def doWork(self):
configHolder = ConfigHolder(self.options.__dict__)
runner = VmManagerFactory.create(self.image, configHolder)
cluster = Cluster(configHolder, runner, self.options.master_vmid)
printAction('Starting cluster')
runner.runInstance()
cluster.deploy()
printStep('Done!')
def runInstance(self, details=False):
self._printContacting()
if Image.isImageId(self.vm_image):
self._checkImageExists(self.vm_image)
self.vm_image = self._prependMarketplaceUrlIfImageId(self.vm_image)
elif Image.isDiskId(self.vm_image):
self.vm_image = self._createDiskUrlIfDiskId(self.vm_image)
elif self._isAliasUrl(self.vm_image):
self.vm_image = self._resolveUrl(self.vm_image)
else:
raise Exceptions.ValidationException('Image reference must be an '
'Alias URL, Marketplace Image ID or Disk ID: %s' %
self.vm_image)
printAction('Starting machine(s)')
self.printDetail('Using VM template file: %s' % self.vmTemplateFile)
vmTpl = self._buildVmTemplate(self.vmTemplateFile)
label = (self.instanceNumber > 1) and 'machines' or 'machine'
printStep('Starting %s %s' % (self.instanceNumber, label))
self.printDetail('on endpoint: %s' % self.endpoint, Util.VERBOSE_LEVEL_DETAILED)
self.printDetail('with template:\n%s' % vmTpl, Util.VERBOSE_LEVEL_DETAILED)
for vmNb in range(self.instanceNumber):
vmId = self.cloud.vmStart(vmTpl)
self.vmIds.append(vmId)
networkName, ip = self.getNetworkDetail(vmId)
self.vmIdsAndNetwork.append((vmId, networkName, ip))
vmIpPretty = '\t%s ip: %s' % (networkName.title(), ip)
printStep('Machine %s (vm ID: %s)\n%s' % (vmNb + 1, vmId, vmIpPretty))
self.instancesDetail.append({'id': vmId, 'ip': ip, 'networkName': networkName})
self._saveVmIds()
printStep('Done!')
if not details:
return self.vmIds
else:
return self.vmIdsAndNetwork
def _startService(self, componentName, componentInstallator):
if self.startComponent:
printAction('Starting %s services' % componentName)
componentInstallator.startServices()
def _setupStep(self, componentName, componentInstallator):
if self.setupStep:
printAction('Setting up %s' % componentName)
componentInstallator.setup()
def _installStep(self, componentName, componentInstallator):
if self.installStep:
printAction('Installing %s' % componentName)
componentInstallator.install()
def deploy(self):
ssh = SSHUtil(self._runner.userPrivateKeyFile, self.cluster_admin)
# Wait until all the images are up and running
vmNetworkDetails = []
vmStartTimeout = 600
# wait until the each machine is up or timeout after 15 minutes
printStep("Waiting for all cluster VMs to be instantiated...")
if self._is_heterogeneous:
printStep("Waiting for master")
self._runner.waitUntilVmRunningOrTimeout(self._master_vmid, vmStartTimeout)
vmNetworkDetails.append(self._runner.getNetworkDetail(self._master_vmid))
for vmId in self._runner.vmIds:
printDebug('Waiting for instance to start running %s' % str(vmId))
self._runner.waitUntilVmRunningOrTimeout(vmId, vmStartTimeout)
vmNetworkDetails.append(self._runner.getNetworkDetail(vmId))
vm_cpu, vm_ram, vm_swap = self._runner.getInstanceResourceValues()
for vmNetwork in vmNetworkDetails:
if vmNetwork[0] == 'public':
host = Host()
host.public_ip = vmNetwork[1]
try:
host.public_dns = socket.gethostbyaddr(host.public_ip)[0]
except:
host.public_dns = host.public_ip
host.cores = vm_cpu
host.ram = vm_ram
host.swap = vm_swap
self.hosts.append(host)
printStep("Waiting for all instances to become accessible...")
failedHosts = []
for host in self.hosts:
hostReady = False
hostFailed = False
while not hostReady and not hostFailed:
if not ssh.waitForConnectivity(host, vmStartTimeout):
printError('Timed out while connecting to %s. Removing from target config. list.' % host.public_ip)
failedHosts.append(host)
hostFailed = True
else:
hostReady = True
if len(failedHosts) > 0:
if self.tolerate_failures:
for host in failedHosts:
self.hosts.remove(host)
else:
printError('Error instantiating some or all of the nodes. Bailing out...')
if self.clean_after_failure:
self._runner.killInstances(self._runner.vmIds)
return 128
master_node = self.hosts[0]
worker_nodes = list(self.hosts)
worker_nodes.remove(master_node)
printInfo('\tMaster is %s' % master_node.public_dns)
for node in worker_nodes:
printInfo('\tWorker: %s' % node.public_dns)
# Configure the hosts
printAction('Configuring nodes')
# Try to install the missing packages
if self.add_packages:
self.doAddPackages(ssh)
# For MPI clusters prepare the machinefile for mpirun
if self.mpi_machine_file:
self.doPrepareMPImachineFile(ssh, worker_nodes)
if self.cluster_user:
# Create a new user and prepare the environments for password-less ssh
self.doCreateClusterUser(ssh, master_node)
# Initialize the shared storage in NFS
if self.shared_folder:
self.doPrepareNFSSharedFolder(ssh, master_node, worker_nodes)
if self.ssh_hostbased:
self.doSetupSSHHostBasedCluster(ssh)
# Update /etc/profile with StratusLab specific environment variables
self.doUpdateEnvironmentVariables(ssh, master_node, worker_nodes)
# Store the list of cluster nodes in a file under /tmp
self.doPrepareNodeList(ssh, worker_nodes)
# Update the /etc/hosts file for all hosts
self.doUpdateHostsFile(ssh, master_node, worker_nodes)
# Start any services defined in rc.cluster-services
self.doStartClusterServices(ssh, master_node)
return 0
