A hack of a delayed_job monitoring script for Munin

delayed_job

#!/usr/bin/env ruby

if ARGV[0] == "autoconf"
  puts "yes"
  exit
end

if ARGV[0] == "config"
puts <<EOT
graph_title delayed_jobs
graph_vlabel jobs
graph_category delayed_job
graph_info jobs completed
graph_args --base 1000 -l 0
EOT
end

require "time"

$dj_logs = {
  :rails_app_name => "/path/to/app/log/delayed_job.log"
}

$now = Time.now

$dj_logs.each do |log_name, log_path|
  File.open(log_path) do |f|
    jobs = {}

    while line = f.gets do
      if m = line.match(/^(\S+): \[.*?\] (\S+) (completed|failed) (.*?)$/)
        ts, job_name, status, rest = m.captures

        jobs[job_name] ||= {}
        jobs[job_name]["completed"] ||= 0
        jobs[job_name]["failed"] ||= 0
        jobs[job_name][:duration] ||= 0

	if ($now - Time.parse(ts)) < 300
          jobs[job_name][status] += 1

          if status == "completed"
            jobs[job_name][:duration] += rest.scan(/after (\S+)/).first.first.to_f 
          end
        end
      end 
    end

    jobs.each do |job_name, job_data|
      lnjn = "#{log_name.to_s}_#{job_name}"

      if ARGV[0] == "config"
        puts "#{lnjn}_completed.label #{log_name.to_s}:#{job_name} completed" 
        puts "#{lnjn}_completed.type GAUGE"
        puts "#{lnjn}_completed.draw LINE1"
        puts "#{lnjn}_completed.min 0"
        puts "#{lnjn}_failed.label #{log_name.to_s}:#{job_name} failed" 
        puts "#{lnjn}_failed.type GAUGE"
        puts "#{lnjn}_failed.draw LINE1"
        puts "#{lnjn}_failed.min 0"
      else
        puts "#{lnjn}_completed.value #{job_data['completed']}"
        puts "#{lnjn}_failed.value #{job_data['failed']}"

        if job_data[:duration] > 0 and job_data["completed"] > 0
          puts "#{lnjn}_duration.value #{job_data['completed'].to_f / job_data[:duration]}"
        else
          puts "#{lnjn}_duration.value 0"
        end
      end
    end
  end
end