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 time
import re
import platform
import logging
import copy

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 serializer
from ganeti import ssconf


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

  Subclasses must follow these rules:
    - implement ExpandNames
    - implement CheckPrereq (except when tasklets are used)
    - implement Exec (except when tasklets are used)
    - 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.

  @ivar dry_run_result: the value (if any) that will be returned to the caller
      in dry-run mode (signalled by opcode dry_run parameter)

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

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

    This needs to be overridden 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.fromkeys(locking.LEVELS, 0)
    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
    self.LogStep = processor.LogStep
    # support for dry-run
    self.dry_run_result = None

    # Tasklets
    self.tasklets = None

    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 acquired 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.

    This function can also define a list of tasklets, which then will be
    executed in order instead of the usual LU-level CheckPrereq and Exec
    functions, if those are not defined by the LU.

    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.

    """
    if self.tasklets is not None:
      for (idx, tl) in enumerate(self.tasklets):
        logging.debug("Checking prerequisites for tasklet %s/%s",
                      idx + 1, len(self.tasklets))
        tl.CheckPrereq()
    else:
      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.

    """
    if self.tasklets is not None:
      for (idx, tl) in enumerate(self.tasklets):
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
        tl.Exec(feedback_fn)
    else:
      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


class Tasklet:
  """Tasklet base class.

  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
  tasklets know nothing about locks.

  Subclasses must follow these rules:
    - Implement CheckPrereq
    - Implement Exec

  """
  def __init__(self, lu):
    self.lu = lu

    # Shortcuts
    self.cfg = lu.cfg
    self.rpc = lu.rpc

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

    This method should check whether the prerequisites for the execution of
    this tasklet 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 parameters to their canonical form if it
    hasn't been done before.

    """
    raise NotImplementedError

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

    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 _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, disk_template, disks,
                          bep, hvp, hypervisor_name):
  """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, mac, mode, link) representing
      the NICs the instance has
  @type disk_template: string
  @param disk_template: the disk template of the instance
  @type disks: list
  @param disks: the list of (size, mode) pairs
  @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
  @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,
    "INSTANCE_DISK_TEMPLATE": disk_template,
    "INSTANCE_HYPERVISOR": hypervisor_name,
  }

  if nics:
    nic_count = len(nics)
    for idx, (ip, mac, mode, link) in enumerate(nics):
      if ip is None:
        ip = ""
      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 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, (size, mode) in enumerate(disks):
      env["INSTANCE_DISK%d_SIZE" % idx] = size
      env["INSTANCE_DISK%d_MODE" % idx] = mode
  else:
    disk_count = 0

  env["INSTANCE_DISK_COUNT"] = disk_count

  for source, kind in [(bep, "BE"), (hvp, "HV")]:
    for key, value in source.items():
      env["INSTANCE_%s_%s" % (kind, key)] = value

  return env


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 LUQueryInstanceData.

  @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 = []
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[constants.PP_DEFAULT]
  for nic in nics:
    ip = nic.ip
    mac = nic.mac
    filled_params = objects.FillDict(c_nicparams, nic.nicparams)
    mode = filled_params[constants.NIC_MODE]
    link = filled_params[constants.NIC_LINK]
    hooks_nics.append((ip, mac, mode, link))
  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_up,
    'memory': bep[constants.BE_MEMORY],
    'vcpus': bep[constants.BE_VCPUS],
    'nics': _NICListToTuple(lu, instance.nics),
    'disk_template': instance.disk_template,
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
    'bep': bep,
    'hvp': hvp,
    'hypervisor_name': instance.hypervisor,
  }
  if override:
    args.update(override)
  return _BuildInstanceHookEnv(**args)


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

  """
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
  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(exceptions)
  if mc_now > mc_max:
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
               (mc_now, mc_max))


def _CheckNicsBridgesExist(lu, target_nics, target_node,
                               profile=constants.PP_DEFAULT):
  """Check that the brigdes needed by a list of nics exist.

  """
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[profile]
  paramslist = [objects.FillDict(c_nicparams, 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)


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 _GetNodeInstancesInner(cfg, fn):
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]


def _GetNodeInstances(cfg, node_name):
  """Returns a list of all primary and secondary instances on a node.

  """

  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)


def _GetNodePrimaryInstances(cfg, node_name):
  """Returns primary instances on a node.

  """
  return _GetNodeInstancesInner(cfg,
                                lambda inst: node_name == inst.primary_node)


def _GetNodeSecondaryInstances(cfg, node_name):
  """Returns secondary instances on a node.

  """
  return _GetNodeInstancesInner(cfg,
                                lambda inst: node_name in inst.secondary_nodes)


def _GetStorageTypeArgs(cfg, storage_type):
  """Returns the arguments for a storage type.

  """
  # Special case for file storage
  if storage_type == constants.ST_FILE:
    # storage.FileStorage wants a list of storage directories
    return [[cfg.GetFileStorageDir()]]

  return []


def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
  faulty = []

  for dev in instance.disks:
    cfg.SetDiskID(dev, node_name)

  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
  result.Raise("Failed to get disk status from node %s" % node_name,
               prereq=prereq)

  for idx, bdev_status in enumerate(result.payload):
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
      faulty.append(idx)

  return faulty


class LUPostInitCluster(LogicalUnit):
  """Logical unit for running hooks after cluster initialization.

  """
  HPATH = "cluster-init"
  HTYPE = constants.HTYPE_CLUSTER
  _OP_REQP = []

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

    """
    env = {"OP_TARGET": self.cfg.GetClusterName()}
    mn = self.cfg.GetMasterNode()
    return env, [], [mn]

  def CheckPrereq(self):
    """No prerequisites to check.

    """
    return True

  def Exec(self, feedback_fn):
    """Nothing to do.

    """
    return True


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

  """
  HPATH = "cluster-destroy"
  HTYPE = constants.HTYPE_CLUSTER
  _OP_REQP = []

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

    """
    env = {"OP_TARGET": self.cfg.GetClusterName()}
    return env, [], []

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

    This checks whether the cluster is empty.

    Any errors are signaled 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()

    # Run post hooks on master node before it's removed
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
    try:
      hm.RunPhase(constants.HOOKS_PHASE_POST, [master])
    except:
      self.LogWarning("Errors occurred running hooks on %s" % master)

    result = self.rpc.call_node_stop_master(master, False)
    result.Raise("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", "verbose", "error_codes", "debug_simulate_errors"]
  REQ_BGL = False

  TCLUSTER = "cluster"
  TNODE = "node"
  TINSTANCE = "instance"

  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
  ENODEDRBD = (TNODE, "ENODEDRBD")
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
  ENODEHV = (TNODE, "ENODEHV")
  ENODELVM = (TNODE, "ENODELVM")
  ENODEN1 = (TNODE, "ENODEN1")
  ENODENET = (TNODE, "ENODENET")
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
  ENODERPC = (TNODE, "ENODERPC")
  ENODESSH = (TNODE, "ENODESSH")
  ENODEVERSION = (TNODE, "ENODEVERSION")

  ETYPE_FIELD = "code"
  ETYPE_ERROR = "ERROR"
  ETYPE_WARNING = "WARNING"

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

  def _Error(self, ecode, item, msg, *args, **kwargs):
    """Format an error message.

    Based on the opcode's error_codes parameter, either format a
    parseable error code, or a simpler error string.

    This must be called only from Exec and functions called from Exec.

    """
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
    itype, etxt = ecode
    # first complete the msg
    if args:
      msg = msg % args
    # then format the whole message
    if self.op.error_codes:
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
    else:
      if item:
        item = " " + item
      else:
        item = ""
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
    # and finally report it via the feedback_fn
    self._feedback_fn("  - %s" % msg)

  def _ErrorIf(self, cond, *args, **kwargs):
    """Log an error message if the passed condition is True.

    """
    cond = bool(cond) or self.op.debug_simulate_errors
    if cond:
      self._Error(*args, **kwargs)
    # do not mark the operation as failed for WARN cases only
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
      self.bad = self.bad or cond

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

    Test list:

      - compares ganeti version
      - checks vg existence 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 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
    @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())

    """
    node = nodeinfo.name
    _ErrorIf = self._ErrorIf

    # main result, node_result should be a non-empty dict
    test = not node_result or not isinstance(node_result, dict)
    _ErrorIf(test, self.ENODERPC, node,
                  "unable to verify node: no data returned")
    if test:
      return

    # compares ganeti version
    local_version = constants.PROTOCOL_VERSION
    remote_version = node_result.get('version', None)
    test = not (remote_version and
                isinstance(remote_version, (list, tuple)) and
                len(remote_version) == 2)
    _ErrorIf(test, self.ENODERPC, node,
             "connection to node returned invalid data")
    if test:
      return

    test = local_version != remote_version[0]
    _ErrorIf(test, self.ENODEVERSION, node,
             "incompatible protocol versions: master %s,"
             " node %s", local_version, remote_version[0])
    if test:
      return

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

    # full package version
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
                  self.ENODEVERSION, node,
                  "software version mismatch: master %s, node %s",
                  constants.RELEASE_VERSION, remote_version[1],
                  code=self.ETYPE_WARNING)

    # checks vg existence and size > 20G
    if vg_name is not None:
      vglist = node_result.get(constants.NV_VGLIST, None)
      test = not vglist
      _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
      if not test:
        vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
                                              constants.MIN_VG_SIZE)
        _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)

    # checks config file checksum

    remote_cksum = node_result.get(constants.NV_FILELIST, None)
    test = not isinstance(remote_cksum, dict)
    _ErrorIf(test, self.ENODEFILECHECK, node,
             "node hasn't returned file checksum data")
    if not test:
      for file_name in file_list:
        node_is_mc = nodeinfo.master_candidate
        must_have = (file_name not in master_files) or node_is_mc
        # missing
        test1 = file_name not in remote_cksum
        # invalid checksum
        test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
        # existing and good
        test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
        _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
                 "file '%s' missing", file_name)
        _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
                 "file '%s' has wrong checksum", file_name)
        # not candidate and this is not a must-have file
        _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
                 "file '%s' should not exist on non master"
                 " candidates (and the file is outdated)", file_name)
        # all good, except non-master/non-must have combination
        _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
                 "file '%s' should not exist"
                 " on non master candidates", file_name)

    # checks ssh to any

    test = constants.NV_NODELIST not in node_result
    _ErrorIf(test, self.ENODESSH, node,
             "node hasn't returned node ssh connectivity data")
    if not test:
      if node_result[constants.NV_NODELIST]:
        for a_node, a_msg in node_result[constants.NV_NODELIST].items():
          _ErrorIf(True, self.ENODESSH, node,
                   "ssh communication with node '%s': %s", a_node, a_msg)

    test = constants.NV_NODENETTEST not in node_result
    _ErrorIf(test, self.ENODENET, node,
             "node hasn't returned node tcp connectivity data")
    if not test:
      if node_result[constants.NV_NODENETTEST]:
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
        for anode in nlist:
          _ErrorIf(True, self.ENODENET, node,
                   "tcp communication with node '%s': %s",
                   anode, node_result[constants.NV_NODENETTEST][anode])

    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
    if isinstance(hyp_result, dict):
      for hv_name, hv_result in hyp_result.iteritems():
        test = hv_result is not None
        _ErrorIf(test, self.ENODEHV, node,
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)

    # check used drbd list
    if vg_name is not None:
      used_minors = node_result.get(constants.NV_DRBDLIST, [])
      test = not isinstance(used_minors, (tuple, list))
      _ErrorIf(test, self.ENODEDRBD, node,
               "cannot parse drbd status file: %s", str(used_minors))
      if not test:
        for minor, (iname, must_exist) in drbd_map.items():
          test = minor not in used_minors and must_exist
          _ErrorIf(test, self.ENODEDRBD, node,
                   "drbd minor %d of instance %s is not active",
                   minor, iname)
        for minor in used_minors:
          test = minor not in drbd_map
          _ErrorIf(test, self.ENODEDRBD, node,
                   "unallocated drbd minor %d is in use", minor)

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

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

    """
    _ErrorIf = self._ErrorIf
    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]:
        test = node not in node_vol_is or volume not in node_vol_is[node]
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
                 "volume %s missing on node %s", volume, node)

    if instanceconfig.admin_up:
      test = ((node_current not in node_instance or
               not instance in node_instance[node_current]) and
              node_current not in n_offline)
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
               "instance not running on its primary node %s",
               node_current)

    for node in node_instance:
      if (not node == node_current):
        test = instance in node_instance[node]
        _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
                 "instance should not run on node %s", node)

  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is):
    """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.

    """
    for node in node_vol_is:
      for volume in node_vol_is[node]:
        test = (node not in node_vol_should or
                volume not in node_vol_should[node])
        self._ErrorIf(test, self.ENODEORPHANLV, node,
                      "volume %s is unknown", volume)

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

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

    """
    for node in node_instance:
      for o_inst in node_instance[node]:
        test = o_inst not in instancelist
        self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
                      "instance %s on node %s should not exist", o_inst, node)

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

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

    """
    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]
        test = nodeinfo['mfree'] < needed_mem
        self._ErrorIf(test, self.ENODEN1, node,
                      "not enough memory on to accommodate"
                      " failovers should peer node %s fail", prinode)

  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 ran 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()
    env = {
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
      }
    for node in self.cfg.GetAllNodesInfo().values():
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())

    return env, [], all_nodes

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

    """
    self.bad = False
    _ErrorIf = self._ErrorIf
    verbose = self.op.verbose
    self._feedback_fn = feedback_fn
    feedback_fn("* Verifying global settings")
    for msg in self.cfg.VerifyConfig():
      _ErrorIf(True, self.ECLUSTERCFG, None, 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_INSTANCELIST: hypervisors,
      constants.NV_VERSION: None,
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
      }
    if vg_name is not None:
      node_verify_param[constants.NV_VGLIST] = None
      node_verify_param[constants.NV_LVLIST] = vg_name
      node_verify_param[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()

    feedback_fn("* Verifying node status")
    for node_i in nodeinfo:
      node = node_i.name

      if node_i.offline:
        if verbose:
          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"
      if verbose:
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))

      msg = all_nvinfo[node].fail_msg
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
      if msg:
        continue

      nresult = all_nvinfo[node].payload
      node_drbd = {}
      for minor, instance in all_drbd_map[node].items():
        test = instance not in instanceinfo
        _ErrorIf(test, self.ECLUSTERCFG, None,
                 "ghost instance '%s' in temporary DRBD map", instance)
          # ghost instance should not be running, but otherwise we
          # don't give double warnings (both ghost instance and
          # unallocated minor in use)
        if test:
          node_drbd[minor] = (instance, False)
        else:
          instance = instanceinfo[instance]
          node_drbd[minor] = (instance.name, instance.admin_up)
      self._VerifyNode(node_i, file_names, local_checksums,
                       nresult, master_files, node_drbd, vg_name)

      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
      if vg_name is None:
        node_volume[node] = {}
      elif isinstance(lvdata, basestring):
        _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
                 utils.SafeEncode(lvdata))
        node_volume[node] = {}
      elif not isinstance(lvdata, dict):
        _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
        continue
      else:
        node_volume[node] = lvdata

      # node_instance
      idata = nresult.get(constants.NV_INSTANCELIST, None)
      test = not isinstance(idata, list)
      _ErrorIf(test, self.ENODEHV, node,
               "rpc call to node failed (instancelist)")
      if test:
        continue

      node_instance[node] = idata