| 1 | require 'statsd-ruby' |
| 2 | require 'docker' |
| 3 | |
| 4 | $statsd = Statsd.new '10.0.0.1', 8125 |
| 5 | |
| 6 | module Docker |
| 7 | class CloseConnectionError < StandardError; end |
| 8 | class Container |
| 9 | def name |
| 10 | info["Names"].first[1..-1] |
| 11 | end |
| 12 | |
| 13 | def stats |
| 14 | path = path_for(:stats) |
| 15 | |
| 16 | result = nil |
| 17 | |
| 18 | streamer = lambda do |chunk, remaining, total| |
| 19 | result ||= chunk |
| 20 | raise CloseConnectionError if result |
| 21 | end |
| 22 | options = { response_block: streamer }.merge(connection.options) |
| 23 | |
| 24 | Excon.get(connection.url + path[1..-1], options) rescue CloseConnectionError |
| 25 | |
| 26 | Docker::Util.parse_json(result) |
| 27 | end |
| 28 | end |
| 29 | end |
| 30 | |
| 31 | def median(array) |
| 32 | sorted = array.sort |
| 33 | len = sorted.length |
| 34 | return ((sorted[(len - 1) / 2] + sorted[len / 2]) / 2.0).to_i |
| 35 | end |
| 36 | |
| 37 | |
| 38 | def analyze_container(container) |
| 39 | |
| 40 | data = container.exec(["ps", "-eo", "rss,args"])[0].join("\n").split("\n") |
| 41 | unicorns = data.grep(/unicorn/).map(&:to_i) |
| 42 | sidekiqs = data.grep(/sidekiq/).map(&:to_i) |
| 43 | |
| 44 | result = {} |
| 45 | |
| 46 | if unicorns.length > 0 |
| 47 | result["unicorn.max_rss"] = unicorns.max |
| 48 | result["unicorn.median_rss"] = median(unicorns) |
| 49 | end |
| 50 | |
| 51 | if sidekiqs.length > 0 |
| 52 | result["sidekiq.max_rss"] = sidekiqs.max |
| 53 | result["sidekiq.median_rss"] = median(sidekiqs) |
| 54 | end |
| 55 | result["total_mem_usage"] = container.stats["memory_stats"]["usage"] |
| 56 | |
| 57 | @prev_stats ||= {} |
| 58 | prev_stats = @prev_stats[container.name] |
| 59 | @prev_stats[container.name] = stats = container.stats |
| 60 | |
| 61 | if prev_stats |
| 62 | cpu_delta = stats["cpu_stats"]["system_cpu_usage"] - prev_stats["cpu_stats"]["system_cpu_usage"] |
| 63 | app_cpu_delta = stats["cpu_stats"]["cpu_usage"]["total_usage"] - prev_stats["cpu_stats"]["cpu_usage"]["total_usage"] |
| 64 | |
| 65 | result["cpu_usage"] = (app_cpu_delta.to_f / cpu_delta.to_f) * stats["cpu_stats"]["cpu_usage"]["percpu_usage"].length * 100.0 |
| 66 | end |
| 67 | |
| 68 | result |
| 69 | |
| 70 | end |
| 71 | |
| 72 | def containers |
| 73 | Docker::Container.all |
| 74 | end |
| 75 | |
| 76 | hostname = Docker.info["Name"] |
| 77 | |
| 78 | STDERR.puts "#{Time.now} Starting Monitor" |
| 79 | |
| 80 | while true |
| 81 | |
| 82 | begin |
| 83 | containers.each do |c| |
| 84 | |
| 85 | analyze_container(c).each do |k, v| |
| 86 | if v && v > 0 |
| 87 | $statsd.gauge "#{hostname}.#{c.name}.#{k}", v |
| 88 | end |
| 89 | end |
| 90 | end |
| 91 | rescue => e |
| 92 | STDERR.puts e |
| 93 | STDERR.puts e.backtrace |
| 94 | end |
| 95 | |
| 96 | sleep 60 |
| 97 | end |
| 98 | |