4 # Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc.
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2 of the License, or
9 # (at your option) any later version.
11 # This program is distributed in the hope that it will be useful, but
12 # WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 # General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, write to the Free Software
18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 """Logical units dealing with storage of instances."""
29 from ganeti import compat
30 from ganeti import constants
31 from ganeti import errors
33 from ganeti import locking
34 from ganeti.masterd import iallocator
35 from ganeti import objects
36 from ganeti import utils
37 from ganeti import opcodes
38 from ganeti import rpc
39 from ganeti.cmdlib.base import LogicalUnit, NoHooksLU, Tasklet
40 from ganeti.cmdlib.common import INSTANCE_DOWN, INSTANCE_NOT_RUNNING, \
41 AnnotateDiskParams, CheckIAllocatorOrNode, ExpandNodeName, \
42 CheckNodeOnline, CheckInstanceNodeGroups, CheckInstanceState, \
43 IsExclusiveStorageEnabledNode, FindFaultyInstanceDisks
44 from ganeti.cmdlib.instance_utils import GetInstanceInfoText, \
45 CopyLockList, ReleaseLocks, CheckNodeVmCapable, \
46 BuildInstanceHookEnvByObject, CheckNodeNotDrained, CheckTargetNodeIPolicy
48 import ganeti.masterd.instance
51 _DISK_TEMPLATE_NAME_PREFIX = {
52 constants.DT_PLAIN: "",
53 constants.DT_RBD: ".rbd",
54 constants.DT_EXT: ".ext",
58 _DISK_TEMPLATE_DEVICE_TYPE = {
59 constants.DT_PLAIN: constants.LD_LV,
60 constants.DT_FILE: constants.LD_FILE,
61 constants.DT_SHARED_FILE: constants.LD_FILE,
62 constants.DT_BLOCK: constants.LD_BLOCKDEV,
63 constants.DT_RBD: constants.LD_RBD,
64 constants.DT_EXT: constants.LD_EXT,
68 def CreateSingleBlockDev(lu, node, instance, device, info, force_open,
70 """Create a single block device on a given node.
72 This will not recurse over children of the device, so they must be
75 @param lu: the lu on whose behalf we execute
76 @param node: the node on which to create the device
77 @type instance: L{objects.Instance}
78 @param instance: the instance which owns the device
79 @type device: L{objects.Disk}
80 @param device: the device to create
81 @param info: the extra 'metadata' we should attach to the device
82 (this will be represented as a LVM tag)
83 @type force_open: boolean
84 @param force_open: this parameter will be passes to the
85 L{backend.BlockdevCreate} function where it specifies
86 whether we run on primary or not, and it affects both
87 the child assembly and the device own Open() execution
88 @type excl_stor: boolean
89 @param excl_stor: Whether exclusive_storage is active for the node
92 lu.cfg.SetDiskID(device, node)
93 result = lu.rpc.call_blockdev_create(node, device, device.size,
94 instance.name, force_open, info,
96 result.Raise("Can't create block device %s on"
97 " node %s for instance %s" % (device, node, instance.name))
98 if device.physical_id is None:
99 device.physical_id = result.payload
102 def _CreateBlockDevInner(lu, node, instance, device, force_create,
103 info, force_open, excl_stor):
104 """Create a tree of block devices on a given node.
106 If this device type has to be created on secondaries, create it and
109 If not, just recurse to children keeping the same 'force' value.
111 @attention: The device has to be annotated already.
113 @param lu: the lu on whose behalf we execute
114 @param node: the node on which to create the device
115 @type instance: L{objects.Instance}
116 @param instance: the instance which owns the device
117 @type device: L{objects.Disk}
118 @param device: the device to create
119 @type force_create: boolean
120 @param force_create: whether to force creation of this device; this
121 will be change to True whenever we find a device which has
122 CreateOnSecondary() attribute
123 @param info: the extra 'metadata' we should attach to the device
124 (this will be represented as a LVM tag)
125 @type force_open: boolean
126 @param force_open: this parameter will be passes to the
127 L{backend.BlockdevCreate} function where it specifies
128 whether we run on primary or not, and it affects both
129 the child assembly and the device own Open() execution
130 @type excl_stor: boolean
131 @param excl_stor: Whether exclusive_storage is active for the node
133 @return: list of created devices
137 if device.CreateOnSecondary():
141 for child in device.children:
142 devs = _CreateBlockDevInner(lu, node, instance, child, force_create,
143 info, force_open, excl_stor)
144 created_devices.extend(devs)
147 return created_devices
149 CreateSingleBlockDev(lu, node, instance, device, info, force_open,
151 # The device has been completely created, so there is no point in keeping
152 # its subdevices in the list. We just add the device itself instead.
153 created_devices = [(node, device)]
154 return created_devices
156 except errors.DeviceCreationError, e:
157 e.created_devices.extend(created_devices)
159 except errors.OpExecError, e:
160 raise errors.DeviceCreationError(str(e), created_devices)
163 def IsExclusiveStorageEnabledNodeName(cfg, nodename):
164 """Whether exclusive_storage is in effect for the given node.
166 @type cfg: L{config.ConfigWriter}
167 @param cfg: The cluster configuration
168 @type nodename: string
169 @param nodename: The node
171 @return: The effective value of exclusive_storage
172 @raise errors.OpPrereqError: if no node exists with the given name
175 ni = cfg.GetNodeInfo(nodename)
177 raise errors.OpPrereqError("Invalid node name %s" % nodename,
179 return IsExclusiveStorageEnabledNode(cfg, ni)
182 def _CreateBlockDev(lu, node, instance, device, force_create, info,
184 """Wrapper around L{_CreateBlockDevInner}.
186 This method annotates the root device first.
189 (disk,) = AnnotateDiskParams(instance, [device], lu.cfg)
190 excl_stor = IsExclusiveStorageEnabledNodeName(lu.cfg, node)
191 return _CreateBlockDevInner(lu, node, instance, disk, force_create, info,
192 force_open, excl_stor)
195 def _UndoCreateDisks(lu, disks_created):
196 """Undo the work performed by L{CreateDisks}.
198 This function is called in case of an error to undo the work of
201 @type lu: L{LogicalUnit}
202 @param lu: the logical unit on whose behalf we execute
203 @param disks_created: the result returned by L{CreateDisks}
206 for (node, disk) in disks_created:
207 lu.cfg.SetDiskID(disk, node)
208 result = lu.rpc.call_blockdev_remove(node, disk)
210 logging.warning("Failed to remove newly-created disk %s on node %s:"
211 " %s", disk, node, result.fail_msg)
214 def CreateDisks(lu, instance, to_skip=None, target_node=None, disks=None):
215 """Create all disks for an instance.
217 This abstracts away some work from AddInstance.
219 @type lu: L{LogicalUnit}
220 @param lu: the logical unit on whose behalf we execute
221 @type instance: L{objects.Instance}
222 @param instance: the instance whose disks we should create
224 @param to_skip: list of indices to skip
225 @type target_node: string
226 @param target_node: if passed, overrides the target node for creation
227 @type disks: list of {objects.Disk}
228 @param disks: the disks to create; if not specified, all the disks of the
230 @return: information about the created disks, to be used to call
232 @raise errors.OpPrereqError: in case of error
235 info = GetInstanceInfoText(instance)
236 if target_node is None:
237 pnode = instance.primary_node
238 all_nodes = instance.all_nodes
244 disks = instance.disks
246 if instance.disk_template in constants.DTS_FILEBASED:
247 file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
248 result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
250 result.Raise("Failed to create directory '%s' on"
251 " node %s" % (file_storage_dir, pnode))
254 for idx, device in enumerate(disks):
255 if to_skip and idx in to_skip:
257 logging.info("Creating disk %s for instance '%s'", idx, instance.name)
258 for node in all_nodes:
259 f_create = node == pnode
261 _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
262 disks_created.append((node, device))
263 except errors.DeviceCreationError, e:
264 logging.warning("Creating disk %s for instance '%s' failed",
266 disks_created.extend(e.created_devices)
267 _UndoCreateDisks(lu, disks_created)
268 raise errors.OpExecError(e.message)
272 def ComputeDiskSizePerVG(disk_template, disks):
273 """Compute disk size requirements in the volume group
276 def _compute(disks, payload):
277 """Universal algorithm.
282 vgs[disk[constants.IDISK_VG]] = \
283 vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
287 # Required free disk space as a function of disk and swap space
289 constants.DT_DISKLESS: {},
290 constants.DT_PLAIN: _compute(disks, 0),
291 # 128 MB are added for drbd metadata for each disk
292 constants.DT_DRBD8: _compute(disks, constants.DRBD_META_SIZE),
293 constants.DT_FILE: {},
294 constants.DT_SHARED_FILE: {},
297 if disk_template not in req_size_dict:
298 raise errors.ProgrammerError("Disk template '%s' size requirement"
299 " is unknown" % disk_template)
301 return req_size_dict[disk_template]
304 def ComputeDisks(op, default_vg):
305 """Computes the instance disks.
307 @param op: The instance opcode
308 @param default_vg: The default_vg to assume
310 @return: The computed disks
314 for disk in op.disks:
315 mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
316 if mode not in constants.DISK_ACCESS_SET:
317 raise errors.OpPrereqError("Invalid disk access mode '%s'" %
318 mode, errors.ECODE_INVAL)
319 size = disk.get(constants.IDISK_SIZE, None)
321 raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
324 except (TypeError, ValueError):
325 raise errors.OpPrereqError("Invalid disk size '%s'" % size,
328 ext_provider = disk.get(constants.IDISK_PROVIDER, None)
329 if ext_provider and op.disk_template != constants.DT_EXT:
330 raise errors.OpPrereqError("The '%s' option is only valid for the %s"
331 " disk template, not %s" %
332 (constants.IDISK_PROVIDER, constants.DT_EXT,
333 op.disk_template), errors.ECODE_INVAL)
335 data_vg = disk.get(constants.IDISK_VG, default_vg)
336 name = disk.get(constants.IDISK_NAME, None)
337 if name is not None and name.lower() == constants.VALUE_NONE:
340 constants.IDISK_SIZE: size,
341 constants.IDISK_MODE: mode,
342 constants.IDISK_VG: data_vg,
343 constants.IDISK_NAME: name,
346 if constants.IDISK_METAVG in disk:
347 new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG]
348 if constants.IDISK_ADOPT in disk:
349 new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
351 # For extstorage, demand the `provider' option and add any
352 # additional parameters (ext-params) to the dict
353 if op.disk_template == constants.DT_EXT:
355 new_disk[constants.IDISK_PROVIDER] = ext_provider
357 if key not in constants.IDISK_PARAMS:
358 new_disk[key] = disk[key]
360 raise errors.OpPrereqError("Missing provider for template '%s'" %
361 constants.DT_EXT, errors.ECODE_INVAL)
363 disks.append(new_disk)
368 def CheckRADOSFreeSpace():
369 """Compute disk size requirements inside the RADOS cluster.
372 # For the RADOS cluster we assume there is always enough space.
376 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
377 iv_name, p_minor, s_minor):
378 """Generate a drbd8 device complete with its children.
381 assert len(vgnames) == len(names) == 2
382 port = lu.cfg.AllocatePort()
383 shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
385 dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
386 logical_id=(vgnames[0], names[0]),
388 dev_data.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
389 dev_meta = objects.Disk(dev_type=constants.LD_LV,
390 size=constants.DRBD_META_SIZE,
391 logical_id=(vgnames[1], names[1]),
393 dev_meta.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
394 drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
395 logical_id=(primary, secondary, port,
398 children=[dev_data, dev_meta],
399 iv_name=iv_name, params={})
400 drbd_dev.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
404 def GenerateDiskTemplate(
405 lu, template_name, instance_name, primary_node, secondary_nodes,
406 disk_info, file_storage_dir, file_driver, base_index,
407 feedback_fn, full_disk_params, _req_file_storage=opcodes.RequireFileStorage,
408 _req_shr_file_storage=opcodes.RequireSharedFileStorage):
409 """Generate the entire disk layout for a given template type.
412 vgname = lu.cfg.GetVGName()
413 disk_count = len(disk_info)
416 if template_name == constants.DT_DISKLESS:
418 elif template_name == constants.DT_DRBD8:
419 if len(secondary_nodes) != 1:
420 raise errors.ProgrammerError("Wrong template configuration")
421 remote_node = secondary_nodes[0]
422 minors = lu.cfg.AllocateDRBDMinor(
423 [primary_node, remote_node] * len(disk_info), instance_name)
425 (drbd_params, _, _) = objects.Disk.ComputeLDParams(template_name,
427 drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
430 for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
431 for i in range(disk_count)]):
432 names.append(lv_prefix + "_data")
433 names.append(lv_prefix + "_meta")
434 for idx, disk in enumerate(disk_info):
435 disk_index = idx + base_index
436 data_vg = disk.get(constants.IDISK_VG, vgname)
437 meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
438 disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
439 disk[constants.IDISK_SIZE],
441 names[idx * 2:idx * 2 + 2],
442 "disk/%d" % disk_index,
443 minors[idx * 2], minors[idx * 2 + 1])
444 disk_dev.mode = disk[constants.IDISK_MODE]
445 disk_dev.name = disk.get(constants.IDISK_NAME, None)
446 disks.append(disk_dev)
449 raise errors.ProgrammerError("Wrong template configuration")
451 if template_name == constants.DT_FILE:
453 elif template_name == constants.DT_SHARED_FILE:
454 _req_shr_file_storage()
456 name_prefix = _DISK_TEMPLATE_NAME_PREFIX.get(template_name, None)
457 if name_prefix is None:
460 names = _GenerateUniqueNames(lu, ["%s.disk%s" %
461 (name_prefix, base_index + i)
462 for i in range(disk_count)])
464 if template_name == constants.DT_PLAIN:
466 def logical_id_fn(idx, _, disk):
467 vg = disk.get(constants.IDISK_VG, vgname)
468 return (vg, names[idx])
470 elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
472 lambda _, disk_index, disk: (file_driver,
473 "%s/disk%d" % (file_storage_dir,
475 elif template_name == constants.DT_BLOCK:
477 lambda idx, disk_index, disk: (constants.BLOCKDEV_DRIVER_MANUAL,
478 disk[constants.IDISK_ADOPT])
479 elif template_name == constants.DT_RBD:
480 logical_id_fn = lambda idx, _, disk: ("rbd", names[idx])
481 elif template_name == constants.DT_EXT:
482 def logical_id_fn(idx, _, disk):
483 provider = disk.get(constants.IDISK_PROVIDER, None)
485 raise errors.ProgrammerError("Disk template is %s, but '%s' is"
486 " not found", constants.DT_EXT,
487 constants.IDISK_PROVIDER)
488 return (provider, names[idx])
490 raise errors.ProgrammerError("Unknown disk template '%s'" % template_name)
492 dev_type = _DISK_TEMPLATE_DEVICE_TYPE[template_name]
494 for idx, disk in enumerate(disk_info):
496 # Only for the Ext template add disk_info to params
497 if template_name == constants.DT_EXT:
498 params[constants.IDISK_PROVIDER] = disk[constants.IDISK_PROVIDER]
500 if key not in constants.IDISK_PARAMS:
501 params[key] = disk[key]
502 disk_index = idx + base_index
503 size = disk[constants.IDISK_SIZE]
504 feedback_fn("* disk %s, size %s" %
505 (disk_index, utils.FormatUnit(size, "h")))
506 disk_dev = objects.Disk(dev_type=dev_type, size=size,
507 logical_id=logical_id_fn(idx, disk_index, disk),
508 iv_name="disk/%d" % disk_index,
509 mode=disk[constants.IDISK_MODE],
511 disk_dev.name = disk.get(constants.IDISK_NAME, None)
512 disk_dev.uuid = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
513 disks.append(disk_dev)
518 class LUInstanceRecreateDisks(LogicalUnit):
519 """Recreate an instance's missing disks.
522 HPATH = "instance-recreate-disks"
523 HTYPE = constants.HTYPE_INSTANCE
526 _MODIFYABLE = compat.UniqueFrozenset([
527 constants.IDISK_SIZE,
528 constants.IDISK_MODE,
531 # New or changed disk parameters may have different semantics
532 assert constants.IDISK_PARAMS == (_MODIFYABLE | frozenset([
533 constants.IDISK_ADOPT,
535 # TODO: Implement support changing VG while recreating
537 constants.IDISK_METAVG,
538 constants.IDISK_PROVIDER,
539 constants.IDISK_NAME,
542 def _RunAllocator(self):
543 """Run the allocator based on input opcode.
546 be_full = self.cfg.GetClusterInfo().FillBE(self.instance)
549 # The allocator should actually run in "relocate" mode, but current
550 # allocators don't support relocating all the nodes of an instance at
551 # the same time. As a workaround we use "allocate" mode, but this is
552 # suboptimal for two reasons:
553 # - The instance name passed to the allocator is present in the list of
554 # existing instances, so there could be a conflict within the
555 # internal structures of the allocator. This doesn't happen with the
556 # current allocators, but it's a liability.
557 # - The allocator counts the resources used by the instance twice: once
558 # because the instance exists already, and once because it tries to
559 # allocate a new instance.
560 # The allocator could choose some of the nodes on which the instance is
561 # running, but that's not a problem. If the instance nodes are broken,
562 # they should be already be marked as drained or offline, and hence
563 # skipped by the allocator. If instance disks have been lost for other
564 # reasons, then recreating the disks on the same nodes should be fine.
565 disk_template = self.instance.disk_template
566 spindle_use = be_full[constants.BE_SPINDLE_USE]
567 req = iallocator.IAReqInstanceAlloc(name=self.op.instance_name,
568 disk_template=disk_template,
569 tags=list(self.instance.GetTags()),
572 vcpus=be_full[constants.BE_VCPUS],
573 memory=be_full[constants.BE_MAXMEM],
574 spindle_use=spindle_use,
575 disks=[{constants.IDISK_SIZE: d.size,
576 constants.IDISK_MODE: d.mode}
577 for d in self.instance.disks],
578 hypervisor=self.instance.hypervisor,
580 ial = iallocator.IAllocator(self.cfg, self.rpc, req)
582 ial.Run(self.op.iallocator)
584 assert req.RequiredNodes() == len(self.instance.all_nodes)
587 raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
588 " %s" % (self.op.iallocator, ial.info),
591 self.op.nodes = ial.result
592 self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
593 self.op.instance_name, self.op.iallocator,
594 utils.CommaJoin(ial.result))
596 def CheckArguments(self):
597 if self.op.disks and ht.TNonNegativeInt(self.op.disks[0]):
598 # Normalize and convert deprecated list of disk indices
599 self.op.disks = [(idx, {}) for idx in sorted(frozenset(self.op.disks))]
601 duplicates = utils.FindDuplicates(map(compat.fst, self.op.disks))
603 raise errors.OpPrereqError("Some disks have been specified more than"
604 " once: %s" % utils.CommaJoin(duplicates),
607 # We don't want _CheckIAllocatorOrNode selecting the default iallocator
608 # when neither iallocator nor nodes are specified
609 if self.op.iallocator or self.op.nodes:
610 CheckIAllocatorOrNode(self, "iallocator", "nodes")
612 for (idx, params) in self.op.disks:
613 utils.ForceDictType(params, constants.IDISK_PARAMS_TYPES)
614 unsupported = frozenset(params.keys()) - self._MODIFYABLE
616 raise errors.OpPrereqError("Parameters for disk %s try to change"
617 " unmodifyable parameter(s): %s" %
618 (idx, utils.CommaJoin(unsupported)),
621 def ExpandNames(self):
622 self._ExpandAndLockInstance()
623 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
626 self.op.nodes = [ExpandNodeName(self.cfg, n) for n in self.op.nodes]
627 self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
629 self.needed_locks[locking.LEVEL_NODE] = []
630 if self.op.iallocator:
631 # iallocator will select a new node in the same group
632 self.needed_locks[locking.LEVEL_NODEGROUP] = []
633 self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
635 self.needed_locks[locking.LEVEL_NODE_RES] = []
637 def DeclareLocks(self, level):
638 if level == locking.LEVEL_NODEGROUP:
639 assert self.op.iallocator is not None
640 assert not self.op.nodes
641 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
642 self.share_locks[locking.LEVEL_NODEGROUP] = 1
643 # Lock the primary group used by the instance optimistically; this
644 # requires going via the node before it's locked, requiring
645 # verification later on
646 self.needed_locks[locking.LEVEL_NODEGROUP] = \
647 self.cfg.GetInstanceNodeGroups(self.op.instance_name, primary_only=True)
649 elif level == locking.LEVEL_NODE:
650 # If an allocator is used, then we lock all the nodes in the current
651 # instance group, as we don't know yet which ones will be selected;
652 # if we replace the nodes without using an allocator, locks are
653 # already declared in ExpandNames; otherwise, we need to lock all the
654 # instance nodes for disk re-creation
655 if self.op.iallocator:
656 assert not self.op.nodes
657 assert not self.needed_locks[locking.LEVEL_NODE]
658 assert len(self.owned_locks(locking.LEVEL_NODEGROUP)) == 1
660 # Lock member nodes of the group of the primary node
661 for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP):
662 self.needed_locks[locking.LEVEL_NODE].extend(
663 self.cfg.GetNodeGroup(group_uuid).members)
665 assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
666 elif not self.op.nodes:
667 self._LockInstancesNodes(primary_only=False)
668 elif level == locking.LEVEL_NODE_RES:
670 self.needed_locks[locking.LEVEL_NODE_RES] = \
671 CopyLockList(self.needed_locks[locking.LEVEL_NODE])
673 def BuildHooksEnv(self):
676 This runs on master, primary and secondary nodes of the instance.
679 return BuildInstanceHookEnvByObject(self, self.instance)
681 def BuildHooksNodes(self):
682 """Build hooks nodes.
685 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
688 def CheckPrereq(self):
689 """Check prerequisites.
691 This checks that the instance is in the cluster and is not running.
694 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
695 assert instance is not None, \
696 "Cannot retrieve locked instance %s" % self.op.instance_name
698 if len(self.op.nodes) != len(instance.all_nodes):
699 raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
700 " %d replacement nodes were specified" %
701 (instance.name, len(instance.all_nodes),
704 assert instance.disk_template != constants.DT_DRBD8 or \
705 len(self.op.nodes) == 2
706 assert instance.disk_template != constants.DT_PLAIN or \
707 len(self.op.nodes) == 1
708 primary_node = self.op.nodes[0]
710 primary_node = instance.primary_node
711 if not self.op.iallocator:
712 CheckNodeOnline(self, primary_node)
714 if instance.disk_template == constants.DT_DISKLESS:
715 raise errors.OpPrereqError("Instance '%s' has no disks" %
716 self.op.instance_name, errors.ECODE_INVAL)
718 # Verify if node group locks are still correct
719 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
721 # Node group locks are acquired only for the primary node (and only
722 # when the allocator is used)
723 CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups,
726 # if we replace nodes *and* the old primary is offline, we don't
727 # check the instance state
728 old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
729 if not ((self.op.iallocator or self.op.nodes) and old_pnode.offline):
730 CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
731 msg="cannot recreate disks")
734 self.disks = dict(self.op.disks)
736 self.disks = dict((idx, {}) for idx in range(len(instance.disks)))
738 maxidx = max(self.disks.keys())
739 if maxidx >= len(instance.disks):
740 raise errors.OpPrereqError("Invalid disk index '%s'" % maxidx,
743 if ((self.op.nodes or self.op.iallocator) and
744 sorted(self.disks.keys()) != range(len(instance.disks))):
745 raise errors.OpPrereqError("Can't recreate disks partially and"
746 " change the nodes at the same time",
749 self.instance = instance
751 if self.op.iallocator:
753 # Release unneeded node and node resource locks
754 ReleaseLocks(self, locking.LEVEL_NODE, keep=self.op.nodes)
755 ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=self.op.nodes)
756 ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
758 assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
760 def Exec(self, feedback_fn):
761 """Recreate the disks.
764 instance = self.instance
766 assert (self.owned_locks(locking.LEVEL_NODE) ==
767 self.owned_locks(locking.LEVEL_NODE_RES))
770 mods = [] # keeps track of needed changes
772 for idx, disk in enumerate(instance.disks):
774 changes = self.disks[idx]
776 # Disk should not be recreated
780 # update secondaries for disks, if needed
781 if self.op.nodes and disk.dev_type == constants.LD_DRBD8:
782 # need to update the nodes and minors
783 assert len(self.op.nodes) == 2
784 assert len(disk.logical_id) == 6 # otherwise disk internals
786 (_, _, old_port, _, _, old_secret) = disk.logical_id
787 new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
788 new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
789 new_minors[0], new_minors[1], old_secret)
790 assert len(disk.logical_id) == len(new_id)
794 mods.append((idx, new_id, changes))
796 # now that we have passed all asserts above, we can apply the mods
797 # in a single run (to avoid partial changes)
798 for idx, new_id, changes in mods:
799 disk = instance.disks[idx]
800 if new_id is not None:
801 assert disk.dev_type == constants.LD_DRBD8
802 disk.logical_id = new_id
804 disk.Update(size=changes.get(constants.IDISK_SIZE, None),
805 mode=changes.get(constants.IDISK_MODE, None))
807 # change primary node, if needed
809 instance.primary_node = self.op.nodes[0]
810 self.LogWarning("Changing the instance's nodes, you will have to"
811 " remove any disks left on the older nodes manually")
814 self.cfg.Update(instance, feedback_fn)
816 # All touched nodes must be locked
817 mylocks = self.owned_locks(locking.LEVEL_NODE)
818 assert mylocks.issuperset(frozenset(instance.all_nodes))
819 new_disks = CreateDisks(self, instance, to_skip=to_skip)
821 # TODO: Release node locks before wiping, or explain why it's not possible
822 if self.cfg.GetClusterInfo().prealloc_wipe_disks:
823 wipedisks = [(idx, disk, 0)
824 for (idx, disk) in enumerate(instance.disks)
825 if idx not in to_skip]
826 WipeOrCleanupDisks(self, instance, disks=wipedisks, cleanup=new_disks)
829 def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
830 """Checks if nodes have enough free disk space in the specified VG.
832 This function checks if all given nodes have the needed amount of
833 free disk. In case any node has less disk or we cannot get the
834 information from the node, this function raises an OpPrereqError
837 @type lu: C{LogicalUnit}
838 @param lu: a logical unit from which we get configuration data
839 @type nodenames: C{list}
840 @param nodenames: the list of node names to check
842 @param vg: the volume group to check
843 @type requested: C{int}
844 @param requested: the amount of disk in MiB to check for
845 @raise errors.OpPrereqError: if the node doesn't have enough disk,
846 or we cannot check the node
849 es_flags = rpc.GetExclusiveStorageForNodeNames(lu.cfg, nodenames)
850 nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None, es_flags)
851 for node in nodenames:
852 info = nodeinfo[node]
853 info.Raise("Cannot get current information from node %s" % node,
854 prereq=True, ecode=errors.ECODE_ENVIRON)
855 (_, (vg_info, ), _) = info.payload
856 vg_free = vg_info.get("vg_free", None)
857 if not isinstance(vg_free, int):
858 raise errors.OpPrereqError("Can't compute free disk space on node"
859 " %s for vg %s, result was '%s'" %
860 (node, vg, vg_free), errors.ECODE_ENVIRON)
861 if requested > vg_free:
862 raise errors.OpPrereqError("Not enough disk space on target node %s"
863 " vg %s: required %d MiB, available %d MiB" %
864 (node, vg, requested, vg_free),
868 def CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
869 """Checks if nodes have enough free disk space in all the VGs.
871 This function checks if all given nodes have the needed amount of
872 free disk. In case any node has less disk or we cannot get the
873 information from the node, this function raises an OpPrereqError
876 @type lu: C{LogicalUnit}
877 @param lu: a logical unit from which we get configuration data
878 @type nodenames: C{list}
879 @param nodenames: the list of node names to check
880 @type req_sizes: C{dict}
881 @param req_sizes: the hash of vg and corresponding amount of disk in
883 @raise errors.OpPrereqError: if the node doesn't have enough disk,
884 or we cannot check the node
887 for vg, req_size in req_sizes.items():
888 _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
891 def _DiskSizeInBytesToMebibytes(lu, size):
892 """Converts a disk size in bytes to mebibytes.
894 Warns and rounds up if the size isn't an even multiple of 1 MiB.
897 (mib, remainder) = divmod(size, 1024 * 1024)
900 lu.LogWarning("Disk size is not an even multiple of 1 MiB; rounding up"
901 " to not overwrite existing data (%s bytes will not be"
902 " wiped)", (1024 * 1024) - remainder)
908 def _CalcEta(time_taken, written, total_size):
909 """Calculates the ETA based on size written and total size.
911 @param time_taken: The time taken so far
912 @param written: amount written so far
913 @param total_size: The total size of data to be written
914 @return: The remaining time in seconds
917 avg_time = time_taken / float(written)
918 return (total_size - written) * avg_time
921 def WipeDisks(lu, instance, disks=None):
922 """Wipes instance disks.
924 @type lu: L{LogicalUnit}
925 @param lu: the logical unit on whose behalf we execute
926 @type instance: L{objects.Instance}
927 @param instance: the instance whose disks we should create
928 @type disks: None or list of tuple of (number, L{objects.Disk}, number)
929 @param disks: Disk details; tuple contains disk index, disk object and the
933 node = instance.primary_node
936 disks = [(idx, disk, 0)
937 for (idx, disk) in enumerate(instance.disks)]
939 for (_, device, _) in disks:
940 lu.cfg.SetDiskID(device, node)
942 logging.info("Pausing synchronization of disks of instance '%s'",
944 result = lu.rpc.call_blockdev_pause_resume_sync(node,
945 (map(compat.snd, disks),
948 result.Raise("Failed to pause disk synchronization on node '%s'" % node)
950 for idx, success in enumerate(result.payload):
952 logging.warn("Pausing synchronization of disk %s of instance '%s'"
953 " failed", idx, instance.name)
956 for (idx, device, offset) in disks:
957 # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
958 # MAX_WIPE_CHUNK at max. Truncating to integer to avoid rounding errors.
960 int(min(constants.MAX_WIPE_CHUNK,
961 device.size / 100.0 * constants.MIN_WIPE_CHUNK_PERCENT))
965 start_time = time.time()
970 info_text = (" (from %s to %s)" %
971 (utils.FormatUnit(offset, "h"),
972 utils.FormatUnit(size, "h")))
974 lu.LogInfo("* Wiping disk %s%s", idx, info_text)
976 logging.info("Wiping disk %d for instance %s on node %s using"
977 " chunk size %s", idx, instance.name, node, wipe_chunk_size)
980 wipe_size = min(wipe_chunk_size, size - offset)
982 logging.debug("Wiping disk %d, offset %s, chunk %s",
983 idx, offset, wipe_size)
985 result = lu.rpc.call_blockdev_wipe(node, (device, instance), offset,
987 result.Raise("Could not wipe disk %d at offset %d for size %d" %
988 (idx, offset, wipe_size))
992 if now - last_output >= 60:
993 eta = _CalcEta(now - start_time, offset, size)
994 lu.LogInfo(" - done: %.1f%% ETA: %s",
995 offset / float(size) * 100, utils.FormatSeconds(eta))
998 logging.info("Resuming synchronization of disks for instance '%s'",
1001 result = lu.rpc.call_blockdev_pause_resume_sync(node,
1002 (map(compat.snd, disks),
1007 lu.LogWarning("Failed to resume disk synchronization on node '%s': %s",
1008 node, result.fail_msg)
1010 for idx, success in enumerate(result.payload):
1012 lu.LogWarning("Resuming synchronization of disk %s of instance '%s'"
1013 " failed", idx, instance.name)
1016 def WipeOrCleanupDisks(lu, instance, disks=None, cleanup=None):
1017 """Wrapper for L{WipeDisks} that handles errors.
1019 @type lu: L{LogicalUnit}
1020 @param lu: the logical unit on whose behalf we execute
1021 @type instance: L{objects.Instance}
1022 @param instance: the instance whose disks we should wipe
1023 @param disks: see L{WipeDisks}
1024 @param cleanup: the result returned by L{CreateDisks}, used for cleanup in
1026 @raise errors.OpPrereqError: in case of failure
1030 WipeDisks(lu, instance, disks=disks)
1031 except errors.OpExecError:
1032 logging.warning("Wiping disks for instance '%s' failed",
1034 _UndoCreateDisks(lu, cleanup)
1038 def ExpandCheckDisks(instance, disks):
1039 """Return the instance disks selected by the disks list
1041 @type disks: list of L{objects.Disk} or None
1042 @param disks: selected disks
1043 @rtype: list of L{objects.Disk}
1044 @return: selected instance disks to act on
1048 return instance.disks
1050 if not set(disks).issubset(instance.disks):
1051 raise errors.ProgrammerError("Can only act on disks belonging to the"
1052 " target instance: expected a subset of %r,"
1053 " got %r" % (instance.disks, disks))
1057 def WaitForSync(lu, instance, disks=None, oneshot=False):
1058 """Sleep and poll for an instance's disk to sync.
1061 if not instance.disks or disks is not None and not disks:
1064 disks = ExpandCheckDisks(instance, disks)
1067 lu.LogInfo("Waiting for instance %s to sync disks", instance.name)
1069 node = instance.primary_node
1072 lu.cfg.SetDiskID(dev, node)
1074 # TODO: Convert to utils.Retry
1077 degr_retries = 10 # in seconds, as we sleep 1 second each time
1081 cumul_degraded = False
1082 rstats = lu.rpc.call_blockdev_getmirrorstatus(node, (disks, instance))
1083 msg = rstats.fail_msg
1085 lu.LogWarning("Can't get any data from node %s: %s", node, msg)
1088 raise errors.RemoteError("Can't contact node %s for mirror data,"
1089 " aborting." % node)
1092 rstats = rstats.payload
1094 for i, mstat in enumerate(rstats):
1096 lu.LogWarning("Can't compute data for node %s/%s",
1097 node, disks[i].iv_name)
1100 cumul_degraded = (cumul_degraded or
1101 (mstat.is_degraded and mstat.sync_percent is None))
1102 if mstat.sync_percent is not None:
1104 if mstat.estimated_time is not None:
1105 rem_time = ("%s remaining (estimated)" %
1106 utils.FormatSeconds(mstat.estimated_time))
1107 max_time = mstat.estimated_time
1109 rem_time = "no time estimate"
1110 lu.LogInfo("- device %s: %5.2f%% done, %s",
1111 disks[i].iv_name, mstat.sync_percent, rem_time)
1113 # if we're done but degraded, let's do a few small retries, to
1114 # make sure we see a stable and not transient situation; therefore
1115 # we force restart of the loop
1116 if (done or oneshot) and cumul_degraded and degr_retries > 0:
1117 logging.info("Degraded disks found, %d retries left", degr_retries)
1125 time.sleep(min(60, max_time))
1128 lu.LogInfo("Instance %s's disks are in sync", instance.name)
1130 return not cumul_degraded
1133 def ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
1134 """Shutdown block devices of an instance.
1136 This does the shutdown on all nodes of the instance.
1138 If the ignore_primary is false, errors on the primary node are
1143 disks = ExpandCheckDisks(instance, disks)
1146 for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
1147 lu.cfg.SetDiskID(top_disk, node)
1148 result = lu.rpc.call_blockdev_shutdown(node, (top_disk, instance))
1149 msg = result.fail_msg
1151 lu.LogWarning("Could not shutdown block device %s on node %s: %s",
1152 disk.iv_name, node, msg)
1153 if ((node == instance.primary_node and not ignore_primary) or
1154 (node != instance.primary_node and not result.offline)):
1159 def _SafeShutdownInstanceDisks(lu, instance, disks=None):
1160 """Shutdown block devices of an instance.
1162 This function checks if an instance is running, before calling
1163 _ShutdownInstanceDisks.
1166 CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
1167 ShutdownInstanceDisks(lu, instance, disks=disks)
1170 def AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
1172 """Prepare the block devices for an instance.
1174 This sets up the block devices on all nodes.
1176 @type lu: L{LogicalUnit}
1177 @param lu: the logical unit on whose behalf we execute
1178 @type instance: L{objects.Instance}
1179 @param instance: the instance for whose disks we assemble
1180 @type disks: list of L{objects.Disk} or None
1181 @param disks: which disks to assemble (or all, if None)
1182 @type ignore_secondaries: boolean
1183 @param ignore_secondaries: if true, errors on secondary nodes
1184 won't result in an error return from the function
1185 @type ignore_size: boolean
1186 @param ignore_size: if true, the current known size of the disk
1187 will not be used during the disk activation, useful for cases
1188 when the size is wrong
1189 @return: False if the operation failed, otherwise a list of
1190 (host, instance_visible_name, node_visible_name)
1191 with the mapping from node devices to instance devices
1196 iname = instance.name
1197 disks = ExpandCheckDisks(instance, disks)
1199 # With the two passes mechanism we try to reduce the window of
1200 # opportunity for the race condition of switching DRBD to primary
1201 # before handshaking occured, but we do not eliminate it
1203 # The proper fix would be to wait (with some limits) until the
1204 # connection has been made and drbd transitions from WFConnection
1205 # into any other network-connected state (Connected, SyncTarget,
1208 # 1st pass, assemble on all nodes in secondary mode
1209 for idx, inst_disk in enumerate(disks):
1210 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
1212 node_disk = node_disk.Copy()
1213 node_disk.UnsetSize()
1214 lu.cfg.SetDiskID(node_disk, node)
1215 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
1217 msg = result.fail_msg
1219 is_offline_secondary = (node in instance.secondary_nodes and
1221 lu.LogWarning("Could not prepare block device %s on node %s"
1222 " (is_primary=False, pass=1): %s",
1223 inst_disk.iv_name, node, msg)
1224 if not (ignore_secondaries or is_offline_secondary):
1227 # FIXME: race condition on drbd migration to primary
1229 # 2nd pass, do only the primary node
1230 for idx, inst_disk in enumerate(disks):
1233 for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
1234 if node != instance.primary_node:
1237 node_disk = node_disk.Copy()
1238 node_disk.UnsetSize()
1239 lu.cfg.SetDiskID(node_disk, node)
1240 result = lu.rpc.call_blockdev_assemble(node, (node_disk, instance), iname,
1242 msg = result.fail_msg
1244 lu.LogWarning("Could not prepare block device %s on node %s"
1245 " (is_primary=True, pass=2): %s",
1246 inst_disk.iv_name, node, msg)
1249 dev_path = result.payload
1251 device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
1253 # leave the disks configured for the primary node
1254 # this is a workaround that would be fixed better by
1255 # improving the logical/physical id handling
1257 lu.cfg.SetDiskID(disk, instance.primary_node)
1259 return disks_ok, device_info
1262 def StartInstanceDisks(lu, instance, force):
1263 """Start the disks of an instance.
1266 disks_ok, _ = AssembleInstanceDisks(lu, instance,
1267 ignore_secondaries=force)
1269 ShutdownInstanceDisks(lu, instance)
1270 if force is not None and not force:
1272 hint=("If the message above refers to a secondary node,"
1273 " you can retry the operation using '--force'"))
1274 raise errors.OpExecError("Disk consistency error")
1277 class LUInstanceGrowDisk(LogicalUnit):
1278 """Grow a disk of an instance.
1282 HTYPE = constants.HTYPE_INSTANCE
1285 def ExpandNames(self):
1286 self._ExpandAndLockInstance()
1287 self.needed_locks[locking.LEVEL_NODE] = []
1288 self.needed_locks[locking.LEVEL_NODE_RES] = []
1289 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
1290 self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
1292 def DeclareLocks(self, level):
1293 if level == locking.LEVEL_NODE:
1294 self._LockInstancesNodes()
1295 elif level == locking.LEVEL_NODE_RES:
1297 self.needed_locks[locking.LEVEL_NODE_RES] = \
1298 CopyLockList(self.needed_locks[locking.LEVEL_NODE])
1300 def BuildHooksEnv(self):
1303 This runs on the master, the primary and all the secondaries.
1307 "DISK": self.op.disk,
1308 "AMOUNT": self.op.amount,
1309 "ABSOLUTE": self.op.absolute,
1311 env.update(BuildInstanceHookEnvByObject(self, self.instance))
1314 def BuildHooksNodes(self):
1315 """Build hooks nodes.
1318 nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
1321 def CheckPrereq(self):
1322 """Check prerequisites.
1324 This checks that the instance is in the cluster.
1327 instance = self.cfg.GetInstanceInfo(self.op.instance_name)
1328 assert instance is not None, \
1329 "Cannot retrieve locked instance %s" % self.op.instance_name
1330 nodenames = list(instance.all_nodes)
1331 for node in nodenames:
1332 CheckNodeOnline(self, node)
1334 self.instance = instance
1336 if instance.disk_template not in constants.DTS_GROWABLE:
1337 raise errors.OpPrereqError("Instance's disk layout does not support"
1338 " growing", errors.ECODE_INVAL)
1340 self.disk = instance.FindDisk(self.op.disk)
1342 if self.op.absolute:
1343 self.target = self.op.amount
1344 self.delta = self.target - self.disk.size
1346 raise errors.OpPrereqError("Requested size (%s) is smaller than "
1347 "current disk size (%s)" %
1348 (utils.FormatUnit(self.target, "h"),
1349 utils.FormatUnit(self.disk.size, "h")),
1352 self.delta = self.op.amount
1353 self.target = self.disk.size + self.delta
1355 raise errors.OpPrereqError("Requested increment (%s) is negative" %
1356 utils.FormatUnit(self.delta, "h"),
1359 self._CheckDiskSpace(nodenames, self.disk.ComputeGrowth(self.delta))
1361 def _CheckDiskSpace(self, nodenames, req_vgspace):
1362 template = self.instance.disk_template
1363 if template not in (constants.DTS_NO_FREE_SPACE_CHECK):
1364 # TODO: check the free disk space for file, when that feature will be
1366 nodes = map(self.cfg.GetNodeInfo, nodenames)
1367 es_nodes = filter(lambda n: IsExclusiveStorageEnabledNode(self.cfg, n),
1370 # With exclusive storage we need to something smarter than just looking
1371 # at free space; for now, let's simply abort the operation.
1372 raise errors.OpPrereqError("Cannot grow disks when exclusive_storage"
1373 " is enabled", errors.ECODE_STATE)
1374 CheckNodesFreeDiskPerVG(self, nodenames, req_vgspace)
1376 def Exec(self, feedback_fn):
1377 """Execute disk grow.
1380 instance = self.instance
1383 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
1384 assert (self.owned_locks(locking.LEVEL_NODE) ==
1385 self.owned_locks(locking.LEVEL_NODE_RES))
1387 wipe_disks = self.cfg.GetClusterInfo().prealloc_wipe_disks
1389 disks_ok, _ = AssembleInstanceDisks(self, self.instance, disks=[disk])
1391 raise errors.OpExecError("Cannot activate block device to grow")
1393 feedback_fn("Growing disk %s of instance '%s' by %s to %s" %
1394 (self.op.disk, instance.name,
1395 utils.FormatUnit(self.delta, "h"),
1396 utils.FormatUnit(self.target, "h")))
1398 # First run all grow ops in dry-run mode
1399 for node in instance.all_nodes:
1400 self.cfg.SetDiskID(disk, node)
1401 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
1403 result.Raise("Dry-run grow request failed to node %s" % node)
1406 # Get disk size from primary node for wiping
1407 result = self.rpc.call_blockdev_getsize(instance.primary_node, [disk])
1408 result.Raise("Failed to retrieve disk size from node '%s'" %
1409 instance.primary_node)
1411 (disk_size_in_bytes, ) = result.payload
1413 if disk_size_in_bytes is None:
1414 raise errors.OpExecError("Failed to retrieve disk size from primary"
1415 " node '%s'" % instance.primary_node)
1417 old_disk_size = _DiskSizeInBytesToMebibytes(self, disk_size_in_bytes)
1419 assert old_disk_size >= disk.size, \
1420 ("Retrieved disk size too small (got %s, should be at least %s)" %
1421 (old_disk_size, disk.size))
1423 old_disk_size = None
1425 # We know that (as far as we can test) operations across different
1426 # nodes will succeed, time to run it for real on the backing storage
1427 for node in instance.all_nodes:
1428 self.cfg.SetDiskID(disk, node)
1429 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
1431 result.Raise("Grow request failed to node %s" % node)
1433 # And now execute it for logical storage, on the primary node
1434 node = instance.primary_node
1435 self.cfg.SetDiskID(disk, node)
1436 result = self.rpc.call_blockdev_grow(node, (disk, instance), self.delta,
1438 result.Raise("Grow request failed to node %s" % node)
1440 disk.RecordGrow(self.delta)
1441 self.cfg.Update(instance, feedback_fn)
1443 # Changes have been recorded, release node lock
1444 ReleaseLocks(self, locking.LEVEL_NODE)
1446 # Downgrade lock while waiting for sync
1447 self.glm.downgrade(locking.LEVEL_INSTANCE)
1449 assert wipe_disks ^ (old_disk_size is None)
1452 assert instance.disks[self.op.disk] == disk
1454 # Wipe newly added disk space
1455 WipeDisks(self, instance,
1456 disks=[(self.op.disk, disk, old_disk_size)])
1458 if self.op.wait_for_sync:
1459 disk_abort = not WaitForSync(self, instance, disks=[disk])
1461 self.LogWarning("Disk syncing has not returned a good status; check"
1463 if instance.admin_state != constants.ADMINST_UP:
1464 _SafeShutdownInstanceDisks(self, instance, disks=[disk])
1465 elif instance.admin_state != constants.ADMINST_UP:
1466 self.LogWarning("Not shutting down the disk even if the instance is"
1467 " not supposed to be running because no wait for"
1468 " sync mode was requested")
1470 assert self.owned_locks(locking.LEVEL_NODE_RES)
1471 assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
1474 class LUInstanceReplaceDisks(LogicalUnit):
1475 """Replace the disks of an instance.
1478 HPATH = "mirrors-replace"
1479 HTYPE = constants.HTYPE_INSTANCE
1482 def CheckArguments(self):
1486 remote_node = self.op.remote_node
1487 ialloc = self.op.iallocator
1488 if self.op.mode == constants.REPLACE_DISK_CHG:
1489 if remote_node is None and ialloc is None:
1490 raise errors.OpPrereqError("When changing the secondary either an"
1491 " iallocator script must be used or the"
1492 " new node given", errors.ECODE_INVAL)
1494 CheckIAllocatorOrNode(self, "iallocator", "remote_node")
1496 elif remote_node is not None or ialloc is not None:
1497 # Not replacing the secondary
1498 raise errors.OpPrereqError("The iallocator and new node options can"
1499 " only be used when changing the"
1500 " secondary node", errors.ECODE_INVAL)
1502 def ExpandNames(self):
1503 self._ExpandAndLockInstance()
1505 assert locking.LEVEL_NODE not in self.needed_locks
1506 assert locking.LEVEL_NODE_RES not in self.needed_locks
1507 assert locking.LEVEL_NODEGROUP not in self.needed_locks
1509 assert self.op.iallocator is None or self.op.remote_node is None, \
1510 "Conflicting options"
1512 if self.op.remote_node is not None:
1513 self.op.remote_node = ExpandNodeName(self.cfg, self.op.remote_node)
1515 # Warning: do not remove the locking of the new secondary here
1516 # unless DRBD8.AddChildren is changed to work in parallel;
1517 # currently it doesn't since parallel invocations of
1518 # FindUnusedMinor will conflict
1519 self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
1520 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
1522 self.needed_locks[locking.LEVEL_NODE] = []
1523 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
1525 if self.op.iallocator is not None:
1526 # iallocator will select a new node in the same group
1527 self.needed_locks[locking.LEVEL_NODEGROUP] = []
1528 self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
1530 self.needed_locks[locking.LEVEL_NODE_RES] = []
1532 self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
1533 self.op.iallocator, self.op.remote_node,
1534 self.op.disks, self.op.early_release,
1535 self.op.ignore_ipolicy)
1537 self.tasklets = [self.replacer]
1539 def DeclareLocks(self, level):
1540 if level == locking.LEVEL_NODEGROUP:
1541 assert self.op.remote_node is None
1542 assert self.op.iallocator is not None
1543 assert not self.needed_locks[locking.LEVEL_NODEGROUP]
1545 self.share_locks[locking.LEVEL_NODEGROUP] = 1
1546 # Lock all groups used by instance optimistically; this requires going
1547 # via the node before it's locked, requiring verification later on
1548 self.needed_locks[locking.LEVEL_NODEGROUP] = \
1549 self.cfg.GetInstanceNodeGroups(self.op.instance_name)
1551 elif level == locking.LEVEL_NODE:
1552 if self.op.iallocator is not None:
1553 assert self.op.remote_node is None
1554 assert not self.needed_locks[locking.LEVEL_NODE]
1555 assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
1557 # Lock member nodes of all locked groups
1558 self.needed_locks[locking.LEVEL_NODE] = \
1560 for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
1561 for node_name in self.cfg.GetNodeGroup(group_uuid).members]
1563 assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
1565 self._LockInstancesNodes()
1567 elif level == locking.LEVEL_NODE_RES:
1569 self.needed_locks[locking.LEVEL_NODE_RES] = \
1570 self.needed_locks[locking.LEVEL_NODE]
1572 def BuildHooksEnv(self):
1575 This runs on the master, the primary and all the secondaries.
1578 instance = self.replacer.instance
1580 "MODE": self.op.mode,
1581 "NEW_SECONDARY": self.op.remote_node,
1582 "OLD_SECONDARY": instance.secondary_nodes[0],
1584 env.update(BuildInstanceHookEnvByObject(self, instance))
1587 def BuildHooksNodes(self):
1588 """Build hooks nodes.
1591 instance = self.replacer.instance
1593 self.cfg.GetMasterNode(),
1594 instance.primary_node,
1596 if self.op.remote_node is not None:
1597 nl.append(self.op.remote_node)
1600 def CheckPrereq(self):
1601 """Check prerequisites.
1604 assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
1605 self.op.iallocator is None)
1607 # Verify if node group locks are still correct
1608 owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
1610 CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
1612 return LogicalUnit.CheckPrereq(self)
1615 class LUInstanceActivateDisks(NoHooksLU):
1616 """Bring up an instance's disks.
1621 def ExpandNames(self):
1622 self._ExpandAndLockInstance()
1623 self.needed_locks[locking.LEVEL_NODE] = []
1624 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
1626 def DeclareLocks(self, level):
1627 if level == locking.LEVEL_NODE:
1628 self._LockInstancesNodes()
1630 def CheckPrereq(self):
1631 """Check prerequisites.
1633 This checks that the instance is in the cluster.
1636 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
1637 assert self.instance is not None, \
1638 "Cannot retrieve locked instance %s" % self.op.instance_name
1639 CheckNodeOnline(self, self.instance.primary_node)
1641 def Exec(self, feedback_fn):
1642 """Activate the disks.
1645 disks_ok, disks_info = \
1646 AssembleInstanceDisks(self, self.instance,
1647 ignore_size=self.op.ignore_size)
1649 raise errors.OpExecError("Cannot activate block devices")
1651 if self.op.wait_for_sync:
1652 if not WaitForSync(self, self.instance):
1653 raise errors.OpExecError("Some disks of the instance are degraded!")
1658 class LUInstanceDeactivateDisks(NoHooksLU):
1659 """Shutdown an instance's disks.
1664 def ExpandNames(self):
1665 self._ExpandAndLockInstance()
1666 self.needed_locks[locking.LEVEL_NODE] = []
1667 self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
1669 def DeclareLocks(self, level):
1670 if level == locking.LEVEL_NODE:
1671 self._LockInstancesNodes()
1673 def CheckPrereq(self):
1674 """Check prerequisites.
1676 This checks that the instance is in the cluster.
1679 self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
1680 assert self.instance is not None, \
1681 "Cannot retrieve locked instance %s" % self.op.instance_name
1683 def Exec(self, feedback_fn):
1684 """Deactivate the disks
1687 instance = self.instance
1689 ShutdownInstanceDisks(self, instance)
1691 _SafeShutdownInstanceDisks(self, instance)
1694 def _CheckDiskConsistencyInner(lu, instance, dev, node, on_primary,
1696 """Check that mirrors are not degraded.
1698 @attention: The device has to be annotated already.
1700 The ldisk parameter, if True, will change the test from the
1701 is_degraded attribute (which represents overall non-ok status for
1702 the device(s)) to the ldisk (representing the local storage status).
1705 lu.cfg.SetDiskID(dev, node)
1709 if on_primary or dev.AssembleOnSecondary():
1710 rstats = lu.rpc.call_blockdev_find(node, dev)
1711 msg = rstats.fail_msg
1713 lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1715 elif not rstats.payload:
1716 lu.LogWarning("Can't find disk on node %s", node)
1720 result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
1722 result = result and not rstats.payload.is_degraded
1725 for child in dev.children:
1726 result = result and _CheckDiskConsistencyInner(lu, instance, child, node,
1732 def CheckDiskConsistency(lu, instance, dev, node, on_primary, ldisk=False):
1733 """Wrapper around L{_CheckDiskConsistencyInner}.
1736 (disk,) = AnnotateDiskParams(instance, [dev], lu.cfg)
1737 return _CheckDiskConsistencyInner(lu, instance, disk, node, on_primary,
1741 def _BlockdevFind(lu, node, dev, instance):
1742 """Wrapper around call_blockdev_find to annotate diskparams.
1744 @param lu: A reference to the lu object
1745 @param node: The node to call out
1746 @param dev: The device to find
1747 @param instance: The instance object the device belongs to
1748 @returns The result of the rpc call
1751 (disk,) = AnnotateDiskParams(instance, [dev], lu.cfg)
1752 return lu.rpc.call_blockdev_find(node, disk)
1755 def _GenerateUniqueNames(lu, exts):
1756 """Generate a suitable LV name.
1758 This will generate a logical volume name for the given instance.
1763 new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
1764 results.append("%s%s" % (new_id, val))
1768 class TLReplaceDisks(Tasklet):
1769 """Replaces disks for an instance.
1771 Note: Locking is not within the scope of this class.
1774 def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
1775 disks, early_release, ignore_ipolicy):
1776 """Initializes this class.
1779 Tasklet.__init__(self, lu)
1782 self.instance_name = instance_name
1784 self.iallocator_name = iallocator_name
1785 self.remote_node = remote_node
1787 self.early_release = early_release
1788 self.ignore_ipolicy = ignore_ipolicy
1791 self.instance = None
1792 self.new_node = None
1793 self.target_node = None
1794 self.other_node = None
1795 self.remote_node_info = None
1796 self.node_secondary_ip = None
1799 def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
1800 """Compute a new secondary node using an IAllocator.
1803 req = iallocator.IAReqRelocate(name=instance_name,
1804 relocate_from=list(relocate_from))
1805 ial = iallocator.IAllocator(lu.cfg, lu.rpc, req)
1807 ial.Run(iallocator_name)
1810 raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
1811 " %s" % (iallocator_name, ial.info),
1814 remote_node_name = ial.result[0]
1816 lu.LogInfo("Selected new secondary for instance '%s': %s",
1817 instance_name, remote_node_name)
1819 return remote_node_name
1821 def _FindFaultyDisks(self, node_name):
1822 """Wrapper for L{FindFaultyInstanceDisks}.
1825 return FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
1828 def _CheckDisksActivated(self, instance):
1829 """Checks if the instance disks are activated.
1831 @param instance: The instance to check disks
1832 @return: True if they are activated, False otherwise
1835 nodes = instance.all_nodes
1837 for idx, dev in enumerate(instance.disks):
1839 self.lu.LogInfo("Checking disk/%d on %s", idx, node)
1840 self.cfg.SetDiskID(dev, node)
1842 result = _BlockdevFind(self, node, dev, instance)
1846 elif result.fail_msg or not result.payload:
1851 def CheckPrereq(self):
1852 """Check prerequisites.
1854 This checks that the instance is in the cluster.
1857 self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
1858 assert instance is not None, \
1859 "Cannot retrieve locked instance %s" % self.instance_name
1861 if instance.disk_template != constants.DT_DRBD8:
1862 raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
1863 " instances", errors.ECODE_INVAL)
1865 if len(instance.secondary_nodes) != 1:
1866 raise errors.OpPrereqError("The instance has a strange layout,"
1867 " expected one secondary but found %d" %
1868 len(instance.secondary_nodes),
1871 instance = self.instance
1872 secondary_node = instance.secondary_nodes[0]
1874 if self.iallocator_name is None:
1875 remote_node = self.remote_node
1877 remote_node = self._RunAllocator(self.lu, self.iallocator_name,
1878 instance.name, instance.secondary_nodes)
1880 if remote_node is None:
1881 self.remote_node_info = None
1883 assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
1884 "Remote node '%s' is not locked" % remote_node
1886 self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
1887 assert self.remote_node_info is not None, \
1888 "Cannot retrieve locked node %s" % remote_node
1890 if remote_node == self.instance.primary_node:
1891 raise errors.OpPrereqError("The specified node is the primary node of"
1892 " the instance", errors.ECODE_INVAL)
1894 if remote_node == secondary_node:
1895 raise errors.OpPrereqError("The specified node is already the"
1896 " secondary node of the instance",
1899 if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
1900 constants.REPLACE_DISK_CHG):
1901 raise errors.OpPrereqError("Cannot specify disks to be replaced",
1904 if self.mode == constants.REPLACE_DISK_AUTO:
1905 if not self._CheckDisksActivated(instance):
1906 raise errors.OpPrereqError("Please run activate-disks on instance %s"
1907 " first" % self.instance_name,
1909 faulty_primary = self._FindFaultyDisks(instance.primary_node)
1910 faulty_secondary = self._FindFaultyDisks(secondary_node)
1912 if faulty_primary and faulty_secondary:
1913 raise errors.OpPrereqError("Instance %s has faulty disks on more than"
1914 " one node and can not be repaired"
1915 " automatically" % self.instance_name,
1919 self.disks = faulty_primary
1920 self.target_node = instance.primary_node
1921 self.other_node = secondary_node
1922 check_nodes = [self.target_node, self.other_node]
1923 elif faulty_secondary:
1924 self.disks = faulty_secondary
1925 self.target_node = secondary_node
1926 self.other_node = instance.primary_node
1927 check_nodes = [self.target_node, self.other_node]
1933 # Non-automatic modes
1934 if self.mode == constants.REPLACE_DISK_PRI:
1935 self.target_node = instance.primary_node
1936 self.other_node = secondary_node
1937 check_nodes = [self.target_node, self.other_node]
1939 elif self.mode == constants.REPLACE_DISK_SEC:
1940 self.target_node = secondary_node
1941 self.other_node = instance.primary_node
1942 check_nodes = [self.target_node, self.other_node]
1944 elif self.mode == constants.REPLACE_DISK_CHG:
1945 self.new_node = remote_node
1946 self.other_node = instance.primary_node
1947 self.target_node = secondary_node
1948 check_nodes = [self.new_node, self.other_node]
1950 CheckNodeNotDrained(self.lu, remote_node)
1951 CheckNodeVmCapable(self.lu, remote_node)
1953 old_node_info = self.cfg.GetNodeInfo(secondary_node)
1954 assert old_node_info is not None
1955 if old_node_info.offline and not self.early_release:
1956 # doesn't make sense to delay the release
1957 self.early_release = True
1958 self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
1959 " early-release mode", secondary_node)
1962 raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
1965 # If not specified all disks should be replaced
1967 self.disks = range(len(self.instance.disks))
1969 # TODO: This is ugly, but right now we can't distinguish between internal
1970 # submitted opcode and external one. We should fix that.
1971 if self.remote_node_info:
1972 # We change the node, lets verify it still meets instance policy
1973 new_group_info = self.cfg.GetNodeGroup(self.remote_node_info.group)
1974 cluster = self.cfg.GetClusterInfo()
1975 ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
1977 CheckTargetNodeIPolicy(self, ipolicy, instance, self.remote_node_info,
1978 self.cfg, ignore=self.ignore_ipolicy)
1980 for node in check_nodes:
1981 CheckNodeOnline(self.lu, node)
1983 touched_nodes = frozenset(node_name for node_name in [self.new_node,
1986 if node_name is not None)
1988 # Release unneeded node and node resource locks
1989 ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
1990 ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
1991 ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
1993 # Release any owned node group
1994 ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
1996 # Check whether disks are valid
1997 for disk_idx in self.disks:
1998 instance.FindDisk(disk_idx)
2000 # Get secondary node IP addresses
2001 self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
2002 in self.cfg.GetMultiNodeInfo(touched_nodes))
2004 def Exec(self, feedback_fn):
2005 """Execute disk replacement.
2007 This dispatches the disk replacement to the appropriate handler.
2011 # Verify owned locks before starting operation
2012 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
2013 assert set(owned_nodes) == set(self.node_secondary_ip), \
2014 ("Incorrect node locks, owning %s, expected %s" %
2015 (owned_nodes, self.node_secondary_ip.keys()))
2016 assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
2017 self.lu.owned_locks(locking.LEVEL_NODE_RES))
2018 assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
2020 owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
2021 assert list(owned_instances) == [self.instance_name], \
2022 "Instance '%s' not locked" % self.instance_name
2024 assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
2025 "Should not own any node group lock at this point"
2028 feedback_fn("No disks need replacement for instance '%s'" %
2032 feedback_fn("Replacing disk(s) %s for instance '%s'" %
2033 (utils.CommaJoin(self.disks), self.instance.name))
2034 feedback_fn("Current primary node: %s" % self.instance.primary_node)
2035 feedback_fn("Current seconary node: %s" %
2036 utils.CommaJoin(self.instance.secondary_nodes))
2038 activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
2040 # Activate the instance disks if we're replacing them on a down instance
2042 StartInstanceDisks(self.lu, self.instance, True)
2045 # Should we replace the secondary node?
2046 if self.new_node is not None:
2047 fn = self._ExecDrbd8Secondary
2049 fn = self._ExecDrbd8DiskOnly
2051 result = fn(feedback_fn)
2053 # Deactivate the instance disks if we're replacing them on a
2056 _SafeShutdownInstanceDisks(self.lu, self.instance)
2058 assert not self.lu.owned_locks(locking.LEVEL_NODE)
2061 # Verify owned locks
2062 owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
2063 nodes = frozenset(self.node_secondary_ip)
2064 assert ((self.early_release and not owned_nodes) or
2065 (not self.early_release and not (set(owned_nodes) - nodes))), \
2066 ("Not owning the correct locks, early_release=%s, owned=%r,"
2067 " nodes=%r" % (self.early_release, owned_nodes, nodes))
2071 def _CheckVolumeGroup(self, nodes):
2072 self.lu.LogInfo("Checking volume groups")
2074 vgname = self.cfg.GetVGName()
2076 # Make sure volume group exists on all involved nodes
2077 results = self.rpc.call_vg_list(nodes)
2079 raise errors.OpExecError("Can't list volume groups on the nodes")
2083 res.Raise("Error checking node %s" % node)
2084 if vgname not in res.payload:
2085 raise errors.OpExecError("Volume group '%s' not found on node %s" %
2088 def _CheckDisksExistence(self, nodes):
2089 # Check disk existence
2090 for idx, dev in enumerate(self.instance.disks):
2091 if idx not in self.disks:
2095 self.lu.LogInfo("Checking disk/%d on %s", idx, node)
2096 self.cfg.SetDiskID(dev, node)
2098 result = _BlockdevFind(self, node, dev, self.instance)
2100 msg = result.fail_msg
2101 if msg or not result.payload:
2103 msg = "disk not found"
2104 raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
2107 def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
2108 for idx, dev in enumerate(self.instance.disks):
2109 if idx not in self.disks:
2112 self.lu.LogInfo("Checking disk/%d consistency on node %s" %
2115 if not CheckDiskConsistency(self.lu, self.instance, dev, node_name,
2116 on_primary, ldisk=ldisk):
2117 raise errors.OpExecError("Node %s has degraded storage, unsafe to"
2118 " replace disks for instance %s" %
2119 (node_name, self.instance.name))
2121 def _CreateNewStorage(self, node_name):
2122 """Create new storage on the primary or secondary node.
2124 This is only used for same-node replaces, not for changing the
2125 secondary node, hence we don't want to modify the existing disk.
2130 disks = AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
2131 for idx, dev in enumerate(disks):
2132 if idx not in self.disks:
2135 self.lu.LogInfo("Adding storage on %s for disk/%d", node_name, idx)
2137 self.cfg.SetDiskID(dev, node_name)
2139 lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
2140 names = _GenerateUniqueNames(self.lu, lv_names)
2142 (data_disk, meta_disk) = dev.children
2143 vg_data = data_disk.logical_id[0]
2144 lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
2145 logical_id=(vg_data, names[0]),
2146 params=data_disk.params)
2147 vg_meta = meta_disk.logical_id[0]
2148 lv_meta = objects.Disk(dev_type=constants.LD_LV,
2149 size=constants.DRBD_META_SIZE,
2150 logical_id=(vg_meta, names[1]),
2151 params=meta_disk.params)
2153 new_lvs = [lv_data, lv_meta]
2154 old_lvs = [child.Copy() for child in dev.children]
2155 iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
2156 excl_stor = IsExclusiveStorageEnabledNodeName(self.lu.cfg, node_name)
2158 # we pass force_create=True to force the LVM creation
2159 for new_lv in new_lvs:
2160 _CreateBlockDevInner(self.lu, node_name, self.instance, new_lv, True,
2161 GetInstanceInfoText(self.instance), False,
2166 def _CheckDevices(self, node_name, iv_names):
2167 for name, (dev, _, _) in iv_names.iteritems():
2168 self.cfg.SetDiskID(dev, node_name)
2170 result = _BlockdevFind(self, node_name, dev, self.instance)
2172 msg = result.fail_msg
2173 if msg or not result.payload:
2175 msg = "disk not found"
2176 raise errors.OpExecError("Can't find DRBD device %s: %s" %
2179 if result.payload.is_degraded:
2180 raise errors.OpExecError("DRBD device %s is degraded!" % name)
2182 def _RemoveOldStorage(self, node_name, iv_names):
2183 for name, (_, old_lvs, _) in iv_names.iteritems():
2184 self.lu.LogInfo("Remove logical volumes for %s", name)
2187 self.cfg.SetDiskID(lv, node_name)
2189 msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
2191 self.lu.LogWarning("Can't remove old LV: %s", msg,
2192 hint="remove unused LVs manually")
2194 def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
2195 """Replace a disk on the primary or secondary for DRBD 8.
2197 The algorithm for replace is quite complicated:
2199 1. for each disk to be replaced:
2201 1. create new LVs on the target node with unique names
2202 1. detach old LVs from the drbd device
2203 1. rename old LVs to name_replaced.<time_t>
2204 1. rename new LVs to old LVs
2205 1. attach the new LVs (with the old names now) to the drbd device
2207 1. wait for sync across all devices
2209 1. for each modified disk:
2211 1. remove old LVs (which have the name name_replaces.<time_t>)
2213 Failures are not very well handled.
2218 # Step: check device activation
2219 self.lu.LogStep(1, steps_total, "Check device existence")
2220 self._CheckDisksExistence([self.other_node, self.target_node])
2221 self._CheckVolumeGroup([self.target_node, self.other_node])
2223 # Step: check other node consistency
2224 self.lu.LogStep(2, steps_total, "Check peer consistency")
2225 self._CheckDisksConsistency(self.other_node,
2226 self.other_node == self.instance.primary_node,
2229 # Step: create new storage
2230 self.lu.LogStep(3, steps_total, "Allocate new storage")
2231 iv_names = self._CreateNewStorage(self.target_node)
2233 # Step: for each lv, detach+rename*2+attach
2234 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
2235 for dev, old_lvs, new_lvs in iv_names.itervalues():
2236 self.lu.LogInfo("Detaching %s drbd from local storage", dev.iv_name)
2238 result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
2240 result.Raise("Can't detach drbd from local storage on node"
2241 " %s for device %s" % (self.target_node, dev.iv_name))
2243 #cfg.Update(instance)
2245 # ok, we created the new LVs, so now we know we have the needed
2246 # storage; as such, we proceed on the target node to rename
2247 # old_lv to _old, and new_lv to old_lv; note that we rename LVs
2248 # using the assumption that logical_id == physical_id (which in
2249 # turn is the unique_id on that node)
2251 # FIXME(iustin): use a better name for the replaced LVs
2252 temp_suffix = int(time.time())
2253 ren_fn = lambda d, suff: (d.physical_id[0],
2254 d.physical_id[1] + "_replaced-%s" % suff)
2256 # Build the rename list based on what LVs exist on the node
2257 rename_old_to_new = []
2258 for to_ren in old_lvs:
2259 result = self.rpc.call_blockdev_find(self.target_node, to_ren)
2260 if not result.fail_msg and result.payload:
2262 rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
2264 self.lu.LogInfo("Renaming the old LVs on the target node")
2265 result = self.rpc.call_blockdev_rename(self.target_node,
2267 result.Raise("Can't rename old LVs on node %s" % self.target_node)
2269 # Now we rename the new LVs to the old LVs
2270 self.lu.LogInfo("Renaming the new LVs on the target node")
2271 rename_new_to_old = [(new, old.physical_id)
2272 for old, new in zip(old_lvs, new_lvs)]
2273 result = self.rpc.call_blockdev_rename(self.target_node,
2275 result.Raise("Can't rename new LVs on node %s" % self.target_node)
2277 # Intermediate steps of in memory modifications
2278 for old, new in zip(old_lvs, new_lvs):
2279 new.logical_id = old.logical_id
2280 self.cfg.SetDiskID(new, self.target_node)
2282 # We need to modify old_lvs so that removal later removes the
2283 # right LVs, not the newly added ones; note that old_lvs is a
2285 for disk in old_lvs:
2286 disk.logical_id = ren_fn(disk, temp_suffix)
2287 self.cfg.SetDiskID(disk, self.target_node)
2289 # Now that the new lvs have the old name, we can add them to the device
2290 self.lu.LogInfo("Adding new mirror component on %s", self.target_node)
2291 result = self.rpc.call_blockdev_addchildren(self.target_node,
2292 (dev, self.instance), new_lvs)
2293 msg = result.fail_msg
2295 for new_lv in new_lvs:
2296 msg2 = self.rpc.call_blockdev_remove(self.target_node,
2299 self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
2300 hint=("cleanup manually the unused logical"
2302 raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
2304 cstep = itertools.count(5)
2306 if self.early_release:
2307 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
2308 self._RemoveOldStorage(self.target_node, iv_names)
2309 # TODO: Check if releasing locks early still makes sense
2310 ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
2312 # Release all resource locks except those used by the instance
2313 ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
2314 keep=self.node_secondary_ip.keys())
2316 # Release all node locks while waiting for sync
2317 ReleaseLocks(self.lu, locking.LEVEL_NODE)
2319 # TODO: Can the instance lock be downgraded here? Take the optional disk
2320 # shutdown in the caller into consideration.
2323 # This can fail as the old devices are degraded and _WaitForSync
2324 # does a combined result over all disks, so we don't check its return value
2325 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
2326 WaitForSync(self.lu, self.instance)
2328 # Check all devices manually
2329 self._CheckDevices(self.instance.primary_node, iv_names)
2331 # Step: remove old storage
2332 if not self.early_release:
2333 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
2334 self._RemoveOldStorage(self.target_node, iv_names)
2336 def _ExecDrbd8Secondary(self, feedback_fn):
2337 """Replace the secondary node for DRBD 8.
2339 The algorithm for replace is quite complicated:
2340 - for all disks of the instance:
2341 - create new LVs on the new node with same names
2342 - shutdown the drbd device on the old secondary
2343 - disconnect the drbd network on the primary
2344 - create the drbd device on the new secondary
2345 - network attach the drbd on the primary, using an artifice:
2346 the drbd code for Attach() will connect to the network if it
2347 finds a device which is connected to the good local disks but
2349 - wait for sync across all devices
2350 - remove all disks from the old secondary
2352 Failures are not very well handled.
2357 pnode = self.instance.primary_node
2359 # Step: check device activation
2360 self.lu.LogStep(1, steps_total, "Check device existence")
2361 self._CheckDisksExistence([self.instance.primary_node])
2362 self._CheckVolumeGroup([self.instance.primary_node])
2364 # Step: check other node consistency
2365 self.lu.LogStep(2, steps_total, "Check peer consistency")
2366 self._CheckDisksConsistency(self.instance.primary_node, True, True)
2368 # Step: create new storage
2369 self.lu.LogStep(3, steps_total, "Allocate new storage")
2370 disks = AnnotateDiskParams(self.instance, self.instance.disks, self.cfg)
2371 excl_stor = IsExclusiveStorageEnabledNodeName(self.lu.cfg, self.new_node)
2372 for idx, dev in enumerate(disks):
2373 self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
2374 (self.new_node, idx))
2375 # we pass force_create=True to force LVM creation
2376 for new_lv in dev.children:
2377 _CreateBlockDevInner(self.lu, self.new_node, self.instance, new_lv,
2378 True, GetInstanceInfoText(self.instance), False,
2381 # Step 4: dbrd minors and drbd setups changes
2382 # after this, we must manually remove the drbd minors on both the
2383 # error and the success paths
2384 self.lu.LogStep(4, steps_total, "Changing drbd configuration")
2385 minors = self.cfg.AllocateDRBDMinor([self.new_node
2386 for dev in self.instance.disks],
2388 logging.debug("Allocated minors %r", minors)
2391 for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
2392 self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
2393 (self.new_node, idx))
2394 # create new devices on new_node; note that we create two IDs:
2395 # one without port, so the drbd will be activated without
2396 # networking information on the new node at this stage, and one
2397 # with network, for the latter activation in step 4
2398 (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
2399 if self.instance.primary_node == o_node1:
2402 assert self.instance.primary_node == o_node2, "Three-node instance?"
2405 new_alone_id = (self.instance.primary_node, self.new_node, None,
2406 p_minor, new_minor, o_secret)
2407 new_net_id = (self.instance.primary_node, self.new_node, o_port,
2408 p_minor, new_minor, o_secret)
2410 iv_names[idx] = (dev, dev.children, new_net_id)
2411 logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
2413 new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
2414 logical_id=new_alone_id,
2415 children=dev.children,
2418 (anno_new_drbd,) = AnnotateDiskParams(self.instance, [new_drbd],
2421 CreateSingleBlockDev(self.lu, self.new_node, self.instance,
2423 GetInstanceInfoText(self.instance), False,
2425 except errors.GenericError:
2426 self.cfg.ReleaseDRBDMinors(self.instance.name)
2429 # We have new devices, shutdown the drbd on the old secondary
2430 for idx, dev in enumerate(self.instance.disks):
2431 self.lu.LogInfo("Shutting down drbd for disk/%d on old node", idx)
2432 self.cfg.SetDiskID(dev, self.target_node)
2433 msg = self.rpc.call_blockdev_shutdown(self.target_node,
2434 (dev, self.instance)).fail_msg
2436 self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
2437 "node: %s" % (idx, msg),
2438 hint=("Please cleanup this device manually as"
2439 " soon as possible"))
2441 self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
2442 result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
2443 self.instance.disks)[pnode]
2445 msg = result.fail_msg
2447 # detaches didn't succeed (unlikely)
2448 self.cfg.ReleaseDRBDMinors(self.instance.name)
2449 raise errors.OpExecError("Can't detach the disks from the network on"
2450 " old node: %s" % (msg,))
2452 # if we managed to detach at least one, we update all the disks of
2453 # the instance to point to the new secondary
2454 self.lu.LogInfo("Updating instance configuration")
2455 for dev, _, new_logical_id in iv_names.itervalues():
2456 dev.logical_id = new_logical_id
2457 self.cfg.SetDiskID(dev, self.instance.primary_node)
2459 self.cfg.Update(self.instance, feedback_fn)
2461 # Release all node locks (the configuration has been updated)
2462 ReleaseLocks(self.lu, locking.LEVEL_NODE)
2464 # and now perform the drbd attach
2465 self.lu.LogInfo("Attaching primary drbds to new secondary"
2466 " (standalone => connected)")
2467 result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
2469 self.node_secondary_ip,
2470 (self.instance.disks, self.instance),
2473 for to_node, to_result in result.items():
2474 msg = to_result.fail_msg
2476 self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
2478 hint=("please do a gnt-instance info to see the"
2479 " status of disks"))
2481 cstep = itertools.count(5)
2483 if self.early_release:
2484 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
2485 self._RemoveOldStorage(self.target_node, iv_names)
2486 # TODO: Check if releasing locks early still makes sense
2487 ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
2489 # Release all resource locks except those used by the instance
2490 ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
2491 keep=self.node_secondary_ip.keys())
2493 # TODO: Can the instance lock be downgraded here? Take the optional disk
2494 # shutdown in the caller into consideration.
2497 # This can fail as the old devices are degraded and _WaitForSync
2498 # does a combined result over all disks, so we don't check its return value
2499 self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
2500 WaitForSync(self.lu, self.instance)
2502 # Check all devices manually
2503 self._CheckDevices(self.instance.primary_node, iv_names)
2505 # Step: remove old storage
2506 if not self.early_release:
2507 self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
2508 self._RemoveOldStorage(self.target_node, iv_names)