab::Metrics::ConnectionRackMiddleware)
end
if Gitlab::Metrics.enabled?
require 'pathname'
require 'influxdb'
......@@ -175,7 +182,7 @@ if Gitlab::Metrics.enabled?
GC::Profiler.enable
Gitlab::Metrics::Sampler.new.start
Gitlab::Metrics::InfluxSampler.initialize_instance.start
module TrackNewRedisConnections
def connect(*args)
......
require 'logger'
module Gitlab
module Metrics
class BaseSampler
def self.initialize_instance(*args)
raise "#{name} singleton instance already initialized" if @instance
@instance = new(*args)
at_exit(&@instance.method(:stop))
@instance
end
def self.instance
@instance
end
attr_reader :running
# interval - The sampling interval in seconds.
def initialize(interval)
interval_half = interval.to_f / 2
@interval = interval
@interval_steps = (-interval_half..interval_half).step(0.1).to_a
@mutex = Mutex.new
end
def enabled?
true
end
def start
return unless enabled?
@mutex.synchronize do
return if running
@running = true
@thread = Thread.new do
sleep(sleep_interval)
while running
safe_sample
sleep(sleep_interval)
end
end
end
end
def stop
@mutex.synchronize do
return unless running
@running = false
if @thread
@thread.wakeup if @thread.alive?
@thread.join
@thread = nil
end
end
end
def safe_sample
sample
rescue => e
Rails.logger.warn("#{self.class}: #{e}, stopping")
stop
end
def sample
raise NotImplementedError
end
# Returns the sleep interval with a random adjustment.
#
# The random adjustment is put in place to ensure we:
#
# 1. Don't generate samples at the exact same interval every time (thus
# potentially missing anything that happens in between samples).
# 2. Don't sample data at the same interval two times in a row.
def sleep_interval
while step = @interval_steps.sample
if step != @last_step
@last_step = step
return @interval + @last_step
end
end
end
end
end
end
module Gitlab
module Metrics
class ConnectionRackMiddleware
def initialize(app)
@app = app
end
def self.rack_request_count
@rack_request_count ||= Gitlab::Metrics.counter(:rack_request, 'Rack request count')
end
def self.rack_response_count
@rack_response_count ||= Gitlab::Metrics.counter(:rack_response, 'Rack response count')
end
def self.rack_uncaught_errors_count
@rack_uncaught_errors_count ||= Gitlab::Metrics.counter(:rack_uncaught_errors, 'Rack connections handling uncaught errors count')
end
def self.rack_execution_time
@rack_execution_time ||= Gitlab::Metrics.histogram(:rack_execution_time, 'Rack connection handling execution time',
{}, [0.05, 0.1, 0.25, 0.5, 0.7, 1, 1.5, 2, 2.5, 3, 5, 7, 10])
end
def call(env)
method = env['REQUEST_METHOD'].downcase
started = Time.now.to_f
begin
ConnectionRackMiddleware.rack_request_count.increment(method: method)
status, headers, body = @app.call(env)
ConnectionRackMiddleware.rack_response_count.increment(method: method, status: status)
[status, headers, body]
rescue
ConnectionRackMiddleware.rack_uncaught_errors_count.increment
raise
ensure
elapsed = Time.now.to_f - started
ConnectionRackMiddleware.rack_execution_time.observe({}, elapsed)
end
end
end
end
end
......@@ -5,13 +5,10 @@ module Gitlab
# This class is used to gather statistics that can't be directly associated
# with a transaction such as system memory usage, garbage collection
# statistics, etc.
class Sampler
class InfluxSampler < BaseSampler
# interval - The sampling interval in seconds.
def initialize(interval = Metrics.settings[:sample_interval])
interval_half = interval.to_f / 2
@interval = interval
@interval_steps = (-interval_half..interval_half).step(0.1).to_a
super(interval)
@last_step = nil
@metrics = []
......@@ -26,18 +23,6 @@ module Gitlab
end
end
def start
Thread.new do
Thread.current.abort_on_exception = true
loop do
sleep(sleep_interval)
sample
end
end
end
def sample
sample_memory_usage
sample_file_descriptors
......@@ -111,23 +96,6 @@ module Gitlab
def sidekiq?
Sidekiq.server?
end
# Returns the sleep interval with a random adjustment.
#
# The random adjustment is put in place to ensure we:
#
# 1. Don't generate samples at the exact same interval every time (thus
# potentially missing anything that happens in between samples).
# 2. Don't sample data at the same interval two times in a row.
def sleep_interval
while step = @interval_steps.sample
if step != @last_step
@last_step = step
return @interval + @last_step
end
end
end
end
end
end
......@@ -29,8 +29,8 @@ module Gitlab
provide_metric(name) || registry.summary(name, docstring, base_labels)
end
def gauge(name, docstring, base_labels = {})
provide_metric(name) || registry.gauge(name, docstring, base_labels)
def gauge(name, docstring, base_labels = {}, multiprocess_mode = :all)
provide_metric(name) || registry.gauge(name, docstring, base_labels, multiprocess_mode)
end
def histogram(name, docstring, base_labels = {}, buckets = ::Prometheus::Client::Histogram::DEFAULT_BUCKETS)
......
module Gitlab
module Metrics
class UnicornSampler < BaseSampler
def initialize(interval)
super(interval)
end
def unicorn_active_connections
@unicorn_active_connections ||= Gitlab::Metrics.gauge(:unicorn_active_connections, 'Unicorn active connections', {}, :max)
end
def unicorn_queued_connections
@unicorn_queued_connections ||= Gitlab::Metrics.gauge(:unicorn_queued_connections, 'Unicorn queued connections', {}, :max)
end
def enabled?
# Raindrops::Linux.tcp_listener_stats is only present on Linux
unicorn_with_listeners? && Raindrops::Linux.respond_to?(:tcp_listener_stats)
end
def sample
Raindrops::Linux.tcp_listener_stats(tcp_listeners).each do |addr, stats|
unicorn_active_connections.set({ type: 'tcp', address: addr }, stats.active)
unicorn_queued_connections.set({ type: 'tcp', address: addr }, stats.queued)
end
Raindrops::Linux.unix_listener_stats(unix_listeners).each do |addr, stats|
unicorn_active_connections.set({ type: 'unix', address: addr }, stats.active)
unicorn_queued_connections.set({ type: 'unix', address: addr }, stats.queued)
end
end
private
def tcp_listeners
@tcp_listeners ||= Unicorn.listener_names.grep(%r{\A[^/]+:\d+\z})
end
def unix_listeners
@unix_listeners ||= Unicorn.listener_names - tcp_listeners
end
def unicorn_with_listeners?
defined?(Unicorn) && Unicorn.listener_names.any?
end
end
end
end
require 'spec_helper'
require_relative '../../config/initializers/8_metrics'
describe 'instrument_classes', lib: true do
let(:config) { double(:config) }
let(:unicorn_sampler) { double(:unicorn_sampler) }
let(:influx_sampler) { double(:influx_sampler) }
before do
allow(config).to receive(:instrument_method)
allow(config).to receive(:instrument_methods)
allow(config).to receive(:instrument_instance_method)
allow(config).to receive(:instrument_instance_methods)
allow(Gitlab::Metrics::UnicornSampler).to receive(:initialize_instance).and_return(unicorn_sampler)
allow(Gitlab::Metrics::InfluxSampler).to receive(:initialize_instance).and_return(influx_sampler)
allow(unicorn_sampler).to receive(:start)
allow(influx_sampler).to receive(:start)
allow(Gitlab::Application).to receive(:configure)
end
it 'can autoload and instrument all files' do
require_relative '../../config/initializers/8_metrics'
expect { instrument_classes(config) }.not_to raise_error
end
end
require 'spec_helper'
describe Gitlab::Metrics::ConnectionRackMiddleware do
let(:app) { double('app') }
subject { described_class.new(app) }
around do |example|
Timecop.freeze { example.run }
end
describe '#call' do
let(:status) { 100 }
let(:env) { { 'REQUEST_METHOD' => 'GET' } }
let(:stack_result) { [status, {}, 'body'] }
before do
allow(app).to receive(:call).and_return(stack_result)
end
context '@app.call succeeds with 200' do
before do
allow(app).to receive(:call).and_return([200, nil, nil])
end
it 'increments response count with status label' do
expect(described_class).to receive_message_chain(:rack_response_count, :increment).with(include(status: 200, method: 'get'))
subject.call(env)
end
it 'increments requests count' do
expect(described_class).to receive_message_chain(:rack_request_count, :increment).with(method: 'get')
subject.call(env)
end
it 'measures execution time' do
execution_time = 10
allow(app).to receive(:call) do |*args|
Timecop.freeze(execution_time.seconds)
end
expect(described_class).to receive_message_chain(:rack_execution_time, :observe).with({}, execution_time)
subject.call(env)
end
end
context '@app.call throws exception' do
let(:rack_response_count) { double('rack_response_count') }
before do
allow(app).to receive(:call).and_raise(StandardError)
allow(described_class).to receive(:rack_response_count).and_return(rack_response_count)
end
it 'increments exceptions count' do
expect(described_class).to receive_message_chain(:rack_uncaught_errors_count, :increment)
expect { subject.call(env) }.to raise_error(StandardError)
end
it 'increments requests count' do
expect(described_class).to receive_message_chain(:rack_request_count, :increment).with(method: 'get')
expect { subject.call(env) }.to raise_error(StandardError)
end
it "does't increment response count" do
expect(described_class.rack_response_count).not_to receive(:increment)
expect { subject.call(env) }.to raise_error(StandardError)
end
it 'measures execution time' do
execution_time = 10
allow(app).to receive(:call) do |*args|
Timecop.freeze(execution_time.seconds)
raise StandardError
end
expect(described_class).to receive_message_chain(:rack_execution_time, :observe).with({}, execution_time)
expect { subject.call(env) }.to raise_error(StandardError)
end
end
end
end
require 'spec_helper'
describe Gitlab::Metrics::Sampler do
describe Gitlab::Metrics::InfluxSampler do
let(:sampler) { described_class.new(5) }
after do
......@@ -8,10 +8,10 @@ describe Gitlab::Metrics::Sampler do
end
describe '#start' do
it 'gathers a sample at a given interval' do
expect(sampler).to receive(:sleep).with(a_kind_of(Numeric))
expect(sampler).to receive(:sample)
expect(sampler).to receive(:loop).and_yield
it 'runs once and gathers a sample at a given interval' do
expect(sampler).to receive(:sleep).with(a_kind_of(Numeric)).twice
expect(sampler).to receive(:sample).once
expect(sampler).to receive(:running).and_return(false, true, false)
sampler.start.join
end
......
require 'spec_helper'
describe Gitlab::Metrics::UnicornSampler do
subject { described_class.new(1.second) }
describe '#sample' do
let(:unicorn) { double('unicorn') }
let(:raindrops) { double('raindrops') }
let(:stats) { double('stats') }
before do
stub_const('Unicorn', unicorn)
stub_const('Raindrops::Linux', raindrops)
allow(raindrops).to receive(:unix_listener_stats).and_return({})
allow(raindrops).to receive(:tcp_listener_stats).and_return({})
end
context 'unicorn listens on unix sockets' do
let(:socket_address) { '/some/sock' }
let(:sockets) { [socket_address] }
before do
allow(unicorn).to receive(:listener_names).and_return(sockets)
end
it 'samples socket data' do
expect(raindrops).to receive(:unix_listener_stats).with(sockets)
subject.sample
end
context 'stats collected' do
before do
allow(stats).to receive(:active).and_return('active')
allow(stats).to receive(:queued).and_return('queued')
allow(raindrops).to receive(:unix_listener_stats).and_return({ socket_address => stats })
end
it 'updates metrics type unix and with addr' do
labels = { type: 'unix', address: socket_address }
expect(subject).to receive_message_chain(:unicorn_active_connections, :set).with(labels, 'active')
expect(subject).to receive_message_chain(:unicorn_queued_connections, :set).with(labels, 'queued')
subject.sample
end
end
end
context 'unicorn listens on tcp sockets' do
let(:tcp_socket_address) { '0.0.0.0:8080' }
let(:tcp_sockets) { [tcp_socket_address] }
before do
allow(unicorn).to receive(:listener_names).and_return(tcp_sockets)
end
it 'samples socket data' do
expect(raindrops).to receive(:tcp_listener_stats).with(tcp_sockets)
subject.sample
end
context 'stats collected' do
before do
allow(stats).to receive(:active).and_return('active')
allow(stats).to receive(:queued).and_return('queued')
allow(raindrops).to receive(:tcp_listener_stats).and_return({ tcp_socket_address => stats })
end
it 'updates metrics type unix and with addr' do
labels = { type: 'tcp', address: tcp_socket_address }
expect(subject).to receive_message_chain(:unicorn_active_connections, :set).with(labels, 'active')
expect(subject).to receive_message_chain(:unicorn_queued_connections, :set).with(labels, 'queued')
subject.sample
end
end
end
end
describe '#start' do
context 'when enabled' do
before do
allow(subject).to receive(:enabled?).and_return(true)
end
it 'creates new thread' do
expect(Thread).to receive(:new)
subject.start
end
end
context 'when disabled' do
before do
allow(subject).to receive(:enabled?).and_return(false)
end
it "doesn't create new thread" do
expect(Thread).not_to receive(:new)
subject.start
end
end
end
end
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment