--- /dev/null
+#
+#
+
+# 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
+# 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.
+
+
+"""Logical units dealing with instances."""
+
+import OpenSSL
+import copy
+import logging
+import os
+
+from ganeti import compat
+from ganeti import constants
+from ganeti import errors
+from ganeti import ht
+from ganeti import hypervisor
+from ganeti import locking
+from ganeti.masterd import iallocator
+from ganeti import masterd
+from ganeti import netutils
+from ganeti import objects
+from ganeti import opcodes
+from ganeti import pathutils
+from ganeti import rpc
+from ganeti import utils
+
+from ganeti.cmdlib.base import NoHooksLU, LogicalUnit, ResultWithJobs
+
+from ganeti.cmdlib.common import INSTANCE_DOWN, \
+ INSTANCE_NOT_RUNNING, CAN_CHANGE_INSTANCE_OFFLINE, CheckNodeOnline, \
+ ShareAll, GetDefaultIAllocator, CheckInstanceNodeGroups, \
+ LoadNodeEvacResult, CheckIAllocatorOrNode, CheckParamsNotGlobal, \
+ IsExclusiveStorageEnabledNode, CheckHVParams, CheckOSParams, \
+ AnnotateDiskParams, GetUpdatedParams, ExpandInstanceName, \
+ ComputeIPolicySpecViolation, CheckInstanceState, ExpandNodeName
+from ganeti.cmdlib.instance_storage import CreateDisks, \
- CheckNodesFreeDiskPerVG, WipeDisks, WaitForSync, \
++ CheckNodesFreeDiskPerVG, WipeDisks, WipeOrCleanupDisks, WaitForSync, \
+ IsExclusiveStorageEnabledNodeName, CreateSingleBlockDev, ComputeDisks, \
+ CheckRADOSFreeSpace, ComputeDiskSizePerVG, GenerateDiskTemplate, \
- CreateBlockDev, StartInstanceDisks, ShutdownInstanceDisks, \
- AssembleInstanceDisks
++ StartInstanceDisks, ShutdownInstanceDisks, AssembleInstanceDisks
+from ganeti.cmdlib.instance_utils import BuildInstanceHookEnvByObject, \
+ GetClusterDomainSecret, BuildInstanceHookEnv, NICListToTuple, \
+ NICToTuple, CheckNodeNotDrained, RemoveInstance, CopyLockList, \
+ ReleaseLocks, CheckNodeVmCapable, CheckTargetNodeIPolicy, \
+ GetInstanceInfoText, RemoveDisks, CheckNodeFreeMemory, \
+ CheckInstanceBridgesExist, CheckNicsBridgesExist, CheckNodeHasOS
+
+import ganeti.masterd.instance
+
+
+#: Type description for changes as returned by L{_ApplyContainerMods}'s
+#: callbacks
+_TApplyContModsCbChanges = \
+ ht.TMaybeListOf(ht.TAnd(ht.TIsLength(2), ht.TItems([
+ ht.TNonEmptyString,
+ ht.TAny,
+ ])))
+
+
+def _CheckHostnameSane(lu, name):
+ """Ensures that a given hostname resolves to a 'sane' name.
+
+ The given name is required to be a prefix of the resolved hostname,
+ to prevent accidental mismatches.
+
+ @param lu: the logical unit on behalf of which we're checking
+ @param name: the name we should resolve and check
+ @return: the resolved hostname object
+
+ """
+ hostname = netutils.GetHostname(name=name)
+ if hostname.name != name:
+ lu.LogInfo("Resolved given name '%s' to '%s'", name, hostname.name)
+ if not utils.MatchNameComponent(name, [hostname.name]):
+ raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
+ " same as given hostname '%s'") %
+ (hostname.name, name), errors.ECODE_INVAL)
+ return hostname
+
+
+def _CheckOpportunisticLocking(op):
+ """Generate error if opportunistic locking is not possible.
+
+ """
+ if op.opportunistic_locking and not op.iallocator:
+ raise errors.OpPrereqError("Opportunistic locking is only available in"
+ " combination with an instance allocator",
+ errors.ECODE_INVAL)
+
+
+def _CreateInstanceAllocRequest(op, disks, nics, beparams, node_whitelist):
+ """Wrapper around IAReqInstanceAlloc.
+
+ @param op: The instance opcode
+ @param disks: The computed disks
+ @param nics: The computed nics
+ @param beparams: The full filled beparams
+ @param node_whitelist: List of nodes which should appear as online to the
+ allocator (unless the node is already marked offline)
+
+ @returns: A filled L{iallocator.IAReqInstanceAlloc}
+
+ """
+ spindle_use = beparams[constants.BE_SPINDLE_USE]
+ return iallocator.IAReqInstanceAlloc(name=op.instance_name,
+ disk_template=op.disk_template,
+ tags=op.tags,
+ os=op.os_type,
+ vcpus=beparams[constants.BE_VCPUS],
+ memory=beparams[constants.BE_MAXMEM],
+ spindle_use=spindle_use,
+ disks=disks,
+ nics=[n.ToDict() for n in nics],
+ hypervisor=op.hypervisor,
+ node_whitelist=node_whitelist)
+
+
+def _ComputeFullBeParams(op, cluster):
+ """Computes the full beparams.
+
+ @param op: The instance opcode
+ @param cluster: The cluster config object
+
+ @return: The fully filled beparams
+
+ """
+ default_beparams = cluster.beparams[constants.PP_DEFAULT]
+ for param, value in op.beparams.iteritems():
+ if value == constants.VALUE_AUTO:
+ op.beparams[param] = default_beparams[param]
+ objects.UpgradeBeParams(op.beparams)
+ utils.ForceDictType(op.beparams, constants.BES_PARAMETER_TYPES)
+ return cluster.SimpleFillBE(op.beparams)
+
+
+def _ComputeNics(op, cluster, default_ip, cfg, ec_id):
+ """Computes the nics.
+
+ @param op: The instance opcode
+ @param cluster: Cluster configuration object
+ @param default_ip: The default ip to assign
+ @param cfg: An instance of the configuration object
+ @param ec_id: Execution context ID
+
+ @returns: The build up nics
+
+ """
+ nics = []
+ for nic in op.nics:
+ nic_mode_req = nic.get(constants.INIC_MODE, None)
+ nic_mode = nic_mode_req
+ if nic_mode is None or nic_mode == constants.VALUE_AUTO:
+ nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
+
+ net = nic.get(constants.INIC_NETWORK, None)
+ link = nic.get(constants.NIC_LINK, None)
+ ip = nic.get(constants.INIC_IP, None)
+
+ if net is None or net.lower() == constants.VALUE_NONE:
+ net = None
+ else:
+ if nic_mode_req is not None or link is not None:
+ raise errors.OpPrereqError("If network is given, no mode or link"
+ " is allowed to be passed",
+ errors.ECODE_INVAL)
+
+ # ip validity checks
+ if ip is None or ip.lower() == constants.VALUE_NONE:
+ nic_ip = None
+ elif ip.lower() == constants.VALUE_AUTO:
+ if not op.name_check:
+ raise errors.OpPrereqError("IP address set to auto but name checks"
+ " have been skipped",
+ errors.ECODE_INVAL)
+ nic_ip = default_ip
+ else:
+ # We defer pool operations until later, so that the iallocator has
+ # filled in the instance's node(s) dimara
+ if ip.lower() == constants.NIC_IP_POOL:
+ if net is None:
+ raise errors.OpPrereqError("if ip=pool, parameter network"
+ " must be passed too",
+ errors.ECODE_INVAL)
+
+ elif not netutils.IPAddress.IsValid(ip):
+ raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
+ errors.ECODE_INVAL)
+
+ nic_ip = ip
+
+ # TODO: check the ip address for uniqueness
+ if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
+ raise errors.OpPrereqError("Routed nic mode requires an ip address",
+ errors.ECODE_INVAL)
+
+ # MAC address verification
+ mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
+ if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ mac = utils.NormalizeAndValidateMac(mac)
+
+ try:
+ # TODO: We need to factor this out
+ cfg.ReserveMAC(mac, ec_id)
+ except errors.ReservationError:
+ raise errors.OpPrereqError("MAC address %s already in use"
+ " in cluster" % mac,
+ errors.ECODE_NOTUNIQUE)
+
+ # Build nic parameters
+ nicparams = {}
+ if nic_mode_req:
+ nicparams[constants.NIC_MODE] = nic_mode
+ if link:
+ nicparams[constants.NIC_LINK] = link
+
+ check_params = cluster.SimpleFillNIC(nicparams)
+ objects.NIC.CheckParameterSyntax(check_params)
+ net_uuid = cfg.LookupNetwork(net)
+ name = nic.get(constants.INIC_NAME, None)
+ if name is not None and name.lower() == constants.VALUE_NONE:
+ name = None
+ nic_obj = objects.NIC(mac=mac, ip=nic_ip, name=name,
+ network=net_uuid, nicparams=nicparams)
+ nic_obj.uuid = cfg.GenerateUniqueID(ec_id)
+ nics.append(nic_obj)
+
+ return nics
+
+
+def _CheckForConflictingIp(lu, ip, node):
+ """In case of conflicting IP address raise error.
+
+ @type ip: string
+ @param ip: IP address
+ @type node: string
+ @param node: node name
+
+ """
+ (conf_net, _) = lu.cfg.CheckIPInNodeGroup(ip, node)
+ if conf_net is not None:
+ raise errors.OpPrereqError(("The requested IP address (%s) belongs to"
+ " network %s, but the target NIC does not." %
+ (ip, conf_net)),
+ errors.ECODE_STATE)
+
+ return (None, None)
+
+
+def _ComputeIPolicyInstanceSpecViolation(
+ ipolicy, instance_spec, disk_template,
+ _compute_fn=ComputeIPolicySpecViolation):
+ """Compute if instance specs meets the specs of ipolicy.
+
+ @type ipolicy: dict
+ @param ipolicy: The ipolicy to verify against
+ @param instance_spec: dict
+ @param instance_spec: The instance spec to verify
+ @type disk_template: string
+ @param disk_template: the disk template of the instance
+ @param _compute_fn: The function to verify ipolicy (unittest only)
+ @see: L{ComputeIPolicySpecViolation}
+
+ """
+ mem_size = instance_spec.get(constants.ISPEC_MEM_SIZE, None)
+ cpu_count = instance_spec.get(constants.ISPEC_CPU_COUNT, None)
+ disk_count = instance_spec.get(constants.ISPEC_DISK_COUNT, 0)
+ disk_sizes = instance_spec.get(constants.ISPEC_DISK_SIZE, [])
+ nic_count = instance_spec.get(constants.ISPEC_NIC_COUNT, 0)
+ spindle_use = instance_spec.get(constants.ISPEC_SPINDLE_USE, None)
+
+ return _compute_fn(ipolicy, mem_size, cpu_count, disk_count, nic_count,
+ disk_sizes, spindle_use, disk_template)
+
+
+def _CheckOSVariant(os_obj, name):
+ """Check whether an OS name conforms to the os variants specification.
+
+ @type os_obj: L{objects.OS}
+ @param os_obj: OS object to check
+ @type name: string
+ @param name: OS name passed by the user, to check for validity
+
+ """
+ variant = objects.OS.GetVariant(name)
+ if not os_obj.supported_variants:
+ if variant:
+ raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
+ " passed)" % (os_obj.name, variant),
+ errors.ECODE_INVAL)
+ return
+ if not variant:
+ raise errors.OpPrereqError("OS name must include a variant",
+ errors.ECODE_INVAL)
+
+ if variant not in os_obj.supported_variants:
+ raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
+
+
+class LUInstanceCreate(LogicalUnit):
+ """Create an instance.
+
+ """
+ HPATH = "instance-add"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ """Check arguments.
+
+ """
+ # do not require name_check to ease forward/backward compatibility
+ # for tools
+ if self.op.no_install and self.op.start:
+ self.LogInfo("No-installation mode selected, disabling startup")
+ self.op.start = False
+ # validate/normalize the instance name
+ self.op.instance_name = \
+ netutils.Hostname.GetNormalizedName(self.op.instance_name)
+
+ if self.op.ip_check and not self.op.name_check:
+ # TODO: make the ip check more flexible and not depend on the name check
+ raise errors.OpPrereqError("Cannot do IP address check without a name"
+ " check", errors.ECODE_INVAL)
+
+ # check nics' parameter names
+ for nic in self.op.nics:
+ utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
+ # check that NIC's parameters names are unique and valid
+ utils.ValidateDeviceNames("NIC", self.op.nics)
+
+ # check that disk's names are unique and valid
+ utils.ValidateDeviceNames("disk", self.op.disks)
+
+ cluster = self.cfg.GetClusterInfo()
+ if not self.op.disk_template in cluster.enabled_disk_templates:
+ raise errors.OpPrereqError("Cannot create an instance with disk template"
+ " '%s', because it is not enabled in the"
+ " cluster. Enabled disk templates are: %s." %
+ (self.op.disk_template,
+ ",".join(cluster.enabled_disk_templates)))
+
+ # check disks. parameter names and consistent adopt/no-adopt strategy
+ has_adopt = has_no_adopt = False
+ for disk in self.op.disks:
+ if self.op.disk_template != constants.DT_EXT:
+ utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
+ if constants.IDISK_ADOPT in disk:
+ has_adopt = True
+ else:
+ has_no_adopt = True
+ if has_adopt and has_no_adopt:
+ raise errors.OpPrereqError("Either all disks are adopted or none is",
+ errors.ECODE_INVAL)
+ if has_adopt:
+ if self.op.disk_template not in constants.DTS_MAY_ADOPT:
+ raise errors.OpPrereqError("Disk adoption is not supported for the"
+ " '%s' disk template" %
+ self.op.disk_template,
+ errors.ECODE_INVAL)
+ if self.op.iallocator is not None:
+ raise errors.OpPrereqError("Disk adoption not allowed with an"
+ " iallocator script", errors.ECODE_INVAL)
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ raise errors.OpPrereqError("Disk adoption not allowed for"
+ " instance import", errors.ECODE_INVAL)
+ else:
+ if self.op.disk_template in constants.DTS_MUST_ADOPT:
+ raise errors.OpPrereqError("Disk template %s requires disk adoption,"
+ " but no 'adopt' parameter given" %
+ self.op.disk_template,
+ errors.ECODE_INVAL)
+
+ self.adopt_disks = has_adopt
+
+ # instance name verification
+ if self.op.name_check:
+ self.hostname1 = _CheckHostnameSane(self, self.op.instance_name)
+ self.op.instance_name = self.hostname1.name
+ # used in CheckPrereq for ip ping check
+ self.check_ip = self.hostname1.ip
+ else:
+ self.check_ip = None
+
+ # file storage checks
+ if (self.op.file_driver and
+ not self.op.file_driver in constants.FILE_DRIVER):
+ raise errors.OpPrereqError("Invalid file driver name '%s'" %
+ self.op.file_driver, errors.ECODE_INVAL)
+
+ if self.op.disk_template == constants.DT_FILE:
+ opcodes.RequireFileStorage()
+ elif self.op.disk_template == constants.DT_SHARED_FILE:
+ opcodes.RequireSharedFileStorage()
+
+ ### Node/iallocator related checks
+ CheckIAllocatorOrNode(self, "iallocator", "pnode")
+
+ if self.op.pnode is not None:
+ if self.op.disk_template in constants.DTS_INT_MIRROR:
+ if self.op.snode is None:
+ raise errors.OpPrereqError("The networked disk templates need"
+ " a mirror node", errors.ECODE_INVAL)
+ elif self.op.snode:
+ self.LogWarning("Secondary node will be ignored on non-mirrored disk"
+ " template")
+ self.op.snode = None
+
+ _CheckOpportunisticLocking(self.op)
+
+ self._cds = GetClusterDomainSecret()
+
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ # On import force_variant must be True, because if we forced it at
+ # initial install, our only chance when importing it back is that it
+ # works again!
+ self.op.force_variant = True
+
+ if self.op.no_install:
+ self.LogInfo("No-installation mode has no effect during import")
+
+ elif self.op.mode == constants.INSTANCE_CREATE:
+ if self.op.os_type is None:
+ raise errors.OpPrereqError("No guest OS specified",
+ errors.ECODE_INVAL)
+ if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
+ raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
+ " installation" % self.op.os_type,
+ errors.ECODE_STATE)
+ if self.op.disk_template is None:
+ raise errors.OpPrereqError("No disk template specified",
+ errors.ECODE_INVAL)
+
+ elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
+ # Check handshake to ensure both clusters have the same domain secret
+ src_handshake = self.op.source_handshake
+ if not src_handshake:
+ raise errors.OpPrereqError("Missing source handshake",
+ errors.ECODE_INVAL)
+
+ errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
+ src_handshake)
+ if errmsg:
+ raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
+ errors.ECODE_INVAL)
+
+ # Load and check source CA
+ self.source_x509_ca_pem = self.op.source_x509_ca
+ if not self.source_x509_ca_pem:
+ raise errors.OpPrereqError("Missing source X509 CA",
+ errors.ECODE_INVAL)
+
+ try:
+ (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
+ self._cds)
+ except OpenSSL.crypto.Error, err:
+ raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
+ (err, ), errors.ECODE_INVAL)
+
+ (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
+ if errcode is not None:
+ raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
+ errors.ECODE_INVAL)
+
+ self.source_x509_ca = cert
+
+ src_instance_name = self.op.source_instance_name
+ if not src_instance_name:
+ raise errors.OpPrereqError("Missing source instance name",
+ errors.ECODE_INVAL)
+
+ self.source_instance_name = \
+ netutils.GetHostname(name=src_instance_name).name
+
+ else:
+ raise errors.OpPrereqError("Invalid instance creation mode %r" %
+ self.op.mode, errors.ECODE_INVAL)
+
+ def ExpandNames(self):
+ """ExpandNames for CreateInstance.
+
+ Figure out the right locks for instance creation.
+
+ """
+ self.needed_locks = {}
+
+ instance_name = self.op.instance_name
+ # this is just a preventive check, but someone might still add this
+ # instance in the meantime, and creation will fail at lock-add time
+ if instance_name in self.cfg.GetInstanceList():
+ raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
+ instance_name, errors.ECODE_EXISTS)
+
+ self.add_locks[locking.LEVEL_INSTANCE] = instance_name
+
+ if self.op.iallocator:
+ # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
+ # specifying a group on instance creation and then selecting nodes from
+ # that group
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
+ if self.op.opportunistic_locking:
+ self.opportunistic_locks[locking.LEVEL_NODE] = True
+ self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
+ else:
+ self.op.pnode = ExpandNodeName(self.cfg, self.op.pnode)
+ nodelist = [self.op.pnode]
+ if self.op.snode is not None:
+ self.op.snode = ExpandNodeName(self.cfg, self.op.snode)
+ nodelist.append(self.op.snode)
+ self.needed_locks[locking.LEVEL_NODE] = nodelist
+
+ # in case of import lock the source node too
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ src_node = self.op.src_node
+ src_path = self.op.src_path
+
+ if src_path is None:
+ self.op.src_path = src_path = self.op.instance_name
+
+ if src_node is None:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+ self.op.src_node = None
+ if os.path.isabs(src_path):
+ raise errors.OpPrereqError("Importing an instance from a path"
+ " requires a source node option",
+ errors.ECODE_INVAL)
+ else:
+ self.op.src_node = src_node = ExpandNodeName(self.cfg, src_node)
+ if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
+ self.needed_locks[locking.LEVEL_NODE].append(src_node)
+ if not os.path.isabs(src_path):
+ self.op.src_path = src_path = \
+ utils.PathJoin(pathutils.EXPORT_DIR, src_path)
+
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def _RunAllocator(self):
+ """Run the allocator based on input opcode.
+
+ """
+ if self.op.opportunistic_locking:
+ # Only consider nodes for which a lock is held
+ node_whitelist = list(self.owned_locks(locking.LEVEL_NODE))
+ else:
+ node_whitelist = None
+
+ #TODO Export network to iallocator so that it chooses a pnode
+ # in a nodegroup that has the desired network connected to
+ req = _CreateInstanceAllocRequest(self.op, self.disks,
+ self.nics, self.be_full,
+ node_whitelist)
+ ial = iallocator.IAllocator(self.cfg, self.rpc, req)
+
+ ial.Run(self.op.iallocator)
+
+ if not ial.success:
+ # When opportunistic locks are used only a temporary failure is generated
+ if self.op.opportunistic_locking:
+ ecode = errors.ECODE_TEMP_NORES
+ else:
+ ecode = errors.ECODE_NORES
+
+ raise errors.OpPrereqError("Can't compute nodes using"
+ " iallocator '%s': %s" %
+ (self.op.iallocator, ial.info),
+ ecode)
+
+ self.op.pnode = ial.result[0]
+ self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
+ self.op.instance_name, self.op.iallocator,
+ utils.CommaJoin(ial.result))
+
+ assert req.RequiredNodes() in (1, 2), "Wrong node count from iallocator"
+
+ if req.RequiredNodes() == 2:
+ self.op.snode = ial.result[1]
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ env = {
+ "ADD_MODE": self.op.mode,
+ }
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ env["SRC_NODE"] = self.op.src_node
+ env["SRC_PATH"] = self.op.src_path
+ env["SRC_IMAGES"] = self.src_images
+
+ env.update(BuildInstanceHookEnv(
+ name=self.op.instance_name,
+ primary_node=self.op.pnode,
+ secondary_nodes=self.secondaries,
+ status=self.op.start,
+ os_type=self.op.os_type,
+ minmem=self.be_full[constants.BE_MINMEM],
+ maxmem=self.be_full[constants.BE_MAXMEM],
+ vcpus=self.be_full[constants.BE_VCPUS],
+ nics=NICListToTuple(self, self.nics),
+ disk_template=self.op.disk_template,
+ disks=[(d[constants.IDISK_NAME], d[constants.IDISK_SIZE],
+ d[constants.IDISK_MODE]) for d in self.disks],
+ bep=self.be_full,
+ hvp=self.hv_full,
+ hypervisor_name=self.op.hypervisor,
+ tags=self.op.tags,
+ ))
+
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
+ return nl, nl
+
+ def _ReadExportInfo(self):
+ """Reads the export information from disk.
+
+ It will override the opcode source node and path with the actual
+ information, if these two were not specified before.
+
+ @return: the export information
+
+ """
+ assert self.op.mode == constants.INSTANCE_IMPORT
+
+ src_node = self.op.src_node
+ src_path = self.op.src_path
+
+ if src_node is None:
+ locked_nodes = self.owned_locks(locking.LEVEL_NODE)
+ exp_list = self.rpc.call_export_list(locked_nodes)
+ found = False
+ for node in exp_list:
+ if exp_list[node].fail_msg:
+ continue
+ if src_path in exp_list[node].payload:
+ found = True
+ self.op.src_node = src_node = node
+ self.op.src_path = src_path = utils.PathJoin(pathutils.EXPORT_DIR,
+ src_path)
+ break
+ if not found:
+ raise errors.OpPrereqError("No export found for relative path %s" %
+ src_path, errors.ECODE_INVAL)
+
+ CheckNodeOnline(self, src_node)
+ result = self.rpc.call_export_info(src_node, src_path)
+ result.Raise("No export or invalid export found in dir %s" % src_path)
+
+ export_info = objects.SerializableConfigParser.Loads(str(result.payload))
+ if not export_info.has_section(constants.INISECT_EXP):
+ raise errors.ProgrammerError("Corrupted export config",
+ errors.ECODE_ENVIRON)
+
+ ei_version = export_info.get(constants.INISECT_EXP, "version")
+ if (int(ei_version) != constants.EXPORT_VERSION):
+ raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
+ (ei_version, constants.EXPORT_VERSION),
+ errors.ECODE_ENVIRON)
+ return export_info
+
+ def _ReadExportParams(self, einfo):
+ """Use export parameters as defaults.
+
+ In case the opcode doesn't specify (as in override) some instance
+ parameters, then try to use them from the export information, if
+ that declares them.
+
+ """
+ self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
+
+ if self.op.disk_template is None:
+ if einfo.has_option(constants.INISECT_INS, "disk_template"):
+ self.op.disk_template = einfo.get(constants.INISECT_INS,
+ "disk_template")
+ if self.op.disk_template not in constants.DISK_TEMPLATES:
+ raise errors.OpPrereqError("Disk template specified in configuration"
+ " file is not one of the allowed values:"
+ " %s" %
+ " ".join(constants.DISK_TEMPLATES),
+ errors.ECODE_INVAL)
+ else:
+ raise errors.OpPrereqError("No disk template specified and the export"
+ " is missing the disk_template information",
+ errors.ECODE_INVAL)
+
+ if not self.op.disks:
+ disks = []
+ # TODO: import the disk iv_name too
+ for idx in range(constants.MAX_DISKS):
+ if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
+ disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
+ disks.append({constants.IDISK_SIZE: disk_sz})
+ self.op.disks = disks
+ if not disks and self.op.disk_template != constants.DT_DISKLESS:
+ raise errors.OpPrereqError("No disk info specified and the export"
+ " is missing the disk information",
+ errors.ECODE_INVAL)
+
+ if not self.op.nics:
+ nics = []
+ for idx in range(constants.MAX_NICS):
+ if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
+ ndict = {}
+ for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
+ v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
+ ndict[name] = v
+ nics.append(ndict)
+ else:
+ break
+ self.op.nics = nics
+
+ if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
+ self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
+
+ if (self.op.hypervisor is None and
+ einfo.has_option(constants.INISECT_INS, "hypervisor")):
+ self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
+
+ if einfo.has_section(constants.INISECT_HYP):
+ # use the export parameters but do not override the ones
+ # specified by the user
+ for name, value in einfo.items(constants.INISECT_HYP):
+ if name not in self.op.hvparams:
+ self.op.hvparams[name] = value
+
+ if einfo.has_section(constants.INISECT_BEP):
+ # use the parameters, without overriding
+ for name, value in einfo.items(constants.INISECT_BEP):
+ if name not in self.op.beparams:
+ self.op.beparams[name] = value
+ # Compatibility for the old "memory" be param
+ if name == constants.BE_MEMORY:
+ if constants.BE_MAXMEM not in self.op.beparams:
+ self.op.beparams[constants.BE_MAXMEM] = value
+ if constants.BE_MINMEM not in self.op.beparams:
+ self.op.beparams[constants.BE_MINMEM] = value
+ else:
+ # try to read the parameters old style, from the main section
+ for name in constants.BES_PARAMETERS:
+ if (name not in self.op.beparams and
+ einfo.has_option(constants.INISECT_INS, name)):
+ self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
+
+ if einfo.has_section(constants.INISECT_OSP):
+ # use the parameters, without overriding
+ for name, value in einfo.items(constants.INISECT_OSP):
+ if name not in self.op.osparams:
+ self.op.osparams[name] = value
+
+ def _RevertToDefaults(self, cluster):
+ """Revert the instance parameters to the default values.
+
+ """
+ # hvparams
+ hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
+ for name in self.op.hvparams.keys():
+ if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
+ del self.op.hvparams[name]
+ # beparams
+ be_defs = cluster.SimpleFillBE({})
+ for name in self.op.beparams.keys():
+ if name in be_defs and be_defs[name] == self.op.beparams[name]:
+ del self.op.beparams[name]
+ # nic params
+ nic_defs = cluster.SimpleFillNIC({})
+ for nic in self.op.nics:
+ for name in constants.NICS_PARAMETERS:
+ if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
+ del nic[name]
+ # osparams
+ os_defs = cluster.SimpleFillOS(self.op.os_type, {})
+ for name in self.op.osparams.keys():
+ if name in os_defs and os_defs[name] == self.op.osparams[name]:
+ del self.op.osparams[name]
+
+ def _CalculateFileStorageDir(self):
+ """Calculate final instance file storage dir.
+
+ """
+ # file storage dir calculation/check
+ self.instance_file_storage_dir = None
+ if self.op.disk_template in constants.DTS_FILEBASED:
+ # build the full file storage dir path
+ joinargs = []
+
+ if self.op.disk_template == constants.DT_SHARED_FILE:
+ get_fsd_fn = self.cfg.GetSharedFileStorageDir
+ else:
+ get_fsd_fn = self.cfg.GetFileStorageDir
+
+ cfg_storagedir = get_fsd_fn()
+ if not cfg_storagedir:
+ raise errors.OpPrereqError("Cluster file storage dir not defined",
+ errors.ECODE_STATE)
+ joinargs.append(cfg_storagedir)
+
+ if self.op.file_storage_dir is not None:
+ joinargs.append(self.op.file_storage_dir)
+
+ joinargs.append(self.op.instance_name)
+
+ # pylint: disable=W0142
+ self.instance_file_storage_dir = utils.PathJoin(*joinargs)
+
+ def CheckPrereq(self): # pylint: disable=R0914
+ """Check prerequisites.
+
+ """
+ self._CalculateFileStorageDir()
+
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ export_info = self._ReadExportInfo()
+ self._ReadExportParams(export_info)
+ self._old_instance_name = export_info.get(constants.INISECT_INS, "name")
+ else:
+ self._old_instance_name = None
+
+ if (not self.cfg.GetVGName() and
+ self.op.disk_template not in constants.DTS_NOT_LVM):
+ raise errors.OpPrereqError("Cluster does not support lvm-based"
+ " instances", errors.ECODE_STATE)
+
+ if (self.op.hypervisor is None or
+ self.op.hypervisor == constants.VALUE_AUTO):
+ self.op.hypervisor = self.cfg.GetHypervisorType()
+
+ cluster = self.cfg.GetClusterInfo()
+ enabled_hvs = cluster.enabled_hypervisors
+ if self.op.hypervisor not in enabled_hvs:
+ raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
+ " cluster (%s)" %
+ (self.op.hypervisor, ",".join(enabled_hvs)),
+ errors.ECODE_STATE)
+
+ # Check tag validity
+ for tag in self.op.tags:
+ objects.TaggableObject.ValidateTag(tag)
+
+ # check hypervisor parameter syntax (locally)
+ utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
+ filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
+ self.op.hvparams)
+ hv_type = hypervisor.GetHypervisorClass(self.op.hypervisor)
+ hv_type.CheckParameterSyntax(filled_hvp)
+ self.hv_full = filled_hvp
+ # check that we don't specify global parameters on an instance
+ CheckParamsNotGlobal(self.op.hvparams, constants.HVC_GLOBALS, "hypervisor",
+ "instance", "cluster")
+
+ # fill and remember the beparams dict
+ self.be_full = _ComputeFullBeParams(self.op, cluster)
+
+ # build os parameters
+ self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
+
+ # now that hvp/bep are in final format, let's reset to defaults,
+ # if told to do so
+ if self.op.identify_defaults:
+ self._RevertToDefaults(cluster)
+
+ # NIC buildup
+ self.nics = _ComputeNics(self.op, cluster, self.check_ip, self.cfg,
+ self.proc.GetECId())
+
+ # disk checks/pre-build
+ default_vg = self.cfg.GetVGName()
+ self.disks = ComputeDisks(self.op, default_vg)
+
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ disk_images = []
+ for idx in range(len(self.disks)):
+ option = "disk%d_dump" % idx
+ if export_info.has_option(constants.INISECT_INS, option):
+ # FIXME: are the old os-es, disk sizes, etc. useful?
+ export_name = export_info.get(constants.INISECT_INS, option)
+ image = utils.PathJoin(self.op.src_path, export_name)
+ disk_images.append(image)
+ else:
+ disk_images.append(False)
+
+ self.src_images = disk_images
+
+ if self.op.instance_name == self._old_instance_name:
+ for idx, nic in enumerate(self.nics):
+ if nic.mac == constants.VALUE_AUTO:
+ nic_mac_ini = "nic%d_mac" % idx
+ nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
+
+ # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
+
+ # ip ping checks (we use the same ip that was resolved in ExpandNames)
+ if self.op.ip_check:
+ if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
+ raise errors.OpPrereqError("IP %s of instance %s already in use" %
+ (self.check_ip, self.op.instance_name),
+ errors.ECODE_NOTUNIQUE)
+
+ #### mac address generation
+ # By generating here the mac address both the allocator and the hooks get
+ # the real final mac address rather than the 'auto' or 'generate' value.
+ # There is a race condition between the generation and the instance object
+ # creation, which means that we know the mac is valid now, but we're not
+ # sure it will be when we actually add the instance. If things go bad
+ # adding the instance will abort because of a duplicate mac, and the
+ # creation job will fail.
+ for nic in self.nics:
+ if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ nic.mac = self.cfg.GenerateMAC(nic.network, self.proc.GetECId())
+
+ #### allocator run
+
+ if self.op.iallocator is not None:
+ self._RunAllocator()
+
+ # Release all unneeded node locks
+ keep_locks = filter(None, [self.op.pnode, self.op.snode, self.op.src_node])
+ ReleaseLocks(self, locking.LEVEL_NODE, keep=keep_locks)
+ ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=keep_locks)
+ ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES)), \
+ "Node locks differ from node resource locks"
+
+ #### node related checks
+
+ # check primary node
+ self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
+ assert self.pnode is not None, \
+ "Cannot retrieve locked node %s" % self.op.pnode
+ if pnode.offline:
+ raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
+ pnode.name, errors.ECODE_STATE)
+ if pnode.drained:
+ raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
+ pnode.name, errors.ECODE_STATE)
+ if not pnode.vm_capable:
+ raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
+ " '%s'" % pnode.name, errors.ECODE_STATE)
+
+ self.secondaries = []
+
+ # Fill in any IPs from IP pools. This must happen here, because we need to
+ # know the nic's primary node, as specified by the iallocator
+ for idx, nic in enumerate(self.nics):
+ net_uuid = nic.network
+ if net_uuid is not None:
+ nobj = self.cfg.GetNetwork(net_uuid)
+ netparams = self.cfg.GetGroupNetParams(net_uuid, self.pnode.name)
+ if netparams is None:
+ raise errors.OpPrereqError("No netparams found for network"
+ " %s. Propably not connected to"
+ " node's %s nodegroup" %
+ (nobj.name, self.pnode.name),
+ errors.ECODE_INVAL)
+ self.LogInfo("NIC/%d inherits netparams %s" %
+ (idx, netparams.values()))
+ nic.nicparams = dict(netparams)
+ if nic.ip is not None:
+ if nic.ip.lower() == constants.NIC_IP_POOL:
+ try:
+ nic.ip = self.cfg.GenerateIp(net_uuid, self.proc.GetECId())
+ except errors.ReservationError:
+ raise errors.OpPrereqError("Unable to get a free IP for NIC %d"
+ " from the address pool" % idx,
+ errors.ECODE_STATE)
+ self.LogInfo("Chose IP %s from network %s", nic.ip, nobj.name)
+ else:
+ try:
+ self.cfg.ReserveIp(net_uuid, nic.ip, self.proc.GetECId())
+ except errors.ReservationError:
+ raise errors.OpPrereqError("IP address %s already in use"
+ " or does not belong to network %s" %
+ (nic.ip, nobj.name),
+ errors.ECODE_NOTUNIQUE)
+
+ # net is None, ip None or given
+ elif self.op.conflicts_check:
+ _CheckForConflictingIp(self, nic.ip, self.pnode.name)
+
+ # mirror node verification
+ if self.op.disk_template in constants.DTS_INT_MIRROR:
+ if self.op.snode == pnode.name:
+ raise errors.OpPrereqError("The secondary node cannot be the"
+ " primary node", errors.ECODE_INVAL)
+ CheckNodeOnline(self, self.op.snode)
+ CheckNodeNotDrained(self, self.op.snode)
+ CheckNodeVmCapable(self, self.op.snode)
+ self.secondaries.append(self.op.snode)
+
+ snode = self.cfg.GetNodeInfo(self.op.snode)
+ if pnode.group != snode.group:
+ self.LogWarning("The primary and secondary nodes are in two"
+ " different node groups; the disk parameters"
+ " from the first disk's node group will be"
+ " used")
+
+ if not self.op.disk_template in constants.DTS_EXCL_STORAGE:
+ nodes = [pnode]
+ if self.op.disk_template in constants.DTS_INT_MIRROR:
+ nodes.append(snode)
+ has_es = lambda n: IsExclusiveStorageEnabledNode(self.cfg, n)
+ if compat.any(map(has_es, nodes)):
+ raise errors.OpPrereqError("Disk template %s not supported with"
+ " exclusive storage" % self.op.disk_template,
+ errors.ECODE_STATE)
+
+ nodenames = [pnode.name] + self.secondaries
+
+ if not self.adopt_disks:
+ if self.op.disk_template == constants.DT_RBD:
+ # _CheckRADOSFreeSpace() is just a placeholder.
+ # Any function that checks prerequisites can be placed here.
+ # Check if there is enough space on the RADOS cluster.
+ CheckRADOSFreeSpace()
+ elif self.op.disk_template == constants.DT_EXT:
+ # FIXME: Function that checks prereqs if needed
+ pass
+ else:
+ # Check lv size requirements, if not adopting
+ req_sizes = ComputeDiskSizePerVG(self.op.disk_template, self.disks)
+ CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
+
+ elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
+ all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
+ disk[constants.IDISK_ADOPT])
+ for disk in self.disks])
+ if len(all_lvs) != len(self.disks):
+ raise errors.OpPrereqError("Duplicate volume names given for adoption",
+ errors.ECODE_INVAL)
+ for lv_name in all_lvs:
+ try:
+ # FIXME: lv_name here is "vg/lv" need to ensure that other calls
+ # to ReserveLV uses the same syntax
+ self.cfg.ReserveLV(lv_name, self.proc.GetECId())
+ except errors.ReservationError:
+ raise errors.OpPrereqError("LV named %s used by another instance" %
+ lv_name, errors.ECODE_NOTUNIQUE)
+
+ vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
+ vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
+
+ node_lvs = self.rpc.call_lv_list([pnode.name],
+ vg_names.payload.keys())[pnode.name]
+ node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
+ node_lvs = node_lvs.payload
+
+ delta = all_lvs.difference(node_lvs.keys())
+ if delta:
+ raise errors.OpPrereqError("Missing logical volume(s): %s" %
+ utils.CommaJoin(delta),
+ errors.ECODE_INVAL)
+ online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
+ if online_lvs:
+ raise errors.OpPrereqError("Online logical volumes found, cannot"
+ " adopt: %s" % utils.CommaJoin(online_lvs),
+ errors.ECODE_STATE)
+ # update the size of disk based on what is found
+ for dsk in self.disks:
+ dsk[constants.IDISK_SIZE] = \
+ int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
+ dsk[constants.IDISK_ADOPT])][0]))
+
+ elif self.op.disk_template == constants.DT_BLOCK:
+ # Normalize and de-duplicate device paths
+ all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
+ for disk in self.disks])
+ if len(all_disks) != len(self.disks):
+ raise errors.OpPrereqError("Duplicate disk names given for adoption",
+ errors.ECODE_INVAL)
+ baddisks = [d for d in all_disks
+ if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
+ if baddisks:
+ raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
+ " cannot be adopted" %
+ (utils.CommaJoin(baddisks),
+ constants.ADOPTABLE_BLOCKDEV_ROOT),
+ errors.ECODE_INVAL)
+
+ node_disks = self.rpc.call_bdev_sizes([pnode.name],
+ list(all_disks))[pnode.name]
+ node_disks.Raise("Cannot get block device information from node %s" %
+ pnode.name)
+ node_disks = node_disks.payload
+ delta = all_disks.difference(node_disks.keys())
+ if delta:
+ raise errors.OpPrereqError("Missing block device(s): %s" %
+ utils.CommaJoin(delta),
+ errors.ECODE_INVAL)
+ for dsk in self.disks:
+ dsk[constants.IDISK_SIZE] = \
+ int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
+
+ # Verify instance specs
+ spindle_use = self.be_full.get(constants.BE_SPINDLE_USE, None)
+ ispec = {
+ constants.ISPEC_MEM_SIZE: self.be_full.get(constants.BE_MAXMEM, None),
+ constants.ISPEC_CPU_COUNT: self.be_full.get(constants.BE_VCPUS, None),
+ constants.ISPEC_DISK_COUNT: len(self.disks),
+ constants.ISPEC_DISK_SIZE: [disk[constants.IDISK_SIZE]
+ for disk in self.disks],
+ constants.ISPEC_NIC_COUNT: len(self.nics),
+ constants.ISPEC_SPINDLE_USE: spindle_use,
+ }
+
+ group_info = self.cfg.GetNodeGroup(pnode.group)
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster, group_info)
+ res = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec,
+ self.op.disk_template)
+ if not self.op.ignore_ipolicy and res:
+ msg = ("Instance allocation to group %s (%s) violates policy: %s" %
+ (pnode.group, group_info.name, utils.CommaJoin(res)))
+ raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
+
+ CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
+
+ CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
+ # check OS parameters (remotely)
+ CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
+
+ CheckNicsBridgesExist(self, self.nics, self.pnode.name)
+
+ #TODO: _CheckExtParams (remotely)
+ # Check parameters for extstorage
+
+ # memory check on primary node
+ #TODO(dynmem): use MINMEM for checking
+ if self.op.start:
+ CheckNodeFreeMemory(self, self.pnode.name,
+ "creating instance %s" % self.op.instance_name,
+ self.be_full[constants.BE_MAXMEM],
+ self.op.hypervisor)
+
+ self.dry_run_result = list(nodenames)
+
+ def Exec(self, feedback_fn):
+ """Create and add the instance to the cluster.
+
+ """
+ instance = self.op.instance_name
+ pnode_name = self.pnode.name
+
+ assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Node locks differ from node resource locks"
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
+ ht_kind = self.op.hypervisor
+ if ht_kind in constants.HTS_REQ_PORT:
+ network_port = self.cfg.AllocatePort()
+ else:
+ network_port = None
+
+ # This is ugly but we got a chicken-egg problem here
+ # We can only take the group disk parameters, as the instance
+ # has no disks yet (we are generating them right here).
+ node = self.cfg.GetNodeInfo(pnode_name)
+ nodegroup = self.cfg.GetNodeGroup(node.group)
+ disks = GenerateDiskTemplate(self,
+ self.op.disk_template,
+ instance, pnode_name,
+ self.secondaries,
+ self.disks,
+ self.instance_file_storage_dir,
+ self.op.file_driver,
+ 0,
+ feedback_fn,
+ self.cfg.GetGroupDiskParams(nodegroup))
+
+ iobj = objects.Instance(name=instance, os=self.op.os_type,
+ primary_node=pnode_name,
+ nics=self.nics, disks=disks,
+ disk_template=self.op.disk_template,
+ admin_state=constants.ADMINST_DOWN,
+ network_port=network_port,
+ beparams=self.op.beparams,
+ hvparams=self.op.hvparams,
+ hypervisor=self.op.hypervisor,
+ osparams=self.op.osparams,
+ )
+
+ if self.op.tags:
+ for tag in self.op.tags:
+ iobj.AddTag(tag)
+
+ if self.adopt_disks:
+ if self.op.disk_template == constants.DT_PLAIN:
+ # rename LVs to the newly-generated names; we need to construct
+ # 'fake' LV disks with the old data, plus the new unique_id
+ tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
+ rename_to = []
+ for t_dsk, a_dsk in zip(tmp_disks, self.disks):
+ rename_to.append(t_dsk.logical_id)
+ t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
+ self.cfg.SetDiskID(t_dsk, pnode_name)
+ result = self.rpc.call_blockdev_rename(pnode_name,
+ zip(tmp_disks, rename_to))
+ result.Raise("Failed to rename adoped LVs")
+ else:
+ feedback_fn("* creating instance disks...")
+ try:
+ CreateDisks(self, iobj)
+ except errors.OpExecError:
+ self.LogWarning("Device creation failed")
+ self.cfg.ReleaseDRBDMinors(instance)
+ raise
+
+ feedback_fn("adding instance %s to cluster config" % instance)
+
+ self.cfg.AddInstance(iobj, self.proc.GetECId())
+
+ # Declare that we don't want to remove the instance lock anymore, as we've
+ # added the instance to the config
+ del self.remove_locks[locking.LEVEL_INSTANCE]
+
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ # Release unused nodes
+ ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
+ else:
+ # Release all nodes
+ ReleaseLocks(self, locking.LEVEL_NODE)
+
+ disk_abort = False
+ if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
+ feedback_fn("* wiping instance disks...")
+ try:
+ WipeDisks(self, iobj)
+ except errors.OpExecError, err:
+ logging.exception("Wiping disks failed")
+ self.LogWarning("Wiping instance disks failed (%s)", err)
+ disk_abort = True
+
+ if disk_abort:
+ # Something is already wrong with the disks, don't do anything else
+ pass
+ elif self.op.wait_for_sync:
+ disk_abort = not WaitForSync(self, iobj)
+ elif iobj.disk_template in constants.DTS_INT_MIRROR:
+ # make sure the disks are not degraded (still sync-ing is ok)
+ feedback_fn("* checking mirrors status")
+ disk_abort = not WaitForSync(self, iobj, oneshot=True)
+ else:
+ disk_abort = False
+
+ if disk_abort:
+ RemoveDisks(self, iobj)
+ self.cfg.RemoveInstance(iobj.name)
+ # Make sure the instance lock gets removed
+ self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
+ raise errors.OpExecError("There are some degraded disks for"
+ " this instance")
+
+ # Release all node resource locks
+ ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
+ if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
+ # we need to set the disks ID to the primary node, since the
+ # preceding code might or might have not done it, depending on
+ # disk template and other options
+ for disk in iobj.disks:
+ self.cfg.SetDiskID(disk, pnode_name)
+ if self.op.mode == constants.INSTANCE_CREATE:
+ if not self.op.no_install:
+ pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
+ not self.op.wait_for_sync)
+ if pause_sync:
+ feedback_fn("* pausing disk sync to install instance OS")
+ result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+ (iobj.disks,
+ iobj), True)
+ for idx, success in enumerate(result.payload):
+ if not success:
+ logging.warn("pause-sync of instance %s for disk %d failed",
+ instance, idx)
+
+ feedback_fn("* running the instance OS create scripts...")
+ # FIXME: pass debug option from opcode to backend
+ os_add_result = \
+ self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
+ self.op.debug_level)
+ if pause_sync:
+ feedback_fn("* resuming disk sync")
+ result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+ (iobj.disks,
+ iobj), False)
+ for idx, success in enumerate(result.payload):
+ if not success:
+ logging.warn("resume-sync of instance %s for disk %d failed",
+ instance, idx)
+
+ os_add_result.Raise("Could not add os for instance %s"
+ " on node %s" % (instance, pnode_name))
+
+ else:
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ feedback_fn("* running the instance OS import scripts...")
+
+ transfers = []
+
+ for idx, image in enumerate(self.src_images):
+ if not image:
+ continue
+
+ # FIXME: pass debug option from opcode to backend
+ dt = masterd.instance.DiskTransfer("disk/%s" % idx,
+ constants.IEIO_FILE, (image, ),
+ constants.IEIO_SCRIPT,
+ (iobj.disks[idx], idx),
+ None)
+ transfers.append(dt)
+
+ import_result = \
+ masterd.instance.TransferInstanceData(self, feedback_fn,
+ self.op.src_node, pnode_name,
+ self.pnode.secondary_ip,
+ iobj, transfers)
+ if not compat.all(import_result):
+ self.LogWarning("Some disks for instance %s on node %s were not"
+ " imported successfully" % (instance, pnode_name))
+
+ rename_from = self._old_instance_name
+
+ elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
+ feedback_fn("* preparing remote import...")
+ # The source cluster will stop the instance before attempting to make
+ # a connection. In some cases stopping an instance can take a long
+ # time, hence the shutdown timeout is added to the connection
+ # timeout.
+ connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
+ self.op.source_shutdown_timeout)
+ timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
+
+ assert iobj.primary_node == self.pnode.name
+ disk_results = \
+ masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
+ self.source_x509_ca,
+ self._cds, timeouts)
+ if not compat.all(disk_results):
+ # TODO: Should the instance still be started, even if some disks
+ # failed to import (valid for local imports, too)?
+ self.LogWarning("Some disks for instance %s on node %s were not"
+ " imported successfully" % (instance, pnode_name))
+
+ rename_from = self.source_instance_name
+
+ else:
+ # also checked in the prereq part
+ raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
+ % self.op.mode)
+
+ # Run rename script on newly imported instance
+ assert iobj.name == instance
+ feedback_fn("Running rename script for %s" % instance)
+ result = self.rpc.call_instance_run_rename(pnode_name, iobj,
+ rename_from,
+ self.op.debug_level)
+ if result.fail_msg:
+ self.LogWarning("Failed to run rename script for %s on node"
+ " %s: %s" % (instance, pnode_name, result.fail_msg))
+
+ assert not self.owned_locks(locking.LEVEL_NODE_RES)
+
+ if self.op.start:
+ iobj.admin_state = constants.ADMINST_UP
+ self.cfg.Update(iobj, feedback_fn)
+ logging.info("Starting instance %s on node %s", instance, pnode_name)
+ feedback_fn("* starting instance...")
+ result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
+ False, self.op.reason)
+ result.Raise("Could not start instance")
+
+ return list(iobj.all_nodes)
+
+
+class LUInstanceRename(LogicalUnit):
+ """Rename an instance.
+
+ """
+ HPATH = "instance-rename"
+ HTYPE = constants.HTYPE_INSTANCE
+
+ def CheckArguments(self):
+ """Check arguments.
+
+ """
+ if self.op.ip_check and not self.op.name_check:
+ # TODO: make the ip check more flexible and not depend on the name check
+ raise errors.OpPrereqError("IP address check requires a name check",
+ errors.ECODE_INVAL)
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ env = BuildInstanceHookEnvByObject(self, self.instance)
+ env["INSTANCE_NEW_NAME"] = self.op.new_name
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
+ return (nl, nl)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster and is not running.
+
+ """
+ self.op.instance_name = ExpandInstanceName(self.cfg,
+ self.op.instance_name)
+ instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert instance is not None
+ CheckNodeOnline(self, instance.primary_node)
+ CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+ msg="cannot rename")
+ self.instance = instance
+
+ new_name = self.op.new_name
+ if self.op.name_check:
+ hostname = _CheckHostnameSane(self, new_name)
+ new_name = self.op.new_name = hostname.name
+ if (self.op.ip_check and
+ netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
+ raise errors.OpPrereqError("IP %s of instance %s already in use" %
+ (hostname.ip, new_name),
+ errors.ECODE_NOTUNIQUE)
+
+ instance_list = self.cfg.GetInstanceList()
+ if new_name in instance_list and new_name != instance.name:
+ raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
+ new_name, errors.ECODE_EXISTS)
+
+ def Exec(self, feedback_fn):
+ """Rename the instance.
+
+ """
+ inst = self.instance
+ old_name = inst.name
+
+ rename_file_storage = False
+ if (inst.disk_template in constants.DTS_FILEBASED and
+ self.op.new_name != inst.name):
+ old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
+ rename_file_storage = True
+
+ self.cfg.RenameInstance(inst.name, self.op.new_name)
+ # Change the instance lock. This is definitely safe while we hold the BGL.
+ # Otherwise the new lock would have to be added in acquired mode.
+ assert self.REQ_BGL
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
+ self.glm.remove(locking.LEVEL_INSTANCE, old_name)
+ self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
+
+ # re-read the instance from the configuration after rename
+ inst = self.cfg.GetInstanceInfo(self.op.new_name)
+
+ if rename_file_storage:
+ new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
+ result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
+ old_file_storage_dir,
+ new_file_storage_dir)
+ result.Raise("Could not rename on node %s directory '%s' to '%s'"
+ " (but the instance has been renamed in Ganeti)" %
+ (inst.primary_node, old_file_storage_dir,
+ new_file_storage_dir))
+
+ StartInstanceDisks(self, inst, None)
+ # update info on disks
+ info = GetInstanceInfoText(inst)
+ for (idx, disk) in enumerate(inst.disks):
+ for node in inst.all_nodes:
+ self.cfg.SetDiskID(disk, node)
+ result = self.rpc.call_blockdev_setinfo(node, disk, info)
+ if result.fail_msg:
+ self.LogWarning("Error setting info on node %s for disk %s: %s",
+ node, idx, result.fail_msg)
+ try:
+ result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
+ old_name, self.op.debug_level)
+ msg = result.fail_msg
+ if msg:
+ msg = ("Could not run OS rename script for instance %s on node %s"
+ " (but the instance has been renamed in Ganeti): %s" %
+ (inst.name, inst.primary_node, msg))
+ self.LogWarning(msg)
+ finally:
+ ShutdownInstanceDisks(self, inst)
+
+ return inst.name
+
+
+class LUInstanceRemove(LogicalUnit):
+ """Remove an instance.
+
+ """
+ HPATH = "instance-remove"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ env = BuildInstanceHookEnvByObject(self, self.instance)
+ env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode()]
+ nl_post = list(self.instance.all_nodes) + nl
+ return (nl, nl_post)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+
+ def Exec(self, feedback_fn):
+ """Remove the instance.
+
+ """
+ instance = self.instance
+ logging.info("Shutting down instance %s on node %s",
+ instance.name, instance.primary_node)
+
+ result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
+ self.op.shutdown_timeout,
+ self.op.reason)
+ msg = result.fail_msg
+ if msg:
+ if self.op.ignore_failures:
+ feedback_fn("Warning: can't shutdown instance: %s" % msg)
+ else:
+ raise errors.OpExecError("Could not shutdown instance %s on"
+ " node %s: %s" %
+ (instance.name, instance.primary_node, msg))
+
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+ assert not (set(instance.all_nodes) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Not owning correct locks"
+
+ RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
+
+
+class LUInstanceMove(LogicalUnit):
+ """Move an instance by data-copying.
+
+ """
+ HPATH = "instance-move"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ target_node = ExpandNodeName(self.cfg, self.op.target_node)
+ self.op.target_node = target_node
+ self.needed_locks[locking.LEVEL_NODE] = [target_node]
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes(primary_only=True)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ env = {
+ "TARGET_NODE": self.op.target_node,
+ "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
+ }
+ env.update(BuildInstanceHookEnvByObject(self, self.instance))
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [
+ self.cfg.GetMasterNode(),
+ self.instance.primary_node,
+ self.op.target_node,
+ ]
+ return (nl, nl)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+
+ if instance.disk_template not in constants.DTS_COPYABLE:
+ raise errors.OpPrereqError("Disk template %s not suitable for copying" %
+ instance.disk_template, errors.ECODE_STATE)
+
+ node = self.cfg.GetNodeInfo(self.op.target_node)
+ assert node is not None, \
+ "Cannot retrieve locked node %s" % self.op.target_node
+
+ self.target_node = target_node = node.name
+
+ if target_node == instance.primary_node:
+ raise errors.OpPrereqError("Instance %s is already on the node %s" %
+ (instance.name, target_node),
+ errors.ECODE_STATE)
+
+ bep = self.cfg.GetClusterInfo().FillBE(instance)
+
+ for idx, dsk in enumerate(instance.disks):
+ if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
+ raise errors.OpPrereqError("Instance disk %d has a complex layout,"
+ " cannot copy" % idx, errors.ECODE_STATE)
+
+ CheckNodeOnline(self, target_node)
+ CheckNodeNotDrained(self, target_node)
+ CheckNodeVmCapable(self, target_node)
+ cluster = self.cfg.GetClusterInfo()
+ group_info = self.cfg.GetNodeGroup(node.group)
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster, group_info)
+ CheckTargetNodeIPolicy(self, ipolicy, instance, node, self.cfg,
+ ignore=self.op.ignore_ipolicy)
+
+ if instance.admin_state == constants.ADMINST_UP:
+ # check memory requirements on the secondary node
+ CheckNodeFreeMemory(self, target_node,
+ "failing over instance %s" %
+ instance.name, bep[constants.BE_MAXMEM],
+ instance.hypervisor)
+ else:
+ self.LogInfo("Not checking memory on the secondary node as"
+ " instance will not be started")
+
+ # check bridge existance
+ CheckInstanceBridgesExist(self, instance, node=target_node)
+
+ def Exec(self, feedback_fn):
+ """Move an instance.
+
+ The move is done by shutting it down on its present node, copying
+ the data over (slow) and starting it on the new node.
+
+ """
+ instance = self.instance
+
+ source_node = instance.primary_node
+ target_node = self.target_node
+
+ self.LogInfo("Shutting down instance %s on source node %s",
+ instance.name, source_node)
+
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
+ result = self.rpc.call_instance_shutdown(source_node, instance,
+ self.op.shutdown_timeout,
+ self.op.reason)
+ msg = result.fail_msg
+ if msg:
+ if self.op.ignore_consistency:
+ self.LogWarning("Could not shutdown instance %s on node %s."
+ " Proceeding anyway. Please make sure node"
+ " %s is down. Error details: %s",
+ instance.name, source_node, source_node, msg)
+ else:
+ raise errors.OpExecError("Could not shutdown instance %s on"
+ " node %s: %s" %
+ (instance.name, source_node, msg))
+
+ # create the target disks
+ try:
+ CreateDisks(self, instance, target_node=target_node)
+ except errors.OpExecError:
+ self.LogWarning("Device creation failed")
+ self.cfg.ReleaseDRBDMinors(instance.name)
+ raise
+
+ cluster_name = self.cfg.GetClusterInfo().cluster_name
+
+ errs = []
+ # activate, get path, copy the data over
+ for idx, disk in enumerate(instance.disks):
+ self.LogInfo("Copying data for disk %d", idx)
+ result = self.rpc.call_blockdev_assemble(target_node, (disk, instance),
+ instance.name, True, idx)
+ if result.fail_msg:
+ self.LogWarning("Can't assemble newly created disk %d: %s",
+ idx, result.fail_msg)
+ errs.append(result.fail_msg)
+ break
+ dev_path = result.payload
+ result = self.rpc.call_blockdev_export(source_node, (disk, instance),
+ target_node, dev_path,
+ cluster_name)
+ if result.fail_msg:
+ self.LogWarning("Can't copy data over for disk %d: %s",
+ idx, result.fail_msg)
+ errs.append(result.fail_msg)
+ break
+
+ if errs:
+ self.LogWarning("Some disks failed to copy, aborting")
+ try:
+ RemoveDisks(self, instance, target_node=target_node)
+ finally:
+ self.cfg.ReleaseDRBDMinors(instance.name)
+ raise errors.OpExecError("Errors during disk copy: %s" %
+ (",".join(errs),))
+
+ instance.primary_node = target_node
+ self.cfg.Update(instance, feedback_fn)
+
+ self.LogInfo("Removing the disks on the original node")
+ RemoveDisks(self, instance, target_node=source_node)
+
+ # Only start the instance if it's marked as up
+ if instance.admin_state == constants.ADMINST_UP:
+ self.LogInfo("Starting instance %s on node %s",
+ instance.name, target_node)
+
+ disks_ok, _ = AssembleInstanceDisks(self, instance,
+ ignore_secondaries=True)
+ if not disks_ok:
+ ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Can't activate the instance's disks")
+
+ result = self.rpc.call_instance_start(target_node,
+ (instance, None, None), False,
+ self.op.reason)
+ msg = result.fail_msg
+ if msg:
+ ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance %s on node %s: %s" %
+ (instance.name, target_node, msg))
+
+
+class LUInstanceMultiAlloc(NoHooksLU):
+ """Allocates multiple instances at the same time.
+
+ """
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ """Check arguments.
+
+ """
+ nodes = []
+ for inst in self.op.instances:
+ if inst.iallocator is not None:
+ raise errors.OpPrereqError("iallocator are not allowed to be set on"
+ " instance objects", errors.ECODE_INVAL)
+ nodes.append(bool(inst.pnode))
+ if inst.disk_template in constants.DTS_INT_MIRROR:
+ nodes.append(bool(inst.snode))
+
+ has_nodes = compat.any(nodes)
+ if compat.all(nodes) ^ has_nodes:
+ raise errors.OpPrereqError("There are instance objects providing"
+ " pnode/snode while others do not",
+ errors.ECODE_INVAL)
+
+ if self.op.iallocator is None:
+ default_iallocator = self.cfg.GetDefaultIAllocator()
+ if default_iallocator and has_nodes:
+ self.op.iallocator = default_iallocator
+ else:
+ raise errors.OpPrereqError("No iallocator or nodes on the instances"
+ " given and no cluster-wide default"
+ " iallocator found; please specify either"
+ " an iallocator or nodes on the instances"
+ " or set a cluster-wide default iallocator",
+ errors.ECODE_INVAL)
+
+ _CheckOpportunisticLocking(self.op)
+
+ dups = utils.FindDuplicates([op.instance_name for op in self.op.instances])
+ if dups:
+ raise errors.OpPrereqError("There are duplicate instance names: %s" %
+ utils.CommaJoin(dups), errors.ECODE_INVAL)
+
+ def ExpandNames(self):
+ """Calculate the locks.
+
+ """
+ self.share_locks = ShareAll()
+ self.needed_locks = {
+ # iallocator will select nodes and even if no iallocator is used,
+ # collisions with LUInstanceCreate should be avoided
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
+
+ if self.op.iallocator:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
+
+ if self.op.opportunistic_locking:
+ self.opportunistic_locks[locking.LEVEL_NODE] = True
+ self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
+ else:
+ nodeslist = []
+ for inst in self.op.instances:
+ inst.pnode = ExpandNodeName(self.cfg, inst.pnode)
+ nodeslist.append(inst.pnode)
+ if inst.snode is not None:
+ inst.snode = ExpandNodeName(self.cfg, inst.snode)
+ nodeslist.append(inst.snode)
+
+ self.needed_locks[locking.LEVEL_NODE] = nodeslist
+ # Lock resources of instance's primary and secondary nodes (copy to
+ # prevent accidential modification)
+ self.needed_locks[locking.LEVEL_NODE_RES] = list(nodeslist)
+
+ def CheckPrereq(self):
+ """Check prerequisite.
+
+ """
+ cluster = self.cfg.GetClusterInfo()
+ default_vg = self.cfg.GetVGName()
+ ec_id = self.proc.GetECId()
+
+ if self.op.opportunistic_locking:
+ # Only consider nodes for which a lock is held
+ node_whitelist = list(self.owned_locks(locking.LEVEL_NODE))
+ else:
+ node_whitelist = None
+
+ insts = [_CreateInstanceAllocRequest(op, ComputeDisks(op, default_vg),
+ _ComputeNics(op, cluster, None,
+ self.cfg, ec_id),
+ _ComputeFullBeParams(op, cluster),
+ node_whitelist)
+ for op in self.op.instances]
+
+ req = iallocator.IAReqMultiInstanceAlloc(instances=insts)
+ ial = iallocator.IAllocator(self.cfg, self.rpc, req)
+
+ ial.Run(self.op.iallocator)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute nodes using"
+ " iallocator '%s': %s" %
+ (self.op.iallocator, ial.info),
+ errors.ECODE_NORES)
+
+ self.ia_result = ial.result
+
+ if self.op.dry_run:
+ self.dry_run_result = objects.FillDict(self._ConstructPartialResult(), {
+ constants.JOB_IDS_KEY: [],
+ })
+
+ def _ConstructPartialResult(self):
+ """Contructs the partial result.
+
+ """
+ (allocatable, failed) = self.ia_result
+ return {
+ opcodes.OpInstanceMultiAlloc.ALLOCATABLE_KEY:
+ map(compat.fst, allocatable),
+ opcodes.OpInstanceMultiAlloc.FAILED_KEY: failed,
+ }
+
+ def Exec(self, feedback_fn):
+ """Executes the opcode.
+
+ """
+ op2inst = dict((op.instance_name, op) for op in self.op.instances)
+ (allocatable, failed) = self.ia_result
+
+ jobs = []
+ for (name, nodes) in allocatable:
+ op = op2inst.pop(name)
+
+ if len(nodes) > 1:
+ (op.pnode, op.snode) = nodes
+ else:
+ (op.pnode,) = nodes
+
+ jobs.append([op])
+
+ missing = set(op2inst.keys()) - set(failed)
+ assert not missing, \
+ "Iallocator did return incomplete result: %s" % utils.CommaJoin(missing)
+
+ return ResultWithJobs(jobs, **self._ConstructPartialResult())
+
+
+class _InstNicModPrivate:
+ """Data structure for network interface modifications.
+
+ Used by L{LUInstanceSetParams}.
+
+ """
+ def __init__(self):
+ self.params = None
+ self.filled = None
+
+
+def _PrepareContainerMods(mods, private_fn):
+ """Prepares a list of container modifications by adding a private data field.
+
+ @type mods: list of tuples; (operation, index, parameters)
+ @param mods: List of modifications
+ @type private_fn: callable or None
+ @param private_fn: Callable for constructing a private data field for a
+ modification
+ @rtype: list
+
+ """
+ if private_fn is None:
+ fn = lambda: None
+ else:
+ fn = private_fn
+
+ return [(op, idx, params, fn()) for (op, idx, params) in mods]
+
+
+def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
+ """Checks if nodes have enough physical CPUs
+
+ This function checks if all given nodes have the needed number of
+ physical CPUs. In case any node has less CPUs or we cannot get the
+ information from the node, this function raises an OpPrereqError
+ exception.
+
+ @type lu: C{LogicalUnit}
+ @param lu: a logical unit from which we get configuration data
+ @type nodenames: C{list}
+ @param nodenames: the list of node names to check
+ @type requested: C{int}
+ @param requested: the minimum acceptable number of physical CPUs
+ @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
+ or we cannot check the node
+
+ """
+ nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name], None)
+ for node in nodenames:
+ info = nodeinfo[node]
+ info.Raise("Cannot get current information from node %s" % node,
+ prereq=True, ecode=errors.ECODE_ENVIRON)
+ (_, _, (hv_info, )) = info.payload
+ num_cpus = hv_info.get("cpu_total", None)
+ if not isinstance(num_cpus, int):
+ raise errors.OpPrereqError("Can't compute the number of physical CPUs"
+ " on node %s, result was '%s'" %
+ (node, num_cpus), errors.ECODE_ENVIRON)
+ if requested > num_cpus:
+ raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
+ "required" % (node, num_cpus, requested),
+ errors.ECODE_NORES)
+
+
+def GetItemFromContainer(identifier, kind, container):
+ """Return the item refered by the identifier.
+
+ @type identifier: string
+ @param identifier: Item index or name or UUID
+ @type kind: string
+ @param kind: One-word item description
+ @type container: list
+ @param container: Container to get the item from
+
+ """
+ # Index
+ try:
+ idx = int(identifier)
+ if idx == -1:
+ # Append
+ absidx = len(container) - 1
+ elif idx < 0:
+ raise IndexError("Not accepting negative indices other than -1")
+ elif idx > len(container):
+ raise IndexError("Got %s index %s, but there are only %s" %
+ (kind, idx, len(container)))
+ else:
+ absidx = idx
+ return (absidx, container[idx])
+ except ValueError:
+ pass
+
+ for idx, item in enumerate(container):
+ if item.uuid == identifier or item.name == identifier:
+ return (idx, item)
+
+ raise errors.OpPrereqError("Cannot find %s with identifier %s" %
+ (kind, identifier), errors.ECODE_NOENT)
+
+
+def _ApplyContainerMods(kind, container, chgdesc, mods,
+ create_fn, modify_fn, remove_fn):
+ """Applies descriptions in C{mods} to C{container}.
+
+ @type kind: string
+ @param kind: One-word item description
+ @type container: list
+ @param container: Container to modify
+ @type chgdesc: None or list
+ @param chgdesc: List of applied changes
+ @type mods: list
+ @param mods: Modifications as returned by L{_PrepareContainerMods}
+ @type create_fn: callable
+ @param create_fn: Callback for creating a new item (L{constants.DDM_ADD});
+ receives absolute item index, parameters and private data object as added
+ by L{_PrepareContainerMods}, returns tuple containing new item and changes
+ as list
+ @type modify_fn: callable
+ @param modify_fn: Callback for modifying an existing item
+ (L{constants.DDM_MODIFY}); receives absolute item index, item, parameters
+ and private data object as added by L{_PrepareContainerMods}, returns
+ changes as list
+ @type remove_fn: callable
+ @param remove_fn: Callback on removing item; receives absolute item index,
+ item and private data object as added by L{_PrepareContainerMods}
+
+ """
+ for (op, identifier, params, private) in mods:
+ changes = None
+
+ if op == constants.DDM_ADD:
+ # Calculate where item will be added
+ # When adding an item, identifier can only be an index
+ try:
+ idx = int(identifier)
+ except ValueError:
+ raise errors.OpPrereqError("Only possitive integer or -1 is accepted as"
+ " identifier for %s" % constants.DDM_ADD,
+ errors.ECODE_INVAL)
+ if idx == -1:
+ addidx = len(container)
+ else:
+ if idx < 0:
+ raise IndexError("Not accepting negative indices other than -1")
+ elif idx > len(container):
+ raise IndexError("Got %s index %s, but there are only %s" %
+ (kind, idx, len(container)))
+ addidx = idx
+
+ if create_fn is None:
+ item = params
+ else:
+ (item, changes) = create_fn(addidx, params, private)
+
+ if idx == -1:
+ container.append(item)
+ else:
+ assert idx >= 0
+ assert idx <= len(container)
+ # list.insert does so before the specified index
+ container.insert(idx, item)
+ else:
+ # Retrieve existing item
+ (absidx, item) = GetItemFromContainer(identifier, kind, container)
+
+ if op == constants.DDM_REMOVE:
+ assert not params
+
+ if remove_fn is not None:
+ remove_fn(absidx, item, private)
+
+ changes = [("%s/%s" % (kind, absidx), "remove")]
+
+ assert container[absidx] == item
+ del container[absidx]
+ elif op == constants.DDM_MODIFY:
+ if modify_fn is not None:
+ changes = modify_fn(absidx, item, params, private)
+ else:
+ raise errors.ProgrammerError("Unhandled operation '%s'" % op)
+
+ assert _TApplyContModsCbChanges(changes)
+
+ if not (chgdesc is None or changes is None):
+ chgdesc.extend(changes)
+
+
+def _UpdateIvNames(base_index, disks):
+ """Updates the C{iv_name} attribute of disks.
+
+ @type disks: list of L{objects.Disk}
+
+ """
+ for (idx, disk) in enumerate(disks):
+ disk.iv_name = "disk/%s" % (base_index + idx, )
+
+
+class LUInstanceSetParams(LogicalUnit):
+ """Modifies an instances's parameters.
+
+ """
+ HPATH = "instance-modify"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ @staticmethod
+ def _UpgradeDiskNicMods(kind, mods, verify_fn):
+ assert ht.TList(mods)
+ assert not mods or len(mods[0]) in (2, 3)
+
+ if mods and len(mods[0]) == 2:
+ result = []
+
+ addremove = 0
+ for op, params in mods:
+ if op in (constants.DDM_ADD, constants.DDM_REMOVE):
+ result.append((op, -1, params))
+ addremove += 1
+
+ if addremove > 1:
+ raise errors.OpPrereqError("Only one %s add or remove operation is"
+ " supported at a time" % kind,
+ errors.ECODE_INVAL)
+ else:
+ result.append((constants.DDM_MODIFY, op, params))
+
+ assert verify_fn(result)
+ else:
+ result = mods
+
+ return result
+
+ @staticmethod
+ def _CheckMods(kind, mods, key_types, item_fn):
+ """Ensures requested disk/NIC modifications are valid.
+
+ """
+ for (op, _, params) in mods:
+ assert ht.TDict(params)
+
+ # If 'key_types' is an empty dict, we assume we have an
+ # 'ext' template and thus do not ForceDictType
+ if key_types:
+ utils.ForceDictType(params, key_types)
+
+ if op == constants.DDM_REMOVE:
+ if params:
+ raise errors.OpPrereqError("No settings should be passed when"
+ " removing a %s" % kind,
+ errors.ECODE_INVAL)
+ elif op in (constants.DDM_ADD, constants.DDM_MODIFY):
+ item_fn(op, params)
+ else:
+ raise errors.ProgrammerError("Unhandled operation '%s'" % op)
+
+ @staticmethod
+ def _VerifyDiskModification(op, params):
+ """Verifies a disk modification.
+
+ """
+ if op == constants.DDM_ADD:
+ mode = params.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
+ if mode not in constants.DISK_ACCESS_SET:
+ raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
+ errors.ECODE_INVAL)
+
+ size = params.get(constants.IDISK_SIZE, None)
+ if size is None:
+ raise errors.OpPrereqError("Required disk parameter '%s' missing" %
+ constants.IDISK_SIZE, errors.ECODE_INVAL)
+
+ try:
+ size = int(size)
+ except (TypeError, ValueError), err:
+ raise errors.OpPrereqError("Invalid disk size parameter: %s" % err,
+ errors.ECODE_INVAL)
+
+ params[constants.IDISK_SIZE] = size
+ name = params.get(constants.IDISK_NAME, None)
+ if name is not None and name.lower() == constants.VALUE_NONE:
+ params[constants.IDISK_NAME] = None
+
+ elif op == constants.DDM_MODIFY:
+ if constants.IDISK_SIZE in params:
+ raise errors.OpPrereqError("Disk size change not possible, use"
+ " grow-disk", errors.ECODE_INVAL)
+ if len(params) > 2:
+ raise errors.OpPrereqError("Disk modification doesn't support"
+ " additional arbitrary parameters",
+ errors.ECODE_INVAL)
+ name = params.get(constants.IDISK_NAME, None)
+ if name is not None and name.lower() == constants.VALUE_NONE:
+ params[constants.IDISK_NAME] = None
+
+ @staticmethod
+ def _VerifyNicModification(op, params):
+ """Verifies a network interface modification.
+
+ """
+ if op in (constants.DDM_ADD, constants.DDM_MODIFY):
+ ip = params.get(constants.INIC_IP, None)
+ name = params.get(constants.INIC_NAME, None)
+ req_net = params.get(constants.INIC_NETWORK, None)
+ link = params.get(constants.NIC_LINK, None)
+ mode = params.get(constants.NIC_MODE, None)
+ if name is not None and name.lower() == constants.VALUE_NONE:
+ params[constants.INIC_NAME] = None
+ if req_net is not None:
+ if req_net.lower() == constants.VALUE_NONE:
+ params[constants.INIC_NETWORK] = None
+ req_net = None
+ elif link is not None or mode is not None:
+ raise errors.OpPrereqError("If network is given"
+ " mode or link should not",
+ errors.ECODE_INVAL)
+
+ if op == constants.DDM_ADD:
+ macaddr = params.get(constants.INIC_MAC, None)
+ if macaddr is None:
+ params[constants.INIC_MAC] = constants.VALUE_AUTO
+
+ if ip is not None:
+ if ip.lower() == constants.VALUE_NONE:
+ params[constants.INIC_IP] = None
+ else:
+ if ip.lower() == constants.NIC_IP_POOL:
+ if op == constants.DDM_ADD and req_net is None:
+ raise errors.OpPrereqError("If ip=pool, parameter network"
+ " cannot be none",
+ errors.ECODE_INVAL)
+ else:
+ if not netutils.IPAddress.IsValid(ip):
+ raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
+ errors.ECODE_INVAL)
+
+ if constants.INIC_MAC in params:
+ macaddr = params[constants.INIC_MAC]
+ if macaddr not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ macaddr = utils.NormalizeAndValidateMac(macaddr)
+
+ if op == constants.DDM_MODIFY and macaddr == constants.VALUE_AUTO:
+ raise errors.OpPrereqError("'auto' is not a valid MAC address when"
+ " modifying an existing NIC",
+ errors.ECODE_INVAL)
+
+ def CheckArguments(self):
+ if not (self.op.nics or self.op.disks or self.op.disk_template or
+ self.op.hvparams or self.op.beparams or self.op.os_name or
+ self.op.offline is not None or self.op.runtime_mem or
+ self.op.pnode):
+ raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
+
+ if self.op.hvparams:
+ CheckParamsNotGlobal(self.op.hvparams, constants.HVC_GLOBALS,
+ "hypervisor", "instance", "cluster")
+
+ self.op.disks = self._UpgradeDiskNicMods(
+ "disk", self.op.disks, opcodes.OpInstanceSetParams.TestDiskModifications)
+ self.op.nics = self._UpgradeDiskNicMods(
+ "NIC", self.op.nics, opcodes.OpInstanceSetParams.TestNicModifications)
+
+ if self.op.disks and self.op.disk_template is not None:
+ raise errors.OpPrereqError("Disk template conversion and other disk"
+ " changes not supported at the same time",
+ errors.ECODE_INVAL)
+
+ if (self.op.disk_template and
+ self.op.disk_template in constants.DTS_INT_MIRROR and
+ self.op.remote_node is None):
+ raise errors.OpPrereqError("Changing the disk template to a mirrored"
+ " one requires specifying a secondary node",
+ errors.ECODE_INVAL)
+
+ # Check NIC modifications
+ self._CheckMods("NIC", self.op.nics, constants.INIC_PARAMS_TYPES,
+ self._VerifyNicModification)
+
+ if self.op.pnode:
+ self.op.pnode = ExpandNodeName(self.cfg, self.op.pnode)
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ # Can't even acquire node locks in shared mode as upcoming changes in
+ # Ganeti 2.6 will start to modify the node object on disk conversion
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ # Look node group to look up the ipolicy
+ self.share_locks[locking.LEVEL_NODEGROUP] = 1
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+ # Acquire locks for the instance's nodegroups optimistically. Needs
+ # to be verified in CheckPrereq
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+ elif level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+ if self.op.disk_template and self.op.remote_node:
+ self.op.remote_node = ExpandNodeName(self.cfg, self.op.remote_node)
+ self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
+ elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, primary and secondaries.
+
+ """
+ args = {}
+ if constants.BE_MINMEM in self.be_new:
+ args["minmem"] = self.be_new[constants.BE_MINMEM]
+ if constants.BE_MAXMEM in self.be_new:
+ args["maxmem"] = self.be_new[constants.BE_MAXMEM]
+ if constants.BE_VCPUS in self.be_new:
+ args["vcpus"] = self.be_new[constants.BE_VCPUS]
+ # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
+ # information at all.
+
+ if self._new_nics is not None:
+ nics = []
+
+ for nic in self._new_nics:
+ n = copy.deepcopy(nic)
+ nicparams = self.cluster.SimpleFillNIC(n.nicparams)
+ n.nicparams = nicparams
+ nics.append(NICToTuple(self, n))
+
+ args["nics"] = nics
+
+ env = BuildInstanceHookEnvByObject(self, self.instance, override=args)
+ if self.op.disk_template:
+ env["NEW_DISK_TEMPLATE"] = self.op.disk_template
+ if self.op.runtime_mem:
+ env["RUNTIME_MEMORY"] = self.op.runtime_mem
+
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
+ return (nl, nl)
+
+ def _PrepareNicModification(self, params, private, old_ip, old_net_uuid,
+ old_params, cluster, pnode):
+
+ update_params_dict = dict([(key, params[key])
+ for key in constants.NICS_PARAMETERS
+ if key in params])
+
+ req_link = update_params_dict.get(constants.NIC_LINK, None)
+ req_mode = update_params_dict.get(constants.NIC_MODE, None)
+
+ new_net_uuid = None
+ new_net_uuid_or_name = params.get(constants.INIC_NETWORK, old_net_uuid)
+ if new_net_uuid_or_name:
+ new_net_uuid = self.cfg.LookupNetwork(new_net_uuid_or_name)
+ new_net_obj = self.cfg.GetNetwork(new_net_uuid)
+
+ if old_net_uuid:
+ old_net_obj = self.cfg.GetNetwork(old_net_uuid)
+
+ if new_net_uuid:
+ netparams = self.cfg.GetGroupNetParams(new_net_uuid, pnode)
+ if not netparams:
+ raise errors.OpPrereqError("No netparams found for the network"
+ " %s, probably not connected" %
+ new_net_obj.name, errors.ECODE_INVAL)
+ new_params = dict(netparams)
+ else:
+ new_params = GetUpdatedParams(old_params, update_params_dict)
+
+ utils.ForceDictType(new_params, constants.NICS_PARAMETER_TYPES)
+
+ new_filled_params = cluster.SimpleFillNIC(new_params)
+ objects.NIC.CheckParameterSyntax(new_filled_params)
+
+ new_mode = new_filled_params[constants.NIC_MODE]
+ if new_mode == constants.NIC_MODE_BRIDGED:
+ bridge = new_filled_params[constants.NIC_LINK]
+ msg = self.rpc.call_bridges_exist(pnode, [bridge]).fail_msg
+ if msg:
+ msg = "Error checking bridges on node '%s': %s" % (pnode, msg)
+ if self.op.force:
+ self.warn.append(msg)
+ else:
+ raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
+
+ elif new_mode == constants.NIC_MODE_ROUTED:
+ ip = params.get(constants.INIC_IP, old_ip)
+ if ip is None:
+ raise errors.OpPrereqError("Cannot set the NIC IP address to None"
+ " on a routed NIC", errors.ECODE_INVAL)
+
+ elif new_mode == constants.NIC_MODE_OVS:
+ # TODO: check OVS link
+ self.LogInfo("OVS links are currently not checked for correctness")
+
+ if constants.INIC_MAC in params:
+ mac = params[constants.INIC_MAC]
+ if mac is None:
+ raise errors.OpPrereqError("Cannot unset the NIC MAC address",
+ errors.ECODE_INVAL)
+ elif mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ # otherwise generate the MAC address
+ params[constants.INIC_MAC] = \
+ self.cfg.GenerateMAC(new_net_uuid, self.proc.GetECId())
+ else:
+ # or validate/reserve the current one
+ try:
+ self.cfg.ReserveMAC(mac, self.proc.GetECId())
+ except errors.ReservationError:
+ raise errors.OpPrereqError("MAC address '%s' already in use"
+ " in cluster" % mac,
+ errors.ECODE_NOTUNIQUE)
+ elif new_net_uuid != old_net_uuid:
+
+ def get_net_prefix(net_uuid):
+ mac_prefix = None
+ if net_uuid:
+ nobj = self.cfg.GetNetwork(net_uuid)
+ mac_prefix = nobj.mac_prefix
+
+ return mac_prefix
+
+ new_prefix = get_net_prefix(new_net_uuid)
+ old_prefix = get_net_prefix(old_net_uuid)
+ if old_prefix != new_prefix:
+ params[constants.INIC_MAC] = \
+ self.cfg.GenerateMAC(new_net_uuid, self.proc.GetECId())
+
+ # if there is a change in (ip, network) tuple
+ new_ip = params.get(constants.INIC_IP, old_ip)
+ if (new_ip, new_net_uuid) != (old_ip, old_net_uuid):
+ if new_ip:
+ # if IP is pool then require a network and generate one IP
+ if new_ip.lower() == constants.NIC_IP_POOL:
+ if new_net_uuid:
+ try:
+ new_ip = self.cfg.GenerateIp(new_net_uuid, self.proc.GetECId())
+ except errors.ReservationError:
+ raise errors.OpPrereqError("Unable to get a free IP"
+ " from the address pool",
+ errors.ECODE_STATE)
+ self.LogInfo("Chose IP %s from network %s",
+ new_ip,
+ new_net_obj.name)
+ params[constants.INIC_IP] = new_ip
+ else:
+ raise errors.OpPrereqError("ip=pool, but no network found",
+ errors.ECODE_INVAL)
+ # Reserve new IP if in the new network if any
+ elif new_net_uuid:
+ try:
+ self.cfg.ReserveIp(new_net_uuid, new_ip, self.proc.GetECId())
+ self.LogInfo("Reserving IP %s in network %s",
+ new_ip, new_net_obj.name)
+ except errors.ReservationError:
+ raise errors.OpPrereqError("IP %s not available in network %s" %
+ (new_ip, new_net_obj.name),
+ errors.ECODE_NOTUNIQUE)
+ # new network is None so check if new IP is a conflicting IP
+ elif self.op.conflicts_check:
+ _CheckForConflictingIp(self, new_ip, pnode)
+
+ # release old IP if old network is not None
+ if old_ip and old_net_uuid:
+ try:
+ self.cfg.ReleaseIp(old_net_uuid, old_ip, self.proc.GetECId())
+ except errors.AddressPoolError:
+ logging.warning("Release IP %s not contained in network %s",
+ old_ip, old_net_obj.name)
+
+ # there are no changes in (ip, network) tuple and old network is not None
+ elif (old_net_uuid is not None and
+ (req_link is not None or req_mode is not None)):
+ raise errors.OpPrereqError("Not allowed to change link or mode of"
+ " a NIC that is connected to a network",
+ errors.ECODE_INVAL)
+
+ private.params = new_params
+ private.filled = new_filled_params
+
+ def _PreCheckDiskTemplate(self, pnode_info):
+ """CheckPrereq checks related to a new disk template."""
+ # Arguments are passed to avoid configuration lookups
+ instance = self.instance
+ pnode = instance.primary_node
+ cluster = self.cluster
+ if instance.disk_template == self.op.disk_template:
+ raise errors.OpPrereqError("Instance already has disk template %s" %
+ instance.disk_template, errors.ECODE_INVAL)
+
+ if (instance.disk_template,
+ self.op.disk_template) not in self._DISK_CONVERSIONS:
+ raise errors.OpPrereqError("Unsupported disk template conversion from"
+ " %s to %s" % (instance.disk_template,
+ self.op.disk_template),
+ errors.ECODE_INVAL)
+ CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot change disk template")
+ if self.op.disk_template in constants.DTS_INT_MIRROR:
+ if self.op.remote_node == pnode:
+ raise errors.OpPrereqError("Given new secondary node %s is the same"
+ " as the primary node of the instance" %
+ self.op.remote_node, errors.ECODE_STATE)
+ CheckNodeOnline(self, self.op.remote_node)
+ CheckNodeNotDrained(self, self.op.remote_node)
+ # FIXME: here we assume that the old instance type is DT_PLAIN
+ assert instance.disk_template == constants.DT_PLAIN
+ disks = [{constants.IDISK_SIZE: d.size,
+ constants.IDISK_VG: d.logical_id[0]}
+ for d in instance.disks]
+ required = ComputeDiskSizePerVG(self.op.disk_template, disks)
+ CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
+
+ snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
+ snode_group = self.cfg.GetNodeGroup(snode_info.group)
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
+ snode_group)
+ CheckTargetNodeIPolicy(self, ipolicy, instance, snode_info, self.cfg,
+ ignore=self.op.ignore_ipolicy)
+ if pnode_info.group != snode_info.group:
+ self.LogWarning("The primary and secondary nodes are in two"
+ " different node groups; the disk parameters"
+ " from the first disk's node group will be"
+ " used")
+
+ if not self.op.disk_template in constants.DTS_EXCL_STORAGE:
+ # Make sure none of the nodes require exclusive storage
+ nodes = [pnode_info]
+ if self.op.disk_template in constants.DTS_INT_MIRROR:
+ assert snode_info
+ nodes.append(snode_info)
+ has_es = lambda n: IsExclusiveStorageEnabledNode(self.cfg, n)
+ if compat.any(map(has_es, nodes)):
+ errmsg = ("Cannot convert disk template from %s to %s when exclusive"
+ " storage is enabled" % (instance.disk_template,
+ self.op.disk_template))
+ raise errors.OpPrereqError(errmsg, errors.ECODE_STATE)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This only checks the instance list against the existing names.
+
+ """
+ assert self.op.instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
+ instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
+ cluster = self.cluster = self.cfg.GetClusterInfo()
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+
+ pnode = instance.primary_node
+
+ self.warn = []
+
+ if (self.op.pnode is not None and self.op.pnode != pnode and
+ not self.op.force):
+ # verify that the instance is not up
+ instance_info = self.rpc.call_instance_info(pnode, instance.name,
+ instance.hypervisor)
+ if instance_info.fail_msg:
+ self.warn.append("Can't get instance runtime information: %s" %
+ instance_info.fail_msg)
+ elif instance_info.payload:
+ raise errors.OpPrereqError("Instance is still running on %s" % pnode,
+ errors.ECODE_STATE)
+
+ assert pnode in self.owned_locks(locking.LEVEL_NODE)
+ nodelist = list(instance.all_nodes)
+ pnode_info = self.cfg.GetNodeInfo(pnode)
+ self.diskparams = self.cfg.GetInstanceDiskParams(instance)
+
+ #_CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
+ assert pnode_info.group in self.owned_locks(locking.LEVEL_NODEGROUP)
+ group_info = self.cfg.GetNodeGroup(pnode_info.group)
+
+ # dictionary with instance information after the modification
+ ispec = {}
+
+ # Check disk modifications. This is done here and not in CheckArguments
+ # (as with NICs), because we need to know the instance's disk template
+ if instance.disk_template == constants.DT_EXT:
+ self._CheckMods("disk", self.op.disks, {},
+ self._VerifyDiskModification)
+ else:
+ self._CheckMods("disk", self.op.disks, constants.IDISK_PARAMS_TYPES,
+ self._VerifyDiskModification)
+
+ # Prepare disk/NIC modifications
+ self.diskmod = _PrepareContainerMods(self.op.disks, None)
+ self.nicmod = _PrepareContainerMods(self.op.nics, _InstNicModPrivate)
+
+ # Check the validity of the `provider' parameter
+ if instance.disk_template in constants.DT_EXT:
+ for mod in self.diskmod:
+ ext_provider = mod[2].get(constants.IDISK_PROVIDER, None)
+ if mod[0] == constants.DDM_ADD:
+ if ext_provider is None:
+ raise errors.OpPrereqError("Instance template is '%s' and parameter"
+ " '%s' missing, during disk add" %
+ (constants.DT_EXT,
+ constants.IDISK_PROVIDER),
+ errors.ECODE_NOENT)
+ elif mod[0] == constants.DDM_MODIFY:
+ if ext_provider:
+ raise errors.OpPrereqError("Parameter '%s' is invalid during disk"
+ " modification" %
+ constants.IDISK_PROVIDER,
+ errors.ECODE_INVAL)
+ else:
+ for mod in self.diskmod:
+ ext_provider = mod[2].get(constants.IDISK_PROVIDER, None)
+ if ext_provider is not None:
+ raise errors.OpPrereqError("Parameter '%s' is only valid for"
+ " instances of type '%s'" %
+ (constants.IDISK_PROVIDER,
+ constants.DT_EXT),
+ errors.ECODE_INVAL)
+
+ # OS change
+ if self.op.os_name and not self.op.force:
+ CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
+ self.op.force_variant)
+ instance_os = self.op.os_name
+ else:
+ instance_os = instance.os
+
+ assert not (self.op.disk_template and self.op.disks), \
+ "Can't modify disk template and apply disk changes at the same time"
+
+ if self.op.disk_template:
+ self._PreCheckDiskTemplate(pnode_info)
+
+ # hvparams processing
+ if self.op.hvparams:
+ hv_type = instance.hypervisor
+ i_hvdict = GetUpdatedParams(instance.hvparams, self.op.hvparams)
+ utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
+ hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
+
+ # local check
+ hypervisor.GetHypervisorClass(hv_type).CheckParameterSyntax(hv_new)
+ CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
+ self.hv_proposed = self.hv_new = hv_new # the new actual values
+ self.hv_inst = i_hvdict # the new dict (without defaults)
+ else:
+ self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
+ instance.hvparams)
+ self.hv_new = self.hv_inst = {}
+
+ # beparams processing
+ if self.op.beparams:
+ i_bedict = GetUpdatedParams(instance.beparams, self.op.beparams,
+ use_none=True)
+ objects.UpgradeBeParams(i_bedict)
+ utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
+ be_new = cluster.SimpleFillBE(i_bedict)
+ self.be_proposed = self.be_new = be_new # the new actual values
+ self.be_inst = i_bedict # the new dict (without defaults)
+ else:
+ self.be_new = self.be_inst = {}
+ self.be_proposed = cluster.SimpleFillBE(instance.beparams)
+ be_old = cluster.FillBE(instance)
+
+ # CPU param validation -- checking every time a parameter is
+ # changed to cover all cases where either CPU mask or vcpus have
+ # changed
+ if (constants.BE_VCPUS in self.be_proposed and
+ constants.HV_CPU_MASK in self.hv_proposed):
+ cpu_list = \
+ utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
+ # Verify mask is consistent with number of vCPUs. Can skip this
+ # test if only 1 entry in the CPU mask, which means same mask
+ # is applied to all vCPUs.
+ if (len(cpu_list) > 1 and
+ len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
+ raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
+ " CPU mask [%s]" %
+ (self.be_proposed[constants.BE_VCPUS],
+ self.hv_proposed[constants.HV_CPU_MASK]),
+ errors.ECODE_INVAL)
+
+ # Only perform this test if a new CPU mask is given
+ if constants.HV_CPU_MASK in self.hv_new:
+ # Calculate the largest CPU number requested
+ max_requested_cpu = max(map(max, cpu_list))
+ # Check that all of the instance's nodes have enough physical CPUs to
+ # satisfy the requested CPU mask
+ _CheckNodesPhysicalCPUs(self, instance.all_nodes,
+ max_requested_cpu + 1, instance.hypervisor)
+
+ # osparams processing
+ if self.op.osparams:
+ i_osdict = GetUpdatedParams(instance.osparams, self.op.osparams)
+ CheckOSParams(self, True, nodelist, instance_os, i_osdict)
+ self.os_inst = i_osdict # the new dict (without defaults)
+ else:
+ self.os_inst = {}
+
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
+ be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
+ mem_check_list = [pnode]
+ if be_new[constants.BE_AUTO_BALANCE]:
+ # either we changed auto_balance to yes or it was from before
+ mem_check_list.extend(instance.secondary_nodes)
+ instance_info = self.rpc.call_instance_info(pnode, instance.name,
+ instance.hypervisor)
+ nodeinfo = self.rpc.call_node_info(mem_check_list, None,
+ [instance.hypervisor], False)
+ pninfo = nodeinfo[pnode]
+ msg = pninfo.fail_msg
+ if msg:
+ # Assume the primary node is unreachable and go ahead
+ self.warn.append("Can't get info from primary node %s: %s" %
+ (pnode, msg))
+ else:
+ (_, _, (pnhvinfo, )) = pninfo.payload
+ if not isinstance(pnhvinfo.get("memory_free", None), int):
+ self.warn.append("Node data from primary node %s doesn't contain"
+ " free memory information" % pnode)
+ elif instance_info.fail_msg:
+ self.warn.append("Can't get instance runtime information: %s" %
+ instance_info.fail_msg)
+ else:
+ if instance_info.payload:
+ current_mem = int(instance_info.payload["memory"])
+ else:
+ # Assume instance not running
+ # (there is a slight race condition here, but it's not very
+ # probable, and we have no other way to check)
+ # TODO: Describe race condition
+ current_mem = 0
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
+ pnhvinfo["memory_free"])
+ if miss_mem > 0:
+ raise errors.OpPrereqError("This change will prevent the instance"
+ " from starting, due to %d MB of memory"
+ " missing on its primary node" %
+ miss_mem, errors.ECODE_NORES)
+
+ if be_new[constants.BE_AUTO_BALANCE]:
+ for node, nres in nodeinfo.items():
+ if node not in instance.secondary_nodes:
+ continue
+ nres.Raise("Can't get info from secondary node %s" % node,
+ prereq=True, ecode=errors.ECODE_STATE)
+ (_, _, (nhvinfo, )) = nres.payload
+ if not isinstance(nhvinfo.get("memory_free", None), int):
+ raise errors.OpPrereqError("Secondary node %s didn't return free"
+ " memory information" % node,
+ errors.ECODE_STATE)
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
+ raise errors.OpPrereqError("This change will prevent the instance"
+ " from failover to its secondary node"
+ " %s, due to not enough memory" % node,
+ errors.ECODE_STATE)
+
+ if self.op.runtime_mem:
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node)
+ if not remote_info.payload: # not running already
+ raise errors.OpPrereqError("Instance %s is not running" %
+ instance.name, errors.ECODE_STATE)
+
+ current_memory = remote_info.payload["memory"]
+ if (not self.op.force and
+ (self.op.runtime_mem > self.be_proposed[constants.BE_MAXMEM] or
+ self.op.runtime_mem < self.be_proposed[constants.BE_MINMEM])):
+ raise errors.OpPrereqError("Instance %s must have memory between %d"
+ " and %d MB of memory unless --force is"
+ " given" %
+ (instance.name,
+ self.be_proposed[constants.BE_MINMEM],
+ self.be_proposed[constants.BE_MAXMEM]),
+ errors.ECODE_INVAL)
+
+ delta = self.op.runtime_mem - current_memory
+ if delta > 0:
+ CheckNodeFreeMemory(self, instance.primary_node,
+ "ballooning memory for instance %s" %
+ instance.name, delta, instance.hypervisor)
+
+ if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
+ raise errors.OpPrereqError("Disk operations not supported for"
+ " diskless instances", errors.ECODE_INVAL)
+
+ def _PrepareNicCreate(_, params, private):
+ self._PrepareNicModification(params, private, None, None,
+ {}, cluster, pnode)
+ return (None, None)
+
+ def _PrepareNicMod(_, nic, params, private):
+ self._PrepareNicModification(params, private, nic.ip, nic.network,
+ nic.nicparams, cluster, pnode)
+ return None
+
+ def _PrepareNicRemove(_, params, __):
+ ip = params.ip
+ net = params.network
+ if net is not None and ip is not None:
+ self.cfg.ReleaseIp(net, ip, self.proc.GetECId())
+
+ # Verify NIC changes (operating on copy)
+ nics = instance.nics[:]
+ _ApplyContainerMods("NIC", nics, None, self.nicmod,
+ _PrepareNicCreate, _PrepareNicMod, _PrepareNicRemove)
+ if len(nics) > constants.MAX_NICS:
+ raise errors.OpPrereqError("Instance has too many network interfaces"
+ " (%d), cannot add more" % constants.MAX_NICS,
+ errors.ECODE_STATE)
+
+ def _PrepareDiskMod(_, disk, params, __):
+ disk.name = params.get(constants.IDISK_NAME, None)
+
+ # Verify disk changes (operating on a copy)
+ disks = copy.deepcopy(instance.disks)
+ _ApplyContainerMods("disk", disks, None, self.diskmod, None,
+ _PrepareDiskMod, None)
+ utils.ValidateDeviceNames("disk", disks)
+ if len(disks) > constants.MAX_DISKS:
+ raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
+ " more" % constants.MAX_DISKS,
+ errors.ECODE_STATE)
+ disk_sizes = [disk.size for disk in instance.disks]
+ disk_sizes.extend(params["size"] for (op, idx, params, private) in
+ self.diskmod if op == constants.DDM_ADD)
+ ispec[constants.ISPEC_DISK_COUNT] = len(disk_sizes)
+ ispec[constants.ISPEC_DISK_SIZE] = disk_sizes
+
+ if self.op.offline is not None and self.op.offline:
+ CheckInstanceState(self, instance, CAN_CHANGE_INSTANCE_OFFLINE,
+ msg="can't change to offline")
+
+ # Pre-compute NIC changes (necessary to use result in hooks)
+ self._nic_chgdesc = []
+ if self.nicmod:
+ # Operate on copies as this is still in prereq
+ nics = [nic.Copy() for nic in instance.nics]
+ _ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
+ self._CreateNewNic, self._ApplyNicMods, None)
+ # Verify that NIC names are unique and valid
+ utils.ValidateDeviceNames("NIC", nics)
+ self._new_nics = nics
+ ispec[constants.ISPEC_NIC_COUNT] = len(self._new_nics)
+ else:
+ self._new_nics = None
+ ispec[constants.ISPEC_NIC_COUNT] = len(instance.nics)
+
+ if not self.op.ignore_ipolicy:
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
+ group_info)
+
+ # Fill ispec with backend parameters
+ ispec[constants.ISPEC_SPINDLE_USE] = \
+ self.be_new.get(constants.BE_SPINDLE_USE, None)
+ ispec[constants.ISPEC_CPU_COUNT] = self.be_new.get(constants.BE_VCPUS,
+ None)
+
+ # Copy ispec to verify parameters with min/max values separately
+ if self.op.disk_template:
+ new_disk_template = self.op.disk_template
+ else:
+ new_disk_template = instance.disk_template
+ ispec_max = ispec.copy()
+ ispec_max[constants.ISPEC_MEM_SIZE] = \
+ self.be_new.get(constants.BE_MAXMEM, None)
+ res_max = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_max,
+ new_disk_template)
+ ispec_min = ispec.copy()
+ ispec_min[constants.ISPEC_MEM_SIZE] = \
+ self.be_new.get(constants.BE_MINMEM, None)
+ res_min = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_min,
+ new_disk_template)
+
+ if (res_max or res_min):
+ # FIXME: Improve error message by including information about whether
+ # the upper or lower limit of the parameter fails the ipolicy.
+ msg = ("Instance allocation to group %s (%s) violates policy: %s" %
+ (group_info, group_info.name,
+ utils.CommaJoin(set(res_max + res_min))))
+ raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
+
+ def _ConvertPlainToDrbd(self, feedback_fn):
+ """Converts an instance from plain to drbd.
+
+ """
+ feedback_fn("Converting template to drbd")
+ instance = self.instance
+ pnode = instance.primary_node
+ snode = self.op.remote_node
+
+ assert instance.disk_template == constants.DT_PLAIN
+
+ # create a fake disk info for _GenerateDiskTemplate
+ disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
+ constants.IDISK_VG: d.logical_id[0],
+ constants.IDISK_NAME: d.name}
+ for d in instance.disks]
+ new_disks = GenerateDiskTemplate(self, self.op.disk_template,
+ instance.name, pnode, [snode],
+ disk_info, None, None, 0, feedback_fn,
+ self.diskparams)
+ anno_disks = rpc.AnnotateDiskParams(constants.DT_DRBD8, new_disks,
+ self.diskparams)
+ p_excl_stor = IsExclusiveStorageEnabledNodeName(self.cfg, pnode)
+ s_excl_stor = IsExclusiveStorageEnabledNodeName(self.cfg, snode)
+ info = GetInstanceInfoText(instance)
+ feedback_fn("Creating additional volumes...")
+ # first, create the missing data and meta devices
+ for disk in anno_disks:
+ # unfortunately this is... not too nice
+ CreateSingleBlockDev(self, pnode, instance, disk.children[1],
+ info, True, p_excl_stor)
+ for child in disk.children:
+ CreateSingleBlockDev(self, snode, instance, child, info, True,
+ s_excl_stor)
+ # at this stage, all new LVs have been created, we can rename the
+ # old ones
+ feedback_fn("Renaming original volumes...")
+ rename_list = [(o, n.children[0].logical_id)
+ for (o, n) in zip(instance.disks, new_disks)]
+ result = self.rpc.call_blockdev_rename(pnode, rename_list)
+ result.Raise("Failed to rename original LVs")
+
+ feedback_fn("Initializing DRBD devices...")
+ # all child devices are in place, we can now create the DRBD devices
+ try:
+ for disk in anno_disks:
+ for (node, excl_stor) in [(pnode, p_excl_stor), (snode, s_excl_stor)]:
+ f_create = node == pnode
+ CreateSingleBlockDev(self, node, instance, disk, info, f_create,
+ excl_stor)
+ except errors.GenericError, e:
+ feedback_fn("Initializing of DRBD devices failed;"
+ " renaming back original volumes...")
+ for disk in new_disks:
+ self.cfg.SetDiskID(disk, pnode)
+ rename_back_list = [(n.children[0], o.logical_id)
+ for (n, o) in zip(new_disks, instance.disks)]
+ result = self.rpc.call_blockdev_rename(pnode, rename_back_list)
+ result.Raise("Failed to rename LVs back after error %s" % str(e))
+ raise
+
+ # at this point, the instance has been modified
+ instance.disk_template = constants.DT_DRBD8
+ instance.disks = new_disks
+ self.cfg.Update(instance, feedback_fn)
+
+ # Release node locks while waiting for sync
+ ReleaseLocks(self, locking.LEVEL_NODE)
+
+ # disks are created, waiting for sync
+ disk_abort = not WaitForSync(self, instance,
+ oneshot=not self.op.wait_for_sync)
+ if disk_abort:
+ raise errors.OpExecError("There are some degraded disks for"
+ " this instance, please cleanup manually")
+
+ # Node resource locks will be released by caller
+
+ def _ConvertDrbdToPlain(self, feedback_fn):
+ """Converts an instance from drbd to plain.
+
+ """
+ instance = self.instance
+
+ assert len(instance.secondary_nodes) == 1
+ assert instance.disk_template == constants.DT_DRBD8
+
+ pnode = instance.primary_node
+ snode = instance.secondary_nodes[0]
+ feedback_fn("Converting template to plain")
+
+ old_disks = AnnotateDiskParams(instance, instance.disks, self.cfg)
+ new_disks = [d.children[0] for d in instance.disks]
+
+ # copy over size, mode and name
+ for parent, child in zip(old_disks, new_disks):
+ child.size = parent.size
+ child.mode = parent.mode
+ child.name = parent.name
+
+ # this is a DRBD disk, return its port to the pool
+ # NOTE: this must be done right before the call to cfg.Update!
+ for disk in old_disks:
+ tcp_port = disk.logical_id[2]
+ self.cfg.AddTcpUdpPort(tcp_port)
+
+ # update instance structure
+ instance.disks = new_disks
+ instance.disk_template = constants.DT_PLAIN
+ _UpdateIvNames(0, instance.disks)
+ self.cfg.Update(instance, feedback_fn)
+
+ # Release locks in case removing disks takes a while
+ ReleaseLocks(self, locking.LEVEL_NODE)
+
+ feedback_fn("Removing volumes on the secondary node...")
+ for disk in old_disks:
+ self.cfg.SetDiskID(disk, snode)
+ msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
+ if msg:
+ self.LogWarning("Could not remove block device %s on node %s,"
+ " continuing anyway: %s", disk.iv_name, snode, msg)
+
+ feedback_fn("Removing unneeded volumes on the primary node...")
+ for idx, disk in enumerate(old_disks):
+ meta = disk.children[1]
+ self.cfg.SetDiskID(meta, pnode)
+ msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
+ if msg:
+ self.LogWarning("Could not remove metadata for disk %d on node %s,"
+ " continuing anyway: %s", idx, pnode, msg)
+
+ def _CreateNewDisk(self, idx, params, _):
+ """Creates a new disk.
+
+ """
+ instance = self.instance
+
+ # add a new disk
+ if instance.disk_template in constants.DTS_FILEBASED:
+ (file_driver, file_path) = instance.disks[0].logical_id
+ file_path = os.path.dirname(file_path)
+ else:
+ file_driver = file_path = None
+
+ disk = \
+ GenerateDiskTemplate(self, instance.disk_template, instance.name,
+ instance.primary_node, instance.secondary_nodes,
+ [params], file_path, file_driver, idx,
+ self.Log, self.diskparams)[0]
+
- info = GetInstanceInfoText(instance)
-
- logging.info("Creating volume %s for instance %s",
- disk.iv_name, instance.name)
- # Note: this needs to be kept in sync with _CreateDisks
- #HARDCODE
- for node in instance.all_nodes:
- f_create = (node == instance.primary_node)
- try:
- CreateBlockDev(self, node, instance, disk, f_create, info, f_create)
- except errors.OpExecError, err:
- self.LogWarning("Failed to create volume %s (%s) on node '%s': %s",
- disk.iv_name, disk, node, err)
++ new_disks = CreateDisks(self, instance, disks=[disk])
+
+ if self.cluster.prealloc_wipe_disks:
+ # Wipe new disk
- WipeDisks(self, instance,
- disks=[(idx, disk, 0)])
++ WipeOrCleanupDisks(self, instance,
++ disks=[(idx, disk, 0)],
++ cleanup=new_disks)
+
+ return (disk, [
+ ("disk/%d" % idx, "add:size=%s,mode=%s" % (disk.size, disk.mode)),
+ ])
+
+ @staticmethod
+ def _ModifyDisk(idx, disk, params, _):
+ """Modifies a disk.
+
+ """
+ changes = []
+ mode = params.get(constants.IDISK_MODE, None)
+ if mode:
+ disk.mode = mode
+ changes.append(("disk.mode/%d" % idx, disk.mode))
+
+ name = params.get(constants.IDISK_NAME, None)
+ disk.name = name
+ changes.append(("disk.name/%d" % idx, disk.name))
+
+ return changes
+
+ def _RemoveDisk(self, idx, root, _):
+ """Removes a disk.
+
+ """
+ (anno_disk,) = AnnotateDiskParams(self.instance, [root], self.cfg)
+ for node, disk in anno_disk.ComputeNodeTree(self.instance.primary_node):
+ self.cfg.SetDiskID(disk, node)
+ msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
+ if msg:
+ self.LogWarning("Could not remove disk/%d on node '%s': %s,"
+ " continuing anyway", idx, node, msg)
+
+ # if this is a DRBD disk, return its port to the pool
+ if root.dev_type in constants.LDS_DRBD:
+ self.cfg.AddTcpUdpPort(root.logical_id[2])
+
+ def _CreateNewNic(self, idx, params, private):
+ """Creates data structure for a new network interface.
+
+ """
+ mac = params[constants.INIC_MAC]
+ ip = params.get(constants.INIC_IP, None)
+ net = params.get(constants.INIC_NETWORK, None)
+ name = params.get(constants.INIC_NAME, None)
+ net_uuid = self.cfg.LookupNetwork(net)
+ #TODO: not private.filled?? can a nic have no nicparams??
+ nicparams = private.filled
+ nobj = objects.NIC(mac=mac, ip=ip, network=net_uuid, name=name,
+ nicparams=nicparams)
+ nobj.uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
+
+ return (nobj, [
+ ("nic.%d" % idx,
+ "add:mac=%s,ip=%s,mode=%s,link=%s,network=%s" %
+ (mac, ip, private.filled[constants.NIC_MODE],
+ private.filled[constants.NIC_LINK],
+ net)),
+ ])
+
+ def _ApplyNicMods(self, idx, nic, params, private):
+ """Modifies a network interface.
+
+ """
+ changes = []
+
+ for key in [constants.INIC_MAC, constants.INIC_IP, constants.INIC_NAME]:
+ if key in params:
+ changes.append(("nic.%s/%d" % (key, idx), params[key]))
+ setattr(nic, key, params[key])
+
+ new_net = params.get(constants.INIC_NETWORK, nic.network)
+ new_net_uuid = self.cfg.LookupNetwork(new_net)
+ if new_net_uuid != nic.network:
+ changes.append(("nic.network/%d" % idx, new_net))
+ nic.network = new_net_uuid
+
+ if private.filled:
+ nic.nicparams = private.filled
+
+ for (key, val) in nic.nicparams.items():
+ changes.append(("nic.%s/%d" % (key, idx), val))
+
+ return changes
+
+ def Exec(self, feedback_fn):
+ """Modifies an instance.
+
+ All parameters take effect only at the next restart of the instance.
+
+ """
+ # Process here the warnings from CheckPrereq, as we don't have a
+ # feedback_fn there.
+ # TODO: Replace with self.LogWarning
+ for warn in self.warn:
+ feedback_fn("WARNING: %s" % warn)
+
+ assert ((self.op.disk_template is None) ^
+ bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
+ "Not owning any node resource locks"
+
+ result = []
+ instance = self.instance
+
+ # New primary node
+ if self.op.pnode:
+ instance.primary_node = self.op.pnode
+
+ # runtime memory
+ if self.op.runtime_mem:
+ rpcres = self.rpc.call_instance_balloon_memory(instance.primary_node,
+ instance,
+ self.op.runtime_mem)
+ rpcres.Raise("Cannot modify instance runtime memory")
+ result.append(("runtime_memory", self.op.runtime_mem))
+
+ # Apply disk changes
+ _ApplyContainerMods("disk", instance.disks, result, self.diskmod,
+ self._CreateNewDisk, self._ModifyDisk,
+ self._RemoveDisk)
+ _UpdateIvNames(0, instance.disks)
+
+ if self.op.disk_template:
+ if __debug__:
+ check_nodes = set(instance.all_nodes)
+ if self.op.remote_node:
+ check_nodes.add(self.op.remote_node)
+ for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
+ owned = self.owned_locks(level)
+ assert not (check_nodes - owned), \
+ ("Not owning the correct locks, owning %r, expected at least %r" %
+ (owned, check_nodes))
+
+ r_shut = ShutdownInstanceDisks(self, instance)
+ if not r_shut:
+ raise errors.OpExecError("Cannot shutdown instance disks, unable to"
+ " proceed with disk template conversion")
+ mode = (instance.disk_template, self.op.disk_template)
+ try:
+ self._DISK_CONVERSIONS[mode](self, feedback_fn)
+ except:
+ self.cfg.ReleaseDRBDMinors(instance.name)
+ raise
+ result.append(("disk_template", self.op.disk_template))
+
+ assert instance.disk_template == self.op.disk_template, \
+ ("Expected disk template '%s', found '%s'" %
+ (self.op.disk_template, instance.disk_template))
+
+ # Release node and resource locks if there are any (they might already have
+ # been released during disk conversion)
+ ReleaseLocks(self, locking.LEVEL_NODE)
+ ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
+ # Apply NIC changes
+ if self._new_nics is not None:
+ instance.nics = self._new_nics
+ result.extend(self._nic_chgdesc)
+
+ # hvparams changes
+ if self.op.hvparams:
+ instance.hvparams = self.hv_inst
+ for key, val in self.op.hvparams.iteritems():
+ result.append(("hv/%s" % key, val))
+
+ # beparams changes
+ if self.op.beparams:
+ instance.beparams = self.be_inst
+ for key, val in self.op.beparams.iteritems():
+ result.append(("be/%s" % key, val))
+
+ # OS change
+ if self.op.os_name:
+ instance.os = self.op.os_name
+
+ # osparams changes
+ if self.op.osparams:
+ instance.osparams = self.os_inst
+ for key, val in self.op.osparams.iteritems():
+ result.append(("os/%s" % key, val))
+
+ if self.op.offline is None:
+ # Ignore
+ pass
+ elif self.op.offline:
+ # Mark instance as offline
+ self.cfg.MarkInstanceOffline(instance.name)
+ result.append(("admin_state", constants.ADMINST_OFFLINE))
+ else:
+ # Mark instance as online, but stopped
+ self.cfg.MarkInstanceDown(instance.name)
+ result.append(("admin_state", constants.ADMINST_DOWN))
+
+ self.cfg.Update(instance, feedback_fn, self.proc.GetECId())
+
+ assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "All node locks should have been released by now"
+
+ return result
+
+ _DISK_CONVERSIONS = {
+ (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
+ (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
+ }
+
+
+class LUInstanceChangeGroup(LogicalUnit):
+ HPATH = "instance-change-group"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.share_locks = ShareAll()
+
+ self.needed_locks = {
+ locking.LEVEL_NODEGROUP: [],
+ locking.LEVEL_NODE: [],
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
+
+ self._ExpandAndLockInstance()
+
+ if self.op.target_groups:
+ self.req_target_uuids = map(self.cfg.LookupNodeGroup,
+ self.op.target_groups)
+ else:
+ self.req_target_uuids = None
+
+ self.op.iallocator = GetDefaultIAllocator(self.cfg, self.op.iallocator)
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+ if self.req_target_uuids:
+ lock_groups = set(self.req_target_uuids)
+
+ # Lock all groups used by instance optimistically; this requires going
+ # via the node before it's locked, requiring verification later on
+ instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+ lock_groups.update(instance_groups)
+ else:
+ # No target groups, need to lock all of them
+ lock_groups = locking.ALL_SET
+
+ self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
+
+ elif level == locking.LEVEL_NODE:
+ if self.req_target_uuids:
+ # Lock all nodes used by instances
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ self._LockInstancesNodes()
+
+ # Lock all nodes in all potential target groups
+ lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name))
+ member_nodes = [node_name
+ for group in lock_groups
+ for node_name in self.cfg.GetNodeGroup(group).members]
+ self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
+ else:
+ # Lock all nodes as all groups are potential targets
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+ def CheckPrereq(self):
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
+
+ assert (self.req_target_uuids is None or
+ owned_groups.issuperset(self.req_target_uuids))
+ assert owned_instances == set([self.op.instance_name])
+
+ # Get instance information
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
+ # Check if node groups for locked instance are still correct
+ assert owned_nodes.issuperset(self.instance.all_nodes), \
+ ("Instance %s's nodes changed while we kept the lock" %
+ self.op.instance_name)
+
+ inst_groups = CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
+ owned_groups)
+
+ if self.req_target_uuids:
+ # User requested specific target groups
+ self.target_uuids = frozenset(self.req_target_uuids)
+ else:
+ # All groups except those used by the instance are potential targets
+ self.target_uuids = owned_groups - inst_groups
+
+ conflicting_groups = self.target_uuids & inst_groups
+ if conflicting_groups:
+ raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
+ " used by the instance '%s'" %
+ (utils.CommaJoin(conflicting_groups),
+ self.op.instance_name),
+ errors.ECODE_INVAL)
+
+ if not self.target_uuids:
+ raise errors.OpPrereqError("There are no possible target groups",
+ errors.ECODE_INVAL)
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ """
+ assert self.target_uuids
+
+ env = {
+ "TARGET_GROUPS": " ".join(self.target_uuids),
+ }
+
+ env.update(BuildInstanceHookEnvByObject(self, self.instance))
+
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ mn = self.cfg.GetMasterNode()
+ return ([mn], [mn])
+
+ def Exec(self, feedback_fn):
+ instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
+
+ assert instances == [self.op.instance_name], "Instance not locked"
+
+ req = iallocator.IAReqGroupChange(instances=instances,
+ target_groups=list(self.target_uuids))
+ ial = iallocator.IAllocator(self.cfg, self.rpc, req)
+
+ ial.Run(self.op.iallocator)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute solution for changing group of"
+ " instance '%s' using iallocator '%s': %s" %
+ (self.op.instance_name, self.op.iallocator,
+ ial.info), errors.ECODE_NORES)
+
+ jobs = LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
+
+ self.LogInfo("Iallocator returned %s job(s) for changing group of"
+ " instance '%s'", len(jobs), self.op.instance_name)
+
+ return ResultWithJobs(jobs)
--- /dev/null
+#
+#
+
+# 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
+# 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.
+
+
+"""Logical units dealing with storage of instances."""
+
+import itertools
+import logging
+import os
+import time
+
+from ganeti import compat
+from ganeti import constants
+from ganeti import errors
+from ganeti import ht
+from ganeti import locking
+from ganeti.masterd import iallocator
+from ganeti import objects
+from ganeti import utils
+from ganeti import opcodes
+from ganeti import rpc
+from ganeti.cmdlib.base import LogicalUnit, NoHooksLU, Tasklet
+from ganeti.cmdlib.common import INSTANCE_DOWN, INSTANCE_NOT_RUNNING, \
+ AnnotateDiskParams, CheckIAllocatorOrNode, ExpandNodeName, \
+ CheckNodeOnline, CheckInstanceNodeGroups, CheckInstanceState, \
+ IsExclusiveStorageEnabledNode, FindFaultyInstanceDisks
+from ganeti.cmdlib.instance_utils import GetInstanceInfoText, \
+ CopyLockList, ReleaseLocks, CheckNodeVmCapable, \
+ BuildInstanceHookEnvByObject, CheckNodeNotDrained, CheckTargetNodeIPolicy
+
+import ganeti.masterd.instance
+
+
+_DISK_TEMPLATE_NAME_PREFIX = {
+ constants.DT_PLAIN: "",
+ constants.DT_RBD: ".rbd",
+ constants.DT_EXT: ".ext",
+ }
+
+
+_DISK_TEMPLATE_DEVICE_TYPE = {
+ constants.DT_PLAIN: constants.LD_LV,
+ constants.DT_FILE: constants.LD_FILE,
+ constants.DT_SHARED_FILE: constants.LD_FILE,
+ constants.DT_BLOCK: constants.LD_BLOCKDEV,
+ constants.DT_RBD: constants.LD_RBD,
+ constants.DT_EXT: constants.LD_EXT,
+ }
+
+
+def CreateSingleBlockDev(lu, node, instance, device, info, force_open,
+ excl_stor):
+ """Create a single block device on a given node.
+
+ This will not recurse over children of the device, so they must be
+ created in advance.
+
+ @param lu: the lu on whose behalf we execute
+ @param node: the node on which to create the device
+ @type instance: L{objects.Instance}
+ @param instance: the instance which owns the device
+ @type device: L{objects.Disk}
+ @param device: the device to create
+ @param info: the extra 'metadata' we should attach to the device
+ (this will be represented as a LVM tag)
+ @type force_open: boolean
+ @param force_open: this parameter will be passes to the
+ L{backend.BlockdevCreate} function where it specifies
+ whether we run on primary or not, and it affects both
+ the child assembly and the device own Open() execution
+ @type excl_stor: boolean
+ @param excl_stor: Whether exclusive_storage is active for the node
+
+ """
+ lu.cfg.SetDiskID(device, node)
+ result = lu.rpc.call_blockdev_create(node, device, device.size,
+ instance.name, force_open, info,
+ excl_stor)
+ result.Raise("Can't create block device %s on"
+ " node %s for instance %s" % (device, node, instance.name))
+ if device.physical_id is None:
+ device.physical_id = result.payload
+
+
+def _CreateBlockDevInner(lu, node, instance, device, force_create,
+ info, force_open, excl_stor):
+ """Create a tree of block devices on a given node.
+
+ If this device type has to be created on secondaries, create it and
+ all its children.
+
+ If not, just recurse to children keeping the same 'force' value.
+
+ @attention: The device has to be annotated already.
+
+ @param lu: the lu on whose behalf we execute
+ @param node: the node on which to create the device
+ @type instance: L{objects.Instance}
+ @param instance: the instance which owns the device
+ @type device: L{objects.Disk}
+ @param device: the device to create
+ @type force_create: boolean
+ @param force_create: whether to force creation of this device; this
+ will be change to True whenever we find a device which has
+ CreateOnSecondary() attribute
+ @param info: the extra 'metadata' we should attach to the device
+ (this will be represented as a LVM tag)
+ @type force_open: boolean
+ @param force_open: this parameter will be passes to the
+ L{backend.BlockdevCreate} function where it specifies
+ whether we run on primary or not, and it affects both
+ the child assembly and the device own Open() execution
+ @type excl_stor: boolean
+ @param excl_stor: Whether exclusive_storage is active for the node
+
+ @return: list of created devices
+ """
+ created_devices = []
+ try:
+ if device.CreateOnSecondary():
+ force_create = True
+
+ if device.children:
+ for child in device.children:
+ devs = _CreateBlockDevInner(lu, node, instance, child, force_create,
+ info, force_open, excl_stor)
+ created_devices.extend(devs)
+
+ if not force_create:
+ return created_devices
+
+ CreateSingleBlockDev(lu, node, instance, device, info, force_open,
+ excl_stor)
+ # The device has been completely created, so there is no point in keeping
+ # its subdevices in the list. We just add the device itself instead.
+ created_devices = [(node, device)]
+ return created_devices
+
+ except errors.DeviceCreationError, e:
+ e.created_devices.extend(created_devices)
+ raise e
+ except errors.OpExecError, e:
+ raise errors.DeviceCreationError(str(e), created_devices)
+
+
+def IsExclusiveStorageEnabledNodeName(cfg, nodename):
+ """Whether exclusive_storage is in effect for the given node.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: The cluster configuration
+ @type nodename: string
+ @param nodename: The node
+ @rtype: bool
+ @return: The effective value of exclusive_storage
+ @raise errors.OpPrereqError: if no node exists with the given name
+
+ """
+ ni = cfg.GetNodeInfo(nodename)
+ if ni is None:
+ raise errors.OpPrereqError("Invalid node name %s" % nodename,
+ errors.ECODE_NOENT)
+ return IsExclusiveStorageEnabledNode(cfg, ni)
+
+
- def CreateBlockDev(lu, node, instance, device, force_create, info,
++def _CreateBlockDev(lu, node, instance, device, force_create, info,
+ force_open):
+ """Wrapper around L{_CreateBlockDevInner}.
+
+ This method annotates the root device first.
+
+ """
+ (disk,) = AnnotateDiskParams(instance, [device], lu.cfg)
+ excl_stor = IsExclusiveStorageEnabledNodeName(lu.cfg, node)
+ return _CreateBlockDevInner(lu, node, instance, disk, force_create, info,
+ force_open, excl_stor)
+
+
- def CreateDisks(lu, instance, to_skip=None, target_node=None):
++def _UndoCreateDisks(lu, disks_created):
++ """Undo the work performed by L{CreateDisks}.
++
++ This function is called in case of an error to undo the work of
++ L{CreateDisks}.
++
++ @type lu: L{LogicalUnit}
++ @param lu: the logical unit on whose behalf we execute
++ @param disks_created: the result returned by L{CreateDisks}
++
++ """
++ for (node, disk) in disks_created:
++ lu.cfg.SetDiskID(disk, node)
++ result = lu.rpc.call_blockdev_remove(node, disk)
++ if result.fail_msg:
++ logging.warning("Failed to remove newly-created disk %s on node %s:"
++ " %s", disk, node, result.fail_msg)
++
++
++def CreateDisks(lu, instance, to_skip=None, target_node=None, disks=None):
+ """Create all disks for an instance.
+
+ This abstracts away some work from AddInstance.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance whose disks we should create
+ @type to_skip: list
+ @param to_skip: list of indices to skip
+ @type target_node: string
+ @param target_node: if passed, overrides the target node for creation
- @rtype: boolean
- @return: the success of the creation
++ @type disks: list of {objects.Disk}
++ @param disks: the disks to create; if not specified, all the disks of the
++ instance are created
++ @return: information about the created disks, to be used to call
++ L{_UndoCreateDisks}
++ @raise errors.OpPrereqError: in case of error
+
+ """
+ info = GetInstanceInfoText(instance)
+ if target_node is None:
+ pnode = instance.primary_node
+ all_nodes = instance.all_nodes
+ else:
+ pnode = target_node
+ all_nodes = [pnode]
+
++ if disks is None:
++ disks = instance.disks
++
+ if instance.disk_template in constants.DTS_FILEBASED:
+ file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
+ result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
+
+ result.Raise("Failed to create directory '%s' on"
+ " node %s" % (file_storage_dir, pnode))
+
+ disks_created = []
- # Note: this needs to be kept in sync with adding of disks in
- # LUInstanceSetParams
- for idx, device in enumerate(instance.disks):
++ for idx, device in enumerate(disks):
+ if to_skip and idx in to_skip:
+ continue
+ logging.info("Creating disk %s for instance '%s'", idx, instance.name)
- #HARDCODE
+ for node in all_nodes:
+ f_create = node == pnode
+ try:
- CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
++ _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
+ disks_created.append((node, device))
- except errors.OpExecError:
- logging.warning("Creating disk %s for instance '%s' failed",
- idx, instance.name)
+ except errors.DeviceCreationError, e:
+ logging.warning("Creating disk %s for instance '%s' failed",
+ idx, instance.name)
+ disks_created.extend(e.created_devices)
- for (node, disk) in disks_created:
- lu.cfg.SetDiskID(disk, node)
- result = lu.rpc.call_blockdev_remove(node, disk)
- if result.fail_msg:
- logging.warning("Failed to remove newly-created disk %s on node %s:"
- " %s", device, node, result.fail_msg)
++ _UndoCreateDisks(lu, disks_created)
+ raise errors.OpExecError(e.message)
++ return disks_created
+
+
+def ComputeDiskSizePerVG(disk_template, disks):
+ """Compute disk size requirements in the volume group
+
+ """
+ def _compute(disks, payload):
+ """Universal algorithm.
+
+ """
+ vgs = {}
+ for disk in disks:
+ vgs[disk[constants.IDISK_VG]] = \
+ vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
+
+ return vgs
+
+ # Required free disk space as a function of disk and swap space
+ req_size_dict = {
+ constants.DT_DISKLESS: {},
+ constants.DT_PLAIN: _compute(disks, 0),
+ # 128 MB are added for drbd metadata for each disk
+ constants.DT_DRBD8: _compute(disks, constants.DRBD_META_SIZE),
+ constants.DT_FILE: {},
+ constants.DT_SHARED_FILE: {},
+ }
+
+ if disk_template not in req_size_dict:
+ raise errors.ProgrammerError("Disk template '%s' size requirement"
+ " is unknown" % disk_template)
+
+ return req_size_dict[disk_template]
+
+
+def ComputeDisks(op, default_vg):
+ """Computes the instance disks.
+
+ @param op: The instance opcode
+ @param default_vg: The default_vg to assume
+
+ @return: The computed disks
+
+ """
+ disks = []
+ for disk in op.disks:
+ mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
+ if mode not in constants.DISK_ACCESS_SET:
+ raise errors.OpPrereqError("Invalid disk access mode '%s'" %
+ mode, errors.ECODE_INVAL)
+ size = disk.get(constants.IDISK_SIZE, None)
+ if size is None:
+ raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
+ try:
+ size = int(size)
+ except (TypeError, ValueError):
+ raise errors.OpPrereqError("Invalid disk size '%s'" % size,
+ errors.ECODE_INVAL)
+
+ ext_provider = disk.get(constants.IDISK_PROVIDER, None)
+ if ext_provider and op.disk_template != constants.DT_EXT:
+ raise errors.OpPrereqError("The '%s' option is only valid for the %s"
+ " disk template, not %s" %
+ (constants.IDISK_PROVIDER, constants.DT_EXT,
+ op.disk_template), errors.ECODE_INVAL)
+
+ data_vg = disk.get(constants.IDISK_VG, default_vg)
+ name = disk.get(constants.IDISK_NAME, None)
+ if name is not None and name.lower() == constants.VALUE_NONE:
+ name = None
+ new_disk = {
+ constants.IDISK_SIZE: size,
+ constants.IDISK_MODE: mode,
+ constants.IDISK_VG: data_vg,
+ constants.IDISK_NAME: name,
+ }
+
+ if constants.IDISK_METAVG in disk:
+ new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG]
+ if constants.IDISK_ADOPT in disk:
+ new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
+
+ # For extstorage, demand the `provider' option and add any
+ # additional parameters (ext-params) to the dict
+ if op.disk_template == constants.DT_EXT:
+ if ext_provider:
+ new_disk[constants.IDISK_PROVIDER] = ext_provider
+ for key in disk:
+ if key not in constants.IDISK_PARAMS:
+ new_disk[key] = disk[key]
+ else:
+ raise errors.OpPrereqError("Missing provider for template '%s'" %
+ constants.DT_EXT, errors.ECODE_INVAL)
+
+ disks.append(new_disk)
+
+ return disks
+
+
+def CheckRADOSFreeSpace():
+ """Compute disk size requirements inside the RADOS cluster.
+
+ """
+ # For the RADOS cluster we assume there is always enough space.
+ pass
+
+
+def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
+ iv_name, p_minor, s_minor):
+ """Generate a drbd8 device complete with its children.
+
+ """
+ assert len(vgnames) == len(names) == 2
+ port = lu.cfg.AllocatePort()
+ shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
+
+ dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
+ logical_id=(vgnames[0], names[0]),
+ params={})
+ dev_data.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
+ dev_meta = objects.Disk(dev_type=constants.LD_LV,
+ size=constants.DRBD_META_SIZE,
+ logical_id=(vgnames[1], names[1]),
+ params={})
+ dev_meta.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
+ drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
+ logical_id=(primary, secondary, port,
+ p_minor, s_minor,
+ shared_secret),
+ children=[dev_data, dev_meta],
+ iv_name=iv_name, params={})
+ drbd_dev.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
+ return drbd_dev
+
+
+def GenerateDiskTemplate(
+ lu, template_name, instance_name, primary_node, secondary_nodes,
+ disk_info, file_storage_dir, file_driver, base_index,
+ feedback_fn, full_disk_params, _req_file_storage=opcodes.RequireFileStorage,
+ _req_shr_file_storage=opcodes.RequireSharedFileStorage):
+ """Generate the entire disk layout for a given template type.
+
+ """
+ vgname = lu.cfg.GetVGName()
+ disk_count = len(disk_info)
+ disks = []
+
+ if template_name == constants.DT_DISKLESS:
+ pass
+ elif template_name == constants.DT_DRBD8:
+ if len(secondary_nodes) != 1:
+ raise errors.ProgrammerError("Wrong template configuration")
+ remote_node = secondary_nodes[0]
+ minors = lu.cfg.AllocateDRBDMinor(
+ [primary_node, remote_node] * len(disk_info), instance_name)
+
+ (drbd_params, _, _) = objects.Disk.ComputeLDParams(template_name,
+ full_disk_params)
+ drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
+
+ names = []
+ for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
+ for i in range(disk_count)]):
+ names.append(lv_prefix + "_data")
+ names.append(lv_prefix + "_meta")
+ for idx, disk in enumerate(disk_info):
+ disk_index = idx + base_index
+ data_vg = disk.get(constants.IDISK_VG, vgname)
+ meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
+ disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
+ disk[constants.IDISK_SIZE],
+ [data_vg, meta_vg],
+ names[idx * 2:idx * 2 + 2],
+ "disk/%d" % disk_index,
+ minors[idx * 2], minors[idx * 2 + 1])
+ disk_dev.mode = disk[constants.IDISK_MODE]
+ disk_dev.name = disk.get(constants.IDISK_NAME, None)
+ disks.append(disk_dev)
+ else:
+ if secondary_nodes:
+ raise errors.ProgrammerError("Wrong template configuration")
+
+ if template_name == constants.DT_FILE:
+ _req_file_storage()
+ elif template_name == constants.DT_SHARED_FILE:
+ _req_shr_file_storage()
+
+ name_prefix = _DISK_TEMPLATE_NAME_PREFIX.get(template_name, None)
+ if name_prefix is None:
+ names = None
+ else:
+ names = _GenerateUniqueNames(lu, ["%s.disk%s" %
+ (name_prefix, base_index + i)
+ for i in range(disk_count)])
+
+ if template_name == constants.DT_PLAIN:
+
+ def logical_id_fn(idx, _, disk):
+ vg = disk.get(constants.IDISK_VG, vgname)
+ return (vg, names[idx])
+
+ elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
+ logical_id_fn = \
+ lambda _, disk_index, disk: (file_driver,
+ "%s/disk%d" % (file_storage_dir,
+ disk_index))
+ elif template_name == constants.DT_BLOCK:
+ logical_id_fn = \
+ lambda idx, disk_index, disk: (constants.BLOCKDEV_DRIVER_MANUAL,
+ disk[constants.IDISK_ADOPT])
+ elif template_name == constants.DT_RBD:
+ logical_id_fn = lambda idx, _, disk: ("rbd", names[idx])
+ elif template_name == constants.DT_EXT:
+ def logical_id_fn(idx, _, disk):
+ provider = disk.get(constants.IDISK_PROVIDER, None)
+ if provider is None:
+ raise errors.ProgrammerError("Disk template is %s, but '%s' is"
+ " not found", constants.DT_EXT,
+ constants.IDISK_PROVIDER)
+ return (provider, names[idx])
+ else:
+ raise errors.ProgrammerError("Unknown disk template '%s'" % template_name)
+
+ dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name]
+
+ for idx, disk in enumerate(disk_info):
+ params = {}
+ # Only for the Ext template add disk_info to params
+ if template_name == constants.DT_EXT:
+ params[constants.IDISK_PROVIDER] = disk[constants.IDISK_PROVIDER]
+ for key in disk:
+ if key not in constants.IDISK_PARAMS:
+ params[key] = disk[key]
+ disk_index = idx + base_index
+ size = disk[constants.IDISK_SIZE]
+ feedback_fn("* disk %s, size %s" %
+ (disk_index, utils.FormatUnit(size, "h")))
+ disk_dev = objects.Disk(dev_type=dev_type, size=size,
+ logical_id=logical_id_fn(idx, disk_index, disk),
+ iv_name="disk/%d" % disk_index,
+ mode=disk[constants.IDISK_MODE],
+ params=params)
+ disk_dev.name = disk.get(constants.IDISK_NAME, None)
+ disk_dev.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
+ disks.append(disk_dev)
+
+ return disks
+
+
+class LUInstanceRecreateDisks(LogicalUnit):
+ """Recreate an instance's missing disks.
+
+ """
+ HPATH = "instance-recreate-disks"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ _MODIFYABLE = compat.UniqueFrozenset([
+ constants.IDISK_SIZE,
+ constants.IDISK_MODE,
+ ])
+
+ # New or changed disk parameters may have different semantics
+ assert constants.IDISK_PARAMS == (_MODIFYABLE | frozenset([
+ constants.IDISK_ADOPT,
+
+ # TODO: Implement support changing VG while recreating
+ constants.IDISK_VG,
+ constants.IDISK_METAVG,
+ constants.IDISK_PROVIDER,
+ constants.IDISK_NAME,
+ ]))
+
+ def _RunAllocator(self):
+ """Run the allocator based on input opcode.
+
+ """
+ be_full = self.cfg.GetClusterInfo().FillBE(self.instance)
+
+ # FIXME
+ # The allocator should actually run in "relocate" mode, but current
+ # allocators don't support relocating all the nodes of an instance at
+ # the same time. As a workaround we use "allocate" mode, but this is
+ # suboptimal for two reasons:
+ # - The instance name passed to the allocator is present in the list of
+ # existing instances, so there could be a conflict within the
+ # internal structures of the allocator. This doesn't happen with the
+ # current allocators, but it's a liability.
+ # - The allocator counts the resources used by the instance twice: once
+ # because the instance exists already, and once because it tries to
+ # allocate a new instance.
+ # The allocator could choose some of the nodes on which the instance is
+ # running, but that's not a problem. If the instance nodes are broken,
+ # they should be already be marked as drained or offline, and hence
+ # skipped by the allocator. If instance disks have been lost for other
+ # reasons, then recreating the disks on the same nodes should be fine.
+ disk_template = self.instance.disk_template
+ spindle_use = be_full[constants.BE_SPINDLE_USE]
+ req = iallocator.IAReqInstanceAlloc(name=self.op.instance_name,
+ disk_template=disk_template,
+ tags=list(self.instance.GetTags()),
+ os=self.instance.os,
+ nics=[{}],
+ vcpus=be_full[constants.BE_VCPUS],
+ memory=be_full[constants.BE_MAXMEM],
+ spindle_use=spindle_use,
+ disks=[{constants.IDISK_SIZE: d.size,
+ constants.IDISK_MODE: d.mode}
+ for d in self.instance.disks],
+ hypervisor=self.instance.hypervisor,
+ node_whitelist=None)
+ ial = iallocator.IAllocator(self.cfg, self.rpc, req)
+
+ ial.Run(self.op.iallocator)
+
+ assert req.RequiredNodes() == len(self.instance.all_nodes)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
+ " %s" % (self.op.iallocator, ial.info),
+ errors.ECODE_NORES)
+
+ self.op.nodes = ial.result
+ self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
+ self.op.instance_name, self.op.iallocator,
+ utils.CommaJoin(ial.result))
+
+ def CheckArguments(self):
+ if self.op.disks and ht.TNonNegativeInt(self.op.disks[0]):
+ # Normalize and convert deprecated list of disk indices
+ self.op.disks = [(idx, {}) for idx in sorted(frozenset(self.op.disks))]
+
+ duplicates = utils.FindDuplicates(map(compat.fst, self.op.disks))
+ if duplicates:
+ raise errors.OpPrereqError("Some disks have been specified more than"
+ " once: %s" % utils.CommaJoin(duplicates),
+ errors.ECODE_INVAL)
+
+ # We don't want _CheckIAllocatorOrNode selecting the default iallocator
+ # when neither iallocator nor nodes are specified
+ if self.op.iallocator or self.op.nodes:
+ CheckIAllocatorOrNode(self, "iallocator", "nodes")
+
+ for (idx, params) in self.op.disks:
+ utils.ForceDictType(params, constants.IDISK_PARAMS_TYPES)
+ unsupported = frozenset(params.keys()) - self._MODIFYABLE
+ if unsupported:
+ raise errors.OpPrereqError("Parameters for disk %s try to change"
+ " unmodifyable parameter(s): %s" %
+ (idx, utils.CommaJoin(unsupported)),
+ errors.ECODE_INVAL)
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
+ if self.op.nodes:
+ self.op.nodes = [ExpandNodeName(self.cfg, n) for n in self.op.nodes]
+ self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
+ else:
+ self.needed_locks[locking.LEVEL_NODE] = []
+ if self.op.iallocator:
+ # iallocator will select a new node in the same group
+ self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert self.op.iallocator is not None
+ assert not self.op.nodes
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+ self.share_locks[locking.LEVEL_NODEGROUP] = 1
+ # Lock the primary group used by the instance optimistically; this
+ # requires going via the node before it's locked, requiring
+ # verification later on
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name, primary_only=True)
+
+ elif level == locking.LEVEL_NODE:
+ # If an allocator is used, then we lock all the nodes in the current
+ # instance group, as we don't know yet which ones will be selected;
+ # if we replace the nodes without using an allocator, locks are
+ # already declared in ExpandNames; otherwise, we need to lock all the
+ # instance nodes for disk re-creation
+ if self.op.iallocator:
+ assert not self.op.nodes
+ assert not self.needed_locks[locking.LEVEL_NODE]
+ assert len(self.owned_locks(locking.LEVEL_NODEGROUP)) == 1
+
+ # Lock member nodes of the group of the primary node
+ for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP):
+ self.needed_locks[locking.LEVEL_NODE].extend(
+ self.cfg.GetNodeGroup(group_uuid).members)
+
+ assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
+ elif not self.op.nodes:
+ self._LockInstancesNodes(primary_only=False)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ return BuildInstanceHookEnvByObject(self, self.instance)
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
+ return (nl, nl)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster and is not running.
+
+ """
+ instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+ if self.op.nodes:
+ if len(self.op.nodes) != len(instance.all_nodes):
+ raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
+ " %d replacement nodes were specified" %
+ (instance.name, len(instance.all_nodes),
+ len(self.op.nodes)),
+ errors.ECODE_INVAL)
+ assert instance.disk_template != constants.DT_DRBD8 or \
+ len(self.op.nodes) == 2
+ assert instance.disk_template != constants.DT_PLAIN or \
+ len(self.op.nodes) == 1
+ primary_node = self.op.nodes[0]
+ else:
+ primary_node = instance.primary_node
+ if not self.op.iallocator:
+ CheckNodeOnline(self, primary_node)
+
+ if instance.disk_template == constants.DT_DISKLESS:
+ raise errors.OpPrereqError("Instance '%s' has no disks" %
+ self.op.instance_name, errors.ECODE_INVAL)
+
+ # Verify if node group locks are still correct
+ owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
+ if owned_groups:
+ # Node group locks are acquired only for the primary node (and only
+ # when the allocator is used)
+ CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups,
+ primary_only=True)
+
+ # if we replace nodes *and* the old primary is offline, we don't
+ # check the instance state
+ old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
+ if not ((self.op.iallocator or self.op.nodes) and old_pnode.offline):
+ CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+ msg="cannot recreate disks")
+
+ if self.op.disks:
+ self.disks = dict(self.op.disks)
+ else:
+ self.disks = dict((idx, {}) for idx in range(len(instance.disks)))
+
+ maxidx = max(self.disks.keys())
+ if maxidx >= len(instance.disks):
+ raise errors.OpPrereqError("Invalid disk index '%s'" % maxidx,
+ errors.ECODE_INVAL)
+
+ if ((self.op.nodes or self.op.iallocator) and
+ sorted(self.disks.keys()) != range(len(instance.disks))):
+ raise errors.OpPrereqError("Can't recreate disks partially and"
+ " change the nodes at the same time",
+ errors.ECODE_INVAL)
+
+ self.instance = instance
+
+ if self.op.iallocator:
+ self._RunAllocator()
+ # Release unneeded node and node resource locks
+ ReleaseLocks(self, locking.LEVEL_NODE, keep=self.op.nodes)
+ ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=self.op.nodes)
+ ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
+ def Exec(self, feedback_fn):
+ """Recreate the disks.
+
+ """
+ instance = self.instance
+
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
+ to_skip = []
+ mods = [] # keeps track of needed changes
+
+ for idx, disk in enumerate(instance.disks):
+ try:
+ changes = self.disks[idx]
+ except KeyError:
+ # Disk should not be recreated
+ to_skip.append(idx)
+ continue
+
+ # update secondaries for disks, if needed
+ if self.op.nodes and disk.dev_type == constants.LD_DRBD8:
+ # need to update the nodes and minors
+ assert len(self.op.nodes) == 2
+ assert len(disk.logical_id) == 6 # otherwise disk internals
+ # have changed
+ (_, _, old_port, _, _, old_secret) = disk.logical_id
+ new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
+ new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
+ new_minors[0], new_minors[1], old_secret)
+ assert len(disk.logical_id) == len(new_id)
+ else:
+ new_id = None
+
+ mods.append((idx, new_id, changes))
+
+ # now that we have passed all asserts above, we can apply the mods
+ # in a single run (to avoid partial changes)
+ for idx, new_id, changes in mods:
+ disk = instance.disks[idx]
+ if new_id is not None:
+ assert disk.dev_type == constants.LD_DRBD8
+ disk.logical_id = new_id
+ if changes:
+ disk.Update(size=changes.get(constants.IDISK_SIZE, None),
+ mode=changes.get(constants.IDISK_MODE, None))
+
+ # change primary node, if needed
+ if self.op.nodes:
+ instance.primary_node = self.op.nodes[0]
+ self.LogWarning("Changing the instance's nodes, you will have to"
+ " remove any disks left on the older nodes manually")
+
+ if self.op.nodes:
+ self.cfg.Update(instance, feedback_fn)
+
+ # All touched nodes must be locked
+ mylocks = self.owned_locks(locking.LEVEL_NODE)
+ assert mylocks.issuperset(frozenset(instance.all_nodes))
- CreateDisks(self, instance, to_skip=to_skip)
++ new_disks = CreateDisks(self, instance, to_skip=to_skip)
++
++ # TODO: Release node locks before wiping, or explain why it's not possible
++ if self.cfg.GetClusterInfo().prealloc_wipe_disks:
++ wipedisks = [(idx, disk, 0)
++ for (idx, disk) in enumerate(instance.disks)
++ if idx not in to_skip]
++ WipeOrCleanupDisks(self, instance, disks=wipedisks, cleanup=new_disks)
+
+
+def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
+ """Checks if nodes have enough free disk space in the specified VG.
+
+ This function checks if all given nodes have the needed amount of
+ free disk. In case any node has less disk or we cannot get the
+ information from the node, this function raises an OpPrereqError
+ exception.
+
+ @type lu: C{LogicalUnit}
+ @param lu: a logical unit from which we get configuration data
+ @type nodenames: C{list}
+ @param nodenames: the list of node names to check
+ @type vg: C{str}
+ @param vg: the volume group to check
+ @type requested: C{int}
+ @param requested: the amount of disk in MiB to check for
+ @raise errors.OpPrereqError: if the node doesn't have enough disk,
+ or we cannot check the node
+
+ """
+ es_flags = rpc.GetExclusiveStorageForNodeNames(lu.cfg, nodenames)
+ nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None, es_flags)
+ for node in nodenames:
+ info = nodeinfo[node]
+ info.Raise("Cannot get current information from node %s" % node,
+ prereq=True, ecode=errors.ECODE_ENVIRON)
+ (_, (vg_info, ), _) = info.payload
+ vg_free = vg_info.get("vg_free", None)
+ if not isinstance(vg_free, int):
+ raise errors.OpPrereqError("Can't compute free disk space on node"
+ " %s for vg %s, result was '%s'" %
+ (node, vg, vg_free), errors.ECODE_ENVIRON)
+ if requested > vg_free:
+ raise errors.OpPrereqError("Not enough disk space on target node %s"
+ " vg %s: required %d MiB, available %d MiB" %
+ (node, vg, requested, vg_free),
+ errors.ECODE_NORES)
+
+
+def CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
+ """Checks if nodes have enough free disk space in all the VGs.
+
+ This function checks if all given nodes have the needed amount of
+ free disk. In case any node has less disk or we cannot get the
+ information from the node, this function raises an OpPrereqError
+ exception.
+
+ @type lu: C{LogicalUnit}
+ @param lu: a logical unit from which we get configuration data
+ @type nodenames: C{list}
+ @param nodenames: the list of node names to check
+ @type req_sizes: C{dict}
+ @param req_sizes: the hash of vg and corresponding amount of disk in
+ MiB to check for
+ @raise errors.OpPrereqError: if the node doesn't have enough disk,
+ or we cannot check the node
+
+ """
+ for vg, req_size in req_sizes.items():
+ _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
+
+
+def _DiskSizeInBytesToMebibytes(lu, size):
+ """Converts a disk size in bytes to mebibytes.
+
+ Warns and rounds up if the size isn't an even multiple of 1 MiB.
+
+ """
+ (mib, remainder) = divmod(size, 1024 * 1024)
+
+ if remainder != 0:
+ lu.LogWarning("Disk size is not an even multiple of 1 MiB; rounding up"
+ " to not overwrite existing data (%s bytes will not be"
+ " wiped)", (1024 * 1024) - remainder)
+ mib += 1
+
+ return mib
+
+
+def _CalcEta(time_taken, written, total_size):
+ """Calculates the ETA based on size written and total size.
+
+ @param time_taken: The time taken so far
+ @param written: amount written so far
+ @param total_size: The total size of data to be written
+ @return: The remaining time in seconds
+
+ """
+ avg_time = time_taken / float(written)
+ return (total_size - written) * avg_time
+
+
+def WipeDisks(lu, instance, disks=None):
+ """Wipes instance disks.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance whose disks we should create
+ @type disks: None or list of tuple of (number, L{objects.Disk}, number)
+ @param disks: Disk details; tuple contains disk index, disk object and the
+ start offset
+
+ """
+ node = instance.primary_node
+
+ if disks is None:
+ disks = [(idx, disk, 0)
+ for (idx, disk) in enumerate(instance.disks)]
+
+ for (_, device, _) in disks:
+ lu.cfg.SetDiskID(device, node)
+
+ logging.info("Pausing synchronization of disks of instance '%s'",
+ instance.name)
+ result = lu.rpc.call_blockdev_pause_resume_sync(node,
+ (map(compat.snd, disks),
+ instance),
+ True)
+ result.Raise("Failed to pause disk synchronization on node '%s'" % node)
+
+ for idx, success in enumerate(result.payload):
+ if not success:
+ logging.warn("Pausing synchronization of disk %s of instance '%s'"
+ " failed", idx, instance.name)
+
+ try:
+ for (idx, device, offset) in disks:
+ # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
+ # MAX_WIPE_CHUNK at max. Truncating to integer to avoid rounding errors.
+ wipe_chunk_size = \
+ int(min(constants.MAX_WIPE_CHUNK,
+ device.size / 100.0 * constants.MIN_WIPE_CHUNK_PERCENT))
+
+ size = device.size
+ last_output = 0
+ start_time = time.time()
+
+ if offset == 0:
+ info_text = ""
+ else:
+ info_text = (" (from %s to %s)" %
+ (utils.FormatUnit(offset, "h"),
+ utils.FormatUnit(size, "h")))
+
+ lu.LogInfo("* Wiping disk %s%s", idx, info_text)
+
+ logging.info("Wiping disk %d for instance %s on node %s using"
+ " chunk size %s", idx, instance.name, node, wipe_chunk_size)
+
+ while offset < size:
+ wipe_size = min(wipe_chunk_size, size - offset)
+
+ logging.debug("Wiping disk %d, offset %s, chunk %s",
+ idx, offset, wipe_size)
+
+ result = lu.rpc.call_blockdev_wipe(node, (device, instance), offset,
+ wipe_size)
+ result.Raise("Could not wipe disk %d at offset %d for size %d" %
+ (idx, offset, wipe_size))
+
+ now = time.time()
+ offset += wipe_size
+ if now - last_output >= 60:
+ eta = _CalcEta(now - start_time, offset, size)
+ lu.LogInfo(" - done: %.1f%% ETA: %s",
+ offset / float(size) * 100, utils.FormatSeconds(eta))
+ last_output = now
+ finally:
+ logging.info("Resuming synchronization of disks for instance '%s'",
+ instance.name)
+
+ result = lu.rpc.call_blockdev_pause_resume_sync(node,
+ (map(compat.snd, disks),
+ instance),
+ False)
+
+ if result.fail_msg:
+ lu.LogWarning("Failed to resume disk synchronization on node '%s': %s",
+ node, result.fail_msg)
+ else:
+ for idx, success in enumerate(result.payload):
+ if not success:
+ lu.LogWarning("Resuming synchronization of disk %s of instance '%s'"
+ " failed", idx, instance.name)
+
+
++def WipeOrCleanupDisks(lu, instance, disks=None, cleanup=None):
++ """Wrapper for L{WipeDisks} that handles errors.
++
++ @type lu: L{LogicalUnit}
++ @param lu: the logical unit on whose behalf we execute
++ @type instance: L{objects.Instance}
++ @param instance: the instance whose disks we should wipe
++ @param disks: see L{WipeDisks}
++ @param cleanup: the result returned by L{CreateDisks}, used for cleanup in
++ case of error
++ @raise errors.OpPrereqError: in case of failure
++
++ """
++ try:
++ WipeDisks(lu, instance, disks=disks)
++ except errors.OpExecError:
++ logging.warning("Wiping disks for instance '%s' failed",
++ instance.name)
++ _UndoCreateDisks(lu, cleanup)
++ raise
++
++
+def ExpandCheckDisks(instance, disks):
+ """Return the instance disks selected by the disks list
+
+ @type disks: list of L{objects.Disk} or None
+ @param disks: selected disks
+ @rtype: list of L{objects.Disk}
+ @return: selected instance disks to act on
+
+ """
+ if disks is None:
+ return instance.disks
+ else:
+ if not set(disks).issubset(instance.disks):
+ raise errors.ProgrammerError("Can only act on disks belonging to the"
+ " target instance")
+ return disks
+
+
+def WaitForSync(lu, instance, disks=None, oneshot=False):
+ """Sleep and poll for an instance's disk to sync.
+
+ """
+ if not instance.disks or disks is not None and not disks:
+ return True
+
+ disks = ExpandCheckDisks(instance, disks)
+
+ if not oneshot:
+ lu.LogInfo("Waiting for instance %s to sync disks", instance.name)
+
+ node = instance.primary_node
+
+ for dev in disks:
+ lu.cfg.SetDiskID(dev, node)
+
+ # TODO: Convert to utils.Retry
+
+ retries = 0
+ degr_retries = 10 # in seconds, as we sleep 1 second each time
+ while True:
+ max_time = 0
+ done = True
+ cumul_degraded = False
+ rstats = lu.rpc.call_blockdev_getmirrorstatus(node, (disks, instance))
+ msg = rstats.fail_msg
+ if msg:
+ lu.LogWarning("Can't get any data from node %s: %s", node, msg)
+ retries += 1
+ if retries >= 10:
+ raise errors.RemoteError("Can't contact node %s for mirror data,"
+ " aborting." % node)
+ time.sleep(6)
+ continue
+ rstats = rstats.payload
+ retries = 0
+ for i, mstat in enumerate(rstats):
+ if mstat is None:
+ lu.LogWarning("Can't compute data for node %s/%s",
+ node, disks[i].iv_name)
+ continue
+
+ cumul_degraded = (cumul_degraded or
+ (mstat.is_degraded and mstat.sync_percent is None))
+ if mstat.sync_percent is not None:
+ done = False
+ if mstat.estimated_time is not None:
+ rem_time = ("%s remaining (estimated)" %
+ utils.FormatSeconds(mstat.estimated_time))
+ max_time = mstat.estimated_time
+ else:
+ rem_time = "no time estimate"
+ lu.LogInfo("- device %s: %5.2f%% done, %s",
+ disks[i].iv_name, mstat.sync_percent, rem_time)
+
+ # if we're done but degraded, let's do a few small retries, to
+ # make sure we see a stable and not transient situation; therefore
+ # we force restart of the loop
+ if (done or oneshot) and cumul_degraded and degr_retries > 0:
+ logging.info("Degraded disks found, %d retries left", degr_retries)
+ degr_retries -= 1
+ time.sleep(1)
+ continue
+
+ if done or oneshot:
+ break
+
+ time.sleep(min(60, max_time))
+
+ if done:
+ lu.LogInfo("Instance %s's disks are in sync", instance.name)
+
+ return not cumul_degraded
+
+
+def ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
+ """Shutdown block devices of an instance.
+
+ This does the shutdown on all nodes of the instance.
+
+ If the ignore_primary is false, errors on the primary node are
+ ignored.
+
+ """
+ all_result = True
+ disks = ExpandCheckDisks(instance, disks)
+
+ for disk in disks:
+ for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
+ lu.cfg.SetDiskID(top_disk, node)
+ result = lu.rpc.call_blockdev_shutdown(node, (top_disk, instance))
+ msg = result.fail_msg
+ if msg:
+ lu.LogWarning("Could not shutdown block device %s on node %s: %s",
+ disk.iv_name, node, msg)
+ if ((node == instance.primary_node and not ignore_primary) or
+ (node != instance.primary_node and not result.offline)):
+ all_result = False
+ return all_result
+
+
+def _SafeShutdownInstanceDisks(lu, instance, disks=None):
+ """Shutdown block devices of an instance.
+
+ This function checks if an instance is running, before calling
+ _ShutdownInstanceDisks.
+
+ """
+ CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
+ ShutdownInstanceDisks(lu, instance, disks=disks)
+
+
+def AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
+ ignore_size=False):
+ """Prepare the block devices for an instance.
+
+ This sets up the block devices on all nodes.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance for whose disks we assemble
+ @type disks: list of L{objects.Disk} or None
+ @param disks: which disks to assemble (or all, if None)
+ @type ignore_secondaries: boolean
+ @param ignore_secondaries: if true, errors on secondary nodes
+ won't result in an error return from the function
+ @type ignore_size: boolean
+ @param ignore_size: if true, the current known size of the disk
+ will not be used during the disk activation, useful for cases
+ when the size is wrong
+ @return: False if the operation failed, otherwise a list of
+ (host, instance_visible_name, node_visible_name)
+ with the mapping from node devices to instance devices
+
+ """
+ device_info = []
+ disks_ok = True
+ iname = instance.name
+ disks = ExpandCheckDisks(instance, disks)
+
+ # With the two passes mechanism we try to reduce the window of
+ # opportunity for the race condition of switching DRBD to primary
+ # before handshaking occured, but we do not eliminate it
+
+ # The proper fix would be to wait (with some limits) until the
+ # connection has been made and drbd transitions from WFConnection
+ # into any other network-connected state (Connected, SyncTarget,
+ # SyncSource, etc.)
+
+ # 1st pass, assemble on all nodes in secondary mode
+ for idx, inst_disk in enumerate(disks):
+ for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
+ if ignore_size:
+ node_disk = node_disk.Copy()
+ node_disk.UnsetSize()
+ lu.cfg.SetDiskID(node_disk, node)
+ result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
+ False, idx)
+ msg = result.fail_msg
+ if msg:
+ is_offline_secondary = (node in instance.secondary_nodes and
+ result.offline)
+ lu.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=False, pass=1): %s",
+ inst_disk.iv_name, node, msg)
+ if not (ignore_secondaries or is_offline_secondary):
+ disks_ok = False
+
+ # FIXME: race condition on drbd migration to primary
+
+ # 2nd pass, do only the primary node
+ for idx, inst_disk in enumerate(disks):
+ dev_path = None
+
+ for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
+ if node != instance.primary_node:
+ continue
+ if ignore_size:
+ node_disk = node_disk.Copy()
+ node_disk.UnsetSize()
+ lu.cfg.SetDiskID(node_disk, node)
+ result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
+ True, idx)
+ msg = result.fail_msg
+ if msg:
+ lu.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=True, pass=2): %s",
+ inst_disk.iv_name, node, msg)
+ disks_ok = False
+ else:
+ dev_path = result.payload
+
+ device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
+
+ # leave the disks configured for the primary node
+ # this is a workaround that would be fixed better by
+ # improving the logical/physical id handling
+ for disk in disks:
+ lu.cfg.SetDiskID(disk, instance.primary_node)
+
+ return disks_ok, device_info
+
+
+def StartInstanceDisks(lu, instance, force):
+ """Start the disks of an instance.
+
+ """
+ disks_ok, _ = AssembleInstanceDisks(lu, instance,
+ ignore_secondaries=force)
+ if not disks_ok:
+ ShutdownInstanceDisks(lu, instance)
+ if force is not None and not force:
+ lu.LogWarning("",
+ hint=("If the message above refers to a secondary node,"
+ " you can retry the operation using '--force'"))
+ raise errors.OpExecError("Disk consistency error")
+
+
+class LUInstanceGrowDisk(LogicalUnit):
+ """Grow a disk of an instance.
+
+ """
+ HPATH = "disk-grow"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, the primary and all the secondaries.
+
+ """
+ env = {
+ "DISK": self.op.disk,
+ "AMOUNT": self.op.amount,
+ "ABSOLUTE": self.op.absolute,
+ }
+ env.update(BuildInstanceHookEnvByObject(self, self.instance))
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
+ return (nl, nl)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+ nodenames = list(instance.all_nodes)
+ for node in nodenames:
+ CheckNodeOnline(self, node)
+
+ self.instance = instance
+
+ if instance.disk_template not in constants.DTS_GROWABLE:
+ raise errors.OpPrereqError("Instance's disk layout does not support"
+ " growing", errors.ECODE_INVAL)
+
+ self.disk = instance.FindDisk(self.op.disk)
+
+ if self.op.absolute:
+ self.target = self.op.amount
+ self.delta = self.target - self.disk.size
+ if self.delta < 0:
+ raise errors.OpPrereqError("Requested size (%s) is smaller than "
+ "current disk size (%s)" %
+ (utils.FormatUnit(self.target, "h"),
+ utils.FormatUnit(self.disk.size, "h")),
+ errors.ECODE_STATE)
+ else:
+ self.delta = self.op.amount
+ self.target = self.disk.size + self.delta
+ if self.delta < 0:
+ raise errors.OpPrereqError("Requested increment (%s) is negative" %
+ utils.FormatUnit(self.delta, "h"),
+ errors.ECODE_INVAL)
+
+ self._CheckDiskSpace(nodenames, self.disk.ComputeGrowth(self.delta))
+
+ def _CheckDiskSpace(self, nodenames, req_vgspace):
+ template = self.instance.disk_template
+ if template not in (constants.DTS_NO_FREE_SPACE_CHECK):
+ # TODO: check the free disk space for file, when that feature will be
+ # supported
+ nodes = map(self.cfg.GetNodeInfo, nodenames)
+ es_nodes = filter(lambda n: IsExclusiveStorageEnabledNode(self.cfg, n),
+ nodes)
+ if es_nodes:
+ # With exclusive storage we need to something smarter than just looking
+ # at free space; for now, let's simply abort the operation.
+ raise errors.OpPrereqError("Cannot grow disks when exclusive_storage"
+ " is enabled", errors.ECODE_STATE)
+ CheckNodesFreeDiskPerVG(self, nodenames, req_vgspace)
+
+ def Exec(self, feedback_fn):
+ """Execute disk grow.
+
+ """
+ instance = self.instance
+ disk = self.disk
+
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
+ wipe_disks = self.cfg.GetClusterInfo().prealloc_wipe_disks
+
+ disks_ok, _ = AssembleInstanceDisks(self, self.instance, disks=[disk])
+ if not disks_ok:
+ raise errors.OpExecError("Cannot activate block device to grow")
+
+ feedback_fn("Growing disk %s of instance '%s' by %s to %s" %
+ (self.op.disk, instance.name,
+ utils.FormatUnit(self.delta, "h"),
+ utils.FormatUnit(self.target, "h")))
+
+ # First run all grow ops in dry-run mode
+ for node in instance.all_nodes:
+ self.cfg.SetDiskID(disk, node)
+ result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
+ True, True)
+ result.Raise("Dry-run grow request failed to node %s" % node)
+
+ if wipe_disks:
+ # Get disk size from primary node for wiping
+ result = self.rpc.call_blockdev_getsize(instance.primary_node, [disk])
+ result.Raise("Failed to retrieve disk size from node '%s'" %
+ instance.primary_node)
+
+ (disk_size_in_bytes, ) = result.payload
+
+ if disk_size_in_bytes is None:
+ raise errors.OpExecError("Failed to retrieve disk size from primary"
+ " node '%s'" % instance.primary_node)
+
+ old_disk_size = _DiskSizeInBytesToMebibytes(self, disk_size_in_bytes)
+
+ assert old_disk_size >= disk.size, \
+ ("Retrieved disk size too small (got %s, should be at least %s)" %
+ (old_disk_size, disk.size))
+ else:
+ old_disk_size = None
+
+ # We know that (as far as we can test) operations across different
+ # nodes will succeed, time to run it for real on the backing storage
+ for node in instance.all_nodes:
+ self.cfg.SetDiskID(disk, node)
+ result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
+ False, True)
+ result.Raise("Grow request failed to node %s" % node)
+
+ # And now execute it for logical storage, on the primary node
+ node = instance.primary_node
+ self.cfg.SetDiskID(disk, node)
+ result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
+ False, False)
+ result.Raise("Grow request failed to node %s" % node)
+
+ disk.RecordGrow(self.delta)
+ self.cfg.Update(instance, feedback_fn)
+
+ # Changes have been recorded, release node lock
+ ReleaseLocks(self, locking.LEVEL_NODE)
+
+ # Downgrade lock while waiting for sync
+ self.glm.downgrade(locking.LEVEL_INSTANCE)
+
+ assert wipe_disks ^ (old_disk_size is None)
+
+ if wipe_disks:
+ assert instance.disks[self.op.disk] == disk
+
+ # Wipe newly added disk space
+ WipeDisks(self, instance,
+ disks=[(self.op.disk, disk, old_disk_size)])
+
+ if self.op.wait_for_sync:
+ disk_abort = not WaitForSync(self, instance, disks=[disk])
+ if disk_abort:
+ self.LogWarning("Disk syncing has not returned a good status; check"
+ " the instance")
+ if instance.admin_state != constants.ADMINST_UP:
+ _SafeShutdownInstanceDisks(self, instance, disks=[disk])
+ elif instance.admin_state != constants.ADMINST_UP:
+ self.LogWarning("Not shutting down the disk even if the instance is"
+ " not supposed to be running because no wait for"
+ " sync mode was requested")
+
+ assert self.owned_locks(locking.LEVEL_NODE_RES)
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+
+
+class LUInstanceReplaceDisks(LogicalUnit):
+ """Replace the disks of an instance.
+
+ """
+ HPATH = "mirrors-replace"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ """Check arguments.
+
+ """
+ remote_node = self.op.remote_node
+ ialloc = self.op.iallocator
+ if self.op.mode == constants.REPLACE_DISK_CHG:
+ if remote_node is None and ialloc is None:
+ raise errors.OpPrereqError("When changing the secondary either an"
+ " iallocator script must be used or the"
+ " new node given", errors.ECODE_INVAL)
+ else:
+ CheckIAllocatorOrNode(self, "iallocator", "remote_node")
+
+ elif remote_node is not None or ialloc is not None:
+ # Not replacing the secondary
+ raise errors.OpPrereqError("The iallocator and new node options can"
+ " only be used when changing the"
+ " secondary node", errors.ECODE_INVAL)
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+
+ assert locking.LEVEL_NODE not in self.needed_locks
+ assert locking.LEVEL_NODE_RES not in self.needed_locks
+ assert locking.LEVEL_NODEGROUP not in self.needed_locks
+
+ assert self.op.iallocator is None or self.op.remote_node is None, \
+ "Conflicting options"
+
+ if self.op.remote_node is not None:
+ self.op.remote_node = ExpandNodeName(self.cfg, self.op.remote_node)
+
+ # Warning: do not remove the locking of the new secondary here
+ # unless DRBD8.AddChildren is changed to work in parallel;
+ # currently it doesn't since parallel invocations of
+ # FindUnusedMinor will conflict
+ self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ else:
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ if self.op.iallocator is not None:
+ # iallocator will select a new node in the same group
+ self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+
+ self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
+ self.op.iallocator, self.op.remote_node,
+ self.op.disks, self.op.early_release,
+ self.op.ignore_ipolicy)
+
+ self.tasklets = [self.replacer]
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert self.op.remote_node is None
+ assert self.op.iallocator is not None
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+ self.share_locks[locking.LEVEL_NODEGROUP] = 1
+ # Lock all groups used by instance optimistically; this requires going
+ # via the node before it's locked, requiring verification later on
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+
+ elif level == locking.LEVEL_NODE:
+ if self.op.iallocator is not None:
+ assert self.op.remote_node is None
+ assert not self.needed_locks[locking.LEVEL_NODE]
+ assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
+
+ # Lock member nodes of all locked groups
+ self.needed_locks[locking.LEVEL_NODE] = \
+ [node_name
+ for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+ for node_name in self.cfg.GetNodeGroup(group_uuid).members]
+ else:
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
+ self._LockInstancesNodes()
+
+ elif level == locking.LEVEL_NODE_RES:
+ # Reuse node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE]
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, the primary and all the secondaries.
+
+ """
+ instance = self.replacer.instance
+ env = {
+ "MODE": self.op.mode,
+ "NEW_SECONDARY": self.op.remote_node,
+ "OLD_SECONDARY": instance.secondary_nodes[0],
+ }
+ env.update(BuildInstanceHookEnvByObject(self, instance))
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ instance = self.replacer.instance
+ nl = [
+ self.cfg.GetMasterNode(),
+ instance.primary_node,
+ ]
+ if self.op.remote_node is not None:
+ nl.append(self.op.remote_node)
+ return nl, nl
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ """
+ assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
+ self.op.iallocator is None)
+
+ # Verify if node group locks are still correct
+ owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
+ if owned_groups:
+ CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
+
+ return LogicalUnit.CheckPrereq(self)
+
+
+class LUInstanceActivateDisks(NoHooksLU):
+ """Bring up an instance's disks.
+
+ """
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+ CheckNodeOnline(self, self.instance.primary_node)
+
+ def Exec(self, feedback_fn):
+ """Activate the disks.
+
+ """
+ disks_ok, disks_info = \
+ AssembleInstanceDisks(self, self.instance,
+ ignore_size=self.op.ignore_size)
+ if not disks_ok:
+ raise errors.OpExecError("Cannot activate block devices")
+
+ if self.op.wait_for_sync:
+ if not WaitForSync(self, self.instance):
+ raise errors.OpExecError("Some disks of the instance are degraded!")
+
+ return disks_info
+
+
+class LUInstanceDeactivateDisks(NoHooksLU):
+ """Shutdown an instance's disks.
+
+ """
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+
+ def Exec(self, feedback_fn):
+ """Deactivate the disks
+
+ """
+ instance = self.instance
+ if self.op.force:
+ ShutdownInstanceDisks(self, instance)
+ else:
+ _SafeShutdownInstanceDisks(self, instance)
+
+
+def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,
+ ldisk=False):
+ """Check that mirrors are not degraded.
+
+ @attention: The device has to be annotated already.
+
+ The ldisk parameter, if True, will change the test from the
+ is_degraded attribute (which represents overall non-ok status for
+ the device(s)) to the ldisk (representing the local storage status).
+
+ """
+ lu.cfg.SetDiskID(dev, node)
+
+ result = True
+
+ if on_primary or dev.AssembleOnSecondary():
+ rstats = lu.rpc.call_blockdev_find(node, dev)
+ msg = rstats.fail_msg
+ if msg:
+ lu.LogWarning("Can't find disk on node %s: %s", node, msg)
+ result = False
+ elif not rstats.payload:
+ lu.LogWarning("Can't find disk on node %s", node)
+ result = False
+ else:
+ if ldisk:
+ result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
+ else:
+ result = result and not rstats.payload.is_degraded
+
+ if dev.children:
+ for child in dev.children:
+ result = result and _CheckDiskConsistencyInner(lu, instance, child, node,
+ on_primary)
+
+ return result
+
+
+def CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):
+ """Wrapper around L{_CheckDiskConsistencyInner}.
+
+ """
+ (disk,) = AnnotateDiskParams(instance, [dev], lu.cfg)
+ return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,
+ ldisk=ldisk)
+
+
+def _BlockdevFind(lu, node, dev, instance):
+ """Wrapper around call_blockdev_find to annotate diskparams.
+
+ @param lu: A reference to the lu object
+ @param node: The node to call out
+ @param dev: The device to find
+ @param instance: The instance object the device belongs to
+ @returns The result of the rpc call
+
+ """
+ (disk,) = AnnotateDiskParams(instance, [dev], lu.cfg)
+ return lu.rpc.call_blockdev_find(node, disk)
+
+
+def _GenerateUniqueNames(lu, exts):
+ """Generate a suitable LV name.
+
+ This will generate a logical volume name for the given instance.
+
+ """
+ results = []
+ for val in exts:
+ new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
+ results.append("%s%s" % (new_id, val))
+ return results
+
+
+class TLReplaceDisks(Tasklet):
+ """Replaces disks for an instance.
+
+ Note: Locking is not within the scope of this class.
+
+ """
+ def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
+ disks, early_release, ignore_ipolicy):
+ """Initializes this class.
+
+ """
+ Tasklet.__init__(self, lu)
+
+ # Parameters
+ self.instance_name = instance_name
+ self.mode = mode
+ self.iallocator_name = iallocator_name
+ self.remote_node = remote_node
+ self.disks = disks
+ self.early_release = early_release
+ self.ignore_ipolicy = ignore_ipolicy
+
+ # Runtime data
+ self.instance = None
+ self.new_node = None
+ self.target_node = None
+ self.other_node = None
+ self.remote_node_info = None
+ self.node_secondary_ip = None
+
+ @staticmethod
+ def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
+ """Compute a new secondary node using an IAllocator.
+
+ """
+ req = iallocator.IAReqRelocate(name=instance_name,
+ relocate_from=list(relocate_from))
+ ial = iallocator.IAllocator(lu.cfg, lu.rpc, req)
+
+ ial.Run(iallocator_name)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
+ " %s" % (iallocator_name, ial.info),
+ errors.ECODE_NORES)
+
+ remote_node_name = ial.result[0]
+
+ lu.LogInfo("Selected new secondary for instance '%s': %s",
+ instance_name, remote_node_name)
+
+ return remote_node_name
+
+ def _FindFaultyDisks(self, node_name):
+ """Wrapper for L{FindFaultyInstanceDisks}.
+
+ """
+ return FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
+ node_name, True)
+
+ def _CheckDisksActivated(self, instance):
+ """Checks if the instance disks are activated.
+
+ @param instance: The instance to check disks
+ @return: True if they are activated, False otherwise
+
+ """
+ nodes = instance.all_nodes
+
+ for idx, dev in enumerate(instance.disks):
+ for node in nodes:
+ self.lu.LogInfo("Checking disk/%d on %s", idx, node)
+ self.cfg.SetDiskID(dev, node)
+
+ result = _BlockdevFind(self, node, dev, instance)
+
+ if result.offline:
+ continue
+ elif result.fail_msg or not result.payload:
+ return False
+
+ return True
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
+ assert instance is not None, \
+ "Cannot retrieve locked instance %s" % self.instance_name
+
+ if instance.disk_template != constants.DT_DRBD8:
+ raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
+ " instances", errors.ECODE_INVAL)
+
+ if len(instance.secondary_nodes) != 1:
+ raise errors.OpPrereqError("The instance has a strange layout,"
+ " expected one secondary but found %d" %
+ len(instance.secondary_nodes),
+ errors.ECODE_FAULT)
+
+ instance = self.instance
+ secondary_node = instance.secondary_nodes[0]
+
+ if self.iallocator_name is None:
+ remote_node = self.remote_node
+ else:
+ remote_node = self._RunAllocator(self.lu, self.iallocator_name,
+ instance.name, instance.secondary_nodes)
+
+ if remote_node is None:
+ self.remote_node_info = None
+ else:
+ assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
+ "Remote node '%s' is not locked" % remote_node
+
+ self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
+ assert self.remote_node_info is not None, \
+ "Cannot retrieve locked node %s" % remote_node
+
+ if remote_node == self.instance.primary_node:
+ raise errors.OpPrereqError("The specified node is the primary node of"
+ " the instance", errors.ECODE_INVAL)
+
+ if remote_node == secondary_node:
+ raise errors.OpPrereqError("The specified node is already the"
+ " secondary node of the instance",
+ errors.ECODE_INVAL)
+
+ if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
+ constants.REPLACE_DISK_CHG):
+ raise errors.OpPrereqError("Cannot specify disks to be replaced",
+ errors.ECODE_INVAL)
+
+ if self.mode == constants.REPLACE_DISK_AUTO:
+ if not self._CheckDisksActivated(instance):
+ raise errors.OpPrereqError("Please run activate-disks on instance %s"
+ " first" % self.instance_name,
+ errors.ECODE_STATE)
+ faulty_primary = self._FindFaultyDisks(instance.primary_node)
+ faulty_secondary = self._FindFaultyDisks(secondary_node)
+
+ if faulty_primary and faulty_secondary:
+ raise errors.OpPrereqError("Instance %s has faulty disks on more than"
+ " one node and can not be repaired"
+ " automatically" % self.instance_name,
+ errors.ECODE_STATE)
+
+ if faulty_primary:
+ self.disks = faulty_primary
+ self.target_node = instance.primary_node
+ self.other_node = secondary_node
+ check_nodes = [self.target_node, self.other_node]
+ elif faulty_secondary:
+ self.disks = faulty_secondary
+ self.target_node = secondary_node
+ self.other_node = instance.primary_node
+ check_nodes = [self.target_node, self.other_node]
+ else:
+ self.disks = []
+ check_nodes = []
+
+ else:
+ # Non-automatic modes
+ if self.mode == constants.REPLACE_DISK_PRI:
+ self.target_node = instance.primary_node
+ self.other_node = secondary_node
+ check_nodes = [self.target_node, self.other_node]
+
+ elif self.mode == constants.REPLACE_DISK_SEC:
+ self.target_node = secondary_node
+ self.other_node = instance.primary_node
+ check_nodes = [self.target_node, self.other_node]
+
+ elif self.mode == constants.REPLACE_DISK_CHG:
+ self.new_node = remote_node
+ self.other_node = instance.primary_node
+ self.target_node = secondary_node
+ check_nodes = [self.new_node, self.other_node]
+
+ CheckNodeNotDrained(self.lu, remote_node)
+ CheckNodeVmCapable(self.lu, remote_node)
+
+ old_node_info = self.cfg.GetNodeInfo(secondary_node)
+ assert old_node_info is not None
+ if old_node_info.offline and not self.early_release:
+ # doesn't make sense to delay the release
+ self.early_release = True
+ self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
+ " early-release mode", secondary_node)
+
+ else:
+ raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
+ self.mode)
+
+ # If not specified all disks should be replaced
+ if not self.disks:
+ self.disks = range(len(self.instance.disks))
+
+ # TODO: This is ugly, but right now we can't distinguish between internal
+ # submitted opcode and external one. We should fix that.
+ if self.remote_node_info:
+ # We change the node, lets verify it still meets instance policy
+ new_group_info = self.cfg.GetNodeGroup(self.remote_node_info.group)
+ cluster = self.cfg.GetClusterInfo()
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
+ new_group_info)
+ CheckTargetNodeIPolicy(self, ipolicy, instance, self.remote_node_info,
+ self.cfg, ignore=self.ignore_ipolicy)
+
+ for node in check_nodes:
+ CheckNodeOnline(self.lu, node)
+
+ touched_nodes = frozenset(node_name for node_name in [self.new_node,
+ self.other_node,
+ self.target_node]
+ if node_name is not None)
+
+ # Release unneeded node and node resource locks
+ ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
+
+ # Release any owned node group
+ ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
+
+ # Check whether disks are valid
+ for disk_idx in self.disks:
+ instance.FindDisk(disk_idx)
+
+ # Get secondary node IP addresses
+ self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
+ in self.cfg.GetMultiNodeInfo(touched_nodes))
+
+ def Exec(self, feedback_fn):
+ """Execute disk replacement.
+
+ This dispatches the disk replacement to the appropriate handler.
+
+ """
+ if __debug__:
+ # Verify owned locks before starting operation
+ owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
+ assert set(owned_nodes) == set(self.node_secondary_ip), \
+ ("Incorrect node locks, owning %s, expected %s" %
+ (owned_nodes, self.node_secondary_ip.keys()))
+ assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
+ self.lu.owned_locks(locking.LEVEL_NODE_RES))
+ assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
+ owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
+ assert list(owned_instances) == [self.instance_name], \
+ "Instance '%s' not locked" % self.instance_name
+
+ assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
+ "Should not own any node group lock at this point"
+
+ if not self.disks:
+ feedback_fn("No disks need replacement for instance '%s'" %
+ self.instance.name)
+ return
+
+ feedback_fn("Replacing disk(s) %s for instance '%s'" %
+ (utils.CommaJoin(self.disks), self.instance.name))
+ feedback_fn("Current primary node: %s" % self.instance.primary_node)
+ feedback_fn("Current seconary node: %s" %
+ utils.CommaJoin(self.instance.secondary_nodes))
+
+ activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
+
+ # Activate the instance disks if we're replacing them on a down instance
+ if activate_disks:
+ StartInstanceDisks(self.lu, self.instance, True)
+
+ try:
+ # Should we replace the secondary node?
+ if self.new_node is not None:
+ fn = self._ExecDrbd8Secondary
+ else:
+ fn = self._ExecDrbd8DiskOnly
+
+ result = fn(feedback_fn)
+ finally:
+ # Deactivate the instance disks if we're replacing them on a
+ # down instance
+ if activate_disks:
+ _SafeShutdownInstanceDisks(self.lu, self.instance)
+
+ assert not self.lu.owned_locks(locking.LEVEL_NODE)
+
+ if __debug__:
+ # Verify owned locks
+ owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
+ nodes = frozenset(self.node_secondary_ip)
+ assert ((self.early_release and not owned_nodes) or
+ (not self.early_release and not (set(owned_nodes) - nodes))), \
+ ("Not owning the correct locks, early_release=%s, owned=%r,"
+ " nodes=%r" % (self.early_release, owned_nodes, nodes))
+
+ return result
+
+ def _CheckVolumeGroup(self, nodes):
+ self.lu.LogInfo("Checking volume groups")
+
+ vgname = self.cfg.GetVGName()
+
+ # Make sure volume group exists on all involved nodes
+ results = self.rpc.call_vg_list(nodes)
+ if not results:
+ raise errors.OpExecError("Can't list volume groups on the nodes")
+
+ for node in nodes:
+ res = results[node]
+ res.Raise("Error checking node %s" % node)
+ if vgname not in res.payload:
+ raise errors.OpExecError("Volume group '%s' not found on node %s" %
+ (vgname, node))
+
+ def _CheckDisksExistence(self, nodes):
+ # Check disk existence
+ for idx, dev in enumerate(self.instance.disks):
+ if idx not in self.disks:
+ continue
+
+ for node in nodes:
+ self.lu.LogInfo("Checking disk/%d on %s", idx, node)
+ self.cfg.SetDiskID(dev, node)
+
+ result = _BlockdevFind(self, node, dev, self.instance)
+
+ msg = result.fail_msg
+ if msg or not result.payload:
+ if not msg:
+ msg = "disk not found"
+ raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
+ (idx, node, msg))
+
+ def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
+ for idx, dev in enumerate(self.instance.disks):
+ if idx not in self.disks:
+ continue
+
+ self.lu.LogInfo("Checking disk/%d consistency on node %s" %
+ (idx, node_name))
+
+ if not CheckDiskConsistency(self.lu, self.instance, dev, node_name,
+ on_primary, ldisk=ldisk):
+ raise errors.OpExecError("Node %s has degraded storage, unsafe to"
+ " replace disks for instance %s" %
+ (node_name, self.instance.name))
+
+ def _CreateNewStorage(self, node_name):
+ """Create new storage on the primary or secondary node.
+
+ This is only used for same-node replaces, not for changing the
+ secondary node, hence we don't want to modify the existing disk.
+
+ """
+ iv_names = {}
+
+ disks = AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
+ for idx, dev in enumerate(disks):
+ if idx not in self.disks:
+ continue
+
+ self.lu.LogInfo("Adding storage on %s for disk/%d", node_name, idx)
+
+ self.cfg.SetDiskID(dev, node_name)
+
+ lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
+ names = _GenerateUniqueNames(self.lu, lv_names)
+
+ (data_disk, meta_disk) = dev.children
+ vg_data = data_disk.logical_id[0]
+ lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
+ logical_id=(vg_data, names[0]),
+ params=data_disk.params)
+ vg_meta = meta_disk.logical_id[0]
+ lv_meta = objects.Disk(dev_type=constants.LD_LV,
+ size=constants.DRBD_META_SIZE,
+ logical_id=(vg_meta, names[1]),
+ params=meta_disk.params)
+
+ new_lvs = [lv_data, lv_meta]
+ old_lvs = [child.Copy() for child in dev.children]
+ iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
+ excl_stor = IsExclusiveStorageEnabledNodeName(self.lu.cfg, node_name)
+
+ # we pass force_create=True to force the LVM creation
+ for new_lv in new_lvs:
+ _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
+ GetInstanceInfoText(self.instance), False,
+ excl_stor)
+
+ return iv_names
+
+ def _CheckDevices(self, node_name, iv_names):
+ for name, (dev, _, _) in iv_names.iteritems():
+ self.cfg.SetDiskID(dev, node_name)
+
+ result = _BlockdevFind(self, node_name, dev, self.instance)
+
+ msg = result.fail_msg
+ if msg or not result.payload:
+ if not msg:
+ msg = "disk not found"
+ raise errors.OpExecError("Can't find DRBD device %s: %s" %
+ (name, msg))
+
+ if result.payload.is_degraded:
+ raise errors.OpExecError("DRBD device %s is degraded!" % name)
+
+ def _RemoveOldStorage(self, node_name, iv_names):
+ for name, (_, old_lvs, _) in iv_names.iteritems():
+ self.lu.LogInfo("Remove logical volumes for %s", name)
+
+ for lv in old_lvs:
+ self.cfg.SetDiskID(lv, node_name)
+
+ msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
+ if msg:
+ self.lu.LogWarning("Can't remove old LV: %s", msg,
+ hint="remove unused LVs manually")
+
+ def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
+ """Replace a disk on the primary or secondary for DRBD 8.
+
+ The algorithm for replace is quite complicated:
+
+ 1. for each disk to be replaced:
+
+ 1. create new LVs on the target node with unique names
+ 1. detach old LVs from the drbd device
+ 1. rename old LVs to name_replaced.<time_t>
+ 1. rename new LVs to old LVs
+ 1. attach the new LVs (with the old names now) to the drbd device
+
+ 1. wait for sync across all devices
+
+ 1. for each modified disk:
+
+ 1. remove old LVs (which have the name name_replaces.<time_t>)
+
+ Failures are not very well handled.
+
+ """
+ steps_total = 6
+
+ # Step: check device activation
+ self.lu.LogStep(1, steps_total, "Check device existence")
+ self._CheckDisksExistence([self.other_node, self.target_node])
+ self._CheckVolumeGroup([self.target_node, self.other_node])
+
+ # Step: check other node consistency
+ self.lu.LogStep(2, steps_total, "Check peer consistency")
+ self._CheckDisksConsistency(self.other_node,
+ self.other_node == self.instance.primary_node,
+ False)
+
+ # Step: create new storage
+ self.lu.LogStep(3, steps_total, "Allocate new storage")
+ iv_names = self._CreateNewStorage(self.target_node)
+
+ # Step: for each lv, detach+rename*2+attach
+ self.lu.LogStep(4, steps_total, "Changing drbd configuration")
+ for dev, old_lvs, new_lvs in iv_names.itervalues():
+ self.lu.LogInfo("Detaching %s drbd from local storage", dev.iv_name)
+
+ result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
+ old_lvs)
+ result.Raise("Can't detach drbd from local storage on node"
+ " %s for device %s" % (self.target_node, dev.iv_name))
+ #dev.children = []
+ #cfg.Update(instance)
+
+ # ok, we created the new LVs, so now we know we have the needed
+ # storage; as such, we proceed on the target node to rename
+ # old_lv to _old, and new_lv to old_lv; note that we rename LVs
+ # using the assumption that logical_id == physical_id (which in
+ # turn is the unique_id on that node)
+
+ # FIXME(iustin): use a better name for the replaced LVs
+ temp_suffix = int(time.time())
+ ren_fn = lambda d, suff: (d.physical_id[0],
+ d.physical_id[1] + "_replaced-%s" % suff)
+
+ # Build the rename list based on what LVs exist on the node
+ rename_old_to_new = []
+ for to_ren in old_lvs:
+ result = self.rpc.call_blockdev_find(self.target_node, to_ren)
+ if not result.fail_msg and result.payload:
+ # device exists
+ rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
+
+ self.lu.LogInfo("Renaming the old LVs on the target node")
+ result = self.rpc.call_blockdev_rename(self.target_node,
+ rename_old_to_new)
+ result.Raise("Can't rename old LVs on node %s" % self.target_node)
+
+ # Now we rename the new LVs to the old LVs
+ self.lu.LogInfo("Renaming the new LVs on the target node")
+ rename_new_to_old = [(new, old.physical_id)
+ for old, new in zip(old_lvs, new_lvs)]
+ result = self.rpc.call_blockdev_rename(self.target_node,
+ rename_new_to_old)
+ result.Raise("Can't rename new LVs on node %s" % self.target_node)
+
+ # Intermediate steps of in memory modifications
+ for old, new in zip(old_lvs, new_lvs):
+ new.logical_id = old.logical_id
+ self.cfg.SetDiskID(new, self.target_node)
+
+ # We need to modify old_lvs so that removal later removes the
+ # right LVs, not the newly added ones; note that old_lvs is a
+ # copy here
+ for disk in old_lvs:
+ disk.logical_id = ren_fn(disk, temp_suffix)
+ self.cfg.SetDiskID(disk, self.target_node)
+
+ # Now that the new lvs have the old name, we can add them to the device
+ self.lu.LogInfo("Adding new mirror component on %s", self.target_node)
+ result = self.rpc.call_blockdev_addchildren(self.target_node,
+ (dev, self.instance), new_lvs)
+ msg = result.fail_msg
+ if msg:
+ for new_lv in new_lvs:
+ msg2 = self.rpc.call_blockdev_remove(self.target_node,
+ new_lv).fail_msg
+ if msg2:
+ self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
+ hint=("cleanup manually the unused logical"
+ "volumes"))
+ raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
+
+ cstep = itertools.count(5)
+
+ if self.early_release:
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
+ # TODO: Check if releasing locks early still makes sense
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+ else:
+ # Release all resource locks except those used by the instance
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+ keep=self.node_secondary_ip.keys())
+
+ # Release all node locks while waiting for sync
+ ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
+ # TODO: Can the instance lock be downgraded here? Take the optional disk
+ # shutdown in the caller into consideration.
+
+ # Wait for sync
+ # This can fail as the old devices are degraded and _WaitForSync
+ # does a combined result over all disks, so we don't check its return value
+ self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
+ WaitForSync(self.lu, self.instance)
+
+ # Check all devices manually
+ self._CheckDevices(self.instance.primary_node, iv_names)
+
+ # Step: remove old storage
+ if not self.early_release:
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
+
+ def _ExecDrbd8Secondary(self, feedback_fn):
+ """Replace the secondary node for DRBD 8.
+
+ The algorithm for replace is quite complicated:
+ - for all disks of the instance:
+ - create new LVs on the new node with same names
+ - shutdown the drbd device on the old secondary
+ - disconnect the drbd network on the primary
+ - create the drbd device on the new secondary
+ - network attach the drbd on the primary, using an artifice:
+ the drbd code for Attach() will connect to the network if it
+ finds a device which is connected to the good local disks but
+ not network enabled
+ - wait for sync across all devices
+ - remove all disks from the old secondary
+
+ Failures are not very well handled.
+
+ """
+ steps_total = 6
+
+ pnode = self.instance.primary_node
+
+ # Step: check device activation
+ self.lu.LogStep(1, steps_total, "Check device existence")
+ self._CheckDisksExistence([self.instance.primary_node])
+ self._CheckVolumeGroup([self.instance.primary_node])
+
+ # Step: check other node consistency
+ self.lu.LogStep(2, steps_total, "Check peer consistency")
+ self._CheckDisksConsistency(self.instance.primary_node, True, True)
+
+ # Step: create new storage
+ self.lu.LogStep(3, steps_total, "Allocate new storage")
+ disks = AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
+ excl_stor = IsExclusiveStorageEnabledNodeName(self.lu.cfg, self.new_node)
+ for idx, dev in enumerate(disks):
+ self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
+ (self.new_node, idx))
+ # we pass force_create=True to force LVM creation
+ for new_lv in dev.children:
+ _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
+ True, GetInstanceInfoText(self.instance), False,
+ excl_stor)
+
+ # Step 4: dbrd minors and drbd setups changes
+ # after this, we must manually remove the drbd minors on both the
+ # error and the success paths
+ self.lu.LogStep(4, steps_total, "Changing drbd configuration")
+ minors = self.cfg.AllocateDRBDMinor([self.new_node
+ for dev in self.instance.disks],
+ self.instance.name)
+ logging.debug("Allocated minors %r", minors)
+
+ iv_names = {}
+ for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
+ self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
+ (self.new_node, idx))
+ # create new devices on new_node; note that we create two IDs:
+ # one without port, so the drbd will be activated without
+ # networking information on the new node at this stage, and one
+ # with network, for the latter activation in step 4
+ (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
+ if self.instance.primary_node == o_node1:
+ p_minor = o_minor1
+ else:
+ assert self.instance.primary_node == o_node2, "Three-node instance?"
+ p_minor = o_minor2
+
+ new_alone_id = (self.instance.primary_node, self.new_node, None,
+ p_minor, new_minor, o_secret)
+ new_net_id = (self.instance.primary_node, self.new_node, o_port,
+ p_minor, new_minor, o_secret)
+
+ iv_names[idx] = (dev, dev.children, new_net_id)
+ logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
+ new_net_id)
+ new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
+ logical_id=new_alone_id,
+ children=dev.children,
+ size=dev.size,
+ params={})
+ (anno_new_drbd,) = AnnotateDiskParams(self.instance, [new_drbd],
+ self.cfg)
+ try:
+ CreateSingleBlockDev(self.lu, self.new_node, self.instance,
+ anno_new_drbd,
+ GetInstanceInfoText(self.instance), False,
+ excl_stor)
+ except errors.GenericError:
+ self.cfg.ReleaseDRBDMinors(self.instance.name)
+ raise
+
+ # We have new devices, shutdown the drbd on the old secondary
+ for idx, dev in enumerate(self.instance.disks):
+ self.lu.LogInfo("Shutting down drbd for disk/%d on old node", idx)
+ self.cfg.SetDiskID(dev, self.target_node)
+ msg = self.rpc.call_blockdev_shutdown(self.target_node,
+ (dev, self.instance)).fail_msg
+ if msg:
+ self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
+ "node: %s" % (idx, msg),
+ hint=("Please cleanup this device manually as"
+ " soon as possible"))
+
+ self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
+ result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
+ self.instance.disks)[pnode]
+
+ msg = result.fail_msg
+ if msg:
+ # detaches didn't succeed (unlikely)
+ self.cfg.ReleaseDRBDMinors(self.instance.name)
+ raise errors.OpExecError("Can't detach the disks from the network on"
+ " old node: %s" % (msg,))
+
+ # if we managed to detach at least one, we update all the disks of
+ # the instance to point to the new secondary
+ self.lu.LogInfo("Updating instance configuration")
+ for dev, _, new_logical_id in iv_names.itervalues():
+ dev.logical_id = new_logical_id
+ self.cfg.SetDiskID(dev, self.instance.primary_node)
+
+ self.cfg.Update(self.instance, feedback_fn)
+
+ # Release all node locks (the configuration has been updated)
+ ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
+ # and now perform the drbd attach
+ self.lu.LogInfo("Attaching primary drbds to new secondary"
+ " (standalone => connected)")
+ result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
+ self.new_node],
+ self.node_secondary_ip,
+ (self.instance.disks, self.instance),
+ self.instance.name,
+ False)
+ for to_node, to_result in result.items():
+ msg = to_result.fail_msg
+ if msg:
+ self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
+ to_node, msg,
+ hint=("please do a gnt-instance info to see the"
+ " status of disks"))
+
+ cstep = itertools.count(5)
+
+ if self.early_release:
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
+ # TODO: Check if releasing locks early still makes sense
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+ else:
+ # Release all resource locks except those used by the instance
+ ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+ keep=self.node_secondary_ip.keys())
+
+ # TODO: Can the instance lock be downgraded here? Take the optional disk
+ # shutdown in the caller into consideration.
+
+ # Wait for sync
+ # This can fail as the old devices are degraded and _WaitForSync
+ # does a combined result over all disks, so we don't check its return value
+ self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
+ WaitForSync(self.lu, self.instance)
+
+ # Check all devices manually
+ self._CheckDevices(self.instance.primary_node, iv_names)
+
+ # Step: remove old storage
+ if not self.early_release:
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
projects / ganeti-github.git / commitdiff