#
#
-# Copyright (C) 2006, 2007, 2008, 2009, 2010 Google Inc.
+# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 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-msg=R0904
+# pylint: disable=R0904
# R0904: Too many public methods
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
_config_lock = locking.SharedLock("ConfigWriter")
"""
if data.version != constants.CONFIG_VERSION:
- raise errors.ConfigurationError("Cluster configuration version"
- " mismatch, got %s instead of %s" %
- (data.version,
- constants.CONFIG_VERSION))
+ raise errors.ConfigVersionMismatch(constants.CONFIG_VERSION, data.version)
class TemporaryReservationManager:
self._ec_reserved = {}
def Reserved(self, resource):
- for holder_reserved in self._ec_reserved.items():
+ for holder_reserved in self._ec_reserved.values():
if resource in holder_reserved:
return True
return False
def Reserve(self, ec_id, resource):
if self.Reserved(resource):
- raise errors.ReservationError("Duplicate reservation for resource: %s." %
- (resource))
+ raise errors.ReservationError("Duplicate reservation for resource '%s'"
+ % str(resource))
if ec_id not in self._ec_reserved:
self._ec_reserved[ec_id] = set([resource])
else:
return new_resource
+def _MatchNameComponentIgnoreCase(short_name, names):
+ """Wrapper around L{utils.text.MatchNameComponent}.
+
+ """
+ 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.
@ivar _all_rms: a list of all temporary reservation managers
"""
- def __init__(self, cfg_file=None, offline=False, _getents=runtime.GetEnts):
+ def __init__(self, cfg_file=None, offline=False, _getents=runtime.GetEnts,
+ accept_foreign=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
+ self._cfg_file = pathutils.CLUSTER_CONF_FILE
else:
self._cfg_file = cfg_file
self._getents = _getents
# file than after it was modified
self._my_hostname = netutils.Hostname.GetSysName()
self._last_cluster_serial = -1
- self._OpenConfig()
+ 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
"""Check if the cluster is configured.
"""
- return os.path.exists(constants.CLUSTER_CONF_FILE)
+ return os.path.exists(pathutils.CLUSTER_CONF_FILE)
def _GenerateOneMAC(self):
"""Generate one mac address
return mac
@locking.ssynchronized(_config_lock, shared=1)
+ def GetNdParams(self, node):
+ """Get the node params populated with cluster defaults.
+
+ @type node: L{objects.Node}
+ @param node: The node we want to know the params for
+ @return: A dict with the filled in node params
+
+ """
+ nodegroup = self._UnlockedGetNodeGroup(node.group)
+ return self._config_data.cluster.FillND(node, nodegroup)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ 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)
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GenerateMAC(self, ec_id):
"""Generate a MAC for an instance.
if mac in all_macs:
raise errors.ReservationError("mac already in use")
else:
- self._temporary_macs.Reserve(mac, ec_id)
+ self._temporary_macs.Reserve(ec_id, mac)
@locking.ssynchronized(_config_lock, shared=1)
def ReserveLV(self, lv_name, ec_id):
if lv_name in all_lvs:
raise errors.ReservationError("LV already in use")
else:
- self._temporary_lvs.Reserve(lv_name, ec_id)
+ self._temporary_lvs.Reserve(ec_id, lv_name)
@locking.ssynchronized(_config_lock, shared=1)
def GenerateDRBDSecret(self, ec_id):
configuration errors
"""
+ # pylint: disable=R0914
result = []
seen_macs = []
ports = {}
data = self._config_data
+ cluster = data.cluster
seen_lids = []
seen_pids = []
# global cluster checks
- if not data.cluster.enabled_hypervisors:
+ if not cluster.enabled_hypervisors:
result.append("enabled hypervisors list doesn't have any entries")
- invalid_hvs = set(data.cluster.enabled_hypervisors) - constants.HYPER_TYPES
+ invalid_hvs = set(cluster.enabled_hypervisors) - constants.HYPER_TYPES
if invalid_hvs:
result.append("enabled hypervisors contains invalid entries: %s" %
invalid_hvs)
- missing_hvp = (set(data.cluster.enabled_hypervisors) -
- set(data.cluster.hvparams.keys()))
+ 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 data.cluster.master_node not in data.nodes:
+ if cluster.master_node not in data.nodes:
result.append("cluster has invalid primary node '%s'" %
- data.cluster.master_node)
+ cluster.master_node)
+
+ def _helper(owner, attr, value, template):
+ try:
+ utils.ForceDictType(value, template)
+ except errors.GenericError, err:
+ result.append("%s has invalid %s: %s" % (owner, attr, err))
+
+ def _helper_nic(owner, params):
+ try:
+ objects.NIC.CheckParameterSyntax(params)
+ except errors.ConfigurationError, err:
+ result.append("%s has invalid nicparams: %s" % (owner, err))
+
+ def _helper_ipolicy(owner, params, check_std):
+ try:
+ objects.InstancePolicy.CheckParameterSyntax(params, check_std)
+ except errors.ConfigurationError, err:
+ result.append("%s has invalid instance policy: %s" % (owner, err))
+
+ def _helper_ispecs(owner, params):
+ for key, value in params.items():
+ if key in constants.IPOLICY_ISPECS:
+ fullkey = "ipolicy/" + key
+ _helper(owner, fullkey, 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)))
+
+ # check cluster parameters
+ _helper("cluster", "beparams", cluster.SimpleFillBE({}),
+ constants.BES_PARAMETER_TYPES)
+ _helper("cluster", "nicparams", cluster.SimpleFillNIC({}),
+ constants.NICS_PARAMETER_TYPES)
+ _helper_nic("cluster", cluster.SimpleFillNIC({}))
+ _helper("cluster", "ndparams", cluster.SimpleFillND({}),
+ constants.NDS_PARAMETER_TYPES)
+ _helper_ipolicy("cluster", cluster.SimpleFillIPolicy({}), True)
+ _helper_ispecs("cluster", cluster.SimpleFillIPolicy({}))
# per-instance checks
for instance_name in data.instances:
(instance_name, idx, nic.mac))
else:
seen_macs.append(nic.mac)
+ if nic.nicparams:
+ filled = cluster.SimpleFillNIC(nic.nicparams)
+ owner = "instance %s nic %d" % (instance.name, idx)
+ _helper(owner, "nicparams",
+ filled, constants.NICS_PARAMETER_TYPES)
+ _helper_nic(owner, filled)
+
+ # 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 data.cluster.tcpudp_port_pool:
+ for free_port in cluster.tcpudp_port_pool:
if free_port not in ports:
ports[free_port] = []
ports[free_port].append(("cluster", "port marked as free"))
# highest used tcp port check
if keys:
- if keys[-1] > data.cluster.highest_used_port:
+ if keys[-1] > cluster.highest_used_port:
result.append("Highest used port mismatch, saved %s, computed %s" %
- (data.cluster.highest_used_port, keys[-1]))
+ (cluster.highest_used_port, keys[-1]))
- if not data.nodes[data.cluster.master_node].master_candidate:
+ if not data.nodes[cluster.master_node].master_candidate:
result.append("Master node is not a master candidate")
# master candidate checks
if [node.master_candidate, node.drained, node.offline].count(True) > 1:
result.append("Node %s state is invalid: master_candidate=%s,"
" drain=%s, offline=%s" %
- (node.name, node.master_candidate, node.drain,
+ (node.name, node.master_candidate, node.drained,
node.offline))
+ if node.group not in data.nodegroups:
+ result.append("Node '%s' has invalid group '%s'" %
+ (node.name, node.group))
+ else:
+ _helper("node %s" % node.name, "ndparams",
+ cluster.FillND(node, data.nodegroups[node.group]),
+ constants.NDS_PARAMETER_TYPES)
# nodegroups checks
+ nodegroups_names = set()
for nodegroup_uuid in data.nodegroups:
nodegroup = data.nodegroups[nodegroup_uuid]
if nodegroup.uuid != nodegroup_uuid:
- result.append("nodegroup '%s' (uuid: '%s') indexed by wrong uuid '%s'"
+ result.append("node group '%s' (uuid: '%s') indexed by wrong uuid '%s'"
% (nodegroup.name, nodegroup.uuid, nodegroup_uuid))
+ if utils.UUID_RE.match(nodegroup.name.lower()):
+ result.append("node group '%s' (uuid: '%s') has uuid-like name" %
+ (nodegroup.name, nodegroup.uuid))
+ if nodegroup.name in nodegroups_names:
+ 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)
+ _helper_ispecs(group_name, cluster.SimpleFillIPolicy(nodegroup.ipolicy))
+ if nodegroup.ndparams:
+ _helper(group_name, "ndparams",
+ cluster.SimpleFillND(nodegroup.ndparams),
+ constants.NDS_PARAMETER_TYPES)
# drbd minors check
_, duplicates = self._UnlockedComputeDRBDMap()
" %s and %s" % (minor, node, instance_a, instance_b))
# IP checks
- default_nicparams = data.cluster.nicparams[constants.PP_DEFAULT]
+ default_nicparams = cluster.nicparams[constants.PP_DEFAULT]
ips = {}
def _AddIpAddress(ip, name):
ips.setdefault(ip, []).append(name)
- _AddIpAddress(data.cluster.master_ip, "cluster_ip")
+ _AddIpAddress(cluster.master_ip, "cluster_ip")
for node in data.nodes.values():
_AddIpAddress(node.primary_ip, "node:%s/primary" % node.name)
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_netdev
@locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterNetmask(self):
+ """Get the netmask of the master node for this cluster.
+
+ """
+ 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.file_storage_dir
@locking.ssynchronized(_config_lock, shared=1)
+ def GetSharedFileStorageDir(self):
+ """Get the shared file storage dir for this cluster.
+
+ """
+ return self._config_data.cluster.shared_file_storage_dir
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetHypervisorType(self):
"""Get the hypervisor type for this cluster.
return self._config_data.cluster.primary_ip_family
@locking.ssynchronized(_config_lock, shared=1)
- def LookupNodeGroup(self, target):
- """Lookup a node group.
+ 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.
+
+ This method calls group.UpgradeConfig() to fill any missing attributes
+ according to their default values.
+
+ @type group: L{objects.NodeGroup}
+ @param group: the NodeGroup object to add
+ @type ec_id: string
+ @param ec_id: unique id for the job to use when creating a missing UUID
+ @type check_uuid: bool
+ @param check_uuid: add an UUID to the group if it doesn't have one or, if
+ it does, ensure that it does not exist in the
+ configuration already
+
+ """
+ self._UnlockedAddNodeGroup(group, ec_id, check_uuid)
+ self._WriteConfig()
+
+ def _UnlockedAddNodeGroup(self, group, ec_id, check_uuid):
+ """Add a node group to the configuration.
+
+ """
+ logging.info("Adding node group %s to configuration", group.name)
+
+ # Some code might need to add a node group with a pre-populated UUID
+ # generated with ConfigWriter.GenerateUniqueID(). We allow them to bypass
+ # the "does this UUID" exist already check.
+ if check_uuid:
+ self._EnsureUUID(group, ec_id)
+
+ try:
+ existing_uuid = self._UnlockedLookupNodeGroup(group.name)
+ except errors.OpPrereqError:
+ pass
+ else:
+ raise errors.OpPrereqError("Desired group name '%s' already exists as a"
+ " node group (UUID: %s)" %
+ (group.name, existing_uuid),
+ errors.ECODE_EXISTS)
+
+ group.serial_no = 1
+ group.ctime = group.mtime = time.time()
+ group.UpgradeConfig()
+
+ self._config_data.nodegroups[group.uuid] = group
+ self._config_data.cluster.serial_no += 1
+
+ @locking.ssynchronized(_config_lock)
+ def RemoveNodeGroup(self, group_uuid):
+ """Remove a node group from the configuration.
+
+ @type group_uuid: string
+ @param group_uuid: the UUID of the node group to remove
+
+ """
+ logging.info("Removing node group %s from configuration", group_uuid)
+
+ if group_uuid not in self._config_data.nodegroups:
+ raise errors.ConfigurationError("Unknown node group '%s'" % group_uuid)
+
+ assert len(self._config_data.nodegroups) != 1, \
+ "Group '%s' is the only group, cannot be removed" % group_uuid
+
+ del self._config_data.nodegroups[group_uuid]
+ self._config_data.cluster.serial_no += 1
+ self._WriteConfig()
+
+ def _UnlockedLookupNodeGroup(self, target):
+ """Lookup a node group's UUID.
@type target: string or None
- @param target: group name or uuid or None to look for the default
+ @param target: group name or UUID or None to look for the default
@rtype: string
- @return: nodegroup uuid
+ @return: nodegroup UUID
@raises errors.OpPrereqError: when the target group cannot be found
"""
if target is None:
if len(self._config_data.nodegroups) != 1:
- raise errors.OpPrereqError("More than one nodegroup exists. Target"
- " group must be specified explicitely.")
+ raise errors.OpPrereqError("More than one node group exists. Target"
+ " group must be specified explicitly.")
else:
return self._config_data.nodegroups.keys()[0]
if target in self._config_data.nodegroups:
for nodegroup in self._config_data.nodegroups.values():
if nodegroup.name == target:
return nodegroup.uuid
- raise errors.OpPrereqError("Nodegroup '%s' not found", target)
+ raise errors.OpPrereqError("Node group '%s' not found" % target,
+ errors.ECODE_NOENT)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def LookupNodeGroup(self, target):
+ """Lookup a node group's UUID.
+
+ This function is just a wrapper over L{_UnlockedLookupNodeGroup}.
+
+ @type target: string or None
+ @param target: group name or UUID or None to look for the default
+ @rtype: string
+ @return: nodegroup UUID
+
+ """
+ return self._UnlockedLookupNodeGroup(target)
+
+ def _UnlockedGetNodeGroup(self, uuid):
+ """Lookup a node group.
+
+ @type uuid: string
+ @param uuid: group UUID
+ @rtype: L{objects.NodeGroup} or None
+ @return: nodegroup object, or None if not found
+
+ """
+ if uuid not in self._config_data.nodegroups:
+ return None
+
+ return self._config_data.nodegroups[uuid]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeGroup(self, uuid):
+ """Lookup a node group.
+
+ @type uuid: string
+ @param uuid: group UUID
+ @rtype: L{objects.NodeGroup} or None
+ @return: nodegroup object, or None if not found
+
+ """
+ return self._UnlockedGetNodeGroup(uuid)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetAllNodeGroupsInfo(self):
+ """Get the configuration of all node groups.
+
+ """
+ return dict(self._config_data.nodegroups)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeGroupList(self):
+ """Get a list of node groups.
+
+ """
+ return self._config_data.nodegroups.keys()
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeGroupMembersByNodes(self, nodes):
+ """Get nodes which are member in the same nodegroups as the given nodes.
+
+ """
+ ngfn = lambda node_name: self._UnlockedGetNodeInfo(node_name).group
+ return frozenset(member_name
+ for node_name in nodes
+ 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):
"""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)
+
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.physical_id = disk.logical_id = (disk.logical_id[0],
- utils.PathJoin(file_storage_dir,
- inst.name,
- disk.iv_name))
+ 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._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.
"""
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(),
- case_sensitive=False)
+ # Locking is done in L{ConfigWriter.GetInstanceList}
+ return _MatchNameComponentIgnoreCase(short_name, self.GetInstanceList())
def _UnlockedGetInstanceInfo(self, instance_name):
"""Returns information about an instance.
return self._UnlockedGetInstanceInfo(instance_name)
@locking.ssynchronized(_config_lock, shared=1)
+ def GetInstanceNodeGroups(self, instance_name, primary_only=False):
+ """Returns set of node group UUIDs for instance's nodes.
+
+ @rtype: frozenset
+
+ """
+ instance = self._UnlockedGetInstanceInfo(instance_name)
+ if not instance:
+ raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
+
+ if primary_only:
+ nodes = [instance.primary_node]
+ else:
+ nodes = instance.all_nodes
+
+ return frozenset(self._UnlockedGetNodeInfo(node_name).group
+ for node_name in nodes)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMultiInstanceInfo(self, instances):
+ """Get the configuration of multiple instances.
+
+ @param instances: list of instance names
+ @rtype: list
+ @return: list of tuples (instance, instance_info), where
+ instance_info is what would GetInstanceInfo return for the
+ node, while keeping the original order
+
+ """
+ return [(name, self._UnlockedGetInstanceInfo(name)) for name in instances]
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetAllInstancesInfo(self):
"""Get the configuration of all 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.
node.serial_no = 1
node.ctime = node.mtime = time.time()
- self._UnlockedAddNodeToGroup(node.name, node.nodegroup)
+ self._UnlockedAddNodeToGroup(node.name, node.group)
self._config_data.nodes[node.name] = node
self._config_data.cluster.serial_no += 1
self._WriteConfig()
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.
+ """Attempt to expand an incomplete node name.
"""
- return utils.MatchNameComponent(short_name,
- self._config_data.nodes.keys(),
- case_sensitive=False)
+ # Locking is done in L{ConfigWriter.GetNodeList}
+ return _MatchNameComponentIgnoreCase(short_name, self.GetNodeList())
def _UnlockedGetNodeInfo(self, node_name):
"""Get the configuration of a node, as stored in the config.
"""
return self._UnlockedGetNodeInfo(node_name)
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeInstances(self, node_name):
+ """Get the instances of a node, as stored in the config.
+
+ @param node_name: the node name, e.g. I{node1.example.com}
+
+ @rtype: (list, list)
+ @return: a tuple with two lists: the primary and the secondary instances
+
+ """
+ pri = []
+ sec = []
+ for inst in self._config_data.instances.values():
+ if inst.primary_node == node_name:
+ pri.append(inst.name)
+ if node_name in inst.secondary_nodes:
+ sec.append(inst.name)
+ return (pri, sec)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeGroupInstances(self, uuid, primary_only=False):
+ """Get the instances of a node group.
+
+ @param uuid: Node group UUID
+ @param primary_only: Whether to only consider primary nodes
+ @rtype: frozenset
+ @return: List of instance names in node group
+
+ """
+ if primary_only:
+ nodes_fn = lambda inst: [inst.primary_node]
+ else:
+ nodes_fn = lambda inst: inst.all_nodes
+
+ return frozenset(inst.name
+ for inst in self._config_data.instances.values()
+ for node_name in nodes_fn(inst)
+ if self._UnlockedGetNodeInfo(node_name).group == uuid)
+
def _UnlockedGetNodeList(self):
"""Return the list of nodes which are in the configuration.
return self._UnlockedGetOnlineNodeList()
@locking.ssynchronized(_config_lock, shared=1)
+ def GetVmCapableNodeList(self):
+ """Return the list of nodes which are not vm capable.
+
+ """
+ all_nodes = [self._UnlockedGetNodeInfo(node)
+ for node in self._UnlockedGetNodeList()]
+ return [node.name for node in all_nodes if node.vm_capable]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNonVmCapableNodeList(self):
+ """Return the list of nodes which are not vm capable.
+
+ """
+ all_nodes = [self._UnlockedGetNodeInfo(node)
+ for node in self._UnlockedGetNodeList()]
+ return [node.name for node in all_nodes if not node.vm_capable]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMultiNodeInfo(self, nodes):
+ """Get the configuration of multiple nodes.
+
+ @param nodes: list of node names
+ @rtype: list
+ @return: list of tuples of (node, node_info), where node_info is
+ what would GetNodeInfo return for the node, in the original
+ order
+
+ """
+ return [(name, self._UnlockedGetNodeInfo(name)) for name in nodes]
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetAllNodesInfo(self):
"""Get the configuration of all nodes.
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):
+ """Returns groups for a list of nodes.
+
+ @type nodes: list of string
+ @param nodes: List of node names
+ @rtype: frozenset
+
+ """
+ return frozenset(self._UnlockedGetNodeInfo(name).group for name in nodes)
def _UnlockedGetMasterCandidateStats(self, exceptions=None):
"""Get the number of current and maximum desired and possible candidates.
for node in self._config_data.nodes.values():
if exceptions and node.name in exceptions:
continue
- if not (node.offline or node.drained):
+ if not (node.offline or node.drained) and node.master_capable:
mc_max += 1
if node.master_candidate:
mc_now += 1
break
node = self._config_data.nodes[name]
if (node.master_candidate or node.offline or node.drained or
- node.name in exceptions):
+ node.name in exceptions or not node.master_capable):
continue
mod_list.append(node)
node.master_candidate = True
# when we're adding the first node to it, since we don't keep a lock in
# the meantime. It's ok though, as we'll fail cleanly if the node group
# is not found anymore.
- raise errors.OpExecError("Unknown nodegroup: %s" % nodegroup_uuid)
+ raise errors.OpExecError("Unknown node group: %s" % nodegroup_uuid)
if node_name not in self._config_data.nodegroups[nodegroup_uuid].members:
self._config_data.nodegroups[nodegroup_uuid].members.append(node_name)
"""Remove a given node from its group.
"""
- nodegroup = node.nodegroup
+ nodegroup = node.group
if nodegroup not in self._config_data.nodegroups:
- logging.warning("Warning: node '%s' has a non-existing nodegroup '%s'"
+ logging.warning("Warning: node '%s' has unknown node group '%s'"
" (while being removed from it)", node.name, nodegroup)
nodegroup_obj = self._config_data.nodegroups[nodegroup]
if node.name not in nodegroup_obj.members:
- logging.warning("Warning: node '%s' not a member of its nodegroup '%s'"
+ logging.warning("Warning: node '%s' not a member of its node group '%s'"
" (while being removed from it)", node.name, nodegroup)
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.
self._config_data.nodegroups.values() +
[self._config_data.cluster])
- def _OpenConfig(self):
+ def _OpenConfig(self, accept_foreign):
"""Read the config data from disk.
"""
# 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)")
+ if data.cluster.master_node != self._my_hostname and not accept_foreign:
+ msg = ("The configuration denotes node %s as master, while my"
+ " hostname is %s; opening a foreign configuration is only"
+ " possible in accept_foreign mode" %
+ (data.cluster.master_node, self._my_hostname))
+ raise errors.ConfigurationError(msg)
+
# Upgrade configuration if needed
data.UpgradeConfig()
# And finally run our (custom) config upgrade sequence
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.
item.uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID)
modified = True
if not self._config_data.nodegroups:
- default_nodegroup_uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID)
- default_nodegroup = objects.NodeGroup(
- uuid=default_nodegroup_uuid,
- name="default",
- members=[],
- )
- self._config_data.nodegroups[default_nodegroup_uuid] = default_nodegroup
+ 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.nodegroup:
- node.nodegroup = self.LookupNodeGroup(None)
+ 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.nodegroup)
+ self._UnlockedAddNodeToGroup(node.name, node.group)
if modified:
self._WriteConfig()
# This is ok even if it acquires the internal lock, as _UpgradeConfig is
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)
+ # TODO: Use dedicated resolver talking to config writer for name resolution
+ result = \
+ 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:
txt = serializer.Dump(self._config_data.ToDict())
getents = self._getents()
- utils.WriteFile(destination, data=txt, gid=getents.confd_gid, mode=0640)
+ try:
+ fd = utils.SafeWriteFile(destination, self._cfg_id, data=txt,
+ close=False, gid=getents.confd_gid, mode=0640)
+ except errors.LockError:
+ raise errors.ConfigurationError("The configuration file has been"
+ " modified since the last write, cannot"
+ " update")
+ try:
+ self._cfg_id = utils.GetFileID(fd=fd)
+ finally:
+ os.close(fd)
self.write_count += 1
# 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.RpcRunner.call_write_ssconf_files(
+ result = self._GetRpc(None).call_write_ssconf_files(
self._UnlockedGetOnlineNodeList(),
self._UnlockedGetSsconfValues())
self._config_data.nodegroups.values()]
nodegroups_data = fn(utils.NiceSort(nodegroups))
- return {
+ ssconf_values = {
constants.SS_CLUSTER_NAME: cluster.cluster_name,
constants.SS_CLUSTER_TAGS: cluster_tags,
constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
+ constants.SS_SHARED_FILE_STORAGE_DIR: cluster.shared_file_storage_dir,
constants.SS_MASTER_CANDIDATES: mc_data,
constants.SS_MASTER_CANDIDATES_IPS: mc_ips_data,
constants.SS_MASTER_IP: cluster.master_ip,
constants.SS_MASTER_NETDEV: cluster.master_netdev,
+ constants.SS_MASTER_NETMASK: str(cluster.master_netmask),
constants.SS_MASTER_NODE: cluster.master_node,
constants.SS_NODE_LIST: node_data,
constants.SS_NODE_PRIMARY_IPS: node_pri_ips_data,
constants.SS_UID_POOL: uid_pool,
constants.SS_NODEGROUPS: nodegroups_data,
}
+ bad_values = [(k, v) for k, v in ssconf_values.items()
+ if not isinstance(v, (str, basestring))]
+ if bad_values:
+ err = utils.CommaJoin("%s=%s" % (k, v) for k, v in bad_values)
+ raise errors.ConfigurationError("Some ssconf key(s) have non-string"
+ " values: %s" % err)
+ return ssconf_values
@locking.ssynchronized(_config_lock, shared=1)
def GetSsconfValues(self):
update_serial = True
elif isinstance(target, objects.Instance):
test = target in self._config_data.instances.values()
+ elif isinstance(target, objects.NodeGroup):
+ test = target in self._config_data.nodegroups.values()
else:
raise errors.ProgrammerError("Invalid object type (%s) passed to"
" ConfigWriter.Update" % type(target))