config.py

#
#

# Copyright (C) 2006, 2007 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.


"""Configuration management for Ganeti

This module provides the interface to the Ganeti cluster configuration.

The configuration data is stored on every node but is updated on the master
only. After each update, the master distributes the data to the other nodes.

Currently, the data storage format is JSON. YAML was slow and consuming too
much memory.

"""

import os
import tempfile
import random
import logging

from ganeti import errors
from ganeti import locking
from ganeti import utils
from ganeti import constants
from ganeti import rpc
from ganeti import objects
from ganeti import serializer


_config_lock = locking.SharedLock()


def _ValidateConfig(data):
  """Verifies that a configuration objects looks valid.

  This only verifies the version of the configuration.

  @raise errors.ConfigurationError: if the version differs from what
      we expect

  """
  if data.version != constants.CONFIG_VERSION:
    raise errors.ConfigurationError("Cluster configuration version"
                                    " mismatch, got %s instead of %s" %
                                    (data.version,
                                     constants.CONFIG_VERSION))


class ConfigWriter:
  """The interface to the cluster configuration.

  """
  def __init__(self, cfg_file=None, offline=False):
    self.write_count = 0
    self._lock = _config_lock
    self._config_data = None
    self._offline = offline
    if cfg_file is None:
      self._cfg_file = constants.CLUSTER_CONF_FILE
    else:
      self._cfg_file = cfg_file
    self._temporary_ids = set()
    self._temporary_drbds = {}
    # Note: in order to prevent errors when resolving our name in
    # _DistributeConfig, we compute it here once and reuse it; it's
    # better to raise an error before starting to modify the config
    # file than after it was modified
    self._my_hostname = utils.HostInfo().name
    self._last_cluster_serial = -1
    self._OpenConfig()

  # this method needs to be static, so that we can call it on the class
  @staticmethod
  def IsCluster():
    """Check if the cluster is configured.

    """
    return os.path.exists(constants.CLUSTER_CONF_FILE)

  @locking.ssynchronized(_config_lock, shared=1)
  def GenerateMAC(self):
    """Generate a MAC for an instance.

    This should check the current instances for duplicates.

    """
    prefix = self._config_data.cluster.mac_prefix
    all_macs = self._AllMACs()
    retries = 64
    while retries > 0:
      byte1 = random.randrange(0, 256)
      byte2 = random.randrange(0, 256)
      byte3 = random.randrange(0, 256)
      mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
      if mac not in all_macs:
        break
      retries -= 1
    else:
      raise errors.ConfigurationError("Can't generate unique MAC")
    return mac

  @locking.ssynchronized(_config_lock, shared=1)
  def IsMacInUse(self, mac):
    """Predicate: check if the specified MAC is in use in the Ganeti cluster.

    This only checks instances managed by this cluster, it does not
    check for potential collisions elsewhere.

    """
    all_macs = self._AllMACs()
    return mac in all_macs

  @locking.ssynchronized(_config_lock, shared=1)
  def GenerateDRBDSecret(self):
    """Generate a DRBD secret.

    This checks the current disks for duplicates.

    """
    all_secrets = self._AllDRBDSecrets()
    retries = 64
    while retries > 0:
      secret = utils.GenerateSecret()
      if secret not in all_secrets:
        break
      retries -= 1
    else:
      raise errors.ConfigurationError("Can't generate unique DRBD secret")
    return secret

  def _ComputeAllLVs(self):
    """Compute the list of all LVs.

    """
    lvnames = set()
    for instance in self._config_data.instances.values():
      node_data = instance.MapLVsByNode()
      for lv_list in node_data.values():
        lvnames.update(lv_list)
    return lvnames

  @locking.ssynchronized(_config_lock, shared=1)
  def GenerateUniqueID(self, exceptions=None):
    """Generate an unique disk name.

    This checks the current node, instances and disk names for
    duplicates.

    @param exceptions: a list with some other names which should be checked
        for uniqueness (used for example when you want to get
        more than one id at one time without adding each one in
        turn to the config file)

    @rtype: string
    @return: the unique id

    """
    existing = set()
    existing.update(self._temporary_ids)
    existing.update(self._ComputeAllLVs())
    existing.update(self._config_data.instances.keys())
    existing.update(self._config_data.nodes.keys())
    if exceptions is not None:
      existing.update(exceptions)
    retries = 64
    while retries > 0:
      unique_id = utils.NewUUID()
      if unique_id not in existing and unique_id is not None:
        break
    else:
      raise errors.ConfigurationError("Not able generate an unique ID"
                                      " (last tried ID: %s" % unique_id)
    self._temporary_ids.add(unique_id)
    return unique_id

  def _AllMACs(self):
    """Return all MACs present in the config.

    @rtype: list
    @return: the list of all MACs

    """
    result = []
    for instance in self._config_data.instances.values():
      for nic in instance.nics:
        result.append(nic.mac)

    return result

  def _AllDRBDSecrets(self):
    """Return all DRBD secrets present in the config.

    @rtype: list
    @return: the list of all DRBD secrets

    """
    def helper(disk, result):
      """Recursively gather secrets from this disk."""
      if disk.dev_type == constants.DT_DRBD8:
        result.append(disk.logical_id[5])
      if disk.children:
        for child in disk.children:
          helper(child, result)

    result = []
    for instance in self._config_data.instances.values():
      for disk in instance.disks:
        helper(disk, result)

    return result

  @locking.ssynchronized(_config_lock, shared=1)
  def VerifyConfig(self):
    """Verify function.

    """
    result = []
    seen_macs = []
    ports = {}
    data = self._config_data
    for instance_name in data.instances:
      instance = data.instances[instance_name]
      if instance.primary_node not in data.nodes:
        result.append("instance '%s' has invalid primary node '%s'" %
                      (instance_name, instance.primary_node))
      for snode in instance.secondary_nodes:
        if snode not in data.nodes:
          result.append("instance '%s' has invalid secondary node '%s'" %
                        (instance_name, snode))
      for idx, nic in enumerate(instance.nics):
        if nic.mac in seen_macs:
          result.append("instance '%s' has NIC %d mac %s duplicate" %
                        (instance_name, idx, nic.mac))
        else:
          seen_macs.append(nic.mac)

      # gather the drbd ports for duplicate checks
      for dsk in instance.disks:
        if dsk.dev_type in constants.LDS_DRBD:
          tcp_port = dsk.logical_id[2]
          if tcp_port not in ports:
            ports[tcp_port] = []
          ports[tcp_port].append((instance.name, "drbd disk %s" % dsk.iv_name))
      # gather network port reservation
      net_port = getattr(instance, "network_port", None)
      if net_port is not None:
        if net_port not in ports:
          ports[net_port] = []
        ports[net_port].append((instance.name, "network port"))

    # cluster-wide pool of free ports
    for free_port in data.cluster.tcpudp_port_pool:
      if free_port not in ports:
        ports[free_port] = []
      ports[free_port].append(("cluster", "port marked as free"))

    # compute tcp/udp duplicate ports
    keys = ports.keys()
    keys.sort()
    for pnum in keys:
      pdata = ports[pnum]
      if len(pdata) > 1:
        txt = ", ".join(["%s/%s" % val for val in pdata])
        result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt))

    # highest used tcp port check
    if keys:
      if keys[-1] > data.cluster.highest_used_port:
        result.append("Highest used port mismatch, saved %s, computed %s" %
                      (data.cluster.highest_used_port, keys[-1]))

    if not data.nodes[data.cluster.master_node].master_candidate:
      result.append("Master node is not a master candidate")

    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
    if mc_now < mc_max:
      result.append("Not enough master candidates: actual %d, target %d" %
                    (mc_now, mc_max))

    return result

  def _UnlockedSetDiskID(self, disk, node_name):
    """Convert the unique ID to the ID needed on the target nodes.

    This is used only for drbd, which needs ip/port configuration.

    The routine descends down and updates its children also, because
    this helps when the only the top device is passed to the remote
    node.

    This function is for internal use, when the config lock is already held.

    """
    if disk.children:
      for child in disk.children:
        self._UnlockedSetDiskID(child, node_name)

    if disk.logical_id is None and disk.physical_id is not None:
      return
    if disk.dev_type == constants.LD_DRBD8:
      pnode, snode, port, pminor, sminor, secret = disk.logical_id
      if node_name not in (pnode, snode):
        raise errors.ConfigurationError("DRBD device not knowing node %s" %
                                        node_name)
      pnode_info = self._UnlockedGetNodeInfo(pnode)
      snode_info = self._UnlockedGetNodeInfo(snode)
      if pnode_info is None or snode_info is None:
        raise errors.ConfigurationError("Can't find primary or secondary node"
                                        " for %s" % str(disk))
      p_data = (pnode_info.secondary_ip, port)
      s_data = (snode_info.secondary_ip, port)
      if pnode == node_name:
        disk.physical_id = p_data + s_data + (pminor, secret)
      else: # it must be secondary, we tested above
        disk.physical_id = s_data + p_data + (sminor, secret)
    else:
      disk.physical_id = disk.logical_id
    return

  @locking.ssynchronized(_config_lock)
  def SetDiskID(self, disk, node_name):
    """Convert the unique ID to the ID needed on the target nodes.

    This is used only for drbd, which needs ip/port configuration.

    The routine descends down and updates its children also, because
    this helps when the only the top device is passed to the remote
    node.

    """
    return self._UnlockedSetDiskID(disk, node_name)

  @locking.ssynchronized(_config_lock)
  def AddTcpUdpPort(self, port):
    """Adds a new port to the available port pool.

    """
    if not isinstance(port, int):
      raise errors.ProgrammerError("Invalid type passed for port")

    self._config_data.cluster.tcpudp_port_pool.add(port)
    self._WriteConfig()

  @locking.ssynchronized(_config_lock, shared=1)
  def GetPortList(self):
    """Returns a copy of the current port list.

    """
    return self._config_data.cluster.tcpudp_port_pool.copy()

  @locking.ssynchronized(_config_lock)
  def AllocatePort(self):
    """Allocate a port.

    The port will be taken from the available port pool or from the
    default port range (and in this case we increase
    highest_used_port).

    """
    # If there are TCP/IP ports configured, we use them first.
    if self._config_data.cluster.tcpudp_port_pool:
      port = self._config_data.cluster.tcpudp_port_pool.pop()
    else:
      port = self._config_data.cluster.highest_used_port + 1
      if port >= constants.LAST_DRBD_PORT:
        raise errors.ConfigurationError("The highest used port is greater"
                                        " than %s. Aborting." %
                                        constants.LAST_DRBD_PORT)
      self._config_data.cluster.highest_used_port = port

    self._WriteConfig()
    return port

  def _UnlockedComputeDRBDMap(self):
    """Compute the used DRBD minor/nodes.

    @return: dictionary of node_name: dict of minor: instance_name;
        the returned dict will have all the nodes in it (even if with
        an empty list).

    """
    def _AppendUsedPorts(instance_name, disk, used):
      if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
        nodeA, nodeB, dummy, minorA, minorB = disk.logical_id[:5]
        for node, port in ((nodeA, minorA), (nodeB, minorB)):
          assert node in used, "Instance node not found in node list"
          if port in used[node]:
            raise errors.ProgrammerError("DRBD minor already used:"
                                         " %s/%s, %s/%s" %
                                         (node, port, instance_name,
                                          used[node][port]))

          used[node][port] = instance_name
      if disk.children:
        for child in disk.children:
          _AppendUsedPorts(instance_name, child, used)

    my_dict = dict((node, {}) for node in self._config_data.nodes)
    for (node, minor), instance in self._temporary_drbds.iteritems():
      my_dict[node][minor] = instance
    for instance in self._config_data.instances.itervalues():
      for disk in instance.disks:
        _AppendUsedPorts(instance.name, disk, my_dict)
    return my_dict

  @locking.ssynchronized(_config_lock)
  def ComputeDRBDMap(self):
    """Compute the used DRBD minor/nodes.

    This is just a wrapper over L{_UnlockedComputeDRBDMap}.

    @return: dictionary of node_name: dict of minor: instance_name;
        the returned dict will have all the nodes in it (even if with
        an empty list).

    """
    return self._UnlockedComputeDRBDMap()

  @locking.ssynchronized(_config_lock)
  def AllocateDRBDMinor(self, nodes, instance):
    """Allocate a drbd minor.

    The free minor will be automatically computed from the existing
    devices. A node can be given multiple times in order to allocate
    multiple minors. The result is the list of minors, in the same
    order as the passed nodes.

    """
    d_map = self._UnlockedComputeDRBDMap()
    result = []
    for nname in nodes:
      ndata = d_map[nname]
      if not ndata:
        # no minors used, we can start at 0
        result.append(0)
        ndata[0] = instance
        self._temporary_drbds[(nname, 0)] = instance
        continue
      keys = ndata.keys()
      keys.sort()
      ffree = utils.FirstFree(keys)
      if ffree is None:
        # return the next minor
        # TODO: implement high-limit check
        minor = keys[-1] + 1
      else:
        minor = ffree
      result.append(minor)
      ndata[minor] = instance
      assert (nname, minor) not in self._temporary_drbds, \
             "Attempt to reuse reserved DRBD minor"
      self._temporary_drbds[(nname, minor)] = instance
    logging.debug("Request to allocate drbd minors, input: %s, returning %s",
                  nodes, result)
    return result

  @locking.ssynchronized(_config_lock)
  def ReleaseDRBDMinors(self, instance):
    """Release temporary drbd minors allocated for a given instance.

    This should be called on both the error paths and on the success
    paths (after the instance has been added or updated).

    @type instance: string
    @param instance: the instance for which temporary minors should be
                     released

    """
    for key, name in self._temporary_drbds.items():
      if name == instance:
        del self._temporary_drbds[key]

  @locking.ssynchronized(_config_lock, shared=1)
  def GetConfigVersion(self):
    """Get the configuration version.

    @return: Config version

    """
    return self._config_data.version

  @locking.ssynchronized(_config_lock, shared=1)
  def GetClusterName(self):
    """Get cluster name.

    @return: Cluster name

    """
    return self._config_data.cluster.cluster_name

  @locking.ssynchronized(_config_lock, shared=1)
  def GetMasterNode(self):
    """Get the hostname of the master node for this cluster.

    @return: Master hostname

    """
    return self._config_data.cluster.master_node

  @locking.ssynchronized(_config_lock, shared=1)
  def GetMasterIP(self):
    """Get the IP of the master node for this cluster.

    @return: Master IP

    """
    return self._config_data.cluster.master_ip

  @locking.ssynchronized(_config_lock, shared=1)
  def GetMasterNetdev(self):
    """Get the master network device for this cluster.

    """
    return self._config_data.cluster.master_netdev

  @locking.ssynchronized(_config_lock, shared=1)
  def GetFileStorageDir(self):
    """Get the file storage dir for this cluster.

    """
    return self._config_data.cluster.file_storage_dir

  @locking.ssynchronized(_config_lock, shared=1)
  def GetHypervisorType(self):
    """Get the hypervisor type for this cluster.

    """
    return self._config_data.cluster.default_hypervisor

  @locking.ssynchronized(_config_lock, shared=1)
  def GetHostKey(self):
    """Return the rsa hostkey from the config.

    @rtype: string
    @return: the rsa hostkey

    """
    return self._config_data.cluster.rsahostkeypub

  @locking.ssynchronized(_config_lock)
  def AddInstance(self, instance):
    """Add an instance to the config.

    This should be used after creating a new instance.

    @type instance: L{objects.Instance}
    @param instance: the instance object

    """
    if not isinstance(instance, objects.Instance):
      raise errors.ProgrammerError("Invalid type passed to AddInstance")

    if instance.disk_template != constants.DT_DISKLESS:
      all_lvs = instance.MapLVsByNode()
      logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)

    instance.serial_no = 1
    self._config_data.instances[instance.name] = instance
    self._WriteConfig()

  def _SetInstanceStatus(self, instance_name, status):
    """Set the instance's status to a given value.

    """
    if status not in ("up", "down"):
      raise errors.ProgrammerError("Invalid status '%s' passed to"
                                   " ConfigWriter._SetInstanceStatus()" %
                                   status)

    if instance_name not in self._config_data.instances:
      raise errors.ConfigurationError("Unknown instance '%s'" %
                                      instance_name)
    instance = self._config_data.instances[instance_name]
    if instance.status != status:
      instance.status = status
      instance.serial_no += 1
      self._WriteConfig()

  @locking.ssynchronized(_config_lock)
  def MarkInstanceUp(self, instance_name):
    """Mark the instance status to up in the config.

    """
    self._SetInstanceStatus(instance_name, "up")

  @locking.ssynchronized(_config_lock)
  def RemoveInstance(self, instance_name):
    """Remove the instance from the configuration.

    """
    if instance_name not in self._config_data.instances:
      raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
    del self._config_data.instances[instance_name]
    self._WriteConfig()

  @locking.ssynchronized(_config_lock)
  def RenameInstance(self, old_name, new_name):
    """Rename an instance.

    This needs to be done in ConfigWriter and not by RemoveInstance
    combined with AddInstance as only we can guarantee an atomic
    rename.

    """
    if old_name not in self._config_data.instances:
      raise errors.ConfigurationError("Unknown instance '%s'" % old_name)
    inst = self._config_data.instances[old_name]
    del self._config_data.instances[old_name]
    inst.name = new_name

    for disk in inst.disks:
      if disk.dev_type == constants.LD_FILE:
        # rename the file paths in logical and physical id
        file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1]))
        disk.physical_id = disk.logical_id = (disk.logical_id[0],
                                              os.path.join(file_storage_dir,
                                                           inst.name,
                                                           disk.iv_name))

    self._config_data.instances[inst.name] = inst
    self._WriteConfig()

  @locking.ssynchronized(_config_lock)
  def MarkInstanceDown(self, instance_name):
    """Mark the status of an instance to down in the configuration.

    """
    self._SetInstanceStatus(instance_name, "down")

  def _UnlockedGetInstanceList(self):
    """Get the list of instances.

    This function is for internal use, when the config lock is already held.

    """
    return self._config_data.instances.keys()

  @locking.ssynchronized(_config_lock, shared=1)
  def GetInstanceList(self):
    """Get the list of instances.

    @return: array of instances, ex. ['instance2.example.com',
        'instance1.example.com']

    """
    return self._UnlockedGetInstanceList()

  @locking.ssynchronized(_config_lock, shared=1)
  def ExpandInstanceName(self, short_name):
    """Attempt to expand an incomplete instance name.

    """
    return utils.MatchNameComponent(short_name,
                                    self._config_data.instances.keys())

  def _UnlockedGetInstanceInfo(self, instance_name):
    """Returns informations about an instance.

    This function is for internal use, when the config lock is already held.

    """
    if instance_name not in self._config_data.instances:
      return None

    return self._config_data.instances[instance_name]

  @locking.ssynchronized(_config_lock, shared=1)
  def GetInstanceInfo(self, instance_name):
    """Returns informations about an instance.

    It takes the information from the configuration file. Other informations of
    an instance are taken from the live systems.

    @param instance_name: name of the instance, e.g.
        I{instance1.example.com}

    @rtype: L{objects.Instance}
    @return: the instance object

    """
    return self._UnlockedGetInstanceInfo(instance_name)

  @locking.ssynchronized(_config_lock, shared=1)
  def GetAllInstancesInfo(self):
    """Get the configuration of all instances.

    @rtype: dict
    @returns: dict of (instance, instance_info), where instance_info is what
              would GetInstanceInfo return for the node

    """
    my_dict = dict([(instance, self._UnlockedGetInstanceInfo(instance))
                    for instance in self._UnlockedGetInstanceList()])
    return my_dict

  @locking.ssynchronized(_config_lock)
  def AddNode(self, node):
    """Add a node to the configuration.

    @type node: L{objects.Node}
    @param node: a Node instance

    """
    logging.info("Adding node %s to configuration" % node.name)

    node.serial_no = 1
    self._config_data.nodes[node.name] = node
    self._config_data.cluster.serial_no += 1
    self._WriteConfig()

  @locking.ssynchronized(_config_lock)
  def RemoveNode(self, node_name):
    """Remove a node from the configuration.

    """
    logging.info("Removing node %s from configuration" % node_name)

    if node_name not in self._config_data.nodes:
      raise errors.ConfigurationError("Unknown node '%s'" % node_name)

    del self._config_data.nodes[node_name]
    self._config_data.cluster.serial_no += 1
    self._WriteConfig()

  @locking.ssynchronized(_config_lock, shared=1)
  def ExpandNodeName(self, short_name):
    """Attempt to expand an incomplete instance name.

    """
    return utils.MatchNameComponent(short_name,
                                    self._config_data.nodes.keys())

  def _UnlockedGetNodeInfo(self, node_name):
    """Get the configuration of a node, as stored in the config.

    This function is for internal use, when the config lock is already
    held.

    @param node_name: the node name, e.g. I{node1.example.com}

    @rtype: L{objects.Node}
    @return: the node object

    """
    if node_name not in self._config_data.nodes:
      return None

    return self._config_data.nodes[node_name]


  @locking.ssynchronized(_config_lock, shared=1)
  def GetNodeInfo(self, node_name):
    """Get the configuration of a node, as stored in the config.

    This is just a locked wrapper over L{_UnlockedGetNodeInfo}.

    @param node_name: the node name, e.g. I{node1.example.com}

    @rtype: L{objects.Node}
    @return: the node object

    """
    return self._UnlockedGetNodeInfo(node_name)

  def _UnlockedGetNodeList(self):
    """Return the list of nodes which are in the configuration.

    This function is for internal use, when the config lock is already
    held.

    @rtype: list

    """
    return self._config_data.nodes.keys()


  @locking.ssynchronized(_config_lock, shared=1)
  def GetNodeList(self):
    """Return the list of nodes which are in the configuration.

    """
    return self._UnlockedGetNodeList()

  @locking.ssynchronized(_config_lock, shared=1)
  def GetOnlineNodeList(self):
    """Return the list of nodes which are online.

    """
    all_nodes = [self._UnlockedGetNodeInfo(node)
                 for node in self._UnlockedGetNodeList()]
    return [node.name for node in all_nodes if not node.offline]

  @locking.ssynchronized(_config_lock, shared=1)
  def GetAllNodesInfo(self):
    """Get the configuration of all nodes.

    @rtype: dict
    @return: dict of (node, node_info), where node_info is what
              would GetNodeInfo return for the node

    """
    my_dict = dict([(node, self._UnlockedGetNodeInfo(node))
                    for node in self._UnlockedGetNodeList()])
    return my_dict

  def _UnlockedGetMasterCandidateStats(self):
    """Get the number of current and maximum desired and possible candidates.

    @rtype: tuple
    @return: tuple of (current, desired and possible)

    """
    mc_now = mc_max = 0
    for node in self._config_data.nodes.itervalues():
      if not node.offline:
        mc_max += 1
      if node.master_candidate:
        mc_now += 1
    mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size)
    return (mc_now, mc_max)

  @locking.ssynchronized(_config_lock, shared=1)
  def GetMasterCandidateStats(self):
    """Get the number of current and maximum possible candidates.

    This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.

    @rtype: tuple
    @return: tuple of (current, max)

    """
    return self._UnlockedGetMasterCandidateStats()

  @locking.ssynchronized(_config_lock)
  def MaintainCandidatePool(self):
    """Try to grow the candidate pool to the desired size.

    @rtype: list
    @return: list with the adjusted nodes (L{objects.Node} instances)

    """
    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
    mod_list = []
    if mc_now < mc_max:
      node_list = self._config_data.nodes.keys()
      random.shuffle(node_list)
      for name in node_list:
        if mc_now >= mc_max:
          break
        node = self._config_data.nodes[name]
        if node.master_candidate or node.offline:
          continue
        mod_list.append(node)
        node.master_candidate = True
        node.serial_no += 1
        mc_now += 1
      if mc_now != mc_max:
        # this should not happen
        logging.warning("Warning: MaintainCandidatePool didn't manage to"
                        " fill the candidate pool (%d/%d)", mc_now, mc_max)
      if mod_list:
        self._config_data.cluster.serial_no += 1
        self._WriteConfig()

    return mod_list

  def _BumpSerialNo(self):
    """Bump up the serial number of the config.

    """
    self._config_data.serial_no += 1

  def _OpenConfig(self):
    """Read the config data from disk.

    """
    f = open(self._cfg_file, 'r')
    try:
      try:
        data = objects.ConfigData.FromDict(serializer.Load(f.read()))
      except Exception, err:
        raise errors.ConfigurationError(err)
    finally:
      f.close()

    # Make sure the configuration has the right version
    _ValidateConfig(data)

    if (not hasattr(data, 'cluster') or
        not hasattr(data.cluster, 'rsahostkeypub')):
      raise errors.ConfigurationError("Incomplete configuration"
                                      " (missing cluster.rsahostkeypub)")
    self._config_data = data
    # reset the last serial as -1 so that the next write will cause
    # ssconf update
    self._last_cluster_serial = -1

  def _DistributeConfig(self):
    """Distribute the configuration to the other nodes.

    Currently, this only copies the configuration file. In the future,
    it could be used to encapsulate the 2/3-phase update mechanism.

    """
    if self._offline:
      return True
    bad = False

    node_list = []
    addr_list = []
    myhostname = self._my_hostname
    # we can skip checking whether _UnlockedGetNodeInfo returns None
    # since the node list comes from _UnlocketGetNodeList, and we are
    # called with the lock held, so no modifications should take place
    # in between
    for node_name in self._UnlockedGetNodeList():
      if node_name == myhostname:
        continue
      node_info = self._UnlockedGetNodeInfo(node_name)
      if not node_info.master_candidate:
        continue
      node_list.append(node_info.name)
      addr_list.append(node_info.primary_ip)

    result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file,
                                            address_list=addr_list)
    for node in node_list:
      if not result[node]:
        logging.error("copy of file %s to node %s failed",
                      self._cfg_file, node)
        bad = True
    return not bad

  def _WriteConfig(self, destination=None):
    """Write the configuration data to persistent storage.

    """
    if destination is None:
      destination = self._cfg_file
    self._BumpSerialNo()
    txt = serializer.Dump(self._config_data.ToDict())
    dir_name, file_name = os.path.split(destination)
    fd, name = tempfile.mkstemp('.newconfig', file_name, dir_name)
    f = os.fdopen(fd, 'w')
    try:
      f.write(txt)
      os.fsync(f.fileno())
    finally:
      f.close()
    # we don't need to do os.close(fd) as f.close() did it
    os.rename(name, destination)
    self.write_count += 1

    # and redistribute the config file to master candidates
    self._DistributeConfig()

    # Write ssconf files on all nodes (including locally)
    if self._last_cluster_serial < self._config_data.cluster.serial_no:
      if not self._offline:
        rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(),
                                              self._UnlockedGetSsconfValues())
      self._last_cluster_serial = self._config_data.cluster.serial_no

  def _UnlockedGetSsconfValues(self):
    """Return the values needed by ssconf.

    @rtype: dict
    @return: a dictionary with keys the ssconf names and values their
        associated value

    """
    fn = "\n".join
    node_names = utils.NiceSort(self._UnlockedGetNodeList())
    node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]

    off_data = fn(node.name for node in node_info if node.offline)
    mc_data = fn(node.name for node in node_info if node.master_candidate)
    node_data = fn(node_names)

    cluster = self._config_data.cluster
    return {
      constants.SS_CLUSTER_NAME: cluster.cluster_name,
      constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
      constants.SS_MASTER_CANDIDATES: mc_data,
      constants.SS_MASTER_IP: cluster.master_ip,
      constants.SS_MASTER_NETDEV: cluster.master_netdev,
      constants.SS_MASTER_NODE: cluster.master_node,
      constants.SS_NODE_LIST: node_data,
      constants.SS_OFFLINE_NODES: off_data,
      constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION,
      }

  @locking.ssynchronized(_config_lock)
  def InitConfig(self, version, cluster_config, master_node_config):
    """Create the initial cluster configuration.

    It will contain the current node, which will also be the master
    node, and no instances.

    @type version: int
    @param version: Configuration version
    @type cluster_config: objects.Cluster
    @param cluster_config: Cluster configuration
    @type master_node_config: objects.Node
    @param master_node_config: Master node configuration

    """
    nodes = {
      master_node_config.name: master_node_config,
      }

    self._config_data = objects.ConfigData(version=version,
                                           cluster=cluster_config,
                                           nodes=nodes,
                                           instances={},
                                           serial_no=1)
    self._WriteConfig()

  @locking.ssynchronized(_config_lock, shared=1)
  def GetVGName(self):
    """Return the volume group name.

    """
    return self._config_data.cluster.volume_group_name

  @locking.ssynchronized(_config_lock)
  def SetVGName(self, vg_name):
    """Set the volume group name.

    """
    self._config_data.cluster.volume_group_name = vg_name
    self._config_data.cluster.serial_no += 1
    self._WriteConfig()

  @locking.ssynchronized(_config_lock, shared=1)
  def GetDefBridge(self):
    """Return the default bridge.

    """
    return self._config_data.cluster.default_bridge

  @locking.ssynchronized(_config_lock, shared=1)
  def GetMACPrefix(self):
    """Return the mac prefix.

    """
    return self._config_data.cluster.mac_prefix

  @locking.ssynchronized(_config_lock, shared=1)
  def GetClusterInfo(self):
    """Returns informations about the cluster

    @rtype: L{objects.Cluster}
    @return: the cluster object

    """
    return self._config_data.cluster

  @locking.ssynchronized(_config_lock)
  def Update(self, target):
    """Notify function to be called after updates.

    This function must be called when an object (as returned by
    GetInstanceInfo, GetNodeInfo, GetCluster) has been updated and the
    caller wants the modifications saved to the backing store. Note
    that all modified objects will be saved, but the target argument
    is the one the caller wants to ensure that it's saved.

    @param target: an instance of either L{objects.Cluster},
        L{objects.Node} or L{objects.Instance} which is existing in
        the cluster

    """
    if self._config_data is None:
      raise errors.ProgrammerError("Configuration file not read,"
                                   " cannot save.")
    update_serial = False
    if isinstance(target, objects.Cluster):
      test = target == self._config_data.cluster
    elif isinstance(target, objects.Node):
      test = target in self._config_data.nodes.values()
      update_serial = True
    elif isinstance(target, objects.Instance):
      test = target in self._config_data.instances.values()
    else:
      raise errors.ProgrammerError("Invalid object type (%s) passed to"
                                   " ConfigWriter.Update" % type(target))
    if not test:
      raise errors.ConfigurationError("Configuration updated since object"
                                      " has been read or unknown object")
    target.serial_no += 1

    if update_serial:
      # for node updates, we need to increase the cluster serial too
      self._config_data.cluster.serial_no += 1

    self._WriteConfig()