EC2 Windows CI hackish scripts

ci_hook.bat

@echo off

REM If you edit this file, make sure to run `unix2dos` to fix those fucking CRLF.

REM Let's keep this script as dumb as possible and do all the logic with a true language.
REM Basically, this should be no more than a sort of remote control.

echo Initializing hook...

:begin

cd C:\Users\Administrator\Desktop

echo Waiting for commands...
:wait
  ping -n 2 127.0.0.1 > nul
  <nul set /p =.
if not exist go goto wait

set /p project= <go
del go

echo.
echo Received go signal, running tests for project %project%...

del done 2>nul
echo. > ci.log
echo. > exitcodes
cd C:\Users\Administrator\Desktop\%project%\spec\support

echo Running tests on IE aka I'm Evil...
java -jar JsTestDriver.jar --tests all --port 1234 --reset --browser "C:\Program Files\Internet Explorer\iexplore.exe" >> C:\Users\Administrator\Desktop\ci.log
echo %errorlevel% >> C:\Users\Administrator\Desktop\exitcodes

echo Running tests on Firefox...
java -jar JsTestDriver.jar --tests all --port 1234 --reset --browser "C:\Program Files\Mozilla Firefox\firefox.exe" >> C:\Users\Administrator\Desktop\ci.log
echo %errorlevel% >> C:\Users\Administrator\Desktop\exitcodes

echo Running tests on Safari...
java -jar JsTestDriver.jar --tests all --port 1234 --reset --browser "C:\Program Files\Safari\Safari.exe" >> C:\Users\Administrator\Desktop\ci.log
echo %errorlevel% >> C:\Users\Administrator\Desktop\exitcodes

echo Running tests on Chrome...
java -jar JsTestDriver.jar --tests all --port 1234 --reset --browser "C:\Users\Administrator\AppData\Local\Google\Chrome\Application\chrome.exe" >> C:\Users\Administrator\Desktop\ci.log
echo %errorlevel% >> C:\Users\Administrator\Desktop\exitcodes

echo. > C:\Users\Administrator\Desktop\done

goto begin

config.yml.erb

credentials:
  access_key_id: <%= @credentials['access_key_id'] %>
  secret_access_key: <%= @credentials['secret_access_key'] %>

ci:
  tests_timeout: 300  # give it 5 min to run tests on all browsers
  username:    Administrator
  host:        x.x.x.x  # elastic IP
  instance_id: i-xxxxxxxx
  region:      xxxxxxxx
  local_path:  /home/cruise/.cruise/projects
  remote_path: /cygdrive/c/Users/Administrator/Desktop
  hook_file:   /cygdrive/c/Users/Administrator/AppData/Roaming/Microsoft/Windows/Start\\ Menu/Programs/Startup/ci_hook.bat
  lock_file:   /home/cruise/aws/lock
  ssh_key:     /home/cruise/.ssh/id_rsa
  ssh_port:    xx
  ssh_params:  -i /home/cruise/.ssh/id_rsa -p xx -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no -q
  last_build_file: /home/cruise/aws/last_build

tests_windows.rb

#!/usr/bin/env ruby

require 'yaml'
require 'fileutils'
require 'rubygems'
require 'aws-sdk'
require 'tzinfo'

STDOUT.sync = true
STDERR.sync = true

def get_instance
  ec2 = AWS::EC2.new
  ec2 = ec2.regions[CONFIG['ci']['region']]

  instance = ec2.instances[CONFIG['ci']['instance_id']]
  unless instance.exists?
    raise "Cannot find instance #{CONFIG['ci']['instance_id']}, wrong region?"
  end

  instance
end

def wait_for_status(instance, status, delay = 5)
  print "Waiting for the instance to be #{status.to_s}..."
  until instance.status == status
    sleep delay
    print '.'
  end
  puts
end

def run_or_raise command
  # rsync fails sometimes because of disk I/O issues on cheap Amazon drives
  # so we give it 5 times to succeed
  5.times do
    output = `#{command}`
    return output if $? == 0
    sleep 30
  end
  FileUtils.rm CONFIG['ci']['lock_file']
  raise "Failed 5 times in a row running: '#{command}'"
end

def launch_instance
  instance = get_instance

  if [:pending, :running].include? instance.status
    raise "Instance already running (or booting)."
  end

  puts "Starting instance..."
  instance.start
  wait_for_status instance, :running

  puts "Instance launched"
end

def shutdown_instance
  instance = get_instance
  return unless :running == instance.status

  instance.stop
end

def reboot_instance
  instance = get_instance

  raise "Instance is not running." unless :running == instance.status

  puts "Rebooting instance..."
  # instance.reboot does NOT work!
  instance.stop
  wait_for_status instance, :stopped
  instance.start
  wait_for_status instance, :running

  puts "Instance rebooted"
end

def check_instance
  instance = get_instance

  if instance.status != :running and instance.status != :stopped
    raise "Don't know what to do, exiting."
  end

  unless instance.status == :running
    launch_instance
    wait_for_status instance, :running
  end

  puts "Associating elastic IP..."
  if instance.elastic_ip.nil?
    instance.associate_elastic_ip(CONFIG['ci']['host'])
    sleep 5
  end

  print "Waiting for Windows' post boot..."
  begin
    print '.'
    `nc -z -w 10 #{CONFIG['ci']['host']} #{CONFIG['ci']['ssh_port']}`
  end while $? != 0
  puts
end

def upload_hook
  check_instance
  puts "Uploading hook..."

  local_hook_file = "#{File.dirname(__FILE__)}/#{File.basename(CONFIG['ci']['hook_file'])}"
  output = run_or_raise "rsync -azci -e \"ssh #{CONFIG['ci']['ssh_params']}\" #{local_hook_file} \"#{CONFIG['ci']['username']}@#{CONFIG['ci']['host']}:#{CONFIG['ci']['hook_file']}\""

  # This would be the proper way to do it, but I can't restart the new hook this way.
  #
  # pid = `ssh #{CONFIG['ci']['ssh_params']}" #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} ps -W | grep cmd.exe | awk '{print $1}'`.split("\n").map(&:chop)
  # if pid.count > 1
  #   raise "Oops, didn't get exactly one PID, I don't how to kill the hook."
  # else
  #   `ssh #{CONFIG['ci']['ssh_params']}" #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} /bin/kill -f #{pid.first}`
  # end

  if output.chop.length > 0
    # Hey, it's Windows, born to be rebooted ;)
    puts "New hook uploaded, rebooting to take effect..."
    reboot_instance
  else
    puts "Remote hook already up-to-date."
  end
end

def run_tests project
  check_instance
  puts "Syncing repo..."
  run_or_raise "rsync -azv --delete --timeout=5 -e \"ssh #{CONFIG['ci']['ssh_params']}\" #{CONFIG['ci']['local_path']}/#{project}/work/ #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']}:#{CONFIG['ci']['remote_path']}/#{project}"

  puts "Sending the green flag to run tests to Batch handler..."
  run_or_raise "ssh #{CONFIG['ci']['ssh_params']} #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} -- echo -n '#{project}' \\> /cygdrive/c/Users/Administrator/Desktop/go"

  print "Wait for tests to finish..."
  timeouted = true
  CONFIG['ci']['tests_timeout'].times do
    sleep 1

    `ssh #{CONFIG['ci']['ssh_params']} #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} -- test -f /cygdrive/c/Users/Administrator/Desktop/done 2>/dev/null`
    unless $? == 0
      print "."
    else
      timeouted = false
      break
    end
  end
  puts

  if timeouted
    puts "Tests timeouted"
    exit_codes = [1]
  else
    puts "Tests finished normally."
    exit_codes = run_or_raise "ssh #{CONFIG['ci']['ssh_params']} #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} -- cat /cygdrive/c/Users/Administrator/Desktop/exitcodes"
  end

  # Print log for CruiseControl
  puts run_or_raise "ssh #{CONFIG['ci']['ssh_params']} #{CONFIG['ci']['username']}@#{CONFIG['ci']['host']} -- cat /cygdrive/c/Users/Administrator/Desktop/ci.log"

  # Return sum of error codes
  exit_codes.split.inject(0) { |sum, x| sum + x.to_i } rescue 1
end

###

CONFIG = YAML.load(File.read(
  File.join(File.dirname(__FILE__), "config.yml")
))

AWS.config CONFIG['credentials']


case ARGV[0]
when '--run-tests'

  # Let's go!
  if File.file? CONFIG['ci']['lock_file']
    puts "Another instance of this script is already running!"
    exit 1
  end

  unless ARGV[1]
    puts "You must pass the project to build."
    exit 1
  end

  FileUtils.touch CONFIG['ci']['lock_file']
  FileUtils.touch CONFIG['ci']['last_build_file']

  upload_hook
  exit_code = run_tests(ARGV[1])

  FileUtils.rm CONFIG['ci']['lock_file']

  # Return JsTestDriver exit code to CruiseControl
  exit exit_code

when '--shutdown-instance-if-needed'

  # Tests are running right now, don't shutdown instance
  exit 0 if File.file? CONFIG['ci']['lock_file']

  right_now = TZInfo::Timezone.get('Europe/London').utc_to_local(Time.new.utc)
  case right_now.wday
    when 0, 6 # Sunday, Saturday
      is_working_time = false
    else
      is_working_time = (right_now.hour.between? 10, 18)
  end

  if is_working_time
    # Shutdown if last build was at least one hour ago
    `find #{File.dirname(CONFIG['ci']['last_build_file'])} -name #{File.basename(CONFIG['ci']['last_build_file'])} -mtime 1`
    exit 0 if $? == 0
  end

  shutdown_instance
  exit 0

else

  puts <<-USAGE
You must pass exactly one argument:
  --run-tests <project>
  --shutdown-instance-if-needed
  USAGE

end

exit 1