4 # Copyright (C) 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 """Module implementing the iallocator code."""
24 from ganeti import compat
25 from ganeti import constants
26 from ganeti import errors
28 from ganeti import outils
29 from ganeti import opcodes
30 from ganeti import rpc
31 from ganeti import serializer
32 from ganeti import utils
34 import ganeti.masterd.instance as gmi
37 _STRING_LIST = ht.TListOf(ht.TString)
38 _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
39 # pylint: disable=E1101
40 # Class '...' has no 'OP_ID' member
41 "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
42 opcodes.OpInstanceMigrate.OP_ID,
43 opcodes.OpInstanceReplaceDisks.OP_ID]),
47 ht.TListOf(ht.TAnd(ht.TIsLength(3),
48 ht.TItems([ht.TNonEmptyString,
50 ht.TListOf(ht.TNonEmptyString),
53 ht.TListOf(ht.TAnd(ht.TIsLength(2),
54 ht.TItems([ht.TNonEmptyString,
57 _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
58 ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
60 _INST_NAME = ("name", ht.TNonEmptyString)
63 class _AutoReqParam(outils.AutoSlots):
64 """Meta class for request definitions.
68 def _GetSlots(mcs, attrs):
69 """Extract the slots out of REQ_PARAMS.
72 params = attrs.setdefault("REQ_PARAMS", [])
73 return [slot for (slot, _) in params]
76 class IARequestBase(outils.ValidatedSlots):
77 """A generic IAllocator request object.
80 __metaclass__ = _AutoReqParam
84 REQ_RESULT = NotImplemented
86 def __init__(self, **kwargs):
87 """Constructor for IARequestBase.
89 The constructor takes only keyword arguments and will set
90 attributes on this object based on the passed arguments. As such,
91 it means that you should not pass arguments which are not in the
92 REQ_PARAMS attribute for this class.
95 outils.ValidatedSlots.__init__(self, **kwargs)
100 """Validates all parameters of the request.
103 assert self.MODE in constants.VALID_IALLOCATOR_MODES
105 for (param, validator) in self.REQ_PARAMS:
106 if not hasattr(self, param):
107 raise errors.OpPrereqError("Request is missing '%s' parameter" % param,
110 value = getattr(self, param)
111 if not validator(value):
112 raise errors.OpPrereqError(("Request parameter '%s' has invalid"
113 " type %s/value %s") %
114 (param, type(value), value),
117 def GetRequest(self, cfg):
118 """Gets the request data dict.
120 @param cfg: The configuration instance
123 raise NotImplementedError
125 def ValidateResult(self, ia, result):
126 """Validates the result of an request.
128 @param ia: The IAllocator instance
129 @param result: The IAllocator run result
130 @raises ResultValidationError: If validation fails
133 if ia.success and not self.REQ_RESULT(result):
134 raise errors.ResultValidationError("iallocator returned invalid result,"
135 " expected %s, got %s" %
136 (self.REQ_RESULT, result))
139 class IAReqInstanceAlloc(IARequestBase):
140 """An instance allocation request.
143 # pylint: disable=E1101
144 MODE = constants.IALLOCATOR_MODE_ALLOC
147 ("memory", ht.TNonNegativeInt),
148 ("spindle_use", ht.TNonNegativeInt),
149 ("disks", ht.TListOf(ht.TDict)),
150 ("disk_template", ht.TString),
152 ("tags", _STRING_LIST),
153 ("nics", ht.TListOf(ht.TDict)),
155 ("hypervisor", ht.TString),
156 ("node_whitelist", ht.TMaybeListOf(ht.TNonEmptyString)),
158 REQ_RESULT = ht.TList
160 def RequiredNodes(self):
161 """Calculates the required nodes based on the disk_template.
164 if self.disk_template in constants.DTS_INT_MIRROR:
169 def GetRequest(self, cfg):
170 """Requests a new instance.
172 The checks for the completeness of the opcode must have already been
176 disk_space = gmi.ComputeDiskSize(self.disk_template, self.disks)
180 "disk_template": self.disk_template,
184 "memory": self.memory,
185 "spindle_use": self.spindle_use,
187 "disk_space_total": disk_space,
189 "required_nodes": self.RequiredNodes(),
190 "hypervisor": self.hypervisor,
193 def ValidateResult(self, ia, result):
194 """Validates an single instance allocation request.
197 IARequestBase.ValidateResult(self, ia, result)
199 if ia.success and len(result) != self.RequiredNodes():
200 raise errors.ResultValidationError("iallocator returned invalid number"
201 " of nodes (%s), required %s" %
202 (len(result), self.RequiredNodes()))
205 class IAReqMultiInstanceAlloc(IARequestBase):
206 """An multi instance allocation request.
209 # pylint: disable=E1101
210 MODE = constants.IALLOCATOR_MODE_MULTI_ALLOC
212 ("instances", ht.TListOf(ht.TInstanceOf(IAReqInstanceAlloc))),
215 ht.TListOf(ht.TAnd(ht.TIsLength(2),
216 ht.TItems([ht.TNonEmptyString,
217 ht.TListOf(ht.TNonEmptyString),
219 _MAFAILED = ht.TListOf(ht.TNonEmptyString)
220 REQ_RESULT = ht.TAnd(ht.TList, ht.TIsLength(2),
221 ht.TItems([_MASUCCESS, _MAFAILED]))
223 def GetRequest(self, cfg):
225 "instances": [iareq.GetRequest(cfg) for iareq in self.instances],
229 class IAReqRelocate(IARequestBase):
230 """A relocation request.
233 # pylint: disable=E1101
234 MODE = constants.IALLOCATOR_MODE_RELOC
237 ("relocate_from", _STRING_LIST),
239 REQ_RESULT = ht.TList
241 def GetRequest(self, cfg):
242 """Request an relocation of an instance
244 The checks for the completeness of the opcode must have already been
248 instance = cfg.GetInstanceInfo(self.name)
250 raise errors.ProgrammerError("Unknown instance '%s' passed to"
251 " IAllocator" % self.name)
253 if instance.disk_template not in constants.DTS_MIRRORED:
254 raise errors.OpPrereqError("Can't relocate non-mirrored instances",
257 if (instance.disk_template in constants.DTS_INT_MIRROR and
258 len(instance.secondary_nodes) != 1):
259 raise errors.OpPrereqError("Instance has not exactly one secondary node",
262 disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
263 disk_space = gmi.ComputeDiskSize(instance.disk_template, disk_sizes)
267 "disk_space_total": disk_space,
269 "relocate_from": self.relocate_from,
272 def ValidateResult(self, ia, result):
273 """Validates the result of an relocation request.
276 IARequestBase.ValidateResult(self, ia, result)
278 node2group = dict((name, ndata["group"])
279 for (name, ndata) in ia.in_data["nodes"].items())
281 fn = compat.partial(self._NodesToGroups, node2group,
282 ia.in_data["nodegroups"])
284 instance = ia.cfg.GetInstanceInfo(self.name)
285 request_groups = fn(self.relocate_from + [instance.primary_node])
286 result_groups = fn(result + [instance.primary_node])
288 if ia.success and not set(result_groups).issubset(request_groups):
289 raise errors.ResultValidationError("Groups of nodes returned by"
290 "iallocator (%s) differ from original"
292 (utils.CommaJoin(result_groups),
293 utils.CommaJoin(request_groups)))
296 def _NodesToGroups(node2group, groups, nodes):
297 """Returns a list of unique group names for a list of nodes.
299 @type node2group: dict
300 @param node2group: Map from node name to group UUID
302 @param groups: Group information
304 @param nodes: Node names
311 group_uuid = node2group[node]
313 # Ignore unknown node
317 group = groups[group_uuid]
319 # Can't find group, let's use UUID
320 group_name = group_uuid
322 group_name = group["name"]
324 result.add(group_name)
326 return sorted(result)
329 class IAReqNodeEvac(IARequestBase):
330 """A node evacuation request.
333 # pylint: disable=E1101
334 MODE = constants.IALLOCATOR_MODE_NODE_EVAC
336 ("instances", _STRING_LIST),
337 ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
339 REQ_RESULT = _NEVAC_RESULT
341 def GetRequest(self, cfg):
342 """Get data for node-evacuate requests.
346 "instances": self.instances,
347 "evac_mode": self.evac_mode,
351 class IAReqGroupChange(IARequestBase):
352 """A group change request.
355 # pylint: disable=E1101
356 MODE = constants.IALLOCATOR_MODE_CHG_GROUP
358 ("instances", _STRING_LIST),
359 ("target_groups", _STRING_LIST),
361 REQ_RESULT = _NEVAC_RESULT
363 def GetRequest(self, cfg):
364 """Get data for node-evacuate requests.
368 "instances": self.instances,
369 "target_groups": self.target_groups,
373 class IAllocator(object):
374 """IAllocator framework.
376 An IAllocator instance has three sets of attributes:
377 - cfg that is needed to query the cluster
378 - input data (all members of the _KEYS class attribute are required)
379 - four buffer attributes (in|out_data|text), that represent the
380 input (to the external script) in text and data structure format,
381 and the output from it, again in two formats
382 - the result variables from the script (success, info, nodes) for
386 # pylint: disable=R0902
387 # lots of instance attributes
389 def __init__(self, cfg, rpc_runner, req):
391 self.rpc = rpc_runner
393 # init buffer variables
394 self.in_text = self.out_text = self.in_data = self.out_data = None
396 self.success = self.info = self.result = None
398 self._BuildInputData(req)
400 def _ComputeClusterData(self):
401 """Compute the generic allocator input data.
403 This is the data that is independent of the actual operation.
407 cluster_info = cfg.GetClusterInfo()
410 "version": constants.IALLOCATOR_VERSION,
411 "cluster_name": cfg.GetClusterName(),
412 "cluster_tags": list(cluster_info.GetTags()),
413 "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
414 "ipolicy": cluster_info.ipolicy,
416 ninfo = cfg.GetAllNodesInfo()
417 iinfo = cfg.GetAllInstancesInfo().values()
418 i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
421 node_list = [n.name for n in ninfo.values() if n.vm_capable]
423 if isinstance(self.req, IAReqInstanceAlloc):
424 hypervisor_name = self.req.hypervisor
425 node_whitelist = self.req.node_whitelist
426 elif isinstance(self.req, IAReqRelocate):
427 hypervisor_name = cfg.GetInstanceInfo(self.req.name).hypervisor
428 node_whitelist = None
430 hypervisor_name = cluster_info.primary_hypervisor
431 node_whitelist = None
433 es_flags = rpc.GetExclusiveStorageForNodeNames(cfg, node_list)
434 vg_name = cfg.GetVGName()
435 if vg_name is not None:
441 node_data = self.rpc.call_node_info(node_list, vg_req,
442 [hypervisor_name], es_flags)
444 self.rpc.call_all_instances_info(node_list,
445 cluster_info.enabled_hypervisors)
447 data["nodegroups"] = self._ComputeNodeGroupData(cfg)
449 config_ndata = self._ComputeBasicNodeData(cfg, ninfo, node_whitelist)
450 data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
451 i_list, config_ndata, has_lvm)
452 assert len(data["nodes"]) == len(ninfo), \
453 "Incomplete node data computed"
455 data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
460 def _ComputeNodeGroupData(cfg):
461 """Compute node groups data.
464 cluster = cfg.GetClusterInfo()
467 "alloc_policy": gdata.alloc_policy,
468 "networks": [net_uuid for net_uuid, _ in gdata.networks.items()],
469 "ipolicy": gmi.CalculateGroupIPolicy(cluster, gdata),
470 "tags": list(gdata.GetTags()),
472 for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
477 def _ComputeBasicNodeData(cfg, node_cfg, node_whitelist):
478 """Compute global node data.
481 @returns: a dict of name: (node dict, node config)
484 # fill in static (config-based) values
485 node_results = dict((ninfo.name, {
486 "tags": list(ninfo.GetTags()),
487 "primary_ip": ninfo.primary_ip,
488 "secondary_ip": ninfo.secondary_ip,
489 "offline": (ninfo.offline or
490 not (node_whitelist is None or
491 ninfo.name in node_whitelist)),
492 "drained": ninfo.drained,
493 "master_candidate": ninfo.master_candidate,
494 "group": ninfo.group,
495 "master_capable": ninfo.master_capable,
496 "vm_capable": ninfo.vm_capable,
497 "ndparams": cfg.GetNdParams(ninfo),
499 for ninfo in node_cfg.values())
504 def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
505 node_results, has_lvm):
506 """Compute global node data.
508 @param node_results: the basic node structures as filled from the config
511 #TODO(dynmem): compute the right data on MAX and MIN memory
512 # make a copy of the current dict
513 node_results = dict(node_results)
514 for nname, nresult in node_data.items():
515 assert nname in node_results, "Missing basic data for node %s" % nname
516 ninfo = node_cfg[nname]
518 if not (ninfo.offline or ninfo.drained):
519 nresult.Raise("Can't get data for node %s" % nname)
520 node_iinfo[nname].Raise("Can't get node instance info from node %s" %
522 remote_info = rpc.MakeLegacyNodeInfo(nresult.payload,
523 require_vg_info=has_lvm)
526 if attr not in remote_info:
527 raise errors.OpExecError("Node '%s' didn't return attribute"
528 " '%s'" % (nname, attr))
529 value = remote_info[attr]
530 if not isinstance(value, int):
531 raise errors.OpExecError("Node '%s' returned invalid value"
533 (nname, attr, value))
536 mem_free = get_attr("memory_free")
538 # compute memory used by primary instances
539 i_p_mem = i_p_up_mem = 0
540 for iinfo, beinfo in i_list:
541 if iinfo.primary_node == nname:
542 i_p_mem += beinfo[constants.BE_MAXMEM]
543 if iinfo.name not in node_iinfo[nname].payload:
546 i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
547 i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
548 mem_free -= max(0, i_mem_diff)
550 if iinfo.admin_state == constants.ADMINST_UP:
551 i_p_up_mem += beinfo[constants.BE_MAXMEM]
553 # TODO: replace this with proper storage reporting
555 total_disk = get_attr("vg_size")
556 free_disk = get_attr("vg_free")
558 # we didn't even ask the node for VG status, so use zeros
559 total_disk = free_disk = 0
561 # compute memory used by instances
563 "total_memory": get_attr("memory_total"),
564 "reserved_memory": get_attr("memory_dom0"),
565 "free_memory": mem_free,
566 "total_disk": total_disk,
567 "free_disk": free_disk,
568 "total_cpus": get_attr("cpu_total"),
569 "i_pri_memory": i_p_mem,
570 "i_pri_up_memory": i_p_up_mem,
572 pnr_dyn.update(node_results[nname])
573 node_results[nname] = pnr_dyn
578 def _ComputeInstanceData(cluster_info, i_list):
579 """Compute global instance data.
583 for iinfo, beinfo in i_list:
585 for nic in iinfo.nics:
586 filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
590 "mode": filled_params[constants.NIC_MODE],
591 "link": filled_params[constants.NIC_LINK],
593 if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
594 nic_dict["bridge"] = filled_params[constants.NIC_LINK]
595 nic_data.append(nic_dict)
597 "tags": list(iinfo.GetTags()),
598 "admin_state": iinfo.admin_state,
599 "vcpus": beinfo[constants.BE_VCPUS],
600 "memory": beinfo[constants.BE_MAXMEM],
601 "spindle_use": beinfo[constants.BE_SPINDLE_USE],
603 "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
605 "disks": [{constants.IDISK_SIZE: dsk.size,
606 constants.IDISK_MODE: dsk.mode}
607 for dsk in iinfo.disks],
608 "disk_template": iinfo.disk_template,
609 "disks_active": iinfo.disks_active,
610 "hypervisor": iinfo.hypervisor,
612 pir["disk_space_total"] = gmi.ComputeDiskSize(iinfo.disk_template,
614 instance_data[iinfo.name] = pir
618 def _BuildInputData(self, req):
619 """Build input data structures.
622 self._ComputeClusterData()
624 request = req.GetRequest(self.cfg)
625 request["type"] = req.MODE
626 self.in_data["request"] = request
628 self.in_text = serializer.Dump(self.in_data)
630 def Run(self, name, validate=True, call_fn=None):
631 """Run an instance allocator and return the results.
635 call_fn = self.rpc.call_iallocator_runner
637 result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
638 result.Raise("Failure while running the iallocator script")
640 self.out_text = result.payload
642 self._ValidateResult()
644 def _ValidateResult(self):
645 """Process the allocator results.
647 This will process and if successful save the result in
648 self.out_data and the other parameters.
652 rdict = serializer.Load(self.out_text)
653 except Exception, err:
654 raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
656 if not isinstance(rdict, dict):
657 raise errors.OpExecError("Can't parse iallocator results: not a dict")
659 # TODO: remove backwards compatiblity in later versions
660 if "nodes" in rdict and "result" not in rdict:
661 rdict["result"] = rdict["nodes"]
664 for key in "success", "info", "result":
666 raise errors.OpExecError("Can't parse iallocator results:"
667 " missing key '%s'" % key)
668 setattr(self, key, rdict[key])
670 self.req.ValidateResult(self, self.result)
671 self.out_data = rdict