Fix pid providing for Prometheus

Use relative worker identifier for metrics (instead of Process.pid) and
identify when Unicorn/Puma/Sidekiq is used.
Previously, it was assumed that all metrics are gathered from Unicorn
due to hardcoded implementation which was incorrect.
This commit is contained in:
Aleksei Lipniagov 2019-07-18 13:54:11 +00:00 committed by Kamil Trzciński
parent 91903d3a9e
commit 22e2917b18
3 changed files with 122 additions and 2 deletions

View file

@ -1,5 +1,4 @@
require 'prometheus/client'
require 'prometheus/client/support/unicorn'
# Keep separate directories for separate processes
def prometheus_default_multiproc_dir
@ -23,7 +22,7 @@ Prometheus::Client.configure do |config|
config.multiprocess_files_dir = ENV['prometheus_multiproc_dir'] || prometheus_default_multiproc_dir
config.pid_provider = Prometheus::Client::Support::Unicorn.method(:worker_pid_provider)
config.pid_provider = Prometheus::PidProvider.method(:worker_id)
end
Gitlab::Application.configure do |config|

View file

@ -0,0 +1,42 @@
# frozen_string_literal: true
require 'prometheus/client/support/unicorn'
module Prometheus
module PidProvider
extend self
def worker_id
if Sidekiq.server?
'sidekiq'
elsif defined?(Unicorn::Worker)
"unicorn_#{unicorn_worker_id}"
elsif defined?(::Puma)
"puma_#{puma_worker_id}"
else
"process_#{Process.pid}"
end
end
private
# This is not fully accurate as we don't really know if the nil returned
# is actually means we're on master or not.
# Follow up issue was created to address this problem and
# to introduce more structrured approach to a current process discovery:
# https://gitlab.com/gitlab-org/gitlab-ce/issues/64740
def unicorn_worker_id
::Prometheus::Client::Support::Unicorn.worker_id || 'master'
end
# See the comment for #unicorn_worker_id
def puma_worker_id
match = process_name.match(/cluster worker ([0-9]+):/)
match ? match[1] : 'master'
end
def process_name
$0
end
end
end

View file

@ -0,0 +1,79 @@
# frozen_string_literal: true
require 'fast_spec_helper'
describe Prometheus::PidProvider do
describe '.worker_id' do
subject { described_class.worker_id }
let(:sidekiq_module) { Module.new }
before do
allow(sidekiq_module).to receive(:server?).and_return(false)
stub_const('Sidekiq', sidekiq_module)
end
context 'when running in Sidekiq server mode' do
before do
expect(Sidekiq).to receive(:server?).and_return(true)
end
it { is_expected.to eq 'sidekiq' }
end
context 'when running in Unicorn mode' do
before do
stub_const('Unicorn::Worker', Class.new)
hide_const('Puma')
end
context 'when `Prometheus::Client::Support::Unicorn` provides worker_id' do
before do
expect(::Prometheus::Client::Support::Unicorn).to receive(:worker_id).and_return(1)
end
it { is_expected.to eq 'unicorn_1' }
end
context 'when no worker_id is provided from `Prometheus::Client::Support::Unicorn`' do
before do
expect(::Prometheus::Client::Support::Unicorn).to receive(:worker_id).and_return(nil)
end
it { is_expected.to eq 'unicorn_master' }
end
end
context 'when running in Puma mode' do
before do
stub_const('Puma', Module.new)
hide_const('Unicorn::Worker')
end
context 'when cluster worker id is specified in process name' do
before do
expect(described_class).to receive(:process_name).and_return('puma: cluster worker 1: 17483 [gitlab-puma-worker]')
end
it { is_expected.to eq 'puma_1' }
end
context 'when no worker id is specified in process name' do
before do
expect(described_class).to receive(:process_name).and_return('bin/puma')
end
it { is_expected.to eq 'puma_master' }
end
end
context 'when running in unknown mode' do
before do
hide_const('Puma')
hide_const('Unicorn::Worker')
end
it { is_expected.to eq "process_#{Process.pid}" }
end
end
end