#!/usr/bin/ruby require 'rbvmomi' require 'redis' require 'time' require 'yaml' $:.unshift(File.dirname(__FILE__)) require 'lib/logger' require 'lib/require_relative' require 'lib/vsphere_helper' Dir.chdir(File.dirname(__FILE__)) # Load the configuration file config_file = File.expand_path('vmware-host-pooler.yaml') config = YAML.load_file(config_file) pools = config[:pools] vsphere = config[:vsphere] # Load logger library $logger = Logger.new config[:config]['logfile'] # Load Graphite helper library (if configured) if (defined? config[:config]['graphite']) require 'lib/graphite' $graphite = Graphite.new config[:config]['graphite'] end # Connect to Redis $redis = Redis.new # vSphere object $vsphere = {} # Our thread-tracker object $threads = {} # Check the state of a VM def check_pending_vm vm, pool, timeout Thread.new { host = $vsphere[pool].find_vms(vm)[vm] if (host) if ( (host.summary) and (host.summary.guest) and (host.summary.guest.hostName) and (host.summary.guest.hostName == vm) ) begin Socket.getaddrinfo(vm, nil) rescue end $redis.smove('vmware_host_pool__pending__'+pool, 'vmware_host_pool__ready__'+pool, vm) $logger.log('s', "[>] [#{pool}] '#{vm}' moved to 'ready' queue") else if ( (host.runtime) and (host.runtime.bootTime) and ((((Time.now - host.runtime.bootTime)/60).to_s[/^\d+\.\d{1}/].to_f) > timeout) ) $redis.smove('vmware_host_pool__pending__'+pool, 'vmware_host_pool__completed__'+pool, vm) $logger.log('d', "[!] [#{pool}] '#{vm}' marked as 'failed' after #{timeout} minutes") end end end } end def check_ready_vm vm, pool, ttl Thread.new { host = $vsphere[pool].find_vms(vm)[vm] if (host) if ( (host.runtime) and (host.runtime.powerState) and (host.runtime.powerState != 'poweredOn') ) $redis.smove('vmware_host_pool__ready__'+pool, 'vmware_host_pool__completed__'+pool, vm) $logger.log('d', "[!] [#{pool}] '#{vm}' appears to be powered off or dead") else if ((((Time.now - host.runtime.bootTime)/60).to_s[/^\d+\.\d{1}/].to_f) > ttl) $redis.smove('vmware_host_pool__ready__'+pool, 'vmware_host_pool__completed__'+pool, vm) $logger.log('d', "[!] [#{pool}] '#{vm}' reached end of TTL after #{ttl} minutes") end end end } end def check_running_vm vm, pool, ttl Thread.new { host = $vsphere[pool].find_vms(vm)[vm] if (host) if ( (host.runtime) and (host.runtime.powerState != 'poweredOn') ) $redis.smove('vmware_host_pool__running__'+pool, 'vmware_host_pool__completed__'+pool, vm) $logger.log('d', "[!] [#{pool}] '#{vm}' appears to be powered off or dead") else if ( (host.runtime) and (host.runtime.bootTime) ((((Time.now - host.runtime.bootTime)/60).to_s[/^\d+\.\d{1}/].to_f) > ttl) ) $redis.smove('vmware_host_pool__running__'+pool, 'vmware_host_pool__completed__'+pool, vm) $logger.log('d', "[!] [#{pool}] '#{vm}' reached end of TTL after #{ttl} minutes") end end end } end # Clone a VM def clone_vm template, pool, folder, datastore Thread.new { vm = {} if template =~ /\// templatefolders = template.split('/') vm['template'] = templatefolders.pop end if templatefolders vm[vm['template']] = $vsphere[vm['template']].find_folder(templatefolders.join('/')).find(vm['template']) else raise "Please provide a full path to the template" end if vm['template'].length == 0 raise "Unable to find template '#{vm['template']}'!" end # Generate a randomized hostname o = [('a'..'z'),('0'..'9')].map{|r| r.to_a}.flatten vm['hostname'] = o[rand(25)]+(0...14).map{o[rand(o.length)]}.join # Add VM to Redis inventory ('pending' pool) $redis.sadd('vmware_host_pool__pending__'+vm['template'], vm['hostname']) # Annotate with creation time, origin template, etc. configSpec = RbVmomi::VIM.VirtualMachineConfigSpec( :annotation => 'Base template: ' + vm['template'] + "\n" + 'Creation time: ' + Time.now.strftime("%Y-%m-%d %H:%M") ) # Put the VM in the specified folder and resource pool relocateSpec = RbVmomi::VIM.VirtualMachineRelocateSpec( :datastore => $vsphere[vm['template']].find_datastore(datastore), :pool => $vsphere[vm['template']].find_pool(pool), :diskMoveType => :moveChildMostDiskBacking ) # Create a clone spec spec = RbVmomi::VIM.VirtualMachineCloneSpec( :location => relocateSpec, :config => configSpec, :powerOn => true, :template => false ) # Clone the VM $logger.log('d', "[ ] [#{vm['template']}] '#{vm['hostname']}' is being cloned from '#{vm['template']}'") start = Time.now vm[vm['template']].CloneVM_Task( :folder => $vsphere[vm['template']].find_folder(folder), :name => vm['hostname'], :spec => spec ).wait_for_completion finish = '%.2f' % (Time.now-start) $logger.log('s', "[+] [#{vm['template']}] '#{vm['hostname']}' cloned from '#{vm['template']}' in #{finish} seconds") begin $graphite.log("vcloud.clone.#{vm['template']}", finish) if defined? $graphite rescue end } end # Destroy a VM def destroy_vm vm, pool Thread.new { $redis.srem('vmware_host_pool__completed__'+pool, vm) $redis.hdel('vmware_host_pool__active__'+pool, vm) host = $vsphere[pool].find_vms(vm)[vm] if (host) start = Time.now if ( (host.runtime) and (host.runtime.powerState) and (host.runtime.powerState == 'poweredOn') ) $logger.log('d', "[ ] [#{pool}] '#{vm}' is being shut down") host.PowerOffVM_Task.wait_for_completion end host.Destroy_Task.wait_for_completion finish = '%.2f' % (Time.now-start) $logger.log('s', "[-] [#{pool}] '#{vm}' destroyed in #{finish} seconds") $graphite.log("vcloud.destroy.#{pool}", finish) if defined? $graphite end } end def check_pool pool $logger.log('d', "[*] [#{pool['name']}] starting worker thread") $redis.del('vmware_host_pool__pending__'+pool['name']) $threads[pool['name']] = Thread.new { $vsphere[pool['name']] ||= VsphereHelper.new loop do # INVENTORY inventory = {} begin base = $vsphere[pool['name']].find_pool(pool['pool']) base.vm.each do |vm| if ( (! $redis.sismember('vmware_host_pool__running__'+pool['name'], vm['name'])) and (! $redis.sismember('vmware_host_pool__ready__'+pool['name'], vm['name'])) and (! $redis.sismember('vmware_host_pool__pending__'+pool['name'], vm['name'])) and (! $redis.sismember('vmware_host_pool__completed__'+pool['name'], vm['name'])) and (! $redis.sismember('vmware_host_pool__discovered__'+pool['name'], vm['name'])) ) $redis.sadd('vmware_host_pool__discovered__'+pool['name'], vm['name']) $logger.log('s', "[?] [#{pool['name']}] '#{vm['name']}' added to 'discovered' queue") end inventory[vm['name']] = 1 end rescue end # RUNNING $redis.smembers('vmware_host_pool__running__'+pool['name']).each do |vm| if (inventory[vm]) if (pool['running_ttl']) begin check_running_vm(vm, pool['name'], pool['running_ttl']) rescue end else begin check_running_vm(vm, pool['name'], '720') rescue end end end end # READY $redis.smembers('vmware_host_pool__ready__'+pool['name']).each do |vm| if (inventory[vm]) if (pool['ready_ttl']) begin check_ready_vm(vm, pool['name'], pool['ready_ttl']) rescue end end end end # PENDING $redis.smembers('vmware_host_pool__pending__'+pool['name']).each do |vm| pool['timeout'] ||= 10 if (inventory[vm]) begin check_pending_vm(vm, pool['name'], pool['timeout']) rescue end end end # COMPLETED $redis.smembers('vmware_host_pool__completed__'+pool['name']).each do |vm| if (inventory[vm]) begin destroy_vm(vm, pool['name']) rescue end end end # LONG-RUNNING $redis.smembers('vmware_host_pool__running__'+pool['name']).each do |vm| if ($redis.hget('vmware_host_pool__active__'+pool['name'], vm)) running = (Time.now - Time.parse($redis.hget('vmware_host_pool__active__'+pool['name'], vm)))/60/60 if (running > 12) $redis.smove('vmware_host_pool__running__'+pool['name'], 'vmware_host_pool__completed__'+pool['name'], vm) $logger.log('d', "[!] [#{pool['name']}] '#{vm}' reached end of TTL after 12 hours") end end end # REPOPULATE total = $redis.scard('vmware_host_pool__ready__'+pool['name']) + $redis.scard('vmware_host_pool__pending__'+pool['name']) begin $graphite.log( 'vcloud.ready.'+pool['name'], $redis.scard('vmware_host_pool__ready__'+pool['name']) ) if defined? $graphite rescue end if (total < pool['size']) (1..(pool['size'] - total)).each { |i| begin clone_vm( pool['template'], pool['pool'], pool['folder'], pool['datastore'] ) rescue end } end sleep(1) end } end $logger.log('d', "starting vmware-host-pooler") loop do pools.each do |pool| if (! $threads[pool['name']]) check_pool(pool) else if (! $threads[pool['name']].alive?) $logger.log('d', "[!] [#{pool['name']}] worker thread died, restarting") check_pool(pool) end end end sleep(1) end