gistfile1.rb

#!/usr/bin/env ruby

require 'rubygems'
require 'logger'
require 'net/ssh'
require 'net/smtp'
require 'patron'
require 'timeout'
require 'parseconfig'

##Globals##
config = ParseConfig.new('/etc/artemis.conf')  #config file location

$log_file = config.get_value('log_location') #loggin info
$log = Logger.new('log_file', 'monthly')

#email settings
$email_svr = config.get_value('email_server') #smtp server info
$email_from = config.get_value('email_from') #email address in which email is sent from
$email_to = config.get_value('email_to') #email address where the errors will be sent

svr = config.get_value('servers') #get servers
$svrs = svr.to_s.split(' ') #split the server string into an array

$www_doc = config.get_value('check_html') #get html/php test file
$jsp_doc = config.get_value('check_java') #get jstp test file

www = config.get_value('sites') #parseconf only passes strings
$www_sites = www.to_s.split(' ') #split the www string into an array

jsp = config.get_value('java_sites') #get jsp/tomcat sites
$jsp_sites = jsp.to_s.split(' ') #split jsp sites the string into an array

smtp = config.get_value('smtp_servers') #get smtp servers
$smtp_svr = smtp.to_ssplit(' ') #split smtp server string into an array

$smtp_log = config.get_value('smtp_logs') #get location of smtp logs

class Server
  def initialize(servers = $svrs)
    @servers = servers
  end
end

##Methods##

#check server mount points
def mnt_chk
  server = nil
  $svrs.each do |server|
    Timeout::timeout(20) do
      Net::SSH.start("#{server}", 'tester', :password => "!@retset") do |ssh|
        uname = ssh.exec!("uname")
        $log.info("Checking mounts system #{server}")

        if uname == nil
          $log.error("Cannot connect to #{server}")
          puts(email_prg "Cannot connect to #{server}")
        else
          mntchk = ssh.exec!("mount | awk '{print $5}' | grep nfs | wc | awk '{print $1}'")
          fstabchk = ssh.exec!("cat /etc/fstab | awk '{print $3}' | grep nfs | wc | awk '{print $1}'")
          if mntchk != fstabchk
            $log.error("A mount on #{server} is down.")
            puts(email_prg "A mount on #{server} is down.")
          end
        end
      end
    end
  end
rescue Timeout::Error
  $log.error("Timeout Error: Check network connection and/or NFS mounts on #{server}")
  puts(email_prg "Timeout Error: Check network connection and/or NFS mounts on #{server}")
end

#check syslog for errors
def syslg_chk
  server = nil
  $svrs.each do |server|
    Timeout::timeout(15) do
      Net::SSH.start("#{server}", 'tester', :password => "!@retset") do |ssh|
        uname = ssh.exec!("uname")
        $log.info("Checking logs on #{server}")
        syserr = ssh.exec!("egrep '(reject|warning|error|fatal|panic|err)' #{$smtp_log}")

        #checking smtp logs
        if $smtp_svr != nil
          mailerr = ssh.exec!("egrep '(reject|warning|error|fatal|panic|err)' #{$smtp_log}")
          if mailerr != nil
            $log.error("#{mailerr} on #{$mntsvr}")
            puts(email_prg "#{mailerr} on #{$mntsvr}")
          end
        end
      end
    end
  end
rescue Timeout::Error
  $log.error("Timeout out while checking logs on #{server}")
  puts(email_prg "Timeout out while checking logs on #{server}")
end

#checking websites
def web_chk
  if $www_sites and $www_doc != nil
    $www_sites.each do |site|
      $log.info "Checking site #{site}"
      sess = Patron::Session.new
      sess.timeout = 1
      sess.base_url = "#{site}"
      resp = sess.get($www_doc)

      if resp.status > 400
        $log.error("#{site} is down.")
        $log.error(resp.body)
      else
        $log.info("#{site} is up.")
      end
    end
  end

  if $jsp_sites and $jsp_doc != nil
        $jsp_sites.each do |site|
      $log.info "Checking site #{site}"
      sess = Patron::Session.new
      sess.timeout = 6
      sess.base_url = "#{site}"
      resp = sess.get($jsp_doc)

      if resp.status > 400
        $log.error("#{site} is down:")
        $log.error(resp.body)
      else
        $log.info("#{site} is up.")
      end
    end
  end
end

#email program
def email_prg(msgstr)
  Net::SMTP.start($email_svr, 25) do |smtp|
    smtp.send_message msgstr, '$email_from', '$email_to'
  end
end
  

##Daemon##
  loop do

    if $svrs != nil 
      mnt_chk
      syslg_chk
    end

    if $www_sites or $jsp_sites != nil
      web_chk
    end

    sleep 120
  end

gistfile2.rb

#!/usr/bin/env ruby

require 'net/ssh'

class Server

  attr_accessor :server

  def initialize(server_name)
    @server = server_name
  end


  def mnt_chk(ssh)

    uname = ssh.exec!("uname")

    if uname.nil?
      return "Failed to connect to #{@server} during mount check."
    else
      mntchk = ssh.exec!("mount | awk '{print $5}' | grep nfs | wc | awk '{print $1}'")
      fstabchk = ssh.exec!("cat /etc/fstab | awk '{print $3}' | grep nfs | wc | awk '{print $1}'")

      if mntchk == fstabchk
        true
      else
        return "A mount on #{@server} is down or hung."
      end
    end
  end
  

  def syslg_chk(ssh)

    uname = ssh.exec!("uname")

    if uname.nil?
      return "Failed to connect to #{@server} during syslog check."
    else
      syserr = ssh.exec!("egrep '(reject|warning|error|fatal|panic|err)' #{$smtp_log}")

      if syserr.nil?
        true
      else
        return syserr
      end
    end
  end


  def smtp_chk(ssh)

    uname = ssh.exec!("uname")

    if uname.nil?
      return "Failed to connect to #{@server} during smtp check."
    else
      mailerr = ssh.exec!("egrep '(reject|warning|error|fatal|panic|err)' #{$smtp_log}")

      if mailerr.nil?
        true
      else
        return mailerr
      end
    end
  end

end

gistfile3.rb

#!/usr/bin/env ruby

class Site

  attr_accessor :html_site, :jsp_site

  def initialize(html_site, jsp_site)
    @html_site = html_site
    @jsp_site = jsp_site
  end

  def www_site

    sess = Patron::Session.new
    sess.timeout = 1
    sess.base_url = @html_site
    resp = sess.get($www_doc)

    if resp.status > 400
      $log.error("#{site} is down.")
      $log.error(resp.body)
    else
      $log.info("#{site} is up.")
    end
  end

  def java_site

    sess = Patron::Session.new
    sess.timeout = 6
    sess.base_url = @jsp_site
    resp = sess.get($jsp_doc)

    if resp.status > 400
      $log.error("#{site} is down:")
      $log.error(resp.body)
    else
      $log.info("#{site} is up.")
    end
  end

end

gistfile4.txt

#
#Config File for artemis.rb v1.x (JD Trout)
#
# All settings with multiple entries should be separated by a space only.

# Location of where artemis should log
log_location = "/var/log/artemis.log"

# SMTP Sever used to send email when an error occures (Best to not have this be the same server that you are checking for errors)
email_server = smtp.loni.ucla.edu

# The email address which which you want send from
email_from = artemis@loni.ucla.edu

# The email address where errors will be emailed
email_to = jd.trout@loni.ucla.edu

# Servers that require nfs and syslog checks
servers = viola-rs1.loni.ucla.edu viola-rs2.loni.ucla.edu viola-local.loni.ucla.edu aragon-rs1.loni.ucla.edu aragon-rs2.loni.ucla.edu aragon-rs3.loni.ucla.edu aragon-rs4.loni.ucla.edu aragon-rs5.loni.ucla.edu

# Servers in here will have /var/log/maillog scanned for errors. (Should not be confused with email_server)
smtp_servers = netserv1.loni.ucla.edu netserv2.loni.ucla.edu

# SMTP logs.  The location of the smtp logs.
smtp_log = /var/log/maillog

# HTML sites. This value is for all websites that are not JSP 
sites = www.loni.ucla.edu cms.loni.ucla.edu

#
# This is the URL
# Example:
# check_java = check.html
check_html = check.html

# This value is for JAVA web apps (glassfish/tomcat) sites
java_sites = ida.loni.ucla.edu utilities.loni.ucla.edu services.loni.ucla.edu

#
# This is the URL
# Example:
# check_java = index.jsp
check_java = check.jsp

gistfile5.rb

loop do
  
  if $svrs != nil
    server = nil
    begin
      $svrs.each do |server|
        Timeout::timeout(20) do
          Server.new(server)
          Server.mnt_chk(ssh)
          Server.syslg_chk(ssh)
        end
      end
    rescue Timeout::Error
      $log.error("Timeout Error: Check network connection and/or NFS mounts on #{server}")      
    end

    sleep 120
  end