#
#
-# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 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
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._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():
def GetNdParams(self, node):
"""Get the node params populated with cluster defaults.
- @type node: L{object.Node}
+ @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
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-msg=R0914
+ # pylint: disable=R0914
result = []
seen_macs = []
ports = {}
except errors.ConfigurationError, err:
result.append("%s has invalid nicparams: %s" % (owner, err))
+ def _helper_ipolicy(owner, params):
+ try:
+ objects.InstancePolicy.CheckParameterSyntax(params)
+ 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_nic("cluster", cluster.SimpleFillNIC({}))
_helper("cluster", "ndparams", cluster.SimpleFillND({}),
constants.NDS_PARAMETER_TYPES)
+ _helper_ipolicy("cluster", cluster.SimpleFillIPolicy({}))
+ _helper_ispecs("cluster", cluster.SimpleFillIPolicy({}))
# per-instance checks
for instance_name in data.instances:
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:
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))
+ _helper_ispecs(group_name, cluster.SimpleFillIPolicy(nodegroup.ipolicy))
if nodegroup.ndparams:
- _helper("group %s" % nodegroup.name, "ndparams",
+ _helper(group_name, "ndparams",
cluster.SimpleFillND(nodegroup.ndparams),
constants.NDS_PARAMETER_TYPES)
-
# drbd minors check
_, duplicates = self._UnlockedComputeDRBDMap()
for node, minor, instance_a, instance_b in duplicates:
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.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.
"""
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)
def AddInstance(self, instance, ec_id):
"""Add an instance to the config.
"""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_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 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.
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 [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.
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.
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:
# 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())
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,