#
#
-# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc.
+# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 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
# pylint: disable=R0904
# R0904: Too many public methods
+import copy
import os
import random
import logging
import time
+import itertools
from ganeti import errors
from ganeti import locking
from ganeti import uidpool
from ganeti import netutils
from ganeti import runtime
+from ganeti import pathutils
+from ganeti import network
_config_lock = locking.SharedLock("ConfigWriter")
all_reserved.update(holder_reserved)
return all_reserved
+ def GetECReserved(self, ec_id):
+ """ Used when you want to retrieve all reservations for a specific
+ execution context. E.g when commiting reserved IPs for a specific
+ network.
+
+ """
+ ec_reserved = set()
+ if ec_id in self._ec_reserved:
+ ec_reserved.update(self._ec_reserved[ec_id])
+ return ec_reserved
+
def Generate(self, existing, generate_one_fn, ec_id):
"""Generate a new resource of this type
return utils.MatchNameComponent(short_name, names, case_sensitive=False)
+def _CheckInstanceDiskIvNames(disks):
+ """Checks if instance's disks' C{iv_name} attributes are in order.
+
+ @type disks: list of L{objects.Disk}
+ @param disks: List of disks
+ @rtype: list of tuples; (int, string, string)
+ @return: List of wrongly named disks, each tuple contains disk index,
+ expected and actual name
+
+ """
+ result = []
+
+ for (idx, disk) in enumerate(disks):
+ exp_iv_name = "disk/%s" % idx
+ if disk.iv_name != exp_iv_name:
+ result.append((idx, exp_iv_name, disk.iv_name))
+
+ return result
+
+
class ConfigWriter:
"""The interface to the cluster configuration.
self._config_data = None
self._offline = offline
if cfg_file is None:
- self._cfg_file = constants.CLUSTER_CONF_FILE
+ self._cfg_file = pathutils.CLUSTER_CONF_FILE
else:
self._cfg_file = cfg_file
self._getents = _getents
self._temporary_macs = TemporaryReservationManager()
self._temporary_secrets = TemporaryReservationManager()
self._temporary_lvs = TemporaryReservationManager()
+ self._temporary_ips = TemporaryReservationManager()
self._all_rms = [self._temporary_ids, self._temporary_macs,
- self._temporary_secrets, self._temporary_lvs]
+ self._temporary_secrets, self._temporary_lvs,
+ self._temporary_ips]
# 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
self._my_hostname = netutils.Hostname.GetSysName()
self._last_cluster_serial = -1
self._cfg_id = None
+ self._context = None
self._OpenConfig(accept_foreign)
+ def _GetRpc(self, address_list):
+ """Returns RPC runner for configuration.
+
+ """
+ return rpc.ConfigRunner(self._context, address_list)
+
+ def SetContext(self, context):
+ """Sets Ganeti context.
+
+ """
+ self._context = context
+
# 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)
-
- def _GenerateOneMAC(self):
- """Generate one mac address
-
- """
- prefix = self._config_data.cluster.mac_prefix
- byte1 = random.randrange(0, 256)
- byte2 = random.randrange(0, 256)
- byte3 = random.randrange(0, 256)
- mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
- return mac
+ return os.path.exists(pathutils.CLUSTER_CONF_FILE)
@locking.ssynchronized(_config_lock, shared=1)
def GetNdParams(self, node):
return self._config_data.cluster.FillND(node, nodegroup)
@locking.ssynchronized(_config_lock, shared=1)
- def GenerateMAC(self, ec_id):
+ def GetInstanceDiskParams(self, instance):
+ """Get the disk params populated with inherit chain.
+
+ @type instance: L{objects.Instance}
+ @param instance: The instance we want to know the params for
+ @return: A dict with the filled in disk params
+
+ """
+ node = self._UnlockedGetNodeInfo(instance.primary_node)
+ nodegroup = self._UnlockedGetNodeGroup(node.group)
+ return self._UnlockedGetGroupDiskParams(nodegroup)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetGroupDiskParams(self, group):
+ """Get the disk params populated with inherit chain.
+
+ @type group: L{objects.NodeGroup}
+ @param group: The group we want to know the params for
+ @return: A dict with the filled in disk params
+
+ """
+ return self._UnlockedGetGroupDiskParams(group)
+
+ def _UnlockedGetGroupDiskParams(self, group):
+ """Get the disk params populated with inherit chain down to node-group.
+
+ @type group: L{objects.NodeGroup}
+ @param group: The group we want to know the params for
+ @return: A dict with the filled in disk params
+
+ """
+ return self._config_data.cluster.SimpleFillDP(group.diskparams)
+
+ def _UnlockedGetNetworkMACPrefix(self, net_uuid):
+ """Return the network mac prefix if it exists or the cluster level default.
+
+ """
+ prefix = None
+ if net_uuid:
+ nobj = self._UnlockedGetNetwork(net_uuid)
+ if nobj.mac_prefix:
+ prefix = nobj.mac_prefix
+
+ return prefix
+
+ def _GenerateOneMAC(self, prefix=None):
+ """Return a function that randomly generates a MAC suffic
+ and appends it to the given prefix. If prefix is not given get
+ the cluster level default.
+
+ """
+ if not prefix:
+ prefix = self._config_data.cluster.mac_prefix
+
+ def GenMac():
+ byte1 = random.randrange(0, 256)
+ byte2 = random.randrange(0, 256)
+ byte3 = random.randrange(0, 256)
+ mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
+ return mac
+
+ return GenMac
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GenerateMAC(self, net_uuid, ec_id):
"""Generate a MAC for an instance.
This should check the current instances for duplicates.
"""
existing = self._AllMACs()
- return self._temporary_ids.Generate(existing, self._GenerateOneMAC, ec_id)
+ prefix = self._UnlockedGetNetworkMACPrefix(net_uuid)
+ gen_mac = self._GenerateOneMAC(prefix)
+ return self._temporary_ids.Generate(existing, gen_mac, ec_id)
@locking.ssynchronized(_config_lock, shared=1)
def ReserveMAC(self, mac, ec_id):
else:
self._temporary_macs.Reserve(ec_id, mac)
+ def _UnlockedCommitTemporaryIps(self, ec_id):
+ """Commit all reserved IP address to their respective pools
+
+ """
+ for action, address, net_uuid in self._temporary_ips.GetECReserved(ec_id):
+ self._UnlockedCommitIp(action, net_uuid, address)
+
+ def _UnlockedCommitIp(self, action, net_uuid, address):
+ """Commit a reserved IP address to an IP pool.
+
+ The IP address is taken from the network's IP pool and marked as reserved.
+
+ """
+ nobj = self._UnlockedGetNetwork(net_uuid)
+ pool = network.AddressPool(nobj)
+ if action == constants.RESERVE_ACTION:
+ pool.Reserve(address)
+ elif action == constants.RELEASE_ACTION:
+ pool.Release(address)
+
+ def _UnlockedReleaseIp(self, net_uuid, address, ec_id):
+ """Give a specific IP address back to an IP pool.
+
+ The IP address is returned to the IP pool designated by pool_id and marked
+ as reserved.
+
+ """
+ self._temporary_ips.Reserve(ec_id,
+ (constants.RELEASE_ACTION, address, net_uuid))
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def ReleaseIp(self, net_uuid, address, ec_id):
+ """Give a specified IP address back to an IP pool.
+
+ This is just a wrapper around _UnlockedReleaseIp.
+
+ """
+ if net_uuid:
+ self._UnlockedReleaseIp(net_uuid, address, ec_id)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GenerateIp(self, net_uuid, ec_id):
+ """Find a free IPv4 address for an instance.
+
+ """
+ nobj = self._UnlockedGetNetwork(net_uuid)
+ pool = network.AddressPool(nobj)
+
+ def gen_one():
+ try:
+ ip = pool.GenerateFree()
+ except errors.AddressPoolError:
+ raise errors.ReservationError("Cannot generate IP. Network is full")
+ return (constants.RESERVE_ACTION, ip, net_uuid)
+
+ _, address, _ = self._temporary_ips.Generate([], gen_one, ec_id)
+ return address
+
+ def _UnlockedReserveIp(self, net_uuid, address, ec_id):
+ """Reserve a given IPv4 address for use by an instance.
+
+ """
+ nobj = self._UnlockedGetNetwork(net_uuid)
+ pool = network.AddressPool(nobj)
+ try:
+ isreserved = pool.IsReserved(address)
+ except errors.AddressPoolError:
+ raise errors.ReservationError("IP address not in network")
+ if isreserved:
+ raise errors.ReservationError("IP address already in use")
+
+ return self._temporary_ips.Reserve(ec_id,
+ (constants.RESERVE_ACTION,
+ address, net_uuid))
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def ReserveIp(self, net_uuid, address, ec_id):
+ """Reserve a given IPv4 address for use by an instance.
+
+ """
+ if net_uuid:
+ return self._UnlockedReserveIp(net_uuid, address, ec_id)
+
@locking.ssynchronized(_config_lock, shared=1)
def ReserveLV(self, lv_name, ec_id):
"""Reserve an VG/LV pair for an instance.
invalid_hvs = set(cluster.enabled_hypervisors) - constants.HYPER_TYPES
if invalid_hvs:
result.append("enabled hypervisors contains invalid entries: %s" %
- invalid_hvs)
+ utils.CommaJoin(invalid_hvs))
missing_hvp = (set(cluster.enabled_hypervisors) -
set(cluster.hvparams.keys()))
if missing_hvp:
result.append("hypervisor parameters missing for the enabled"
" hypervisor(s) %s" % utils.CommaJoin(missing_hvp))
+ if not cluster.enabled_disk_templates:
+ result.append("enabled disk templates list doesn't have any entries")
+ invalid_disk_templates = set(cluster.enabled_disk_templates) \
+ - constants.DISK_TEMPLATES
+ if invalid_disk_templates:
+ result.append("enabled disk templates list contains invalid entries:"
+ " %s" % utils.CommaJoin(invalid_disk_templates))
+
if cluster.master_node not in data.nodes:
result.append("cluster has invalid primary node '%s'" %
cluster.master_node)
except errors.ConfigurationError, err:
result.append("%s has invalid nicparams: %s" % (owner, err))
+ def _helper_ipolicy(owner, ipolicy, iscluster):
+ try:
+ objects.InstancePolicy.CheckParameterSyntax(ipolicy, iscluster)
+ except errors.ConfigurationError, err:
+ result.append("%s has invalid instance policy: %s" % (owner, err))
+ for key, value in ipolicy.items():
+ if key == constants.ISPECS_MINMAX:
+ _helper_ispecs(owner, "ipolicy/" + key, value)
+ elif key == constants.ISPECS_STD:
+ _helper(owner, "ipolicy/" + key, value,
+ constants.ISPECS_PARAMETER_TYPES)
+ else:
+ # FIXME: assuming list type
+ if key in constants.IPOLICY_PARAMETERS:
+ exp_type = float
+ else:
+ exp_type = list
+ if not isinstance(value, exp_type):
+ result.append("%s has invalid instance policy: for %s,"
+ " expecting %s, got %s" %
+ (owner, key, exp_type.__name__, type(value)))
+
+ def _helper_ispecs(owner, parentkey, params):
+ for (key, value) in params.items():
+ fullkey = "/".join([parentkey, key])
+ _helper(owner, fullkey, value, constants.ISPECS_PARAMETER_TYPES)
+
# check cluster parameters
_helper("cluster", "beparams", cluster.SimpleFillBE({}),
constants.BES_PARAMETER_TYPES)
_helper_nic("cluster", cluster.SimpleFillNIC({}))
_helper("cluster", "ndparams", cluster.SimpleFillND({}),
constants.NDS_PARAMETER_TYPES)
+ _helper_ipolicy("cluster", cluster.ipolicy, True)
# per-instance checks
for instance_name in data.instances:
filled, constants.NICS_PARAMETER_TYPES)
_helper_nic(owner, filled)
+ # disk template checks
+ if not instance.disk_template in data.cluster.enabled_disk_templates:
+ result.append("instance '%s' uses the disabled disk template '%s'." %
+ (instance_name, instance.disk_template))
+
# parameter checks
if instance.beparams:
_helper("instance %s" % instance.name, "beparams",
cluster.FillBE(instance), constants.BES_PARAMETER_TYPES)
# gather the drbd ports for duplicate checks
- for dsk in instance.disks:
+ for (idx, dsk) in enumerate(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))
+ ports[tcp_port].append((instance.name, "drbd disk %s" % idx))
# gather network port reservation
net_port = getattr(instance, "network_port", None)
if net_port is not None:
(instance.name, idx, msg) for msg in disk.Verify()])
result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids))
+ wrong_names = _CheckInstanceDiskIvNames(instance.disks)
+ if wrong_names:
+ tmp = "; ".join(("name of disk %s should be '%s', but is '%s'" %
+ (idx, exp_name, actual_name))
+ for (idx, exp_name, actual_name) in wrong_names)
+
+ result.append("Instance '%s' has wrongly named disks: %s" %
+ (instance.name, tmp))
+
# cluster-wide pool of free ports
for free_port in cluster.tcpudp_port_pool:
if free_port not in ports:
_helper("node %s" % node.name, "ndparams",
cluster.FillND(node, data.nodegroups[node.group]),
constants.NDS_PARAMETER_TYPES)
+ used_globals = constants.NDC_GLOBALS.intersection(node.ndparams)
+ if used_globals:
+ result.append("Node '%s' has some global parameters set: %s" %
+ (node.name, utils.CommaJoin(used_globals)))
# nodegroups checks
nodegroups_names = set()
result.append("duplicate node group name '%s'" % nodegroup.name)
else:
nodegroups_names.add(nodegroup.name)
+ group_name = "group %s" % nodegroup.name
+ _helper_ipolicy(group_name, cluster.SimpleFillIPolicy(nodegroup.ipolicy),
+ False)
if nodegroup.ndparams:
- _helper("group %s" % nodegroup.name, "ndparams",
+ _helper(group_name, "ndparams",
cluster.SimpleFillND(nodegroup.ndparams),
constants.NDS_PARAMETER_TYPES)
else:
raise errors.ProgrammerError("NIC mode '%s' not handled" % nic_mode)
- _AddIpAddress("%s/%s" % (link, nic.ip),
+ _AddIpAddress("%s/%s/%s" % (link, nic.ip, nic.network),
"instance:%s/nic:%d" % (instance.name, idx))
for ip, owners in ips.items():
def AddTcpUdpPort(self, port):
"""Adds a new port to the available port pool.
+ @warning: this method does not "flush" the configuration (via
+ L{_WriteConfig}); callers should do that themselves once the
+ configuration is stable
+
"""
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):
return self._config_data.cluster.master_netmask
@locking.ssynchronized(_config_lock, shared=1)
+ def GetUseExternalMipScript(self):
+ """Get flag representing whether to use the external master IP setup script.
+
+ """
+ return self._config_data.cluster.use_external_mip_script
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetFileStorageDir(self):
"""Get the file storage dir for this cluster.
"""
return self._config_data.cluster.primary_ip_family
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterNetworkParameters(self):
+ """Get network parameters of the master node.
+
+ @rtype: L{object.MasterNetworkParameters}
+ @return: network parameters of the master node
+
+ """
+ cluster = self._config_data.cluster
+ result = objects.MasterNetworkParameters(
+ name=cluster.master_node, ip=cluster.master_ip,
+ netmask=cluster.master_netmask, netdev=cluster.master_netdev,
+ ip_family=cluster.primary_ip_family)
+
+ return result
+
@locking.ssynchronized(_config_lock)
def AddNodeGroup(self, group, ec_id, check_uuid=True):
"""Add a node group to the configuration.
if target is None:
if len(self._config_data.nodegroups) != 1:
raise errors.OpPrereqError("More than one node group exists. Target"
- " group must be specified explicitely.")
+ " group must be specified explicitly.")
else:
return self._config_data.nodegroups.keys()[0]
if target in self._config_data.nodegroups:
for member_name in
self._UnlockedGetNodeGroup(ngfn(node_name)).members)
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMultiNodeGroupInfo(self, group_uuids):
+ """Get the configuration of multiple node groups.
+
+ @param group_uuids: List of node group UUIDs
+ @rtype: list
+ @return: List of tuples of (group_uuid, group_info)
+
+ """
+ return [(uuid, self._UnlockedGetNodeGroup(uuid)) for uuid in group_uuids]
+
@locking.ssynchronized(_config_lock)
def AddInstance(self, instance, ec_id):
"""Add an instance to the config.
self._config_data.instances[instance.name] = instance
self._config_data.cluster.serial_no += 1
self._UnlockedReleaseDRBDMinors(instance.name)
+ self._UnlockedCommitTemporaryIps(ec_id)
self._WriteConfig()
def _EnsureUUID(self, item, ec_id):
"""Set the instance's status to a given value.
"""
- assert isinstance(status, bool), \
+ assert status in constants.ADMINST_ALL, \
"Invalid status '%s' passed to 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.admin_up != status:
- instance.admin_up = status
+ if instance.admin_state != status:
+ instance.admin_state = status
instance.serial_no += 1
instance.mtime = time.time()
self._WriteConfig()
"""Mark the instance status to up in the config.
"""
- self._SetInstanceStatus(instance_name, True)
+ self._SetInstanceStatus(instance_name, constants.ADMINST_UP)
+
+ @locking.ssynchronized(_config_lock)
+ def MarkInstanceOffline(self, instance_name):
+ """Mark the instance status to down in the config.
+
+ """
+ self._SetInstanceStatus(instance_name, constants.ADMINST_OFFLINE)
@locking.ssynchronized(_config_lock)
def RemoveInstance(self, instance_name):
"""
if instance_name not in self._config_data.instances:
raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
+
+ # If a network port has been allocated to the instance,
+ # return it to the pool of free ports.
+ inst = self._config_data.instances[instance_name]
+ network_port = getattr(inst, "network_port", None)
+ if network_port is not None:
+ self._config_data.cluster.tcpudp_port_pool.add(network_port)
+
+ instance = self._UnlockedGetInstanceInfo(instance_name)
+
+ for nic in instance.nics:
+ if nic.network and nic.ip:
+ # Return all IP addresses to the respective address pools
+ self._UnlockedCommitIp(constants.RELEASE_ACTION, nic.network, nic.ip)
+
del self._config_data.instances[instance_name]
self._config_data.cluster.serial_no += 1
self._WriteConfig()
"""
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]
+
+ # Operate on a copy to not loose instance object in case of a failure
+ inst = self._config_data.instances[old_name].Copy()
inst.name = new_name
- for disk in inst.disks:
+ for (idx, disk) in enumerate(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_fname = "disk%s" % disk.iv_name.split("/")[1]
- disk.physical_id = disk.logical_id = (disk.logical_id[0],
- utils.PathJoin(file_storage_dir,
- inst.name,
- disk_fname))
+ disk.logical_id = (disk.logical_id[0],
+ utils.PathJoin(file_storage_dir, inst.name,
+ "disk%s" % idx))
+ disk.physical_id = disk.logical_id
+
+ # Actually replace instance object
+ del self._config_data.instances[old_name]
+ self._config_data.instances[inst.name] = inst
# Force update of ssconf files
self._config_data.cluster.serial_no += 1
- self._config_data.instances[inst.name] = inst
self._WriteConfig()
@locking.ssynchronized(_config_lock)
"""Mark the status of an instance to down in the configuration.
"""
- self._SetInstanceStatus(instance_name, False)
+ self._SetInstanceStatus(instance_name, constants.ADMINST_DOWN)
def _UnlockedGetInstanceList(self):
"""Get the list of instances.
for node_name in nodes)
@locking.ssynchronized(_config_lock, shared=1)
+ def GetInstanceNetworks(self, instance_name):
+ """Returns set of network UUIDs for instance's nics.
+
+ @rtype: frozenset
+
+ """
+ instance = self._UnlockedGetInstanceInfo(instance_name)
+ if not instance:
+ raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
+
+ networks = set()
+ for nic in instance.nics:
+ if nic.network:
+ networks.add(nic.network)
+
+ return frozenset(networks)
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetMultiInstanceInfo(self, instances):
"""Get the configuration of multiple instances.
for instance in self._UnlockedGetInstanceList()])
return my_dict
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetInstancesInfoByFilter(self, filter_fn):
+ """Get instance configuration with a filter.
+
+ @type filter_fn: callable
+ @param filter_fn: Filter function receiving instance object as parameter,
+ returning boolean. Important: this function is called while the
+ configuration locks is held. It must not do any complex work or call
+ functions potentially leading to a deadlock. Ideally it doesn't call any
+ other functions and just compares instance attributes.
+
+ """
+ return dict((name, inst)
+ for (name, inst) in self._config_data.instances.items()
+ if filter_fn(inst))
+
@locking.ssynchronized(_config_lock)
def AddNode(self, node, ec_id):
"""Add a node to the configuration.
would GetNodeInfo return for the node
"""
- my_dict = dict([(node, self._UnlockedGetNodeInfo(node))
- for node in self._UnlockedGetNodeList()])
- return my_dict
+ return self._UnlockedGetAllNodesInfo()
+
+ def _UnlockedGetAllNodesInfo(self):
+ """Gets configuration of all nodes.
+
+ @note: See L{GetAllNodesInfo}
+
+ """
+ return dict([(node, self._UnlockedGetNodeInfo(node))
+ for node in self._UnlockedGetNodeList()])
@locking.ssynchronized(_config_lock, shared=1)
def GetNodeGroupsFromNodes(self, nodes):
else:
nodegroup_obj.members.remove(node.name)
+ @locking.ssynchronized(_config_lock)
+ def AssignGroupNodes(self, mods):
+ """Changes the group of a number of nodes.
+
+ @type mods: list of tuples; (node name, new group UUID)
+ @param mods: Node membership modifications
+
+ """
+ groups = self._config_data.nodegroups
+ nodes = self._config_data.nodes
+
+ resmod = []
+
+ # Try to resolve names/UUIDs first
+ for (node_name, new_group_uuid) in mods:
+ try:
+ node = nodes[node_name]
+ except KeyError:
+ raise errors.ConfigurationError("Unable to find node '%s'" % node_name)
+
+ if node.group == new_group_uuid:
+ # Node is being assigned to its current group
+ logging.debug("Node '%s' was assigned to its current group (%s)",
+ node_name, node.group)
+ continue
+
+ # Try to find current group of node
+ try:
+ old_group = groups[node.group]
+ except KeyError:
+ raise errors.ConfigurationError("Unable to find old group '%s'" %
+ node.group)
+
+ # Try to find new group for node
+ try:
+ new_group = groups[new_group_uuid]
+ except KeyError:
+ raise errors.ConfigurationError("Unable to find new group '%s'" %
+ new_group_uuid)
+
+ assert node.name in old_group.members, \
+ ("Inconsistent configuration: node '%s' not listed in members for its"
+ " old group '%s'" % (node.name, old_group.uuid))
+ assert node.name not in new_group.members, \
+ ("Inconsistent configuration: node '%s' already listed in members for"
+ " its new group '%s'" % (node.name, new_group.uuid))
+
+ resmod.append((node, old_group, new_group))
+
+ # Apply changes
+ for (node, old_group, new_group) in resmod:
+ assert node.uuid != new_group.uuid and old_group.uuid != new_group.uuid, \
+ "Assigning to current group is not possible"
+
+ node.group = new_group.uuid
+
+ # Update members of involved groups
+ if node.name in old_group.members:
+ old_group.members.remove(node.name)
+ if node.name not in new_group.members:
+ new_group.members.append(node.name)
+
+ # Update timestamps and serials (only once per node/group object)
+ now = time.time()
+ for obj in frozenset(itertools.chain(*resmod)): # pylint: disable=W0142
+ obj.serial_no += 1
+ obj.mtime = now
+
+ # Force ssconf update
+ self._config_data.cluster.serial_no += 1
+
+ self._WriteConfig()
+
def _BumpSerialNo(self):
"""Bump up the serial number of the config.
return (self._config_data.instances.values() +
self._config_data.nodes.values() +
self._config_data.nodegroups.values() +
+ self._config_data.networks.values() +
[self._config_data.cluster])
def _OpenConfig(self, accept_foreign):
# Make sure the configuration has the right version
_ValidateConfig(data)
- if (not hasattr(data, 'cluster') or
- not hasattr(data.cluster, 'rsahostkeypub')):
+ if (not hasattr(data, "cluster") or
+ not hasattr(data.cluster, "rsahostkeypub")):
raise errors.ConfigurationError("Incomplete configuration"
" (missing cluster.rsahostkeypub)")
(data.cluster.master_node, self._my_hostname))
raise errors.ConfigurationError(msg)
- # Upgrade configuration if needed
- data.UpgradeConfig()
-
self._config_data = data
# reset the last serial as -1 so that the next write will cause
# ssconf update
self._last_cluster_serial = -1
- # And finally run our (custom) config upgrade sequence
+ # Upgrade configuration if needed
self._UpgradeConfig()
self._cfg_id = utils.GetFileID(path=self._cfg_file)
def _UpgradeConfig(self):
- """Run upgrade steps that cannot be done purely in the objects.
+ """Run any upgrade steps.
- This is because some data elements need uniqueness across the
- whole configuration, etc.
+ This method performs both in-object upgrades and also update some data
+ elements that need uniqueness across the whole configuration or interact
+ with other objects.
@warning: this function will call L{_WriteConfig()}, but also
L{DropECReservations} so it needs to be called only from a
created first, to avoid causing deadlock.
"""
- modified = False
+ # Keep a copy of the persistent part of _config_data to check for changes
+ # Serialization doesn't guarantee order in dictionaries
+ oldconf = copy.deepcopy(self._config_data.ToDict())
+
+ # In-object upgrades
+ self._config_data.UpgradeConfig()
+
for item in self._AllUUIDObjects():
if item.uuid is None:
item.uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID)
- modified = True
if not self._config_data.nodegroups:
default_nodegroup_name = constants.INITIAL_NODE_GROUP_NAME
default_nodegroup = objects.NodeGroup(name=default_nodegroup_name,
members=[])
self._UnlockedAddNodeGroup(default_nodegroup, _UPGRADE_CONFIG_JID, True)
- modified = True
for node in self._config_data.nodes.values():
if not node.group:
node.group = self.LookupNodeGroup(None)
- modified = True
# This is technically *not* an upgrade, but needs to be done both when
# nodegroups are being added, and upon normally loading the config,
# because the members list of a node group is discarded upon
# serializing/deserializing the object.
self._UnlockedAddNodeToGroup(node.name, node.group)
+
+ modified = (oldconf != self._config_data.ToDict())
if modified:
self._WriteConfig()
# This is ok even if it acquires the internal lock, as _UpgradeConfig is
# only called at config init time, without the lock held
self.DropECReservations(_UPGRADE_CONFIG_JID)
+ else:
+ config_errors = self._UnlockedVerifyConfig()
+ if config_errors:
+ errmsg = ("Loaded configuration data is not consistent: %s" %
+ (utils.CommaJoin(config_errors)))
+ logging.critical(errmsg)
def _DistributeConfig(self, feedback_fn):
"""Distribute the configuration to the other nodes.
# TODO: Use dedicated resolver talking to config writer for name resolution
result = \
- rpc.ConfigRunner(addr_list).call_upload_file(node_list, self._cfg_file)
+ self._GetRpc(addr_list).call_upload_file(node_list, self._cfg_file)
for to_node, to_result in result.items():
msg = to_result.fail_msg
if msg:
# Write ssconf files on all nodes (including locally)
if self._last_cluster_serial < self._config_data.cluster.serial_no:
if not self._offline:
- result = rpc.ConfigRunner(None).call_write_ssconf_files(
+ result = self._GetRpc(None).call_write_ssconf_files(
self._UnlockedGetOnlineNodeList(),
self._UnlockedGetSsconfValues())
nodegroups = ["%s %s" % (nodegroup.uuid, nodegroup.name) for nodegroup in
self._config_data.nodegroups.values()]
nodegroups_data = fn(utils.NiceSort(nodegroups))
+ networks = ["%s %s" % (net.uuid, net.name) for net in
+ self._config_data.networks.values()]
+ networks_data = fn(utils.NiceSort(networks))
ssconf_values = {
constants.SS_CLUSTER_NAME: cluster.cluster_name,
constants.SS_MAINTAIN_NODE_HEALTH: str(cluster.maintain_node_health),
constants.SS_UID_POOL: uid_pool,
constants.SS_NODEGROUPS: nodegroups_data,
+ constants.SS_NETWORKS: networks_data,
}
bad_values = [(k, v) for k, v in ssconf_values.items()
if not isinstance(v, (str, basestring))]
return self._config_data.HasAnyDiskOfType(dev_type)
@locking.ssynchronized(_config_lock)
- def Update(self, target, feedback_fn):
+ def Update(self, target, feedback_fn, ec_id=None):
"""Notify function to be called after updates.
This function must be called when an object (as returned by
test = target in self._config_data.instances.values()
elif isinstance(target, objects.NodeGroup):
test = target in self._config_data.nodegroups.values()
+ elif isinstance(target, objects.Network):
+ test = target in self._config_data.networks.values()
else:
raise errors.ProgrammerError("Invalid object type (%s) passed to"
" ConfigWriter.Update" % type(target))
if isinstance(target, objects.Instance):
self._UnlockedReleaseDRBDMinors(target.name)
+ if ec_id is not None:
+ # Commit all ips reserved by OpInstanceSetParams and OpGroupSetParams
+ self._UnlockedCommitTemporaryIps(ec_id)
+
self._WriteConfig(feedback_fn=feedback_fn)
@locking.ssynchronized(_config_lock)
"""
for rm in self._all_rms:
rm.DropECReservations(ec_id)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetAllNetworksInfo(self):
+ """Get configuration info of all the networks.
+
+ """
+ return dict(self._config_data.networks)
+
+ def _UnlockedGetNetworkList(self):
+ """Get the list of networks.
+
+ This function is for internal use, when the config lock is already held.
+
+ """
+ return self._config_data.networks.keys()
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNetworkList(self):
+ """Get the list of networks.
+
+ @return: array of networks, ex. ["main", "vlan100", "200]
+
+ """
+ return self._UnlockedGetNetworkList()
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNetworkNames(self):
+ """Get a list of network names
+
+ """
+ names = [net.name
+ for net in self._config_data.networks.values()]
+ return names
+
+ def _UnlockedGetNetwork(self, uuid):
+ """Returns information about a network.
+
+ This function is for internal use, when the config lock is already held.
+
+ """
+ if uuid not in self._config_data.networks:
+ return None
+
+ return self._config_data.networks[uuid]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNetwork(self, uuid):
+ """Returns information about a network.
+
+ It takes the information from the configuration file.
+
+ @param uuid: UUID of the network
+
+ @rtype: L{objects.Network}
+ @return: the network object
+
+ """
+ return self._UnlockedGetNetwork(uuid)
+
+ @locking.ssynchronized(_config_lock)
+ def AddNetwork(self, net, ec_id, check_uuid=True):
+ """Add a network to the configuration.
+
+ @type net: L{objects.Network}
+ @param net: the Network object to add
+ @type ec_id: string
+ @param ec_id: unique id for the job to use when creating a missing UUID
+
+ """
+ self._UnlockedAddNetwork(net, ec_id, check_uuid)
+ self._WriteConfig()
+
+ def _UnlockedAddNetwork(self, net, ec_id, check_uuid):
+ """Add a network to the configuration.
+
+ """
+ logging.info("Adding network %s to configuration", net.name)
+
+ if check_uuid:
+ self._EnsureUUID(net, ec_id)
+
+ net.serial_no = 1
+ self._config_data.networks[net.uuid] = net
+ self._config_data.cluster.serial_no += 1
+
+ def _UnlockedLookupNetwork(self, target):
+ """Lookup a network's UUID.
+
+ @type target: string
+ @param target: network name or UUID
+ @rtype: string
+ @return: network UUID
+ @raises errors.OpPrereqError: when the target network cannot be found
+
+ """
+ if target is None:
+ return None
+ if target in self._config_data.networks:
+ return target
+ for net in self._config_data.networks.values():
+ if net.name == target:
+ return net.uuid
+ raise errors.OpPrereqError("Network '%s' not found" % target,
+ errors.ECODE_NOENT)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def LookupNetwork(self, target):
+ """Lookup a network's UUID.
+
+ This function is just a wrapper over L{_UnlockedLookupNetwork}.
+
+ @type target: string
+ @param target: network name or UUID
+ @rtype: string
+ @return: network UUID
+
+ """
+ return self._UnlockedLookupNetwork(target)
+
+ @locking.ssynchronized(_config_lock)
+ def RemoveNetwork(self, network_uuid):
+ """Remove a network from the configuration.
+
+ @type network_uuid: string
+ @param network_uuid: the UUID of the network to remove
+
+ """
+ logging.info("Removing network %s from configuration", network_uuid)
+
+ if network_uuid not in self._config_data.networks:
+ raise errors.ConfigurationError("Unknown network '%s'" % network_uuid)
+
+ del self._config_data.networks[network_uuid]
+ self._config_data.cluster.serial_no += 1
+ self._WriteConfig()
+
+ def _UnlockedGetGroupNetParams(self, net_uuid, node):
+ """Get the netparams (mode, link) of a network.
+
+ Get a network's netparams for a given node.
+
+ @type net_uuid: string
+ @param net_uuid: network uuid
+ @type node: string
+ @param node: node name
+ @rtype: dict or None
+ @return: netparams
+
+ """
+ node_info = self._UnlockedGetNodeInfo(node)
+ nodegroup_info = self._UnlockedGetNodeGroup(node_info.group)
+ netparams = nodegroup_info.networks.get(net_uuid, None)
+
+ return netparams
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetGroupNetParams(self, net_uuid, node):
+ """Locking wrapper of _UnlockedGetGroupNetParams()
+
+ """
+ return self._UnlockedGetGroupNetParams(net_uuid, node)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def CheckIPInNodeGroup(self, ip, node):
+ """Check IP uniqueness in nodegroup.
+
+ Check networks that are connected in the node's node group
+ if ip is contained in any of them. Used when creating/adding
+ a NIC to ensure uniqueness among nodegroups.
+
+ @type ip: string
+ @param ip: ip address
+ @type node: string
+ @param node: node name
+ @rtype: (string, dict) or (None, None)
+ @return: (network name, netparams)
+
+ """
+ if ip is None:
+ return (None, None)
+ node_info = self._UnlockedGetNodeInfo(node)
+ nodegroup_info = self._UnlockedGetNodeGroup(node_info.group)
+ for net_uuid in nodegroup_info.networks.keys():
+ net_info = self._UnlockedGetNetwork(net_uuid)
+ pool = network.AddressPool(net_info)
+ if pool.Contains(ip):
+ return (net_info.name, nodegroup_info.networks[net_uuid])
+
+ return (None, None)