Resque Stalled Alert : this script can check if a resque queue is overflowing or a job is "stuck"

resque_alert.rb

require 'date'
require 'redis'
require 'net/smtp'
require 'resque'
require 'resque/version'

class App
  VERSION = '0.0.1'

  def initialize

    # Development parameters
    redis_host = '127.0.0.1'
    redis_password = 'password'
    redis_port = 6379
    @redis_namespace = 'resque:app_development'
    @verbose = true               # set to false in production
    @veryverbose = true           # set to false in production and in staging
    @cycle = 3                    # check resque every @cycle seconds
    @alert_queue_length_threshold = 300 # over this threshold an alert will be sent
    @alert_job_stuck_threshold = 300    # over this threshold an alert will be sent, in seconds

    #email settings
    #Senders and Recipients
    @send_email = false            # set to true is you want to send alert emails
    @from_mail = '[email protected]'
    @to_mail = '[email protected]'
    @from_name = 'Resque Queue Monitoring Robot'
    @to_name = 'Monitoring team'

    #Servers and Authentication
    @smtp_host   = '127.0.0.1'
    @smtp_port   = 25
    @smtp_domain = 'email.com'
    @smtp_user   = '[email protected]'
    @smtp_pwd    = 'password'

    @redis = Redis.new(:host => redis_host, :port => redis_port, :password => redis_password, :thread_safe => true)

    Resque.redis = @redis
    Resque.redis.namespace = @redis_namespace

    @prev_stat = [0,0,0,0,0,0,0,0,0,0,0,0]
    @sent_job_alert = ''
  end

  def run
    puts "Start at #{DateTime.now}\n\n" if @verbose

    check_queue_length
    check_job_stuck

    puts "\nFinished at #{DateTime.now}" if @verbose

    sleep @cycle
  end

  protected

    def check_queue_length
      queues = Resque.queues
      stat = []
      index = 0
      queues.each do |q|
        size = Resque.size(q)
        prev_size = @prev_stat[index]
        stat << size
        puts (q + ' ; ' + size.to_s) if @verbose
        if prev_size < @alert_queue_length_threshold
          if size > @alert_queue_length_threshold
            alert_queue_length(q,size)
          end
        end
        index += 1
      end
      puts queues.to_s if @veryverbose
      @prev_stat = stat
      write_stat(queues, stat)
    end

    def write_stat(queues, stat)
      puts stat.join(", ") if @veryverbose
      log_file = "/tmp/log_file_#{@rails_env}_#{Date.today.to_s}"
      File.open(log_file, 'a') do |f|
        f.puts (DateTime.now.to_s + ', ' + queues.join(', ') + ', ' + stat.join(", "))
      end
    end

    def alert_queue_length(queue, nb_jobs)
      subj = "#{@rails_env} Warning Resque:#{queue} has #{nb_jobs} jobs"
      msg_body ="the resque #{@redis_namespace} queue #{queue} is #{nb_jobs} jobs long. (alerting over #{@alert_queue_length_threshold} jobs)"
      puts subj  if @verbose
      puts msg_body if @veryverbose
      email(subj, msg_body) if @send_email
      @sent_queue_alert = true
    end

    def check_job_stuck
      workers = Resque.workers

      for worker in (workers = workers.sort_by { |w| w.to_s })
        data = worker.processing || {}
        if data['queue']
          job = data['payload']['class'] + ':' + data['run_at'] + ':' + data[':args'].to_s
          max_tim = DateTime.strptime(data['run_at']).to_time + @alert_job_stuck_threshold
          alert_job_stuck(worker, job) if max_tim < Time.now && !@sent_job_alert.include?(job)
        end
        puts (worker.to_s + ' ; ' + job.to_s) if @verbose
      end
    end

    def alert_job_stuck(worker, job_title)
      subj = "#{@rails_env}  Warning: #{job_title} stuck on #{worker}"
      msg_body ="the job #{job_title} is running for more than #{@alert_job_stuck_threshold/60} minutes on on #{worker}"
      puts subj if @verbose
      puts msg_body if @veryverbose
      email(subj, msg_body) if @send_email
      @sent_job_alert << job_title
    end

    def email(subj, msg_body)
      #The subject and the message
      t = Time.now
      #The date/time should look something like: Thu, 03 Jan 2006 12:33:22 -0700
      msg_date = t.strftime("%a, %d %b %Y %H:%M:%S %z")

      #Compose the message for the email
msg = <<END_OF_MESSAGE
Date: #{@msg_date}
From: #{@from_name} <#{@from_mail}>
To: #{@to_name} <#{@to_mail}>
Subject: #{subj}

#{msg_body}
END_OF_MESSAGE

      Net::SMTP.start(@smtp_host, @smtp_port, @smtp_domain) do |smtp|
        smtp.send_message msg, @smtp_user, @to_mail
      end
    end

end

# Create and run the application
app = App.new
loop do
  app.run
end