Synnefo » snf-ganeti » ganeti-local

import os

import copy

import itertools

import operator

from ganeti import hypervisor

from ganeti import objects

from ganeti import netutils

from ganeti import opcodes

from ganeti import ht

from ganeti import rpc

from ganeti import pathutils

from ganeti import network

from ganeti.masterd import iallocator

  _CheckIAllocatorOrNode, _FindFaultyInstanceDisks

import ganeti.masterd.instance # pylint: disable=W0611

def _IsExclusiveStorageEnabledNodeName(cfg, nodename):

  """Whether exclusive_storage is in effect for the given node.

  @type cfg: L{config.ConfigWriter}

  @param cfg: The cluster configuration

  @type nodename: string

  @param nodename: The node

  @rtype: bool

  @return: The effective value of exclusive_storage

  @raise errors.OpPrereqError: if no node exists with the given name

  """

  ni = cfg.GetNodeInfo(nodename)

  if ni is None:

    raise errors.OpPrereqError("Invalid node name %s" % nodename,

                               errors.ECODE_NOENT)

  return _IsExclusiveStorageEnabledNode(cfg, ni)

def _CopyLockList(names):

  """Makes a copy of a list of lock names.

  Handles L{locking.ALL_SET} correctly.

  """

  if names == locking.ALL_SET:

    return locking.ALL_SET

  else:

    return names[:]

def _ReleaseLocks(lu, level, names=None, keep=None):

  """Releases locks owned by an LU.

  @type lu: L{LogicalUnit}

  @param level: Lock level

  @type names: list or None

  @param names: Names of locks to release

  @type keep: list or None

  @param keep: Names of locks to retain

  """

  assert not (keep is not None and names is not None), \

         "Only one of the 'names' and the 'keep' parameters can be given"

  if names is not None:

    should_release = names.__contains__

  elif keep:

    should_release = lambda name: name not in keep

  else:

    should_release = None

  owned = lu.owned_locks(level)

  if not owned:

    # Not owning any lock at this level, do nothing

    pass

  elif should_release:

    retain = []

    release = []

    # Determine which locks to release

    for name in owned:

      if should_release(name):

        release.append(name)

      else:

        retain.append(name)

    assert len(lu.owned_locks(level)) == (len(retain) + len(release))

    # Release just some locks

    lu.glm.release(level, names=release)

    assert frozenset(lu.owned_locks(level)) == frozenset(retain)

  else:

    # Release everything

    lu.glm.release(level)

    assert not lu.glm.is_owned(level), "No locks should be owned"

def _CheckNodeOnline(lu, node, msg=None):

  """Ensure that a given node is online.

  @param lu: the LU on behalf of which we make the check

  @param node: the node to check

  @param msg: if passed, should be a message to replace the default one

  @raise errors.OpPrereqError: if the node is offline

  """

  if msg is None:

    msg = "Can't use offline node"

  if lu.cfg.GetNodeInfo(node).offline:

    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)

def _CheckNodeNotDrained(lu, node):

  """Ensure that a given node is not drained.

  @param lu: the LU on behalf of which we make the check

  @param node: the node to check

  @raise errors.OpPrereqError: if the node is drained

  """

  if lu.cfg.GetNodeInfo(node).drained:

    raise errors.OpPrereqError("Can't use drained node %s" % node,

                               errors.ECODE_STATE)

def _CheckNodeVmCapable(lu, node):

  """Ensure that a given node is vm capable.

  @param lu: the LU on behalf of which we make the check

  @param node: the node to check

  @raise errors.OpPrereqError: if the node is not vm capable

  """

  if not lu.cfg.GetNodeInfo(node).vm_capable:

    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,

                               errors.ECODE_STATE)

def _CheckNodeHasOS(lu, node, os_name, force_variant):

  """Ensure that a node supports a given OS.

  @param lu: the LU on behalf of which we make the check

  @param node: the node to check

  @param os_name: the OS to query about

  @param force_variant: whether to ignore variant errors

  @raise errors.OpPrereqError: if the node is not supporting the OS

  """

  result = lu.rpc.call_os_get(node, os_name)

  result.Raise("OS '%s' not in supported OS list for node %s" %

               (os_name, node),

               prereq=True, ecode=errors.ECODE_INVAL)

  if not force_variant:

    _CheckOSVariant(result.payload, os_name)

def _GetClusterDomainSecret():

  """Reads the cluster domain secret.

  """

  return utils.ReadOneLineFile(pathutils.CLUSTER_DOMAIN_SECRET_FILE,

                               strict=True)

def _ComputeIPolicyInstanceSpecViolation(

  ipolicy, instance_spec, disk_template,

  _compute_fn=_ComputeIPolicySpecViolation):

  """Compute if instance specs meets the specs of ipolicy.

  @type ipolicy: dict

  @param ipolicy: The ipolicy to verify against

  @param instance_spec: dict

  @param instance_spec: The instance spec to verify

  @type disk_template: string

  @param disk_template: the disk template of the instance

  @param _compute_fn: The function to verify ipolicy (unittest only)

  @see: L{_ComputeIPolicySpecViolation}

  """

  mem_size = instance_spec.get(constants.ISPEC_MEM_SIZE, None)

  cpu_count = instance_spec.get(constants.ISPEC_CPU_COUNT, None)

  disk_count = instance_spec.get(constants.ISPEC_DISK_COUNT, 0)

  disk_sizes = instance_spec.get(constants.ISPEC_DISK_SIZE, [])

  nic_count = instance_spec.get(constants.ISPEC_NIC_COUNT, 0)

  spindle_use = instance_spec.get(constants.ISPEC_SPINDLE_USE, None)

  return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count,

                     disk_sizes, spindle_use, disk_template)

def _ComputeIPolicyNodeViolation(ipolicy, instance, current_group,

                                 target_group, cfg,

                                 _compute_fn=_ComputeIPolicyInstanceViolation):

  """Compute if instance meets the specs of the new target group.

  @param ipolicy: The ipolicy to verify

  @param instance: The instance object to verify

  @param current_group: The current group of the instance

  @param target_group: The new group of the instance

  @type cfg: L{config.ConfigWriter}

  @param cfg: Cluster configuration

  @param _compute_fn: The function to verify ipolicy (unittest only)

  @see: L{_ComputeIPolicySpecViolation}

  """

  if current_group == target_group:

    return []

  else:

    return _compute_fn(ipolicy, instance, cfg)

def _CheckTargetNodeIPolicy(lu, ipolicy, instance, node, cfg, ignore=False,

                            _compute_fn=_ComputeIPolicyNodeViolation):

  """Checks that the target node is correct in terms of instance policy.

  @param ipolicy: The ipolicy to verify

  @param instance: The instance object to verify

  @param node: The new node to relocate

  @type cfg: L{config.ConfigWriter}

  @param cfg: Cluster configuration

  @param ignore: Ignore violations of the ipolicy

  @param _compute_fn: The function to verify ipolicy (unittest only)

  @see: L{_ComputeIPolicySpecViolation}

  """

  primary_node = lu.cfg.GetNodeInfo(instance.primary_node)

  res = _compute_fn(ipolicy, instance, primary_node.group, node.group, cfg)

  if res:

    msg = ("Instance does not meet target node group's (%s) instance"

           " policy: %s") % (node.group, utils.CommaJoin(res))

    if ignore:

      lu.LogWarning(msg)

    else:

      raise errors.OpPrereqError(msg, errors.ECODE_INVAL)

def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,

                          minmem, maxmem, vcpus, nics, disk_template, disks,

                          bep, hvp, hypervisor_name, tags):

  """Builds instance related env variables for hooks

  This builds the hook environment from individual variables.

  @type name: string

  @param name: the name of the instance

  @type primary_node: string

  @param primary_node: the name of the instance's primary node

  @type secondary_nodes: list

  @param secondary_nodes: list of secondary nodes as strings

  @type os_type: string

  @param os_type: the name of the instance's OS

  @type status: string

  @param status: the desired status of the instance

  @type minmem: string

  @param minmem: the minimum memory size of the instance

  @type maxmem: string

  @param maxmem: the maximum memory size of the instance

  @type vcpus: string

  @param vcpus: the count of VCPUs the instance has

  @type nics: list

  @param nics: list of tuples (name, uuid, ip, mac, mode, link, net, netinfo)

      representing the NICs the instance has

  @type disk_template: string

  @param disk_template: the disk template of the instance

  @type disks: list

  @param disks: list of tuples (name, uuid, size, mode)

  @type bep: dict

  @param bep: the backend parameters for the instance

  @type hvp: dict

  @param hvp: the hypervisor parameters for the instance

  @type hypervisor_name: string

  @param hypervisor_name: the hypervisor for the instance

  @type tags: list

  @param tags: list of instance tags as strings

  @rtype: dict

  @return: the hook environment for this instance

  """

  env = {

    "OP_TARGET": name,

    "INSTANCE_NAME": name,

    "INSTANCE_PRIMARY": primary_node,

    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),

    "INSTANCE_OS_TYPE": os_type,

    "INSTANCE_STATUS": status,

    "INSTANCE_MINMEM": minmem,

    "INSTANCE_MAXMEM": maxmem,

    # TODO(2.9) remove deprecated "memory" value

    "INSTANCE_MEMORY": maxmem,

    "INSTANCE_VCPUS": vcpus,

    "INSTANCE_DISK_TEMPLATE": disk_template,

    "INSTANCE_HYPERVISOR": hypervisor_name,

  }

  if nics:

    nic_count = len(nics)

    for idx, (name, _, ip, mac, mode, link, net, netinfo) in enumerate(nics):

      if ip is None:

        ip = ""

      env["INSTANCE_NIC%d_NAME" % idx] = name

      env["INSTANCE_NIC%d_IP" % idx] = ip

      env["INSTANCE_NIC%d_MAC" % idx] = mac

      env["INSTANCE_NIC%d_MODE" % idx] = mode

      env["INSTANCE_NIC%d_LINK" % idx] = link

      if netinfo:

        nobj = objects.Network.FromDict(netinfo)

        env.update(nobj.HooksDict("INSTANCE_NIC%d_" % idx))

      elif network:

        # FIXME: broken network reference: the instance NIC specifies a

        # network, but the relevant network entry was not in the config. This

        # should be made impossible.

        env["INSTANCE_NIC%d_NETWORK_NAME" % idx] = net

      if mode == constants.NIC_MODE_BRIDGED:

        env["INSTANCE_NIC%d_BRIDGE" % idx] = link

  else:

    nic_count = 0

  env["INSTANCE_NIC_COUNT"] = nic_count

  if disks:

    disk_count = len(disks)

    for idx, (name, size, mode) in enumerate(disks):

      env["INSTANCE_DISK%d_NAME" % idx] = name

      env["INSTANCE_DISK%d_SIZE" % idx] = size

      env["INSTANCE_DISK%d_MODE" % idx] = mode

  else:

    disk_count = 0

  env["INSTANCE_DISK_COUNT"] = disk_count

  if not tags:

    tags = []

  env["INSTANCE_TAGS"] = " ".join(tags)

  for source, kind in [(bep, "BE"), (hvp, "HV")]:

    for key, value in source.items():

      env["INSTANCE_%s_%s" % (kind, key)] = value

  return env

def _NICToTuple(lu, nic):

  """Build a tupple of nic information.

  @type lu:  L{LogicalUnit}

  @param lu: the logical unit on whose behalf we execute

  @type nic: L{objects.NIC}

  @param nic: nic to convert to hooks tuple

  """

  cluster = lu.cfg.GetClusterInfo()

  filled_params = cluster.SimpleFillNIC(nic.nicparams)

  mode = filled_params[constants.NIC_MODE]

  link = filled_params[constants.NIC_LINK]

  netinfo = None

  if nic.network:

    nobj = lu.cfg.GetNetwork(nic.network)

    netinfo = objects.Network.ToDict(nobj)

  return (nic.name, nic.uuid, nic.ip, nic.mac, mode, link, nic.network, netinfo)

def _NICListToTuple(lu, nics):

  """Build a list of nic information tuples.

  This list is suitable to be passed to _BuildInstanceHookEnv or as a return

  value in LUInstanceQueryData.

  @type lu:  L{LogicalUnit}

  @param lu: the logical unit on whose behalf we execute

  @type nics: list of L{objects.NIC}

  @param nics: list of nics to convert to hooks tuples

  """

  hooks_nics = []

  for nic in nics:

    hooks_nics.append(_NICToTuple(lu, nic))

  return hooks_nics

def _BuildInstanceHookEnvByObject(lu, instance, override=None):

  """Builds instance related env variables for hooks from an object.

  @type lu: L{LogicalUnit}

  @param lu: the logical unit on whose behalf we execute

  @type instance: L{objects.Instance}

  @param instance: the instance for which we should build the

      environment

  @type override: dict

  @param override: dictionary with key/values that will override

      our values

  @rtype: dict

  @return: the hook environment dictionary

  """

  cluster = lu.cfg.GetClusterInfo()

  bep = cluster.FillBE(instance)

  hvp = cluster.FillHV(instance)

  args = {

    "name": instance.name,

    "primary_node": instance.primary_node,

    "secondary_nodes": instance.secondary_nodes,

    "os_type": instance.os,

    "status": instance.admin_state,

    "maxmem": bep[constants.BE_MAXMEM],

    "minmem": bep[constants.BE_MINMEM],

    "vcpus": bep[constants.BE_VCPUS],

    "nics": _NICListToTuple(lu, instance.nics),

    "disk_template": instance.disk_template,

    "disks": [(disk.name, disk.size, disk.mode)

              for disk in instance.disks],

    "bep": bep,

    "hvp": hvp,

    "hypervisor_name": instance.hypervisor,

    "tags": instance.tags,

  }

  if override:

    args.update(override)

  return _BuildInstanceHookEnv(**args) # pylint: disable=W0142

def _CheckNicsBridgesExist(lu, target_nics, target_node):

  """Check that the brigdes needed by a list of nics exist.

  """

  cluster = lu.cfg.GetClusterInfo()

  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]

  brlist = [params[constants.NIC_LINK] for params in paramslist

            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]

  if brlist:

    result = lu.rpc.call_bridges_exist(target_node, brlist)

    result.Raise("Error checking bridges on destination node '%s'" %

                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)

def _CheckInstanceBridgesExist(lu, instance, node=None):

  """Check that the brigdes needed by an instance exist.

  """

  if node is None:

    node = instance.primary_node

  _CheckNicsBridgesExist(lu, instance.nics, node)

def _CheckOSVariant(os_obj, name):

  """Check whether an OS name conforms to the os variants specification.

  @type os_obj: L{objects.OS}

  @param os_obj: OS object to check

  @type name: string

  @param name: OS name passed by the user, to check for validity

  """

  variant = objects.OS.GetVariant(name)

  if not os_obj.supported_variants:

    if variant:

      raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"

                                 " passed)" % (os_obj.name, variant),

                                 errors.ECODE_INVAL)

    return

  if not variant:

    raise errors.OpPrereqError("OS name must include a variant",

                               errors.ECODE_INVAL)

  if variant not in os_obj.supported_variants:

    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)

def _CheckHostnameSane(lu, name):

  """Ensures that a given hostname resolves to a 'sane' name.

  The given name is required to be a prefix of the resolved hostname,

  to prevent accidental mismatches.

  @param lu: the logical unit on behalf of which we're checking

  @param name: the name we should resolve and check

  @return: the resolved hostname object

  """

  hostname = netutils.GetHostname(name=name)

  if hostname.name != name:

    lu.LogInfo("Resolved given name '%s' to '%s'", name, hostname.name)

  if not utils.MatchNameComponent(name, [hostname.name]):

    raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"

                                " same as given hostname '%s'") %

                                (hostname.name, name), errors.ECODE_INVAL)

  return hostname

def _WaitForSync(lu, instance, disks=None, oneshot=False):

  """Sleep and poll for an instance's disk to sync.

  """

  if not instance.disks or disks is not None and not disks:

    return True

  disks = _ExpandCheckDisks(instance, disks)

  if not oneshot:

    lu.LogInfo("Waiting for instance %s to sync disks", instance.name)

  node = instance.primary_node

  for dev in disks:

    lu.cfg.SetDiskID(dev, node)

  # TODO: Convert to utils.Retry

  retries = 0

  degr_retries = 10 # in seconds, as we sleep 1 second each time

  while True:

    max_time = 0

    done = True

    cumul_degraded = False

    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, (disks, instance))

    msg = rstats.fail_msg

    if msg:

      lu.LogWarning("Can't get any data from node %s: %s", node, msg)

      retries += 1

      if retries >= 10:

        raise errors.RemoteError("Can't contact node %s for mirror data,"

                                 " aborting." % node)

      time.sleep(6)

      continue

    rstats = rstats.payload

    retries = 0

    for i, mstat in enumerate(rstats):

      if mstat is None:

        lu.LogWarning("Can't compute data for node %s/%s",

                           node, disks[i].iv_name)

        continue

      cumul_degraded = (cumul_degraded or

                        (mstat.is_degraded and mstat.sync_percent is None))

      if mstat.sync_percent is not None:

        done = False

        if mstat.estimated_time is not None:

          rem_time = ("%s remaining (estimated)" %

                      utils.FormatSeconds(mstat.estimated_time))

          max_time = mstat.estimated_time

        else:

          rem_time = "no time estimate"

        lu.LogInfo("- device %s: %5.2f%% done, %s",

                   disks[i].iv_name, mstat.sync_percent, rem_time)

    # if we're done but degraded, let's do a few small retries, to

    # make sure we see a stable and not transient situation; therefore

    # we force restart of the loop

    if (done or oneshot) and cumul_degraded and degr_retries > 0:

      logging.info("Degraded disks found, %d retries left", degr_retries)

      degr_retries -= 1

      time.sleep(1)

      continue

    if done or oneshot:

      break

    time.sleep(min(60, max_time))

  if done:

    lu.LogInfo("Instance %s's disks are in sync", instance.name)

  return not cumul_degraded

def _BlockdevFind(lu, node, dev, instance):

  """Wrapper around call_blockdev_find to annotate diskparams.

  @param lu: A reference to the lu object

  @param node: The node to call out

  @param dev: The device to find

  @param instance: The instance object the device belongs to

  @returns The result of the rpc call

  """

  (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)

  return lu.rpc.call_blockdev_find(node, disk)

def _CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):

  """Wrapper around L{_CheckDiskConsistencyInner}.

  """

  (disk,) = _AnnotateDiskParams(instance, [dev], lu.cfg)

  return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,

                                    ldisk=ldisk)

def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,

                               ldisk=False):

  """Check that mirrors are not degraded.

  @attention: The device has to be annotated already.

  The ldisk parameter, if True, will change the test from the

  is_degraded attribute (which represents overall non-ok status for

  the device(s)) to the ldisk (representing the local storage status).

  """

  lu.cfg.SetDiskID(dev, node)

  result = True

  if on_primary or dev.AssembleOnSecondary():

    rstats = lu.rpc.call_blockdev_find(node, dev)

    msg = rstats.fail_msg

    if msg:

      lu.LogWarning("Can't find disk on node %s: %s", node, msg)

      result = False

    elif not rstats.payload:

      lu.LogWarning("Can't find disk on node %s", node)

      result = False

    else:

      if ldisk:

        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY

      else:

        result = result and not rstats.payload.is_degraded

  if dev.children:

    for child in dev.children:

      result = result and _CheckDiskConsistencyInner(lu, instance, child, node,

                                                     on_primary)

  return result

    # Prepare data in requested order

    return [data[name] for name in self._GetNames(lu, pol.keys(), None)

            if name in data]

class LUOsDiagnose(NoHooksLU):

  """Logical unit for OS diagnose/query.

  """

  REQ_BGL = False

  @staticmethod

  def _BuildFilter(fields, names):

    """Builds a filter for querying OSes.

    """

    name_filter = qlang.MakeSimpleFilter("name", names)

    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the

    # respective field is not requested

    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]

                     for fname in ["hidden", "blacklisted"]

                     if fname not in fields]

    if "valid" not in fields:

      status_filter.append([qlang.OP_TRUE, "valid"])

    if status_filter:

      status_filter.insert(0, qlang.OP_AND)

    else:

      status_filter = None

    if name_filter and status_filter:

      return [qlang.OP_AND, name_filter, status_filter]

    elif name_filter:

      return name_filter

    else:

      return status_filter

  def CheckArguments(self):

    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),

                       self.op.output_fields, False)

  def ExpandNames(self):

    self.oq.ExpandNames(self)

  def Exec(self, feedback_fn):

    return self.oq.OldStyleQuery(self)

class _ExtStorageQuery(_QueryBase):

  FIELDS = query.EXTSTORAGE_FIELDS

  def ExpandNames(self, lu):

    # Lock all nodes in shared mode

    # Temporary removal of locks, should be reverted later

    # TODO: reintroduce locks when they are lighter-weight

    lu.needed_locks = {}

    #self.share_locks[locking.LEVEL_NODE] = 1

    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET

    # The following variables interact with _QueryBase._GetNames

    if self.names:

      self.wanted = self.names

    else:

      self.wanted = locking.ALL_SET

    self.do_locking = self.use_locking

  def DeclareLocks(self, lu, level):

    pass

  @staticmethod

  def _DiagnoseByProvider(rlist):

    """Remaps a per-node return list into an a per-provider per-node dictionary

    @param rlist: a map with node names as keys and ExtStorage objects as values

    @rtype: dict

    @return: a dictionary with extstorage providers as keys and as

        value another map, with nodes as keys and tuples of

        (path, status, diagnose, parameters) as values, eg::

          {"provider1": {"node1": [(/usr/lib/..., True, "", [])]

                         "node2": [(/srv/..., False, "missing file")]

                         "node3": [(/srv/..., True, "", [])]

          }

    """

    all_es = {}

    # we build here the list of nodes that didn't fail the RPC (at RPC

    # level), so that nodes with a non-responding node daemon don't

    # make all OSes invalid

    good_nodes = [node_name for node_name in rlist

                  if not rlist[node_name].fail_msg]

    for node_name, nr in rlist.items():

      if nr.fail_msg or not nr.payload:

        continue

      for (name, path, status, diagnose, params) in nr.payload:

        if name not in all_es:

          # build a list of nodes for this os containing empty lists

          # for each node in node_list

          all_es[name] = {}

          for nname in good_nodes:

            all_es[name][nname] = []

        # convert params from [name, help] to (name, help)

        params = [tuple(v) for v in params]

        all_es[name][node_name].append((path, status, diagnose, params))

    return all_es

  def _GetQueryData(self, lu):

    """Computes the list of nodes and their attributes.

    """

    # Locking is not used

    assert not (compat.any(lu.glm.is_owned(level)

                           for level in locking.LEVELS

                           if level != locking.LEVEL_CLUSTER) or

                self.do_locking or self.use_locking)

    valid_nodes = [node.name

                   for node in lu.cfg.GetAllNodesInfo().values()

                   if not node.offline and node.vm_capable]

    pol = self._DiagnoseByProvider(lu.rpc.call_extstorage_diagnose(valid_nodes))

    data = {}

    nodegroup_list = lu.cfg.GetNodeGroupList()

    for (es_name, es_data) in pol.items():

      # For every provider compute the nodegroup validity.

      # To do this we need to check the validity of each node in es_data

      # and then construct the corresponding nodegroup dict:

      #      { nodegroup1: status

      #        nodegroup2: status

      #      }

      ndgrp_data = {}

      for nodegroup in nodegroup_list:

        ndgrp = lu.cfg.GetNodeGroup(nodegroup)

        nodegroup_nodes = ndgrp.members

        nodegroup_name = ndgrp.name

        node_statuses = []

        for node in nodegroup_nodes:

          if node in valid_nodes:

            if es_data[node] != []:

              node_status = es_data[node][0][1]

              node_statuses.append(node_status)

            else:

              node_statuses.append(False)

        if False in node_statuses:

          ndgrp_data[nodegroup_name] = False

        else:

          ndgrp_data[nodegroup_name] = True

      # Compute the provider's parameters

      parameters = set()

      for idx, esl in enumerate(es_data.values()):

        valid = bool(esl and esl[0][1])

        if not valid:

          break

        node_params = esl[0][3]

        if idx == 0:

          # First entry

          parameters.update(node_params)

        else:

          # Filter out inconsistent values

          parameters.intersection_update(node_params)

      params = list(parameters)

      # Now fill all the info for this provider

      info = query.ExtStorageInfo(name=es_name, node_status=es_data,

                                  nodegroup_status=ndgrp_data,

                                  parameters=params)

      data[es_name] = info

    # Prepare data in requested order

    return [data[name] for name in self._GetNames(lu, pol.keys(), None)

            if name in data]

class LUExtStorageDiagnose(NoHooksLU):

  """Logical unit for ExtStorage diagnose/query.

  """

  REQ_BGL = False

  def CheckArguments(self):

    self.eq = _ExtStorageQuery(qlang.MakeSimpleFilter("name", self.op.names),

                               self.op.output_fields, False)

  def ExpandNames(self):

    self.eq.ExpandNames(self)

  def Exec(self, feedback_fn):

    return self.eq.OldStyleQuery(self)

class _InstanceQuery(_QueryBase):

  FIELDS = query.INSTANCE_FIELDS

  def ExpandNames(self, lu):

    lu.needed_locks = {}

    lu.share_locks = _ShareAll()

    if self.names:

      self.wanted = _GetWantedInstances(lu, self.names)

    else:

      self.wanted = locking.ALL_SET

    self.do_locking = (self.use_locking and

                       query.IQ_LIVE in self.requested_data)

    if self.do_locking:

      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted

      lu.needed_locks[locking.LEVEL_NODEGROUP] = []

      lu.needed_locks[locking.LEVEL_NODE] = []

      lu.needed_locks[locking.LEVEL_NETWORK] = []

      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE

    self.do_grouplocks = (self.do_locking and

                          query.IQ_NODES in self.requested_data)

  def DeclareLocks(self, lu, level):

    if self.do_locking:

      if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:

        assert not lu.needed_locks[locking.LEVEL_NODEGROUP]

        # Lock all groups used by instances optimistically; this requires going

        # via the node before it's locked, requiring verification later on

        lu.needed_locks[locking.LEVEL_NODEGROUP] = \

          set(group_uuid

              for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)

              for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))

      elif level == locking.LEVEL_NODE:

        lu._LockInstancesNodes() # pylint: disable=W0212

      elif level == locking.LEVEL_NETWORK:

        lu.needed_locks[locking.LEVEL_NETWORK] = \

          frozenset(net_uuid

                    for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)

                    for net_uuid in lu.cfg.GetInstanceNetworks(instance_name))

  @staticmethod

  def _CheckGroupLocks(lu):

    owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))

    owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))

    # Check if node groups for locked instances are still correct

    for instance_name in owned_instances:

      _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)

  def _GetQueryData(self, lu):

    """Computes the list of instances and their attributes.

    """

    if self.do_grouplocks:

      self._CheckGroupLocks(lu)

    cluster = lu.cfg.GetClusterInfo()

    all_info = lu.cfg.GetAllInstancesInfo()

    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)

    instance_list = [all_info[name] for name in instance_names]

    nodes = frozenset(itertools.chain(*(inst.all_nodes

                                        for inst in instance_list)))

    hv_list = list(set([inst.hypervisor for inst in instance_list]))

    bad_nodes = []

    offline_nodes = []

    wrongnode_inst = set()

    # Gather data as requested

    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):

      live_data = {}

      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)

      for name in nodes:

        result = node_data[name]

        if result.offline:

          # offline nodes will be in both lists

          assert result.fail_msg

          offline_nodes.append(name)

        if result.fail_msg:

          bad_nodes.append(name)

        elif result.payload:

          for inst in result.payload:

            if inst in all_info:

              if all_info[inst].primary_node == name:

                live_data.update(result.payload)

              else:

                wrongnode_inst.add(inst)

            else:

              # orphan instance; we don't list it here as we don't

              # handle this case yet in the output of instance listing

              logging.warning("Orphan instance '%s' found on node %s",

                              inst, name)

        # else no instance is alive

    else:

      live_data = {}

    if query.IQ_DISKUSAGE in self.requested_data:

      gmi = ganeti.masterd.instance

      disk_usage = dict((inst.name,

                         gmi.ComputeDiskSize(inst.disk_template,

                                             [{constants.IDISK_SIZE: disk.size}

                                              for disk in inst.disks]))

                        for inst in instance_list)

    else:

      disk_usage = None

    if query.IQ_CONSOLE in self.requested_data:

      consinfo = {}

      for inst in instance_list:

        if inst.name in live_data:

          # Instance is running

          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)

        else:

          consinfo[inst.name] = None

      assert set(consinfo.keys()) == set(instance_names)

    else:

      consinfo = None

    if query.IQ_NODES in self.requested_data:

      node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),

                                            instance_list)))

      nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))

      groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))

                    for uuid in set(map(operator.attrgetter("group"),

                                        nodes.values())))

    else:

      nodes = None

      groups = None

    if query.IQ_NETWORKS in self.requested_data:

      net_uuids = itertools.chain(*(lu.cfg.GetInstanceNetworks(i.name)

                                    for i in instance_list))

      networks = dict((uuid, lu.cfg.GetNetwork(uuid)) for uuid in net_uuids)

    else:

      networks = None

    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),

                                   disk_usage, offline_nodes, bad_nodes,

                                   live_data, wrongnode_inst, consinfo,

                                   nodes, groups, networks)

class LUQuery(NoHooksLU):

  """Query for resources/items of a certain kind.

  """

  # pylint: disable=W0142

  REQ_BGL = False

  def CheckArguments(self):

    qcls = _GetQueryImplementation(self.op.what)

    self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)

  def ExpandNames(self):

    self.impl.ExpandNames(self)

  def DeclareLocks(self, level):

    self.impl.DeclareLocks(self, level)

  def Exec(self, feedback_fn):

    return self.impl.NewStyleQuery(self)

class LUQueryFields(NoHooksLU):

  """Query for resources/items of a certain kind.

  """

  # pylint: disable=W0142

  REQ_BGL = False

  def CheckArguments(self):

    self.qcls = _GetQueryImplementation(self.op.what)

  def ExpandNames(self):

    self.needed_locks = {}

  def Exec(self, feedback_fn):

    return query.QueryFields(self.qcls.FIELDS, self.op.fields)

class LUInstanceActivateDisks(NoHooksLU):

  """Bring up an instance's disks.

  """

  REQ_BGL = False

  def ExpandNames(self):

    self._ExpandAndLockInstance()

    self.needed_locks[locking.LEVEL_NODE] = []

    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE

  def DeclareLocks(self, level):

    if level == locking.LEVEL_NODE:

      self._LockInstancesNodes()

  def CheckPrereq(self):

    """Check prerequisites.

    This checks that the instance is in the cluster.

    """

    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)

    assert self.instance is not None, \

      "Cannot retrieve locked instance %s" % self.op.instance_name

    _CheckNodeOnline(self, self.instance.primary_node)

  def Exec(self, feedback_fn):

    """Activate the disks.

    """

    disks_ok, disks_info = \

              _AssembleInstanceDisks(self, self.instance,

                                     ignore_size=self.op.ignore_size)

    if not disks_ok:

      raise errors.OpExecError("Cannot activate block devices")

    if self.op.wait_for_sync:

      if not _WaitForSync(self, self.instance):

        raise errors.OpExecError("Some disks of the instance are degraded!")

    return disks_info

def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,

                           ignore_size=False):

  """Prepare the block devices for an instance.

  This sets up the block devices on all nodes.

  @type lu: L{LogicalUnit}

  @param lu: the logical unit on whose behalf we execute

  @type instance: L{objects.Instance}

  @param instance: the instance for whose disks we assemble

  @type disks: list of L{objects.Disk} or None

  @param disks: which disks to assemble (or all, if None)

  @type ignore_secondaries: boolean

  @param ignore_secondaries: if true, errors on secondary nodes

      won't result in an error return from the function

  @type ignore_size: boolean

  @param ignore_size: if true, the current known size of the disk

      will not be used during the disk activation, useful for cases

      when the size is wrong

  @return: False if the operation failed, otherwise a list of

      (host, instance_visible_name, node_visible_name)

      with the mapping from node devices to instance devices

  """

  device_info = []

  disks_ok = True

  iname = instance.name

  disks = _ExpandCheckDisks(instance, disks)

  # With the two passes mechanism we try to reduce the window of

  # opportunity for the race condition of switching DRBD to primary

  # before handshaking occured, but we do not eliminate it

  # The proper fix would be to wait (with some limits) until the

  # connection has been made and drbd transitions from WFConnection

  # into any other network-connected state (Connected, SyncTarget,

  # SyncSource, etc.)

  # 1st pass, assemble on all nodes in secondary mode

  for idx, inst_disk in enumerate(disks):

    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):

      if ignore_size:

        node_disk = node_disk.Copy()

        node_disk.UnsetSize()

      lu.cfg.SetDiskID(node_disk, node)

      result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,

                                             False, idx)

      msg = result.fail_msg

      if msg:

        is_offline_secondary = (node in instance.secondary_nodes and

                                result.offline)

        lu.LogWarning("Could not prepare block device %s on node %s"

                      " (is_primary=False, pass=1): %s",

                      inst_disk.iv_name, node, msg)

        if not (ignore_secondaries or is_offline_secondary):

          disks_ok = False

  # FIXME: race condition on drbd migration to primary

  # 2nd pass, do only the primary node

  for idx, inst_disk in enumerate(disks):

    dev_path = None

    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):

      if node != instance.primary_node:

        continue

      if ignore_size:

        node_disk = node_disk.Copy()

        node_disk.UnsetSize()

      lu.cfg.SetDiskID(node_disk, node)

      result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,

                                             True, idx)

      msg = result.fail_msg

      if msg:

        lu.LogWarning("Could not prepare block device %s on node %s"

                      " (is_primary=True, pass=2): %s",

                      inst_disk.iv_name, node, msg)

        disks_ok = False

      else:

        dev_path = result.payload

    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))

  # leave the disks configured for the primary node

  # this is a workaround that would be fixed better by

  # improving the logical/physical id handling

  for disk in disks:

    lu.cfg.SetDiskID(disk, instance.primary_node)

  return disks_ok, device_info

def _StartInstanceDisks(lu, instance, force):

  """Start the disks of an instance.

  """

  disks_ok, _ = _AssembleInstanceDisks(lu, instance,

                                           ignore_secondaries=force)

  if not disks_ok:

    _ShutdownInstanceDisks(lu, instance)

    if force is not None and not force:

      lu.LogWarning("",

                    hint=("If the message above refers to a secondary node,"

                          " you can retry the operation using '--force'"))

    raise errors.OpExecError("Disk consistency error")

class LUInstanceDeactivateDisks(NoHooksLU):

  """Shutdown an instance's disks.

  """

  REQ_BGL = False

  def ExpandNames(self):

    self._ExpandAndLockInstance()

    self.needed_locks[locking.LEVEL_NODE] = []

    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE

  def DeclareLocks(self, level):

    if level == locking.LEVEL_NODE:

      self._LockInstancesNodes()

  def CheckPrereq(self):

    """Check prerequisites.

    This checks that the instance is in the cluster.

    """

    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)

    assert self.instance is not None, \

      "Cannot retrieve locked instance %s" % self.op.instance_name

  def Exec(self, feedback_fn):

    """Deactivate the disks

    """

    instance = self.instance

    if self.op.force:

      _ShutdownInstanceDisks(self, instance)

    else:

      _SafeShutdownInstanceDisks(self, instance)

def _SafeShutdownInstanceDisks(lu, instance, disks=None):

  """Shutdown block devices of an instance.

  This function checks if an instance is running, before calling

  _ShutdownInstanceDisks.

  """

  _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")

  _ShutdownInstanceDisks(lu, instance, disks=disks)

def _ExpandCheckDisks(instance, disks):

  """Return the instance disks selected by the disks list

  @type disks: list of L{objects.Disk} or None

  @param disks: selected disks

  @rtype: list of L{objects.Disk}

  @return: selected instance disks to act on

  """

  if disks is None:

    return instance.disks

  else:

    if not set(disks).issubset(instance.disks):

      raise errors.ProgrammerError("Can only act on disks belonging to the"

                                   " target instance")

    return disks

def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):

  """Shutdown block devices of an instance.

  This does the shutdown on all nodes of the instance.

  If the ignore_primary is false, errors on the primary node are

  ignored.

  """

  all_result = True

  disks = _ExpandCheckDisks(instance, disks)

  for disk in disks:

    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):

      lu.cfg.SetDiskID(top_disk, node)

      result = lu.rpc.call_blockdev_shutdown(node, (top_disk, instance))

      msg = result.fail_msg

      if msg:

        lu.LogWarning("Could not shutdown block device %s on node %s: %s",

                      disk.iv_name, node, msg)

        if ((node == instance.primary_node and not ignore_primary) or

            (node != instance.primary_node and not result.offline)):

          all_result = False

  return all_result

def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):

  """Checks if a node has enough free memory.

  This function checks if a given node has the needed amount of free

  memory. In case the node has less memory or we cannot get the

  information from the node, this function raises an OpPrereqError

  exception.

  @type lu: C{LogicalUnit}

  @param lu: a logical unit from which we get configuration data

  @type node: C{str}

  @param node: the node to check

  @type reason: C{str}

  @param reason: string to use in the error message

  @type requested: C{int}

  @param requested: the amount of memory in MiB to check for

  @type hypervisor_name: C{str}

  @param hypervisor_name: the hypervisor to ask for memory stats

  @rtype: integer

  @return: node current free memory

  @raise errors.OpPrereqError: if the node doesn't have enough memory, or

      we cannot check the node

  """

  nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name], False)

  nodeinfo[node].Raise("Can't get data from node %s" % node,

                       prereq=True, ecode=errors.ECODE_ENVIRON)

  (_, _, (hv_info, )) = nodeinfo[node].payload

  free_mem = hv_info.get("memory_free", None)

  if not isinstance(free_mem, int):

    raise errors.OpPrereqError("Can't compute free memory on node %s, result"

                               " was '%s'" % (node, free_mem),

                               errors.ECODE_ENVIRON)

  if requested > free_mem:

    raise errors.OpPrereqError("Not enough memory on node %s for %s:"

                               " needed %s MiB, available %s MiB" %

                               (node, reason, requested, free_mem),

                               errors.ECODE_NORES)

  return free_mem

def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):

  """Checks if nodes have enough free disk space in all the VGs.

  This function checks if all given nodes have the needed amount of

  free disk. In case any node has less disk or we cannot get the

  information from the node, this function raises an OpPrereqError

  exception.

  @type lu: C{LogicalUnit}

  @param lu: a logical unit from which we get configuration data

  @type nodenames: C{list}

  @param nodenames: the list of node names to check

  @type req_sizes: C{dict}

  @param req_sizes: the hash of vg and corresponding amount of disk in

      MiB to check for

  @raise errors.OpPrereqError: if the node doesn't have enough disk,

      or we cannot check the node

  """

  for vg, req_size in req_sizes.items():

    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)

def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):

  """Checks if nodes have enough free disk space in the specified VG.

  This function checks if all given nodes have the needed amount of

  free disk. In case any node has less disk or we cannot get the

  information from the node, this function raises an OpPrereqError

  exception.

  @type lu: C{LogicalUnit}

  @param lu: a logical unit from which we get configuration data

  @type nodenames: C{list}

  @param nodenames: the list of node names to check

  @type vg: C{str}

  @param vg: the volume group to check

  @type requested: C{int}

  @param requested: the amount of disk in MiB to check for

  @raise errors.OpPrereqError: if the node doesn't have enough disk,

      or we cannot check the node

  """

  es_flags = rpc.GetExclusiveStorageForNodeNames(lu.cfg, nodenames)

  nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None, es_flags)

  for node in nodenames:

    info = nodeinfo[node]

    info.Raise("Cannot get current information from node %s" % node,

               prereq=True, ecode=errors.ECODE_ENVIRON)

    (_, (vg_info, ), _) = info.payload

    vg_free = vg_info.get("vg_free", None)

    if not isinstance(vg_free, int):

      raise errors.OpPrereqError("Can't compute free disk space on node"

                                 " %s for vg %s, result was '%s'" %

                                 (node, vg, vg_free), errors.ECODE_ENVIRON)

    if requested > vg_free:

      raise errors.OpPrereqError("Not enough disk space on target node %s"

                                 " vg %s: required %d MiB, available %d MiB" %

                                 (node, vg, requested, vg_free),

                                 errors.ECODE_NORES)

def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):

  """Checks if nodes have enough physical CPUs

  This function checks if all given nodes have the needed number of

  physical CPUs. In case any node has less CPUs or we cannot get the

  information from the node, this function raises an OpPrereqError

  exception.

  @type lu: C{LogicalUnit}

  @param lu: a logical unit from which we get configuration data

  @type nodenames: C{list}

  @param nodenames: the list of node names to check

  @type requested: C{int}

  @param requested: the minimum acceptable number of physical CPUs

  @raise errors.OpPrereqError: if the node doesn't have enough CPUs,

      or we cannot check the node

  """

  nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name], None)

  for node in nodenames:

    info = nodeinfo[node]

    info.Raise("Cannot get current information from node %s" % node,

               prereq=True, ecode=errors.ECODE_ENVIRON)

    (_, _, (hv_info, )) = info.payload

    num_cpus = hv_info.get("cpu_total", None)

    if not isinstance(num_cpus, int):

      raise errors.OpPrereqError("Can't compute the number of physical CPUs"

                                 " on node %s, result was '%s'" %

                                 (node, num_cpus), errors.ECODE_ENVIRON)

    if requested > num_cpus:

      raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "

                                 "required" % (node, num_cpus, requested),

                                 errors.ECODE_NORES)

class LUInstanceStartup(LogicalUnit):

  """Starts an instance.

  """

  HPATH = "instance-start"

  HTYPE = constants.HTYPE_INSTANCE

  REQ_BGL = False

  def CheckArguments(self):

    # extra beparams

    if self.op.beparams:

      # fill the beparams dict

      objects.UpgradeBeParams(self.op.beparams)

      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)

  def ExpandNames(self):

    self._ExpandAndLockInstance()

    self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE

  def DeclareLocks(self, level):