cmdlib.py

#
#

# Copyright (C) 2006, 2007, 2008 Google Inc.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
# 02110-1301, USA.


"""Module implementing the master-side code."""

# pylint: disable-msg=W0613,W0201

import os
import os.path
import sha
import time
import tempfile
import re
import platform
import logging
import copy
import random

from ganeti import ssh
from ganeti import utils
from ganeti import errors
from ganeti import hypervisor
from ganeti import locking
from ganeti import constants
from ganeti import objects
from ganeti import opcodes
from ganeti import serializer
from ganeti import ssconf


class LogicalUnit(object):
  """Logical Unit base class.

  Subclasses must follow these rules:
    - implement ExpandNames
    - implement CheckPrereq
    - implement Exec
    - implement BuildHooksEnv
    - redefine HPATH and HTYPE
    - optionally redefine their run requirements:
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively

  Note that all commands require root permissions.

  """
  HPATH = None
  HTYPE = None
  _OP_REQP = []
  REQ_BGL = True

  def __init__(self, processor, op, context, rpc):
    """Constructor for LogicalUnit.

    This needs to be overriden in derived classes in order to check op
    validity.

    """
    self.proc = processor
    self.op = op
    self.cfg = context.cfg
    self.context = context
    self.rpc = rpc
    # Dicts used to declare locking needs to mcpu
    self.needed_locks = None
    self.acquired_locks = {}
    self.share_locks = dict(((i, 0) for i in locking.LEVELS))
    self.add_locks = {}
    self.remove_locks = {}
    # Used to force good behavior when calling helper functions
    self.recalculate_locks = {}
    self.__ssh = None
    # logging
    self.LogWarning = processor.LogWarning
    self.LogInfo = processor.LogInfo

    for attr_name in self._OP_REQP:
      attr_val = getattr(op, attr_name, None)
      if attr_val is None:
        raise errors.OpPrereqError("Required parameter '%s' missing" %
                                   attr_name)
    self.CheckArguments()

  def __GetSSH(self):
    """Returns the SshRunner object

    """
    if not self.__ssh:
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
    return self.__ssh

  ssh = property(fget=__GetSSH)

  def CheckArguments(self):
    """Check syntactic validity for the opcode arguments.

    This method is for doing a simple syntactic check and ensure
    validity of opcode parameters, without any cluster-related
    checks. While the same can be accomplished in ExpandNames and/or
    CheckPrereq, doing these separate is better because:

      - ExpandNames is left as as purely a lock-related function
      - CheckPrereq is run after we have aquired locks (and possible
        waited for them)

    The function is allowed to change the self.op attribute so that
    later methods can no longer worry about missing parameters.

    """
    pass

  def ExpandNames(self):
    """Expand names for this LU.

    This method is called before starting to execute the opcode, and it should
    update all the parameters of the opcode to their canonical form (e.g. a
    short node name must be fully expanded after this method has successfully
    completed). This way locking, hooks, logging, ecc. can work correctly.

    LUs which implement this method must also populate the self.needed_locks
    member, as a dict with lock levels as keys, and a list of needed lock names
    as values. Rules:

      - use an empty dict if you don't need any lock
      - if you don't need any lock at a particular level omit that level
      - don't put anything for the BGL level
      - if you want all locks at a level use locking.ALL_SET as a value

    If you need to share locks (rather than acquire them exclusively) at one
    level you can modify self.share_locks, setting a true value (usually 1) for
    that level. By default locks are not shared.

    Examples::

      # Acquire all nodes and one instance
      self.needed_locks = {
        locking.LEVEL_NODE: locking.ALL_SET,
        locking.LEVEL_INSTANCE: ['instance1.example.tld'],
      }
      # Acquire just two nodes
      self.needed_locks = {
        locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
      }
      # Acquire no locks
      self.needed_locks = {} # No, you can't leave it to the default value None

    """
    # The implementation of this method is mandatory only if the new LU is
    # concurrent, so that old LUs don't need to be changed all at the same
    # time.
    if self.REQ_BGL:
      self.needed_locks = {} # Exclusive LUs don't need locks.
    else:
      raise NotImplementedError

  def DeclareLocks(self, level):
    """Declare LU locking needs for a level

    While most LUs can just declare their locking needs at ExpandNames time,
    sometimes there's the need to calculate some locks after having acquired
    the ones before. This function is called just before acquiring locks at a
    particular level, but after acquiring the ones at lower levels, and permits
    such calculations. It can be used to modify self.needed_locks, and by
    default it does nothing.

    This function is only called if you have something already set in
    self.needed_locks for the level.

    @param level: Locking level which is going to be locked
    @type level: member of ganeti.locking.LEVELS

    """

  def CheckPrereq(self):
    """Check prerequisites for this LU.

    This method should check that the prerequisites for the execution
    of this LU are fulfilled. It can do internode communication, but
    it should be idempotent - no cluster or system changes are
    allowed.

    The method should raise errors.OpPrereqError in case something is
    not fulfilled. Its return value is ignored.

    This method should also update all the parameters of the opcode to
    their canonical form if it hasn't been done by ExpandNames before.

    """
    raise NotImplementedError

  def Exec(self, feedback_fn):
    """Execute the LU.

    This method should implement the actual work. It should raise
    errors.OpExecError for failures that are somewhat dealt with in
    code, or expected.

    """
    raise NotImplementedError

  def BuildHooksEnv(self):
    """Build hooks environment for this LU.

    This method should return a three-node tuple consisting of: a dict
    containing the environment that will be used for running the
    specific hook for this LU, a list of node names on which the hook
    should run before the execution, and a list of node names on which
    the hook should run after the execution.

    The keys of the dict must not have 'GANETI_' prefixed as this will
    be handled in the hooks runner. Also note additional keys will be
    added by the hooks runner. If the LU doesn't define any
    environment, an empty dict (and not None) should be returned.

    No nodes should be returned as an empty list (and not None).

    Note that if the HPATH for a LU class is None, this function will
    not be called.

    """
    raise NotImplementedError

  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
    """Notify the LU about the results of its hooks.

    This method is called every time a hooks phase is executed, and notifies
    the Logical Unit about the hooks' result. The LU can then use it to alter
    its result based on the hooks.  By default the method does nothing and the
    previous result is passed back unchanged but any LU can define it if it
    wants to use the local cluster hook-scripts somehow.

    @param phase: one of L{constants.HOOKS_PHASE_POST} or
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
    @param hook_results: the results of the multi-node hooks rpc call
    @param feedback_fn: function used send feedback back to the caller
    @param lu_result: the previous Exec result this LU had, or None
        in the PRE phase
    @return: the new Exec result, based on the previous result
        and hook results

    """
    return lu_result

  def _ExpandAndLockInstance(self):
    """Helper function to expand and lock an instance.

    Many LUs that work on an instance take its name in self.op.instance_name
    and need to expand it and then declare the expanded name for locking. This
    function does it, and then updates self.op.instance_name to the expanded
    name. It also initializes needed_locks as a dict, if this hasn't been done
    before.

    """
    if self.needed_locks is None:
      self.needed_locks = {}
    else:
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
        "_ExpandAndLockInstance called with instance-level locks set"
    expanded_name = self.cfg.ExpandInstanceName(self.op.instance_name)
    if expanded_name is None:
      raise errors.OpPrereqError("Instance '%s' not known" %
                                  self.op.instance_name)
    self.needed_locks[locking.LEVEL_INSTANCE] = expanded_name
    self.op.instance_name = expanded_name

  def _LockInstancesNodes(self, primary_only=False):
    """Helper function to declare instances' nodes for locking.

    This function should be called after locking one or more instances to lock
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
    with all primary or secondary nodes for instances already locked and
    present in self.needed_locks[locking.LEVEL_INSTANCE].

    It should be called from DeclareLocks, and for safety only works if
    self.recalculate_locks[locking.LEVEL_NODE] is set.

    In the future it may grow parameters to just lock some instance's nodes, or
    to just lock primaries or secondary nodes, if needed.

    If should be called in DeclareLocks in a way similar to::

      if level == locking.LEVEL_NODE:
        self._LockInstancesNodes()

    @type primary_only: boolean
    @param primary_only: only lock primary nodes of locked instances

    """
    assert locking.LEVEL_NODE in self.recalculate_locks, \
      "_LockInstancesNodes helper function called with no nodes to recalculate"

    # TODO: check if we're really been called with the instance locks held

    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
    # future we might want to have different behaviors depending on the value
    # of self.recalculate_locks[locking.LEVEL_NODE]
    wanted_nodes = []
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
      instance = self.context.cfg.GetInstanceInfo(instance_name)
      wanted_nodes.append(instance.primary_node)
      if not primary_only:
        wanted_nodes.extend(instance.secondary_nodes)

    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)

    del self.recalculate_locks[locking.LEVEL_NODE]


class NoHooksLU(LogicalUnit):
  """Simple LU which runs no hooks.

  This LU is intended as a parent for other LogicalUnits which will
  run no hooks, in order to reduce duplicate code.

  """
  HPATH = None
  HTYPE = None


def _GetWantedNodes(lu, nodes):
  """Returns list of checked and expanded node names.

  @type lu: L{LogicalUnit}
  @param lu: the logical unit on whose behalf we execute
  @type nodes: list
  @param nodes: list of node names or None for all nodes
  @rtype: list
  @return: the list of nodes, sorted
  @raise errors.OpProgrammerError: if the nodes parameter is wrong type

  """
  if not isinstance(nodes, list):
    raise errors.OpPrereqError("Invalid argument type 'nodes'")

  if not nodes:
    raise errors.ProgrammerError("_GetWantedNodes should only be called with a"
      " non-empty list of nodes whose name is to be expanded.")

  wanted = []
  for name in nodes:
    node = lu.cfg.ExpandNodeName(name)
    if node is None:
      raise errors.OpPrereqError("No such node name '%s'" % name)
    wanted.append(node)

  return utils.NiceSort(wanted)


def _GetWantedInstances(lu, instances):
  """Returns list of checked and expanded instance names.

  @type lu: L{LogicalUnit}
  @param lu: the logical unit on whose behalf we execute
  @type instances: list
  @param instances: list of instance names or None for all instances
  @rtype: list
  @return: the list of instances, sorted
  @raise errors.OpPrereqError: if the instances parameter is wrong type
  @raise errors.OpPrereqError: if any of the passed instances is not found

  """
  if not isinstance(instances, list):
    raise errors.OpPrereqError("Invalid argument type 'instances'")

  if instances:
    wanted = []

    for name in instances:
      instance = lu.cfg.ExpandInstanceName(name)
      if instance is None:
        raise errors.OpPrereqError("No such instance name '%s'" % name)
      wanted.append(instance)

  else:
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
  return wanted


def _CheckOutputFields(static, dynamic, selected):
  """Checks whether all selected fields are valid.

  @type static: L{utils.FieldSet}
  @param static: static fields set
  @type dynamic: L{utils.FieldSet}
  @param dynamic: dynamic fields set

  """
  f = utils.FieldSet()
  f.Extend(static)
  f.Extend(dynamic)

  delta = f.NonMatching(selected)
  if delta:
    raise errors.OpPrereqError("Unknown output fields selected: %s"
                               % ",".join(delta))


def _CheckBooleanOpField(op, name):
  """Validates boolean opcode parameters.

  This will ensure that an opcode parameter is either a boolean value,
  or None (but that it always exists).

  """
  val = getattr(op, name, None)
  if not (val is None or isinstance(val, bool)):
    raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
                               (name, str(val)))
  setattr(op, name, val)


def _CheckNodeOnline(lu, node):
  """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
  @raise errors.OpPrereqError: if the node is offline

  """
  if lu.cfg.GetNodeInfo(node).offline:
    raise errors.OpPrereqError("Can't use offline node %s" % node)


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)


def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
                          memory, vcpus, nics):
  """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: boolean
  @param status: the should_run status of the instance
  @type memory: string
  @param memory: the 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 (ip, bridge, mac) representing
      the NICs the instance  has
  @rtype: dict
  @return: the hook environment for this instance

  """
  if status:
    str_status = "up"
  else:
    str_status = "down"
  env = {
    "OP_TARGET": name,
    "INSTANCE_NAME": name,
    "INSTANCE_PRIMARY": primary_node,
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
    "INSTANCE_OS_TYPE": os_type,
    "INSTANCE_STATUS": str_status,
    "INSTANCE_MEMORY": memory,
    "INSTANCE_VCPUS": vcpus,
  }

  if nics:
    nic_count = len(nics)
    for idx, (ip, bridge, mac) in enumerate(nics):
      if ip is None:
        ip = ""
      env["INSTANCE_NIC%d_IP" % idx] = ip
      env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
      env["INSTANCE_NIC%d_HWADDR" % idx] = mac
  else:
    nic_count = 0

  env["INSTANCE_NIC_COUNT"] = nic_count

  return env


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

  """
  bep = lu.cfg.GetClusterInfo().FillBE(instance)
  args = {
    'name': instance.name,
    'primary_node': instance.primary_node,
    'secondary_nodes': instance.secondary_nodes,
    'os_type': instance.os,
    'status': instance.admin_up,
    'memory': bep[constants.BE_MEMORY],
    'vcpus': bep[constants.BE_VCPUS],
    'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
  }
  if override:
    args.update(override)
  return _BuildInstanceHookEnv(**args)


def _AdjustCandidatePool(lu):
  """Adjust the candidate pool after node operations.

  """
  mod_list = lu.cfg.MaintainCandidatePool()
  if mod_list:
    lu.LogInfo("Promoted nodes to master candidate role: %s",
               ", ".join(node.name for node in mod_list))
    for name in mod_list:
      lu.context.ReaddNode(name)
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
  if mc_now > mc_max:
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
               (mc_now, mc_max))


def _CheckInstanceBridgesExist(lu, instance):
  """Check that the brigdes needed by an instance exist.

  """
  # check bridges existance
  brlist = [nic.bridge for nic in instance.nics]
  result = lu.rpc.call_bridges_exist(instance.primary_node, brlist)
  result.Raise()
  if not result.data:
    raise errors.OpPrereqError("One or more target bridges %s does not"
                               " exist on destination node '%s'" %
                               (brlist, instance.primary_node))


class LUDestroyCluster(NoHooksLU):
  """Logical unit for destroying the cluster.

  """
  _OP_REQP = []

  def CheckPrereq(self):
    """Check prerequisites.

    This checks whether the cluster is empty.

    Any errors are signalled by raising errors.OpPrereqError.

    """
    master = self.cfg.GetMasterNode()

    nodelist = self.cfg.GetNodeList()
    if len(nodelist) != 1 or nodelist[0] != master:
      raise errors.OpPrereqError("There are still %d node(s) in"
                                 " this cluster." % (len(nodelist) - 1))
    instancelist = self.cfg.GetInstanceList()
    if instancelist:
      raise errors.OpPrereqError("There are still %d instance(s) in"
                                 " this cluster." % len(instancelist))

  def Exec(self, feedback_fn):
    """Destroys the cluster.

    """
    master = self.cfg.GetMasterNode()
    result = self.rpc.call_node_stop_master(master, False)
    result.Raise()
    if not result.data:
      raise errors.OpExecError("Could not disable the master role")
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
    utils.CreateBackup(priv_key)
    utils.CreateBackup(pub_key)
    return master


class LUVerifyCluster(LogicalUnit):
  """Verifies the cluster status.

  """
  HPATH = "cluster-verify"
  HTYPE = constants.HTYPE_CLUSTER
  _OP_REQP = ["skip_checks"]
  REQ_BGL = False

  def ExpandNames(self):
    self.needed_locks = {
      locking.LEVEL_NODE: locking.ALL_SET,
      locking.LEVEL_INSTANCE: locking.ALL_SET,
    }
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))

  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
                  node_result, feedback_fn, master_files,
                  drbd_map):
    """Run multiple tests against a node.

    Test list:

      - compares ganeti version
      - checks vg existance and size > 20G
      - checks config file checksum
      - checks ssh to other nodes

    @type nodeinfo: L{objects.Node}
    @param nodeinfo: the node to check
    @param file_list: required list of files
    @param local_cksum: dictionary of local files and their checksums
    @param node_result: the results from the node
    @param feedback_fn: function used to accumulate results
    @param master_files: list of files that only masters should have
    @param drbd_map: the useddrbd minors for this node, in
        form of minor: (instance, must_exist) which correspond to instances
        and their running status

    """
    node = nodeinfo.name

    # main result, node_result should be a non-empty dict
    if not node_result or not isinstance(node_result, dict):
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
      return True

    # compares ganeti version
    local_version = constants.PROTOCOL_VERSION
    remote_version = node_result.get('version', None)
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
            len(remote_version) == 2):
      feedback_fn("  - ERROR: connection to %s failed" % (node))
      return True

    if local_version != remote_version[0]:
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
                  " node %s %s" % (local_version, node, remote_version[0]))
      return True

    # node seems compatible, we can actually try to look into its results

    bad = False

    # full package version
    if constants.RELEASE_VERSION != remote_version[1]:
      feedback_fn("  - WARNING: software version mismatch: master %s,"
                  " node %s %s" %
                  (constants.RELEASE_VERSION, node, remote_version[1]))

    # checks vg existence and size > 20G

    vglist = node_result.get(constants.NV_VGLIST, None)
    if not vglist:
      feedback_fn("  - ERROR: unable to check volume groups on node %s." %
                      (node,))
      bad = True
    else:
      vgstatus = utils.CheckVolumeGroupSize(vglist, self.cfg.GetVGName(),
                                            constants.MIN_VG_SIZE)
      if vgstatus:
        feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
        bad = True

    # checks config file checksum

    remote_cksum = node_result.get(constants.NV_FILELIST, None)
    if not isinstance(remote_cksum, dict):
      bad = True
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
    else:
      for file_name in file_list:
        node_is_mc = nodeinfo.master_candidate
        must_have_file = file_name not in master_files
        if file_name not in remote_cksum:
          if node_is_mc or must_have_file:
            bad = True
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
        elif remote_cksum[file_name] != local_cksum[file_name]:
          if node_is_mc or must_have_file:
            bad = True
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
          else:
            # not candidate and this is not a must-have file
            bad = True
            feedback_fn("  - ERROR: non master-candidate has old/wrong file"
                        " '%s'" % file_name)
        else:
          # all good, except non-master/non-must have combination
          if not node_is_mc and not must_have_file:
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
                        " candidates" % file_name)

    # checks ssh to any

    if constants.NV_NODELIST not in node_result:
      bad = True
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
    else:
      if node_result[constants.NV_NODELIST]:
        bad = True
        for node in node_result[constants.NV_NODELIST]:
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
                          (node, node_result[constants.NV_NODELIST][node]))

    if constants.NV_NODENETTEST not in node_result:
      bad = True
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
    else:
      if node_result[constants.NV_NODENETTEST]:
        bad = True
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
        for node in nlist:
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
                          (node, node_result[constants.NV_NODENETTEST][node]))

    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
    if isinstance(hyp_result, dict):
      for hv_name, hv_result in hyp_result.iteritems():
        if hv_result is not None:
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
                      (hv_name, hv_result))

    # check used drbd list
    used_minors = node_result.get(constants.NV_DRBDLIST, [])
    if not isinstance(used_minors, (tuple, list)):
      feedback_fn("  - ERROR: cannot parse drbd status file: %s" %
                  str(used_minors))
    else:
      for minor, (iname, must_exist) in drbd_map.items():
        if minor not in used_minors and must_exist:
          feedback_fn("  - ERROR: drbd minor %d of instance %s is not active" %
                      (minor, iname))
          bad = True
      for minor in used_minors:
        if minor not in drbd_map:
          feedback_fn("  - ERROR: unallocated drbd minor %d is in use" % minor)
          bad = True

    return bad

  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
                      node_instance, feedback_fn, n_offline):
    """Verify an instance.

    This function checks to see if the required block devices are
    available on the instance's node.

    """
    bad = False

    node_current = instanceconfig.primary_node

    node_vol_should = {}
    instanceconfig.MapLVsByNode(node_vol_should)

    for node in node_vol_should:
      if node in n_offline:
        # ignore missing volumes on offline nodes
        continue
      for volume in node_vol_should[node]:
        if node not in node_vol_is or volume not in node_vol_is[node]:
          feedback_fn("  - ERROR: volume %s missing on node %s" %
                          (volume, node))
          bad = True

    if instanceconfig.admin_up:
      if ((node_current not in node_instance or
          not instance in node_instance[node_current]) and
          node_current not in n_offline):
        feedback_fn("  - ERROR: instance %s not running on node %s" %
                        (instance, node_current))
        bad = True

    for node in node_instance:
      if (not node == node_current):
        if instance in node_instance[node]:
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
                          (instance, node))
          bad = True

    return bad

  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
    """Verify if there are any unknown volumes in the cluster.

    The .os, .swap and backup volumes are ignored. All other volumes are
    reported as unknown.

    """
    bad = False

    for node in node_vol_is:
      for volume in node_vol_is[node]:
        if node not in node_vol_should or volume not in node_vol_should[node]:
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
                      (volume, node))
          bad = True
    return bad

  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
    """Verify the list of running instances.

    This checks what instances are running but unknown to the cluster.

    """
    bad = False
    for node in node_instance:
      for runninginstance in node_instance[node]:
        if runninginstance not in instancelist:
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
                          (runninginstance, node))
          bad = True
    return bad

  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
    """Verify N+1 Memory Resilience.

    Check that if one single node dies we can still start all the instances it
    was primary for.

    """
    bad = False

    for node, nodeinfo in node_info.iteritems():
      # This code checks that every node which is now listed as secondary has
      # enough memory to host all instances it is supposed to should a single
      # other node in the cluster fail.
      # FIXME: not ready for failover to an arbitrary node
      # FIXME: does not support file-backed instances
      # WARNING: we currently take into account down instances as well as up
      # ones, considering that even if they're down someone might want to start
      # them even in the event of a node failure.
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
        needed_mem = 0
        for instance in instances:
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
          if bep[constants.BE_AUTO_BALANCE]:
            needed_mem += bep[constants.BE_MEMORY]
        if nodeinfo['mfree'] < needed_mem:
          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
                      " failovers should node %s fail" % (node, prinode))
          bad = True
    return bad

  def CheckPrereq(self):
    """Check prerequisites.

    Transform the list of checks we're going to skip into a set and check that
    all its members are valid.

    """
    self.skip_set = frozenset(self.op.skip_checks)
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
      raise errors.OpPrereqError("Invalid checks to be skipped specified")

  def BuildHooksEnv(self):
    """Build hooks env.

    Cluster-Verify hooks just rone in the post phase and their failure makes
    the output be logged in the verify output and the verification to fail.

    """
    all_nodes = self.cfg.GetNodeList()
    # TODO: populate the environment with useful information for verify hooks
    env = {}
    return env, [], all_nodes

  def Exec(self, feedback_fn):
    """Verify integrity of cluster, performing various test on nodes.

    """
    bad = False
    feedback_fn("* Verifying global settings")
    for msg in self.cfg.VerifyConfig():
      feedback_fn("  - ERROR: %s" % msg)

    vg_name = self.cfg.GetVGName()
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
                        for iname in instancelist)
    i_non_redundant = [] # Non redundant instances
    i_non_a_balanced = [] # Non auto-balanced instances
    n_offline = [] # List of offline nodes
    n_drained = [] # List of nodes being drained
    node_volume = {}
    node_instance = {}
    node_info = {}
    instance_cfg = {}

    # FIXME: verify OS list
    # do local checksums
    master_files = [constants.CLUSTER_CONF_FILE]

    file_names = ssconf.SimpleStore().GetFileList()
    file_names.append(constants.SSL_CERT_FILE)
    file_names.append(constants.RAPI_CERT_FILE)
    file_names.extend(master_files)

    local_checksums = utils.FingerprintFiles(file_names)

    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
    node_verify_param = {
      constants.NV_FILELIST: file_names,
      constants.NV_NODELIST: [node.name for node in nodeinfo
                              if not node.offline],
      constants.NV_HYPERVISOR: hypervisors,
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
                                  node.secondary_ip) for node in nodeinfo
                                 if not node.offline],
      constants.NV_LVLIST: vg_name,
      constants.NV_INSTANCELIST: hypervisors,
      constants.NV_VGLIST: None,
      constants.NV_VERSION: None,
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
      constants.NV_DRBDLIST: None,
      }
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
                                           self.cfg.GetClusterName())

    cluster = self.cfg.GetClusterInfo()
    master_node = self.cfg.GetMasterNode()
    all_drbd_map = self.cfg.ComputeDRBDMap()

    for node_i in nodeinfo:
      node = node_i.name
      nresult = all_nvinfo[node].data

      if node_i.offline:
        feedback_fn("* Skipping offline node %s" % (node,))
        n_offline.append(node)
        continue

      if node == master_node:
        ntype = "master"
      elif node_i.master_candidate:
        ntype = "master candidate"
      elif node_i.drained:
        ntype = "drained"
        n_drained.append(node)
      else:
        ntype = "regular"
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))

      if all_nvinfo[node].failed or not isinstance(nresult, dict):
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
        bad = True
        continue

      node_drbd = {}
      for minor, instance in all_drbd_map[node].items():
        instance = instanceinfo[instance]
        node_drbd[minor] = (instance.name, instance.admin_up)
      result = self._VerifyNode(node_i, file_names, local_checksums,
                                nresult, feedback_fn, master_files,
                                node_drbd)
      bad = bad or result

      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
      if isinstance(lvdata, basestring):
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
                    (node, utils.SafeEncode(lvdata)))
        bad = True
        node_volume[node] = {}
      elif not isinstance(lvdata, dict):
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
        bad = True
        continue
      else:
        node_volume[node] = lvdata

      # node_instance
      idata = nresult.get(constants.NV_INSTANCELIST, None)
      if not isinstance(idata, list):
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
                    (node,))
        bad = True
        continue

      node_instance[node] = idata

      # node_info
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
      if not isinstance(nodeinfo, dict):
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
        bad = True
        continue

      try:
        node_info[node] = {
          "mfree": int(nodeinfo['memory_free']),
          "dfree": int(nresult[constants.NV_VGLIST][vg_name]),
          "pinst": [],
          "sinst": [],
          # dictionary holding all instances this node is secondary for,
          # grouped by their primary node. Each key is a cluster node, and each
          # value is a list of instances which have the key as primary and the
          # current node as secondary.  this is handy to calculate N+1 memory
          # availability if you can only failover from a primary to its
          # secondary.
          "sinst-by-pnode": {},
        }
      except ValueError:
        feedback_fn("  - ERROR: invalid value returned from node %s" % (node,))
        bad = True
        continue

    node_vol_should = {}

    for instance in instancelist:
      feedback_fn("* Verifying instance %s" % instance)
      inst_config = instanceinfo[instance]
      result =  self._VerifyInstance(instance, inst_config, node_volume,
                                     node_instance, feedback_fn, n_offline)
      bad = bad or result
      inst_nodes_offline = []

      inst_config.MapLVsByNode(node_vol_should)

      instance_cfg[instance] = inst_config

      pnode = inst_config.primary_node
      if pnode in node_info:
        node_info[pnode]['pinst'].append(instance)
      elif pnode not in n_offline:
        feedback_fn("  - ERROR: instance %s, connection to primary node"
                    " %s failed" % (instance, pnode))
        bad = True

      if pnode in n_offline:
        inst_nodes_offline.append(pnode)

      # If the instance is non-redundant we cannot survive losing its primary
      # node, so we are not N+1 compliant. On the other hand we have no disk
      # templates with more than one secondary so that situation is not well
      # supported either.
      # FIXME: does not support file-backed instances
      if len(inst_config.secondary_nodes) == 0:
        i_non_redundant.append(instance)
      elif len(inst_config.secondary_nodes) > 1:
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
                    % instance)

      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
        i_non_a_balanced.append(instance)

      for snode in inst_config.secondary_nodes:
        if snode in node_info:
          node_info[snode]['sinst'].append(instance)
          if pnode not in node_info[snode]['sinst-by-pnode']:
            node_info[snode]['sinst-by-pnode'][pnode] = []
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
        elif snode not in n_offline:
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
                      " %s failed" % (instance, snode))
          bad = True
        if snode in n_offline:
          inst_nodes_offline.append(snode)

      if inst_nodes_offline:
        # warn that the instance lives on offline nodes, and set bad=True
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
                    ", ".join(inst_nodes_offline))
        bad = True

    feedback_fn("* Verifying orphan volumes")
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
                                       feedback_fn)
    bad = bad or result

    feedback_fn("* Verifying remaining instances")
    result = self._VerifyOrphanInstances(instancelist, node_instance,
                                         feedback_fn)
    bad = bad or result

    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
      feedback_fn("* Verifying N+1 Memory redundancy")
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
      bad = bad or result

    feedback_fn("* Other Notes")
    if i_non_redundant:
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
                  % len(i_non_redundant))

    if i_non_a_balanced:
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
                  % len(i_non_a_balanced))

    if n_offline:
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))

    if n_drained:
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))

    return not bad

  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
    """Analize the post-hooks' result

    This method analyses the hook result, handles it, and sends some
    nicely-formatted feedback back to the user.

    @param phase: one of L{constants.HOOKS_PHASE_POST} or
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
    @param hooks_results: the results of the multi-node hooks rpc call
    @param feedback_fn: function used send feedback back to the caller
    @param lu_result: previous Exec result
    @return: the new Exec result, based on the previous result
        and hook results

    """
    # We only really run POST phase hooks, and are only interested in
    # their results
    if phase == constants.HOOKS_PHASE_POST:
      # Used to change hooks' output to proper indentation
      indent_re = re.compile('^', re.M)
      feedback_fn("* Hooks Results")
      if not hooks_results:
        feedback_fn("  - ERROR: general communication failure")
        lu_result = 1
      else:
        for node_name in hooks_results:
          show_node_header = True
          res = hooks_results[node_name]
          if res.failed or res.data is False or not isinstance(res.data, list):
            if res.offline:
              # no need to warn or set fail return value
              continue
            feedback_fn("    Communication failure in hooks execution")
            lu_result = 1
            continue
          for script, hkr, output in res.data:
            if hkr == constants.HKR_FAIL:
              # The node header is only shown once, if there are
              # failing hooks on that node
              if show_node_header:
                feedback_fn("  Node %s:" % node_name)
                show_node_header = False
              feedback_fn("    ERROR: Script %s failed, output:" % script)
              output = indent_re.sub('      ', output)
              feedback_fn("%s" % output)
              lu_result = 1

      return lu_result


class LUVerifyDisks(NoHooksLU):
  """Verifies the cluster disks status.

  """
  _OP_REQP = []
  REQ_BGL = False

  def ExpandNames(self):
    self.needed_locks = {
      locking.LEVEL_NODE: locking.ALL_SET,
      locking.LEVEL_INSTANCE: locking.ALL_SET,
    }
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))

  def CheckPrereq(self):
    """Check prerequisites.

    This has no prerequisites.

    """
    pass

  def Exec(self, feedback_fn):
    """Verify integrity of cluster disks.

    """
    result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}

    vg_name = self.cfg.GetVGName()
    nodes = utils.NiceSort(self.cfg.GetNodeList())
    instances = [self.cfg.GetInstanceInfo(name)
                 for name in self.cfg.GetInstanceList()]

    nv_dict = {}
    for inst in instances:
      inst_lvs = {}
      if (not inst.admin_up or
          inst.disk_template not in constants.DTS_NET_MIRROR):
        continue
      inst.MapLVsByNode(inst_lvs)
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
      for node, vol_list in inst_lvs.iteritems():
        for vol in vol_list:
          nv_dict[(node, vol)] = inst

    if not nv_dict:
      return result

    node_lvs = self.rpc.call_volume_list(nodes, vg_name)

    to_act = set()
    for node in nodes:
      # node_volume
      lvs = node_lvs[node]
      if lvs.failed:
        if not lvs.offline:
          self.LogWarning("Connection to node %s failed: %s" %
                          (node, lvs.data))
        continue
      lvs = lvs.data
      if isinstance(lvs, basestring):
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
        res_nlvm[node] = lvs
      elif not isinstance(lvs, dict):
        logging.warning("Connection to node %s failed or invalid data"
                        " returned", node)
        res_nodes.append(node)
        continue

      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
        inst = nv_dict.pop((node, lv_name), None)
        if (not lv_online and inst is not None
            and inst.name not in res_instances):
          res_instances.append(inst.name)