import os
import os.path
-import sha
import time
-import tempfile
import re
import platform
+import logging
+import copy
-from ganeti import rpc
from ganeti import ssh
-from ganeti import logger
from ganeti import utils
from ganeti import errors
from ganeti import hypervisor
from ganeti import locking
from ganeti import constants
from ganeti import objects
-from ganeti import opcodes
from ganeti import serializer
+from ganeti import ssconf
class LogicalUnit(object):
Subclasses must follow these rules:
- implement ExpandNames
- - implement CheckPrereq
- - implement Exec
+ - implement CheckPrereq (except when tasklets are used)
+ - implement Exec (except when tasklets are used)
- implement BuildHooksEnv
- redefine HPATH and HTYPE
- optionally redefine their run requirements:
- REQ_MASTER: the LU needs to run on the master node
- REQ_WSSTORE: the LU needs a writable SimpleStore
REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
Note that all commands require root permissions.
+ @ivar dry_run_result: the value (if any) that will be returned to the caller
+ in dry-run mode (signalled by opcode dry_run parameter)
+
"""
HPATH = None
HTYPE = None
_OP_REQP = []
- REQ_MASTER = True
- REQ_WSSTORE = False
REQ_BGL = True
- def __init__(self, processor, op, context, sstore):
+ def __init__(self, processor, op, context, rpc):
"""Constructor for LogicalUnit.
- This needs to be overriden in derived classes in order to check op
+ This needs to be overridden in derived classes in order to check op
validity.
"""
self.proc = processor
self.op = op
self.cfg = context.cfg
- self.sstore = sstore
self.context = context
+ self.rpc = rpc
+ # Dicts used to declare locking needs to mcpu
self.needed_locks = None
self.acquired_locks = {}
- self.share_locks = dict(((i, 0) for i in locking.LEVELS))
+ self.share_locks = dict.fromkeys(locking.LEVELS, 0)
+ self.add_locks = {}
+ self.remove_locks = {}
# Used to force good behavior when calling helper functions
self.recalculate_locks = {}
self.__ssh = None
+ # logging
+ self.LogWarning = processor.LogWarning
+ self.LogInfo = processor.LogInfo
+ self.LogStep = processor.LogStep
+ # support for dry-run
+ self.dry_run_result = None
+
+ # Tasklets
+ self.tasklets = None
for attr_name in self._OP_REQP:
attr_val = getattr(op, attr_name, None)
raise errors.OpPrereqError("Required parameter '%s' missing" %
attr_name)
- if not self.cfg.IsCluster():
- raise errors.OpPrereqError("Cluster not initialized yet,"
- " use 'gnt-cluster init' first.")
- if self.REQ_MASTER:
- master = sstore.GetMasterNode()
- if master != utils.HostInfo().name:
- raise errors.OpPrereqError("Commands must be run on the master"
- " node %s" % master)
+ self.CheckArguments()
def __GetSSH(self):
"""Returns the SshRunner object
"""
if not self.__ssh:
- self.__ssh = ssh.SshRunner(self.sstore)
+ self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
return self.__ssh
ssh = property(fget=__GetSSH)
+ def CheckArguments(self):
+ """Check syntactic validity for the opcode arguments.
+
+ This method is for doing a simple syntactic check and ensure
+ validity of opcode parameters, without any cluster-related
+ checks. While the same can be accomplished in ExpandNames and/or
+ CheckPrereq, doing these separate is better because:
+
+ - ExpandNames is left as as purely a lock-related function
+ - CheckPrereq is run after we have acquired locks (and possible
+ waited for them)
+
+ The function is allowed to change the self.op attribute so that
+ later methods can no longer worry about missing parameters.
+
+ """
+ pass
+
def ExpandNames(self):
"""Expand names for this LU.
LUs which implement this method must also populate the self.needed_locks
member, as a dict with lock levels as keys, and a list of needed lock names
as values. Rules:
- - Use an empty dict if you don't need any lock
- - If you don't need any lock at a particular level omit that level
- - Don't put anything for the BGL level
- - If you want all locks at a level use locking.ALL_SET as a value
+
+ - use an empty dict if you don't need any lock
+ - if you don't need any lock at a particular level omit that level
+ - don't put anything for the BGL level
+ - if you want all locks at a level use locking.ALL_SET as a value
If you need to share locks (rather than acquire them exclusively) at one
level you can modify self.share_locks, setting a true value (usually 1) for
that level. By default locks are not shared.
- Examples:
- # Acquire all nodes and one instance
- self.needed_locks = {
- locking.LEVEL_NODE: locking.ALL_SET,
- locking.LEVEL_INSTANCE: ['instance1.example.tld'],
- }
- # Acquire just two nodes
- self.needed_locks = {
- locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
- }
- # Acquire no locks
- self.needed_locks = {} # No, you can't leave it to the default value None
+ This function can also define a list of tasklets, which then will be
+ executed in order instead of the usual LU-level CheckPrereq and Exec
+ functions, if those are not defined by the LU.
+
+ Examples::
+
+ # Acquire all nodes and one instance
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_INSTANCE: ['instance1.example.tld'],
+ }
+ # Acquire just two nodes
+ self.needed_locks = {
+ locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
+ }
+ # Acquire no locks
+ self.needed_locks = {} # No, you can't leave it to the default value None
"""
# The implementation of this method is mandatory only if the new LU is
their canonical form if it hasn't been done by ExpandNames before.
"""
- raise NotImplementedError
+ if self.tasklets is not None:
+ for (idx, tl) in enumerate(self.tasklets):
+ logging.debug("Checking prerequisites for tasklet %s/%s",
+ idx + 1, len(self.tasklets))
+ tl.CheckPrereq()
+ else:
+ raise NotImplementedError
def Exec(self, feedback_fn):
"""Execute the LU.
code, or expected.
"""
- raise NotImplementedError
+ if self.tasklets is not None:
+ for (idx, tl) in enumerate(self.tasklets):
+ logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
+ tl.Exec(feedback_fn)
+ else:
+ raise NotImplementedError
def BuildHooksEnv(self):
"""Build hooks environment for this LU.
previous result is passed back unchanged but any LU can define it if it
wants to use the local cluster hook-scripts somehow.
- Args:
- phase: the hooks phase that has just been run
- hooks_results: the results of the multi-node hooks rpc call
- feedback_fn: function to send feedback back to the caller
- lu_result: the previous result this LU had, or None in the PRE phase.
+ @param phase: one of L{constants.HOOKS_PHASE_POST} or
+ L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
+ @param hook_results: the results of the multi-node hooks rpc call
+ @param feedback_fn: function used send feedback back to the caller
+ @param lu_result: the previous Exec result this LU had, or None
+ in the PRE phase
+ @return: the new Exec result, based on the previous result
+ and hook results
"""
return lu_result
In the future it may grow parameters to just lock some instance's nodes, or
to just lock primaries or secondary nodes, if needed.
- If should be called in DeclareLocks in a way similar to:
+ If should be called in DeclareLocks in a way similar to::
- if level == locking.LEVEL_NODE:
- self._LockInstancesNodes()
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
@type primary_only: boolean
@param primary_only: only lock primary nodes of locked instances
HTYPE = None
+class Tasklet:
+ """Tasklet base class.
+
+ Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
+ they can mix legacy code with tasklets. Locking needs to be done in the LU,
+ tasklets know nothing about locks.
+
+ Subclasses must follow these rules:
+ - Implement CheckPrereq
+ - Implement Exec
+
+ """
+ def __init__(self, lu):
+ self.lu = lu
+
+ # Shortcuts
+ self.cfg = lu.cfg
+ self.rpc = lu.rpc
+
+ def CheckPrereq(self):
+ """Check prerequisites for this tasklets.
+
+ This method should check whether the prerequisites for the execution of
+ this tasklet are fulfilled. It can do internode communication, but it
+ should be idempotent - no cluster or system changes are allowed.
+
+ The method should raise errors.OpPrereqError in case something is not
+ fulfilled. Its return value is ignored.
+
+ This method should also update all parameters to their canonical form if it
+ hasn't been done before.
+
+ """
+ raise NotImplementedError
+
+ def Exec(self, feedback_fn):
+ """Execute the tasklet.
+
+ This method should implement the actual work. It should raise
+ errors.OpExecError for failures that are somewhat dealt with in code, or
+ expected.
+
+ """
+ raise NotImplementedError
+
+
def _GetWantedNodes(lu, nodes):
"""Returns list of checked and expanded node names.
- Args:
- nodes: List of nodes (strings) or None for all
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type nodes: list
+ @param nodes: list of node names or None for all nodes
+ @rtype: list
+ @return: the list of nodes, sorted
+ @raise errors.OpProgrammerError: if the nodes parameter is wrong type
"""
if not isinstance(nodes, list):
def _GetWantedInstances(lu, instances):
"""Returns list of checked and expanded instance names.
- Args:
- instances: List of instances (strings) or None for all
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instances: list
+ @param instances: list of instance names or None for all instances
+ @rtype: list
+ @return: the list of instances, sorted
+ @raise errors.OpPrereqError: if the instances parameter is wrong type
+ @raise errors.OpPrereqError: if any of the passed instances is not found
"""
if not isinstance(instances, list):
wanted.append(instance)
else:
- wanted = lu.cfg.GetInstanceList()
- return utils.NiceSort(wanted)
+ wanted = utils.NiceSort(lu.cfg.GetInstanceList())
+ return wanted
def _CheckOutputFields(static, dynamic, selected):
"""Checks whether all selected fields are valid.
- Args:
- static: Static fields
- dynamic: Dynamic fields
+ @type static: L{utils.FieldSet}
+ @param static: static fields set
+ @type dynamic: L{utils.FieldSet}
+ @param dynamic: dynamic fields set
"""
- static_fields = frozenset(static)
- dynamic_fields = frozenset(dynamic)
-
- all_fields = static_fields | dynamic_fields
+ f = utils.FieldSet()
+ f.Extend(static)
+ f.Extend(dynamic)
- if not all_fields.issuperset(selected):
+ delta = f.NonMatching(selected)
+ if delta:
raise errors.OpPrereqError("Unknown output fields selected: %s"
- % ",".join(frozenset(selected).
- difference(all_fields)))
+ % ",".join(delta))
+
+
+def _CheckBooleanOpField(op, name):
+ """Validates boolean opcode parameters.
+
+ This will ensure that an opcode parameter is either a boolean value,
+ or None (but that it always exists).
+
+ """
+ val = getattr(op, name, None)
+ if not (val is None or isinstance(val, bool)):
+ raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
+ (name, str(val)))
+ setattr(op, name, val)
+
+
+def _CheckNodeOnline(lu, node):
+ """Ensure that a given node is online.
+
+ @param lu: the LU on behalf of which we make the check
+ @param node: the node to check
+ @raise errors.OpPrereqError: if the node is offline
+
+ """
+ if lu.cfg.GetNodeInfo(node).offline:
+ raise errors.OpPrereqError("Can't use offline node %s" % node)
+
+
+def _CheckNodeNotDrained(lu, node):
+ """Ensure that a given node is not drained.
+
+ @param lu: the LU on behalf of which we make the check
+ @param node: the node to check
+ @raise errors.OpPrereqError: if the node is drained
+
+ """
+ if lu.cfg.GetNodeInfo(node).drained:
+ raise errors.OpPrereqError("Can't use drained node %s" % node)
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
- memory, vcpus, nics):
- """Builds instance related env variables for hooks from single variables.
+ memory, vcpus, nics, disk_template, disks,
+ bep, hvp, hypervisor_name):
+ """Builds instance related env variables for hooks
+
+ This builds the hook environment from individual variables.
+
+ @type name: string
+ @param name: the name of the instance
+ @type primary_node: string
+ @param primary_node: the name of the instance's primary node
+ @type secondary_nodes: list
+ @param secondary_nodes: list of secondary nodes as strings
+ @type os_type: string
+ @param os_type: the name of the instance's OS
+ @type status: boolean
+ @param status: the should_run status of the instance
+ @type memory: string
+ @param memory: the memory size of the instance
+ @type vcpus: string
+ @param vcpus: the count of VCPUs the instance has
+ @type nics: list
+ @param nics: list of tuples (ip, mac, mode, link) representing
+ the NICs the instance has
+ @type disk_template: string
+ @param disk_template: the disk template of the instance
+ @type disks: list
+ @param disks: the list of (size, mode) pairs
+ @type bep: dict
+ @param bep: the backend parameters for the instance
+ @type hvp: dict
+ @param hvp: the hypervisor parameters for the instance
+ @type hypervisor_name: string
+ @param hypervisor_name: the hypervisor for the instance
+ @rtype: dict
+ @return: the hook environment for this instance
- Args:
- secondary_nodes: List of secondary nodes as strings
"""
+ if status:
+ str_status = "up"
+ else:
+ str_status = "down"
env = {
"OP_TARGET": name,
"INSTANCE_NAME": name,
"INSTANCE_PRIMARY": primary_node,
"INSTANCE_SECONDARIES": " ".join(secondary_nodes),
"INSTANCE_OS_TYPE": os_type,
- "INSTANCE_STATUS": status,
+ "INSTANCE_STATUS": str_status,
"INSTANCE_MEMORY": memory,
"INSTANCE_VCPUS": vcpus,
+ "INSTANCE_DISK_TEMPLATE": disk_template,
+ "INSTANCE_HYPERVISOR": hypervisor_name,
}
if nics:
nic_count = len(nics)
- for idx, (ip, bridge, mac) in enumerate(nics):
+ for idx, (ip, mac, mode, link) in enumerate(nics):
if ip is None:
ip = ""
env["INSTANCE_NIC%d_IP" % idx] = ip
- env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
- env["INSTANCE_NIC%d_HWADDR" % idx] = mac
+ env["INSTANCE_NIC%d_MAC" % idx] = mac
+ env["INSTANCE_NIC%d_MODE" % idx] = mode
+ env["INSTANCE_NIC%d_LINK" % idx] = link
+ if mode == constants.NIC_MODE_BRIDGED:
+ env["INSTANCE_NIC%d_BRIDGE" % idx] = link
else:
nic_count = 0
env["INSTANCE_NIC_COUNT"] = nic_count
+ if disks:
+ disk_count = len(disks)
+ for idx, (size, mode) in enumerate(disks):
+ env["INSTANCE_DISK%d_SIZE" % idx] = size
+ env["INSTANCE_DISK%d_MODE" % idx] = mode
+ else:
+ disk_count = 0
+
+ env["INSTANCE_DISK_COUNT"] = disk_count
+
+ for source, kind in [(bep, "BE"), (hvp, "HV")]:
+ for key, value in source.items():
+ env["INSTANCE_%s_%s" % (kind, key)] = value
+
return env
-def _BuildInstanceHookEnvByObject(instance, override=None):
+def _NICListToTuple(lu, nics):
+ """Build a list of nic information tuples.
+
+ This list is suitable to be passed to _BuildInstanceHookEnv or as a return
+ value in LUQueryInstanceData.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type nics: list of L{objects.NIC}
+ @param nics: list of nics to convert to hooks tuples
+
+ """
+ hooks_nics = []
+ c_nicparams = lu.cfg.GetClusterInfo().nicparams[constants.PP_DEFAULT]
+ for nic in nics:
+ ip = nic.ip
+ mac = nic.mac
+ filled_params = objects.FillDict(c_nicparams, nic.nicparams)
+ mode = filled_params[constants.NIC_MODE]
+ link = filled_params[constants.NIC_LINK]
+ hooks_nics.append((ip, mac, mode, link))
+ return hooks_nics
+
+
+def _BuildInstanceHookEnvByObject(lu, instance, override=None):
"""Builds instance related env variables for hooks from an object.
- Args:
- instance: objects.Instance object of instance
- override: dict of values to override
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance for which we should build the
+ environment
+ @type override: dict
+ @param override: dictionary with key/values that will override
+ our values
+ @rtype: dict
+ @return: the hook environment dictionary
+
"""
+ cluster = lu.cfg.GetClusterInfo()
+ bep = cluster.FillBE(instance)
+ hvp = cluster.FillHV(instance)
args = {
'name': instance.name,
'primary_node': instance.primary_node,
'secondary_nodes': instance.secondary_nodes,
'os_type': instance.os,
- 'status': instance.os,
- 'memory': instance.memory,
- 'vcpus': instance.vcpus,
- 'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
+ 'status': instance.admin_up,
+ 'memory': bep[constants.BE_MEMORY],
+ 'vcpus': bep[constants.BE_VCPUS],
+ 'nics': _NICListToTuple(lu, instance.nics),
+ 'disk_template': instance.disk_template,
+ 'disks': [(disk.size, disk.mode) for disk in instance.disks],
+ 'bep': bep,
+ 'hvp': hvp,
+ 'hypervisor_name': instance.hypervisor,
}
if override:
args.update(override)
return _BuildInstanceHookEnv(**args)
-def _CheckInstanceBridgesExist(instance):
+def _AdjustCandidatePool(lu):
+ """Adjust the candidate pool after node operations.
+
+ """
+ mod_list = lu.cfg.MaintainCandidatePool()
+ if mod_list:
+ lu.LogInfo("Promoted nodes to master candidate role: %s",
+ ", ".join(node.name for node in mod_list))
+ for name in mod_list:
+ lu.context.ReaddNode(name)
+ mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
+ if mc_now > mc_max:
+ lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
+ (mc_now, mc_max))
+
+
+def _CheckNicsBridgesExist(lu, target_nics, target_node,
+ profile=constants.PP_DEFAULT):
+ """Check that the brigdes needed by a list of nics exist.
+
+ """
+ c_nicparams = lu.cfg.GetClusterInfo().nicparams[profile]
+ paramslist = [objects.FillDict(c_nicparams, nic.nicparams)
+ for nic in target_nics]
+ brlist = [params[constants.NIC_LINK] for params in paramslist
+ if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
+ if brlist:
+ result = lu.rpc.call_bridges_exist(target_node, brlist)
+ result.Raise("Error checking bridges on destination node '%s'" %
+ target_node, prereq=True)
+
+
+def _CheckInstanceBridgesExist(lu, instance, node=None):
"""Check that the brigdes needed by an instance exist.
"""
- # check bridges existance
- brlist = [nic.bridge for nic in instance.nics]
- if not rpc.call_bridges_exist(instance.primary_node, brlist):
- raise errors.OpPrereqError("one or more target bridges %s does not"
- " exist on destination node '%s'" %
- (brlist, instance.primary_node))
+ if node is None:
+ node = instance.primary_node
+ _CheckNicsBridgesExist(lu, instance.nics, node)
+
+
+def _GetNodeInstancesInner(cfg, fn):
+ return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
+
+
+def _GetNodeInstances(cfg, node_name):
+ """Returns a list of all primary and secondary instances on a node.
+
+ """
+
+ return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
+
+
+def _GetNodePrimaryInstances(cfg, node_name):
+ """Returns primary instances on a node.
+
+ """
+ return _GetNodeInstancesInner(cfg,
+ lambda inst: node_name == inst.primary_node)
+
+
+def _GetNodeSecondaryInstances(cfg, node_name):
+ """Returns secondary instances on a node.
+
+ """
+ return _GetNodeInstancesInner(cfg,
+ lambda inst: node_name in inst.secondary_nodes)
+
+
+def _GetStorageTypeArgs(cfg, storage_type):
+ """Returns the arguments for a storage type.
+
+ """
+ # Special case for file storage
+ if storage_type == constants.ST_FILE:
+ # storage.FileStorage wants a list of storage directories
+ return [[cfg.GetFileStorageDir()]]
+
+ return []
+
+
+def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
+ faulty = []
+
+ for dev in instance.disks:
+ cfg.SetDiskID(dev, node_name)
+
+ result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
+ result.Raise("Failed to get disk status from node %s" % node_name,
+ prereq=prereq)
+
+ for idx, bdev_status in enumerate(result.payload):
+ if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
+ faulty.append(idx)
+
+ return faulty
+
+
+class LUPostInitCluster(LogicalUnit):
+ """Logical unit for running hooks after cluster initialization.
+
+ """
+ HPATH = "cluster-init"
+ HTYPE = constants.HTYPE_CLUSTER
+ _OP_REQP = []
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ """
+ env = {"OP_TARGET": self.cfg.GetClusterName()}
+ mn = self.cfg.GetMasterNode()
+ return env, [], [mn]
+
+ def CheckPrereq(self):
+ """No prerequisites to check.
+
+ """
+ return True
+
+ def Exec(self, feedback_fn):
+ """Nothing to do.
+
+ """
+ return True
-class LUDestroyCluster(NoHooksLU):
+class LUDestroyCluster(LogicalUnit):
"""Logical unit for destroying the cluster.
"""
+ HPATH = "cluster-destroy"
+ HTYPE = constants.HTYPE_CLUSTER
_OP_REQP = []
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ """
+ env = {"OP_TARGET": self.cfg.GetClusterName()}
+ return env, [], []
+
def CheckPrereq(self):
"""Check prerequisites.
This checks whether the cluster is empty.
- Any errors are signalled by raising errors.OpPrereqError.
+ Any errors are signaled by raising errors.OpPrereqError.
"""
- master = self.sstore.GetMasterNode()
+ master = self.cfg.GetMasterNode()
nodelist = self.cfg.GetNodeList()
if len(nodelist) != 1 or nodelist[0] != master:
"""Destroys the cluster.
"""
- master = self.sstore.GetMasterNode()
- if not rpc.call_node_stop_master(master, False):
- raise errors.OpExecError("Could not disable the master role")
+ master = self.cfg.GetMasterNode()
+
+ # Run post hooks on master node before it's removed
+ hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
+ try:
+ hm.RunPhase(constants.HOOKS_PHASE_POST, [master])
+ except:
+ self.LogWarning("Errors occurred running hooks on %s" % master)
+
+ result = self.rpc.call_node_stop_master(master, False)
+ result.Raise("Could not disable the master role")
priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
utils.CreateBackup(priv_key)
utils.CreateBackup(pub_key)
"""
HPATH = "cluster-verify"
HTYPE = constants.HTYPE_CLUSTER
- _OP_REQP = ["skip_checks"]
+ _OP_REQP = ["skip_checks", "verbose", "error_codes", "debug_simulate_errors"]
REQ_BGL = False
+ TCLUSTER = "cluster"
+ TNODE = "node"
+ TINSTANCE = "instance"
+
+ ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
+ EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
+ EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
+ EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
+ EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
+ EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
+ EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
+ ENODEDRBD = (TNODE, "ENODEDRBD")
+ ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
+ ENODEHOOKS = (TNODE, "ENODEHOOKS")
+ ENODEHV = (TNODE, "ENODEHV")
+ ENODELVM = (TNODE, "ENODELVM")
+ ENODEN1 = (TNODE, "ENODEN1")
+ ENODENET = (TNODE, "ENODENET")
+ ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
+ ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
+ ENODERPC = (TNODE, "ENODERPC")
+ ENODESSH = (TNODE, "ENODESSH")
+ ENODEVERSION = (TNODE, "ENODEVERSION")
+
+ ETYPE_FIELD = "code"
+ ETYPE_ERROR = "ERROR"
+ ETYPE_WARNING = "WARNING"
+
def ExpandNames(self):
self.needed_locks = {
locking.LEVEL_NODE: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
}
- self.share_locks = dict(((i, 1) for i in locking.LEVELS))
+ self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+
+ def _Error(self, ecode, item, msg, *args, **kwargs):
+ """Format an error message.
+
+ Based on the opcode's error_codes parameter, either format a
+ parseable error code, or a simpler error string.
+
+ This must be called only from Exec and functions called from Exec.
- def _VerifyNode(self, node, file_list, local_cksum, vglist, node_result,
- remote_version, feedback_fn):
+ """
+ ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
+ itype, etxt = ecode
+ # first complete the msg
+ if args:
+ msg = msg % args
+ # then format the whole message
+ if self.op.error_codes:
+ msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
+ else:
+ if item:
+ item = " " + item
+ else:
+ item = ""
+ msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
+ # and finally report it via the feedback_fn
+ self._feedback_fn(" - %s" % msg)
+
+ def _ErrorIf(self, cond, *args, **kwargs):
+ """Log an error message if the passed condition is True.
+
+ """
+ cond = bool(cond) or self.op.debug_simulate_errors
+ if cond:
+ self._Error(*args, **kwargs)
+ # do not mark the operation as failed for WARN cases only
+ if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
+ self.bad = self.bad or cond
+
+ def _VerifyNode(self, nodeinfo, file_list, local_cksum,
+ node_result, master_files, drbd_map, vg_name):
"""Run multiple tests against a node.
Test list:
+
- compares ganeti version
- - checks vg existance and size > 20G
+ - checks vg existence and size > 20G
- checks config file checksum
- checks ssh to other nodes
- Args:
- node: name of the node to check
- file_list: required list of files
- local_cksum: dictionary of local files and their checksums
+ @type nodeinfo: L{objects.Node}
+ @param nodeinfo: the node to check
+ @param file_list: required list of files
+ @param local_cksum: dictionary of local files and their checksums
+ @param node_result: the results from the node
+ @param master_files: list of files that only masters should have
+ @param drbd_map: the useddrbd minors for this node, in
+ form of minor: (instance, must_exist) which correspond to instances
+ and their running status
+ @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())
"""
+ node = nodeinfo.name
+ _ErrorIf = self._ErrorIf
+
+ # main result, node_result should be a non-empty dict
+ test = not node_result or not isinstance(node_result, dict)
+ _ErrorIf(test, self.ENODERPC, node,
+ "unable to verify node: no data returned")
+ if test:
+ return
+
# compares ganeti version
local_version = constants.PROTOCOL_VERSION
- if not remote_version:
- feedback_fn(" - ERROR: connection to %s failed" % (node))
- return True
-
- if local_version != remote_version:
- feedback_fn(" - ERROR: sw version mismatch: master %s, node(%s) %s" %
- (local_version, node, remote_version))
- return True
-
- # checks vg existance and size > 20G
-
- bad = False
- if not vglist:
- feedback_fn(" - ERROR: unable to check volume groups on node %s." %
- (node,))
- bad = True
- else:
- vgstatus = utils.CheckVolumeGroupSize(vglist, self.cfg.GetVGName(),
- constants.MIN_VG_SIZE)
- if vgstatus:
- feedback_fn(" - ERROR: %s on node %s" % (vgstatus, node))
- bad = True
+ remote_version = node_result.get('version', None)
+ test = not (remote_version and
+ isinstance(remote_version, (list, tuple)) and
+ len(remote_version) == 2)
+ _ErrorIf(test, self.ENODERPC, node,
+ "connection to node returned invalid data")
+ if test:
+ return
+
+ test = local_version != remote_version[0]
+ _ErrorIf(test, self.ENODEVERSION, node,
+ "incompatible protocol versions: master %s,"
+ " node %s", local_version, remote_version[0])
+ if test:
+ return
+
+ # node seems compatible, we can actually try to look into its results
+
+ # full package version
+ self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
+ self.ENODEVERSION, node,
+ "software version mismatch: master %s, node %s",
+ constants.RELEASE_VERSION, remote_version[1],
+ code=self.ETYPE_WARNING)
+
+ # checks vg existence and size > 20G
+ if vg_name is not None:
+ vglist = node_result.get(constants.NV_VGLIST, None)
+ test = not vglist
+ _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
+ if not test:
+ vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
+ constants.MIN_VG_SIZE)
+ _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
# checks config file checksum
- # checks ssh to any
- if 'filelist' not in node_result:
- bad = True
- feedback_fn(" - ERROR: node hasn't returned file checksum data")
- else:
- remote_cksum = node_result['filelist']
+ remote_cksum = node_result.get(constants.NV_FILELIST, None)
+ test = not isinstance(remote_cksum, dict)
+ _ErrorIf(test, self.ENODEFILECHECK, node,
+ "node hasn't returned file checksum data")
+ if not test:
for file_name in file_list:
- if file_name not in remote_cksum:
- bad = True
- feedback_fn(" - ERROR: file '%s' missing" % file_name)
- elif remote_cksum[file_name] != local_cksum[file_name]:
- bad = True
- feedback_fn(" - ERROR: file '%s' has wrong checksum" % file_name)
-
- if 'nodelist' not in node_result:
- bad = True
- feedback_fn(" - ERROR: node hasn't returned node ssh connectivity data")
- else:
- if node_result['nodelist']:
- bad = True
- for node in node_result['nodelist']:
- feedback_fn(" - ERROR: ssh communication with node '%s': %s" %
- (node, node_result['nodelist'][node]))
- if 'node-net-test' not in node_result:
- bad = True
- feedback_fn(" - ERROR: node hasn't returned node tcp connectivity data")
- else:
- if node_result['node-net-test']:
- bad = True
- nlist = utils.NiceSort(node_result['node-net-test'].keys())
- for node in nlist:
- feedback_fn(" - ERROR: tcp communication with node '%s': %s" %
- (node, node_result['node-net-test'][node]))
-
- hyp_result = node_result.get('hypervisor', None)
- if hyp_result is not None:
- feedback_fn(" - ERROR: hypervisor verify failure: '%s'" % hyp_result)
- return bad
+ node_is_mc = nodeinfo.master_candidate
+ must_have = (file_name not in master_files) or node_is_mc
+ # missing
+ test1 = file_name not in remote_cksum
+ # invalid checksum
+ test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
+ # existing and good
+ test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
+ _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
+ "file '%s' missing", file_name)
+ _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
+ "file '%s' has wrong checksum", file_name)
+ # not candidate and this is not a must-have file
+ _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
+ "file '%s' should not exist on non master"
+ " candidates (and the file is outdated)", file_name)
+ # all good, except non-master/non-must have combination
+ _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
+ "file '%s' should not exist"
+ " on non master candidates", file_name)
+
+ # checks ssh to any
+
+ test = constants.NV_NODELIST not in node_result
+ _ErrorIf(test, self.ENODESSH, node,
+ "node hasn't returned node ssh connectivity data")
+ if not test:
+ if node_result[constants.NV_NODELIST]:
+ for a_node, a_msg in node_result[constants.NV_NODELIST].items():
+ _ErrorIf(True, self.ENODESSH, node,
+ "ssh communication with node '%s': %s", a_node, a_msg)
+
+ test = constants.NV_NODENETTEST not in node_result
+ _ErrorIf(test, self.ENODENET, node,
+ "node hasn't returned node tcp connectivity data")
+ if not test:
+ if node_result[constants.NV_NODENETTEST]:
+ nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
+ for anode in nlist:
+ _ErrorIf(True, self.ENODENET, node,
+ "tcp communication with node '%s': %s",
+ anode, node_result[constants.NV_NODENETTEST][anode])
+
+ hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
+ if isinstance(hyp_result, dict):
+ for hv_name, hv_result in hyp_result.iteritems():
+ test = hv_result is not None
+ _ErrorIf(test, self.ENODEHV, node,
+ "hypervisor %s verify failure: '%s'", hv_name, hv_result)
+
+ # check used drbd list
+ if vg_name is not None:
+ used_minors = node_result.get(constants.NV_DRBDLIST, [])
+ test = not isinstance(used_minors, (tuple, list))
+ _ErrorIf(test, self.ENODEDRBD, node,
+ "cannot parse drbd status file: %s", str(used_minors))
+ if not test:
+ for minor, (iname, must_exist) in drbd_map.items():
+ test = minor not in used_minors and must_exist
+ _ErrorIf(test, self.ENODEDRBD, node,
+ "drbd minor %d of instance %s is not active",
+ minor, iname)
+ for minor in used_minors:
+ test = minor not in drbd_map
+ _ErrorIf(test, self.ENODEDRBD, node,
+ "unallocated drbd minor %d is in use", minor)
def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
- node_instance, feedback_fn):
+ node_instance, n_offline):
"""Verify an instance.
This function checks to see if the required block devices are
available on the instance's node.
"""
- bad = False
-
+ _ErrorIf = self._ErrorIf
node_current = instanceconfig.primary_node
node_vol_should = {}
instanceconfig.MapLVsByNode(node_vol_should)
for node in node_vol_should:
+ if node in n_offline:
+ # ignore missing volumes on offline nodes
+ continue
for volume in node_vol_should[node]:
- if node not in node_vol_is or volume not in node_vol_is[node]:
- feedback_fn(" - ERROR: volume %s missing on node %s" %
- (volume, node))
- bad = True
-
- if not instanceconfig.status == 'down':
- if (node_current not in node_instance or
- not instance in node_instance[node_current]):
- feedback_fn(" - ERROR: instance %s not running on node %s" %
- (instance, node_current))
- bad = True
+ test = node not in node_vol_is or volume not in node_vol_is[node]
+ _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
+ "volume %s missing on node %s", volume, node)
+
+ if instanceconfig.admin_up:
+ test = ((node_current not in node_instance or
+ not instance in node_instance[node_current]) and
+ node_current not in n_offline)
+ _ErrorIf(test, self.EINSTANCEDOWN, instance,
+ "instance not running on its primary node %s",
+ node_current)
for node in node_instance:
if (not node == node_current):
- if instance in node_instance[node]:
- feedback_fn(" - ERROR: instance %s should not run on node %s" %
- (instance, node))
- bad = True
+ test = instance in node_instance[node]
+ _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
+ "instance should not run on node %s", node)
- return bad
-
- def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
+ def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is):
"""Verify if there are any unknown volumes in the cluster.
The .os, .swap and backup volumes are ignored. All other volumes are
reported as unknown.
"""
- bad = False
-
for node in node_vol_is:
for volume in node_vol_is[node]:
- if node not in node_vol_should or volume not in node_vol_should[node]:
- feedback_fn(" - ERROR: volume %s on node %s should not exist" %
- (volume, node))
- bad = True
- return bad
+ test = (node not in node_vol_should or
+ volume not in node_vol_should[node])
+ self._ErrorIf(test, self.ENODEORPHANLV, node,
+ "volume %s is unknown", volume)
- def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
+ def _VerifyOrphanInstances(self, instancelist, node_instance):
"""Verify the list of running instances.
This checks what instances are running but unknown to the cluster.
"""
- bad = False
for node in node_instance:
- for runninginstance in node_instance[node]:
- if runninginstance not in instancelist:
- feedback_fn(" - ERROR: instance %s on node %s should not exist" %
- (runninginstance, node))
- bad = True
- return bad
-
- def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
+ for o_inst in node_instance[node]:
+ test = o_inst not in instancelist
+ self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
+ "instance %s on node %s should not exist", o_inst, node)
+
+ def _VerifyNPlusOneMemory(self, node_info, instance_cfg):
"""Verify N+1 Memory Resilience.
Check that if one single node dies we can still start all the instances it
was primary for.
"""
- bad = False
-
for node, nodeinfo in node_info.iteritems():
# This code checks that every node which is now listed as secondary has
# enough memory to host all instances it is supposed to should a single
for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
needed_mem = 0
for instance in instances:
- needed_mem += instance_cfg[instance].memory
- if nodeinfo['mfree'] < needed_mem:
- feedback_fn(" - ERROR: not enough memory on node %s to accomodate"
- " failovers should node %s fail" % (node, prinode))
- bad = True
- return bad
+ bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
+ if bep[constants.BE_AUTO_BALANCE]:
+ needed_mem += bep[constants.BE_MEMORY]
+ test = nodeinfo['mfree'] < needed_mem
+ self._ErrorIf(test, self.ENODEN1, node,
+ "not enough memory on to accommodate"
+ " failovers should peer node %s fail", prinode)
def CheckPrereq(self):
"""Check prerequisites.
def BuildHooksEnv(self):
"""Build hooks env.
- Cluster-Verify hooks just rone in the post phase and their failure makes
+ Cluster-Verify hooks just ran in the post phase and their failure makes
the output be logged in the verify output and the verification to fail.
"""
all_nodes = self.cfg.GetNodeList()
- # TODO: populate the environment with useful information for verify hooks
- env = {}
+ env = {
+ "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
+ }
+ for node in self.cfg.GetAllNodesInfo().values():
+ env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
+
return env, [], all_nodes
def Exec(self, feedback_fn):
"""Verify integrity of cluster, performing various test on nodes.
"""
- bad = False
+ self.bad = False
+ _ErrorIf = self._ErrorIf
+ verbose = self.op.verbose
+ self._feedback_fn = feedback_fn
feedback_fn("* Verifying global settings")
for msg in self.cfg.VerifyConfig():
- feedback_fn(" - ERROR: %s" % msg)
+ _ErrorIf(True, self.ECLUSTERCFG, None, msg)
vg_name = self.cfg.GetVGName()
+ hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
nodelist = utils.NiceSort(self.cfg.GetNodeList())
nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
instancelist = utils.NiceSort(self.cfg.GetInstanceList())
+ instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
+ for iname in instancelist)
i_non_redundant = [] # Non redundant instances
+ i_non_a_balanced = [] # Non auto-balanced instances
+ n_offline = [] # List of offline nodes
+ n_drained = [] # List of nodes being drained
node_volume = {}
node_instance = {}
node_info = {}
# FIXME: verify OS list
# do local checksums
- file_names = list(self.sstore.GetFileList())
+ master_files = [constants.CLUSTER_CONF_FILE]
+
+ file_names = ssconf.SimpleStore().GetFileList()
file_names.append(constants.SSL_CERT_FILE)
- file_names.append(constants.CLUSTER_CONF_FILE)
+ file_names.append(constants.RAPI_CERT_FILE)
+ file_names.extend(master_files)
+
local_checksums = utils.FingerprintFiles(file_names)
feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
- all_volumeinfo = rpc.call_volume_list(nodelist, vg_name)
- all_instanceinfo = rpc.call_instance_list(nodelist)
- all_vglist = rpc.call_vg_list(nodelist)
node_verify_param = {
- 'filelist': file_names,
- 'nodelist': nodelist,
- 'hypervisor': None,
- 'node-net-test': [(node.name, node.primary_ip, node.secondary_ip)
- for node in nodeinfo]
+ constants.NV_FILELIST: file_names,
+ constants.NV_NODELIST: [node.name for node in nodeinfo
+ if not node.offline],
+ constants.NV_HYPERVISOR: hypervisors,
+ constants.NV_NODENETTEST: [(node.name, node.primary_ip,
+ node.secondary_ip) for node in nodeinfo
+ if not node.offline],
+ constants.NV_INSTANCELIST: hypervisors,
+ constants.NV_VERSION: None,
+ constants.NV_HVINFO: self.cfg.GetHypervisorType(),
}
- all_nvinfo = rpc.call_node_verify(nodelist, node_verify_param)
- all_rversion = rpc.call_version(nodelist)
- all_ninfo = rpc.call_node_info(nodelist, self.cfg.GetVGName())
+ if vg_name is not None:
+ node_verify_param[constants.NV_VGLIST] = None
+ node_verify_param[constants.NV_LVLIST] = vg_name
+ node_verify_param[constants.NV_DRBDLIST] = None
+ all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
+ self.cfg.GetClusterName())
+
+ cluster = self.cfg.GetClusterInfo()
+ master_node = self.cfg.GetMasterNode()
+ all_drbd_map = self.cfg.ComputeDRBDMap()
+
+ feedback_fn("* Verifying node status")
+ for node_i in nodeinfo:
+ node = node_i.name
+
+ if node_i.offline:
+ if verbose:
+ feedback_fn("* Skipping offline node %s" % (node,))
+ n_offline.append(node)
+ continue
- for node in nodelist:
- feedback_fn("* Verifying node %s" % node)
- result = self._VerifyNode(node, file_names, local_checksums,
- all_vglist[node], all_nvinfo[node],
- all_rversion[node], feedback_fn)
- bad = bad or result
+ if node == master_node:
+ ntype = "master"
+ elif node_i.master_candidate:
+ ntype = "master candidate"
+ elif node_i.drained:
+ ntype = "drained"
+ n_drained.append(node)
+ else:
+ ntype = "regular"
+ if verbose:
+ feedback_fn("* Verifying node %s (%s)" % (node, ntype))
- # node_volume
- volumeinfo = all_volumeinfo[node]
+ msg = all_nvinfo[node].fail_msg
+ _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
+ if msg:
+ continue
+
+ nresult = all_nvinfo[node].payload
+ node_drbd = {}
+ for minor, instance in all_drbd_map[node].items():
+ test = instance not in instanceinfo
+ _ErrorIf(test, self.ECLUSTERCFG, None,
+ "ghost instance '%s' in temporary DRBD map", instance)
+ # ghost instance should not be running, but otherwise we
+ # don't give double warnings (both ghost instance and
+ # unallocated minor in use)
+ if test:
+ node_drbd[minor] = (instance, False)
+ else:
+ instance = instanceinfo[instance]
+ node_drbd[minor] = (instance.name, instance.admin_up)
+ self._VerifyNode(node_i, file_names, local_checksums,
+ nresult, master_files, node_drbd, vg_name)
- if isinstance(volumeinfo, basestring):
- feedback_fn(" - ERROR: LVM problem on node %s: %s" %
- (node, volumeinfo[-400:].encode('string_escape')))
- bad = True
+ lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
+ if vg_name is None:
node_volume[node] = {}
- elif not isinstance(volumeinfo, dict):
- feedback_fn(" - ERROR: connection to %s failed" % (node,))
- bad = True
+ elif isinstance(lvdata, basestring):
+ _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
+ utils.SafeEncode(lvdata))
+ node_volume[node] = {}
+ elif not isinstance(lvdata, dict):
+ _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
continue
else:
- node_volume[node] = volumeinfo
+ node_volume[node] = lvdata
# node_instance
- nodeinstance = all_instanceinfo[node]
- if type(nodeinstance) != list:
- feedback_fn(" - ERROR: connection to %s failed" % (node,))
- bad = True
+ idata = nresult.get(constants.NV_INSTANCELIST, None)
+ test = not isinstance(idata, list)
+ _ErrorIf(test, self.ENODEHV, node,
+ "rpc call to node failed (instancelist)")
+ if test:
continue
- node_instance[node] = nodeinstance
+ node_instance[node] = idata
# node_info
- nodeinfo = all_ninfo[node]
- if not isinstance(nodeinfo, dict):
- feedback_fn(" - ERROR: connection to %s failed" % (node,))
- bad = True
+ nodeinfo = nresult.get(constants.NV_HVINFO, None)
+ test = not isinstance(nodeinfo, dict)
+ _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
+ if test:
continue
try:
node_info[node] = {
"mfree": int(nodeinfo['memory_free']),
- "dfree": int(nodeinfo['vg_free']),
"pinst": [],
"sinst": [],
# dictionary holding all instances this node is secondary for,
# secondary.
"sinst-by-pnode": {},
}
- except ValueError:
- feedback_fn(" - ERROR: invalid value returned from node %s" % (node,))
- bad = True
+ # FIXME: devise a free space model for file based instances as well
+ if vg_name is not None:
+ test = (constants.NV_VGLIST not in nresult or
+ vg_name not in nresult[constants.NV_VGLIST])
+ _ErrorIf(test, self.ENODELVM, node,
+ "node didn't return data for the volume group '%s'"
+ " - it is either missing or broken", vg_name)
+ if test:
+ continue
+ node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
+ except (ValueError, KeyError):
+ _ErrorIf(True, self.ENODERPC, node,
+ "node returned invalid nodeinfo, check lvm/hypervisor")
continue
node_vol_should = {}
+ feedback_fn("* Verifying instance status")
for instance in instancelist:
- feedback_fn("* Verifying instance %s" % instance)
- inst_config = self.cfg.GetInstanceInfo(instance)
- result = self._VerifyInstance(instance, inst_config, node_volume,
- node_instance, feedback_fn)
- bad = bad or result
+ if verbose:
+ feedback_fn("* Verifying instance %s" % instance)
+ inst_config = instanceinfo[instance]
+ self._VerifyInstance(instance, inst_config, node_volume,
+ node_instance, n_offline)
+ inst_nodes_offline = []
inst_config.MapLVsByNode(node_vol_should)
instance_cfg[instance] = inst_config
pnode = inst_config.primary_node
+ _ErrorIf(pnode not in node_info and pnode not in n_offline,
+ self.ENODERPC, pnode, "instance %s, connection to"
+ " primary node failed", instance)
if pnode in node_info:
node_info[pnode]['pinst'].append(instance)
- else:
- feedback_fn(" - ERROR: instance %s, connection to primary node"
- " %s failed" % (instance, pnode))
- bad = True
+
+ if pnode in n_offline:
+ inst_nodes_offline.append(pnode)
# If the instance is non-redundant we cannot survive losing its primary
# node, so we are not N+1 compliant. On the other hand we have no disk
# FIXME: does not support file-backed instances
if len(inst_config.secondary_nodes) == 0:
i_non_redundant.append(instance)
- elif len(inst_config.secondary_nodes) > 1:
- feedback_fn(" - WARNING: multiple secondaries for instance %s"
- % instance)
+ _ErrorIf(len(inst_config.secondary_nodes) > 1,
+ self.EINSTANCELAYOUT, instance,
+ "instance has multiple secondary nodes", code="WARNING")
+
+ if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
+ i_non_a_balanced.append(instance)
for snode in inst_config.secondary_nodes:
+ _ErrorIf(snode not in node_info and snode not in n_offline,
+ self.ENODERPC, snode,
+ "instance %s, connection to secondary node"
+ "failed", instance)
+
if snode in node_info:
node_info[snode]['sinst'].append(instance)
if pnode not in node_info[snode]['sinst-by-pnode']:
node_info[snode]['sinst-by-pnode'][pnode] = []
node_info[snode]['sinst-by-pnode'][pnode].append(instance)
- else:
- feedback_fn(" - ERROR: instance %s, connection to secondary node"
- " %s failed" % (instance, snode))
+
+ if snode in n_offline:
+ inst_nodes_offline.append(snode)
+
+ # warn that the instance lives on offline nodes
+ _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
+ "instance lives on offline node(s) %s",
+ ", ".join(inst_nodes_offline))
feedback_fn("* Verifying orphan volumes")
- result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
- feedback_fn)
- bad = bad or result
+ self._VerifyOrphanVolumes(node_vol_should, node_volume)
feedback_fn("* Verifying remaining instances")
- result = self._VerifyOrphanInstances(instancelist, node_instance,
- feedback_fn)
- bad = bad or result
+ self._VerifyOrphanInstances(instancelist, node_instance)
if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
feedback_fn("* Verifying N+1 Memory redundancy")
- result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
- bad = bad or result
+ self._VerifyNPlusOneMemory(node_info, instance_cfg)
feedback_fn("* Other Notes")
if i_non_redundant:
feedback_fn(" - NOTICE: %d non-redundant instance(s) found."
% len(i_non_redundant))
- return not bad
+ if i_non_a_balanced:
+ feedback_fn(" - NOTICE: %d non-auto-balanced instance(s) found."
+ % len(i_non_a_balanced))
+
+ if n_offline:
+ feedback_fn(" - NOTICE: %d offline node(s) found." % len(n_offline))
+
+ if n_drained:
+ feedback_fn(" - NOTICE: %d drained node(s) found." % len(n_drained))
+
+ return not self.bad
def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
- """Analize the post-hooks' result, handle it, and send some
+ """Analyze the post-hooks' result
+
+ This method analyses the hook result, handles it, and sends some
nicely-formatted feedback back to the user.
- Args:
- phase: the hooks phase that has just been run
- hooks_results: the results of the multi-node hooks rpc call
- feedback_fn: function to send feedback back to the caller
- lu_result: previous Exec result
+ @param phase: one of L{constants.HOOKS_PHASE_POST} or
+ L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
+ @param hooks_results: the results of the multi-node hooks rpc call
+ @param feedback_fn: function used send feedback back to the caller
+ @param lu_result: previous Exec result
+ @return: the new Exec result, based on the previous result
+ and hook results
"""
# We only really run POST phase hooks, and are only interested in
# Used to change hooks' output to proper indentation
indent_re = re.compile('^', re.M)
feedback_fn("* Hooks Results")
- if not hooks_results:
- feedback_fn(" - ERROR: general communication failure")
- lu_result = 1
- else:
- for node_name in hooks_results:
- show_node_header = True
- res = hooks_results[node_name]
- if res is False or not isinstance(res, list):
- feedback_fn(" Communication failure")
+ assert hooks_results, "invalid result from hooks"
+
+ for node_name in hooks_results:
+ show_node_header = True
+ res = hooks_results[node_name]
+ msg = res.fail_msg
+ test = msg and not res.offline
+ self._ErrorIf(test, self.ENODEHOOKS, node_name,
+ "Communication failure in hooks execution: %s", msg)
+ if test:
+ # override manually lu_result here as _ErrorIf only
+ # overrides self.bad
+ lu_result = 1
+ continue
+ for script, hkr, output in res.payload:
+ test = hkr == constants.HKR_FAIL
+ self._ErrorIf(test, self.ENODEHOOKS, node_name,
+ "Script %s failed, output:", script)
+ if test:
+ output = indent_re.sub(' ', output)
+ feedback_fn("%s" % output)
lu_result = 1
- continue
- for script, hkr, output in res:
- if hkr == constants.HKR_FAIL:
- # The node header is only shown once, if there are
- # failing hooks on that node
- if show_node_header:
- feedback_fn(" Node %s:" % node_name)
- show_node_header = False
- feedback_fn(" ERROR: Script %s failed, output:" % script)
- output = indent_re.sub(' ', output)
- feedback_fn("%s" % output)
- lu_result = 1
return lu_result
locking.LEVEL_NODE: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
}
- self.share_locks = dict(((i, 1) for i in locking.LEVELS))
+ self.share_locks = dict.fromkeys(locking.LEVELS, 1)
def CheckPrereq(self):
"""Check prerequisites.
def Exec(self, feedback_fn):
"""Verify integrity of cluster disks.
+ @rtype: tuple of three items
+ @return: a tuple of (dict of node-to-node_error, list of instances
+ which need activate-disks, dict of instance: (node, volume) for
+ missing volumes
+
"""
- result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}
+ result = res_nodes, res_instances, res_missing = {}, [], {}
vg_name = self.cfg.GetVGName()
nodes = utils.NiceSort(self.cfg.GetNodeList())
nv_dict = {}
for inst in instances:
inst_lvs = {}
- if (inst.status != "up" or
+ if (not inst.admin_up or
inst.disk_template not in constants.DTS_NET_MIRROR):
continue
inst.MapLVsByNode(inst_lvs)
if not nv_dict:
return result
- node_lvs = rpc.call_volume_list(nodes, vg_name)
+ node_lvs = self.rpc.call_lv_list(nodes, vg_name)
- to_act = set()
for node in nodes:
# node_volume
- lvs = node_lvs[node]
-
- if isinstance(lvs, basestring):
- logger.Info("error enumerating LVs on node %s: %s" % (node, lvs))
- res_nlvm[node] = lvs
- elif not isinstance(lvs, dict):
- logger.Info("connection to node %s failed or invalid data returned" %
- (node,))
- res_nodes.append(node)
+ node_res = node_lvs[node]
+ if node_res.offline:
+ continue
+ msg = node_res.fail_msg
+ if msg:
+ logging.warning("Error enumerating LVs on node %s: %s", node, msg)
+ res_nodes[node] = msg
continue
- for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
+ lvs = node_res.payload
+ for lv_name, (_, lv_inactive, lv_online) in lvs.items():
inst = nv_dict.pop((node, lv_name), None)
if (not lv_online and inst is not None
and inst.name not in res_instances):
return result
+class LURepairDiskSizes(NoHooksLU):
+ """Verifies the cluster disks sizes.
+
+ """
+ _OP_REQP = ["instances"]
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ if not isinstance(self.op.instances, list):
+ raise errors.OpPrereqError("Invalid argument type 'instances'")
+
+ if self.op.instances:
+ self.wanted_names = []
+ for name in self.op.instances:
+ full_name = self.cfg.ExpandInstanceName(name)
+ if full_name is None:
+ raise errors.OpPrereqError("Instance '%s' not known" % name)
+ self.wanted_names.append(full_name)
+ self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
+ self.needed_locks = {
+ locking.LEVEL_NODE: [],
+ locking.LEVEL_INSTANCE: self.wanted_names,
+ }
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ else:
+ self.wanted_names = None
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_INSTANCE: locking.ALL_SET,
+ }
+ self.share_locks = dict(((i, 1) for i in locking.LEVELS))
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE and self.wanted_names is not None:
+ self._LockInstancesNodes(primary_only=True)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This only checks the optional instance list against the existing names.
+
+ """
+ if self.wanted_names is None:
+ self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+
+ self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
+ in self.wanted_names]
+
+ def Exec(self, feedback_fn):
+ """Verify the size of cluster disks.
+
+ """
+ # TODO: check child disks too
+ # TODO: check differences in size between primary/secondary nodes
+ per_node_disks = {}
+ for instance in self.wanted_instances:
+ pnode = instance.primary_node
+ if pnode not in per_node_disks:
+ per_node_disks[pnode] = []
+ for idx, disk in enumerate(instance.disks):
+ per_node_disks[pnode].append((instance, idx, disk))
+
+ changed = []
+ for node, dskl in per_node_disks.items():
+ result = self.rpc.call_blockdev_getsizes(node, [v[2] for v in dskl])
+ if result.fail_msg:
+ self.LogWarning("Failure in blockdev_getsizes call to node"
+ " %s, ignoring", node)
+ continue
+ if len(result.data) != len(dskl):
+ self.LogWarning("Invalid result from node %s, ignoring node results",
+ node)
+ continue
+ for ((instance, idx, disk), size) in zip(dskl, result.data):
+ if size is None:
+ self.LogWarning("Disk %d of instance %s did not return size"
+ " information, ignoring", idx, instance.name)
+ continue
+ if not isinstance(size, (int, long)):
+ self.LogWarning("Disk %d of instance %s did not return valid"
+ " size information, ignoring", idx, instance.name)
+ continue
+ size = size >> 20
+ if size != disk.size:
+ self.LogInfo("Disk %d of instance %s has mismatched size,"
+ " correcting: recorded %d, actual %d", idx,
+ instance.name, disk.size, size)
+ disk.size = size
+ self.cfg.Update(instance)
+ changed.append((instance.name, idx, size))
+ return changed
+
+
class LURenameCluster(LogicalUnit):
"""Rename the cluster.
HPATH = "cluster-rename"
HTYPE = constants.HTYPE_CLUSTER
_OP_REQP = ["name"]
- REQ_WSSTORE = True
def BuildHooksEnv(self):
"""Build hooks env.
"""
env = {
- "OP_TARGET": self.sstore.GetClusterName(),
+ "OP_TARGET": self.cfg.GetClusterName(),
"NEW_NAME": self.op.name,
}
- mn = self.sstore.GetMasterNode()
+ mn = self.cfg.GetMasterNode()
return env, [mn], [mn]
def CheckPrereq(self):
new_name = hostname.name
self.ip = new_ip = hostname.ip
- old_name = self.sstore.GetClusterName()
- old_ip = self.sstore.GetMasterIP()
+ old_name = self.cfg.GetClusterName()
+ old_ip = self.cfg.GetMasterIP()
if new_name == old_name and new_ip == old_ip:
raise errors.OpPrereqError("Neither the name nor the IP address of the"
" cluster has changed")
"""
clustername = self.op.name
ip = self.ip
- ss = self.sstore
# shutdown the master IP
- master = ss.GetMasterNode()
- if not rpc.call_node_stop_master(master, False):
- raise errors.OpExecError("Could not disable the master role")
+ master = self.cfg.GetMasterNode()
+ result = self.rpc.call_node_stop_master(master, False)
+ result.Raise("Could not disable the master role")
try:
- # modify the sstore
- ss.SetKey(ss.SS_MASTER_IP, ip)
- ss.SetKey(ss.SS_CLUSTER_NAME, clustername)
-
- # Distribute updated ss config to all nodes
- myself = self.cfg.GetNodeInfo(master)
- dist_nodes = self.cfg.GetNodeList()
- if myself.name in dist_nodes:
- dist_nodes.remove(myself.name)
-
- logger.Debug("Copying updated ssconf data to all nodes")
- for keyname in [ss.SS_CLUSTER_NAME, ss.SS_MASTER_IP]:
- fname = ss.KeyToFilename(keyname)
- result = rpc.call_upload_file(dist_nodes, fname)
- for to_node in dist_nodes:
- if not result[to_node]:
- logger.Error("copy of file %s to node %s failed" %
- (fname, to_node))
+ cluster = self.cfg.GetClusterInfo()
+ cluster.cluster_name = clustername
+ cluster.master_ip = ip
+ self.cfg.Update(cluster)
+
+ # update the known hosts file
+ ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
+ node_list = self.cfg.GetNodeList()
+ try:
+ node_list.remove(master)
+ except ValueError:
+ pass
+ result = self.rpc.call_upload_file(node_list,
+ constants.SSH_KNOWN_HOSTS_FILE)
+ for to_node, to_result in result.iteritems():
+ msg = to_result.fail_msg
+ if msg:
+ msg = ("Copy of file %s to node %s failed: %s" %
+ (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
+ self.proc.LogWarning(msg)
+
finally:
- if not rpc.call_node_start_master(master, False):
- logger.Error("Could not re-enable the master role on the master,"
- " please restart manually.")
+ result = self.rpc.call_node_start_master(master, False, False)
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Could not re-enable the master role on"
+ " the master, please restart manually: %s", msg)
def _RecursiveCheckIfLVMBased(disk):
"""Check if the given disk or its children are lvm-based.
- Args:
- disk: ganeti.objects.Disk object
-
- Returns:
- boolean indicating whether a LD_LV dev_type was found or not
+ @type disk: L{objects.Disk}
+ @param disk: the disk to check
+ @rtype: boolean
+ @return: boolean indicating whether a LD_LV dev_type was found or not
"""
if disk.children:
HPATH = "cluster-modify"
HTYPE = constants.HTYPE_CLUSTER
_OP_REQP = []
+ REQ_BGL = False
- def BuildHooksEnv(self):
- """Build hooks env.
+ def CheckArguments(self):
+ """Check parameters
+
+ """
+ if not hasattr(self.op, "candidate_pool_size"):
+ self.op.candidate_pool_size = None
+ if self.op.candidate_pool_size is not None:
+ try:
+ self.op.candidate_pool_size = int(self.op.candidate_pool_size)
+ except (ValueError, TypeError), err:
+ raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
+ str(err))
+ if self.op.candidate_pool_size < 1:
+ raise errors.OpPrereqError("At least one master candidate needed")
+
+ def ExpandNames(self):
+ # FIXME: in the future maybe other cluster params won't require checking on
+ # all nodes to be modified.
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ }
+ self.share_locks[locking.LEVEL_NODE] = 1
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
"""
env = {
- "OP_TARGET": self.sstore.GetClusterName(),
+ "OP_TARGET": self.cfg.GetClusterName(),
"NEW_VG_NAME": self.op.vg_name,
}
- mn = self.sstore.GetMasterNode()
+ mn = self.cfg.GetMasterNode()
return env, [mn], [mn]
def CheckPrereq(self):
if the given volume group is valid.
"""
- if not self.op.vg_name:
- instances = [self.cfg.GetInstanceInfo(name)
- for name in self.cfg.GetInstanceList()]
+ if self.op.vg_name is not None and not self.op.vg_name:
+ instances = self.cfg.GetAllInstancesInfo().values()
for inst in instances:
for disk in inst.disks:
if _RecursiveCheckIfLVMBased(disk):
raise errors.OpPrereqError("Cannot disable lvm storage while"
" lvm-based instances exist")
+ node_list = self.acquired_locks[locking.LEVEL_NODE]
+
# if vg_name not None, checks given volume group on all nodes
if self.op.vg_name:
- node_list = self.cfg.GetNodeList()
- vglist = rpc.call_vg_list(node_list)
+ vglist = self.rpc.call_vg_list(node_list)
for node in node_list:
- vgstatus = utils.CheckVolumeGroupSize(vglist[node], self.op.vg_name,
+ msg = vglist[node].fail_msg
+ if msg:
+ # ignoring down node
+ self.LogWarning("Error while gathering data on node %s"
+ " (ignoring node): %s", node, msg)
+ continue
+ vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
+ self.op.vg_name,
constants.MIN_VG_SIZE)
if vgstatus:
raise errors.OpPrereqError("Error on node '%s': %s" %
(node, vgstatus))
+ self.cluster = cluster = self.cfg.GetClusterInfo()
+ # validate params changes
+ if self.op.beparams:
+ utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
+ self.new_beparams = objects.FillDict(
+ cluster.beparams[constants.PP_DEFAULT], self.op.beparams)
+
+ if self.op.nicparams:
+ utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
+ self.new_nicparams = objects.FillDict(
+ cluster.nicparams[constants.PP_DEFAULT], self.op.nicparams)
+ objects.NIC.CheckParameterSyntax(self.new_nicparams)
+
+ # hypervisor list/parameters
+ self.new_hvparams = objects.FillDict(cluster.hvparams, {})
+ if self.op.hvparams:
+ if not isinstance(self.op.hvparams, dict):
+ raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
+ for hv_name, hv_dict in self.op.hvparams.items():
+ if hv_name not in self.new_hvparams:
+ self.new_hvparams[hv_name] = hv_dict
+ else:
+ self.new_hvparams[hv_name].update(hv_dict)
+
+ if self.op.enabled_hypervisors is not None:
+ self.hv_list = self.op.enabled_hypervisors
+ if not self.hv_list:
+ raise errors.OpPrereqError("Enabled hypervisors list must contain at"
+ " least one member")
+ invalid_hvs = set(self.hv_list) - constants.HYPER_TYPES
+ if invalid_hvs:
+ raise errors.OpPrereqError("Enabled hypervisors contains invalid"
+ " entries: %s" %
+ utils.CommaJoin(invalid_hvs))
+ else:
+ self.hv_list = cluster.enabled_hypervisors
+
+ if self.op.hvparams or self.op.enabled_hypervisors is not None:
+ # either the enabled list has changed, or the parameters have, validate
+ for hv_name, hv_params in self.new_hvparams.items():
+ if ((self.op.hvparams and hv_name in self.op.hvparams) or
+ (self.op.enabled_hypervisors and
+ hv_name in self.op.enabled_hypervisors)):
+ # either this is a new hypervisor, or its parameters have changed
+ hv_class = hypervisor.GetHypervisor(hv_name)
+ utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
+ hv_class.CheckParameterSyntax(hv_params)
+ _CheckHVParams(self, node_list, hv_name, hv_params)
+
def Exec(self, feedback_fn):
"""Change the parameters of the cluster.
"""
- if self.op.vg_name != self.cfg.GetVGName():
- self.cfg.SetVGName(self.op.vg_name)
- else:
- feedback_fn("Cluster LVM configuration already in desired"
- " state, not changing")
+ if self.op.vg_name is not None:
+ new_volume = self.op.vg_name
+ if not new_volume:
+ new_volume = None
+ if new_volume != self.cfg.GetVGName():
+ self.cfg.SetVGName(new_volume)
+ else:
+ feedback_fn("Cluster LVM configuration already in desired"
+ " state, not changing")
+ if self.op.hvparams:
+ self.cluster.hvparams = self.new_hvparams
+ if self.op.enabled_hypervisors is not None:
+ self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
+ if self.op.beparams:
+ self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
+ if self.op.nicparams:
+ self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
+
+ if self.op.candidate_pool_size is not None:
+ self.cluster.candidate_pool_size = self.op.candidate_pool_size
+ # we need to update the pool size here, otherwise the save will fail
+ _AdjustCandidatePool(self)
+
+ self.cfg.Update(self.cluster)
+
+
+def _RedistributeAncillaryFiles(lu, additional_nodes=None):
+ """Distribute additional files which are part of the cluster configuration.
+
+ ConfigWriter takes care of distributing the config and ssconf files, but
+ there are more files which should be distributed to all nodes. This function
+ makes sure those are copied.
+
+ @param lu: calling logical unit
+ @param additional_nodes: list of nodes not in the config to distribute to
+
+ """
+ # 1. Gather target nodes
+ myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
+ dist_nodes = lu.cfg.GetNodeList()
+ if additional_nodes is not None:
+ dist_nodes.extend(additional_nodes)
+ if myself.name in dist_nodes:
+ dist_nodes.remove(myself.name)
+ # 2. Gather files to distribute
+ dist_files = set([constants.ETC_HOSTS,
+ constants.SSH_KNOWN_HOSTS_FILE,
+ constants.RAPI_CERT_FILE,
+ constants.RAPI_USERS_FILE,
+ constants.HMAC_CLUSTER_KEY,
+ ])
+
+ enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
+ for hv_name in enabled_hypervisors:
+ hv_class = hypervisor.GetHypervisor(hv_name)
+ dist_files.update(hv_class.GetAncillaryFiles())
+
+ # 3. Perform the files upload
+ for fname in dist_files:
+ if os.path.exists(fname):
+ result = lu.rpc.call_upload_file(dist_nodes, fname)
+ for to_node, to_result in result.items():
+ msg = to_result.fail_msg
+ if msg:
+ msg = ("Copy of file %s to node %s failed: %s" %
+ (fname, to_node, msg))
+ lu.proc.LogWarning(msg)
+
+
+class LURedistributeConfig(NoHooksLU):
+ """Force the redistribution of cluster configuration.
+
+ This is a very simple LU.
+
+ """
+ _OP_REQP = []
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ }
+ self.share_locks[locking.LEVEL_NODE] = 1
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ """
+
+ def Exec(self, feedback_fn):
+ """Redistribute the configuration.
+
+ """
+ self.cfg.Update(self.cfg.GetClusterInfo())
+ _RedistributeAncillaryFiles(self)
-def _WaitForSync(cfgw, instance, proc, oneshot=False, unlock=False):
+def _WaitForSync(lu, instance, oneshot=False, unlock=False):
"""Sleep and poll for an instance's disk to sync.
"""
return True
if not oneshot:
- proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
+ lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
node = instance.primary_node
for dev in instance.disks:
- cfgw.SetDiskID(dev, node)
+ lu.cfg.SetDiskID(dev, node)
retries = 0
+ degr_retries = 10 # in seconds, as we sleep 1 second each time
while True:
max_time = 0
done = True
cumul_degraded = False
- rstats = rpc.call_blockdev_getmirrorstatus(node, instance.disks)
- if not rstats:
- proc.LogWarning("Can't get any data from node %s" % node)
+ rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
+ msg = rstats.fail_msg
+ if msg:
+ lu.LogWarning("Can't get any data from node %s: %s", node, msg)
retries += 1
if retries >= 10:
raise errors.RemoteError("Can't contact node %s for mirror data,"
" aborting." % node)
time.sleep(6)
continue
+ rstats = rstats.payload
retries = 0
- for i in range(len(rstats)):
- mstat = rstats[i]
+ for i, mstat in enumerate(rstats):
if mstat is None:
- proc.LogWarning("Can't compute data for node %s/%s" %
- (node, instance.disks[i].iv_name))
+ lu.LogWarning("Can't compute data for node %s/%s",
+ node, instance.disks[i].iv_name)
continue
- # we ignore the ldisk parameter
- perc_done, est_time, is_degraded, _ = mstat
- cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
- if perc_done is not None:
+
+ cumul_degraded = (cumul_degraded or
+ (mstat.is_degraded and mstat.sync_percent is None))
+ if mstat.sync_percent is not None:
done = False
- if est_time is not None:
- rem_time = "%d estimated seconds remaining" % est_time
- max_time = est_time
+ if mstat.estimated_time is not None:
+ rem_time = "%d estimated seconds remaining" % mstat.estimated_time
+ max_time = mstat.estimated_time
else:
rem_time = "no time estimate"
- proc.LogInfo("- device %s: %5.2f%% done, %s" %
- (instance.disks[i].iv_name, perc_done, rem_time))
+ lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
+ (instance.disks[i].iv_name, mstat.sync_percent,
+ rem_time))
+
+ # if we're done but degraded, let's do a few small retries, to
+ # make sure we see a stable and not transient situation; therefore
+ # we force restart of the loop
+ if (done or oneshot) and cumul_degraded and degr_retries > 0:
+ logging.info("Degraded disks found, %d retries left", degr_retries)
+ degr_retries -= 1
+ time.sleep(1)
+ continue
+
if done or oneshot:
break
time.sleep(min(60, max_time))
if done:
- proc.LogInfo("Instance %s's disks are in sync." % instance.name)
+ lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
return not cumul_degraded
-def _CheckDiskConsistency(cfgw, dev, node, on_primary, ldisk=False):
+def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
"""Check that mirrors are not degraded.
The ldisk parameter, if True, will change the test from the
the device(s)) to the ldisk (representing the local storage status).
"""
- cfgw.SetDiskID(dev, node)
- if ldisk:
- idx = 6
- else:
- idx = 5
+ lu.cfg.SetDiskID(dev, node)
result = True
+
if on_primary or dev.AssembleOnSecondary():
- rstats = rpc.call_blockdev_find(node, dev)
- if not rstats:
- logger.ToStderr("Node %s: Disk degraded, not found or node down" % node)
+ rstats = lu.rpc.call_blockdev_find(node, dev)
+ msg = rstats.fail_msg
+ if msg:
+ lu.LogWarning("Can't find disk on node %s: %s", node, msg)
+ result = False
+ elif not rstats.payload:
+ lu.LogWarning("Can't find disk on node %s", node)
result = False
else:
- result = result and (not rstats[idx])
+ if ldisk:
+ result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
+ else:
+ result = result and not rstats.payload.is_degraded
+
if dev.children:
for child in dev.children:
- result = result and _CheckDiskConsistency(cfgw, child, node, on_primary)
+ result = result and _CheckDiskConsistency(lu, child, node, on_primary)
return result
"""
_OP_REQP = ["output_fields", "names"]
REQ_BGL = False
+ _FIELDS_STATIC = utils.FieldSet()
+ _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
def ExpandNames(self):
if self.op.names:
raise errors.OpPrereqError("Selective OS query not supported")
- self.dynamic_fields = frozenset(["name", "valid", "node_status"])
- _CheckOutputFields(static=[],
- dynamic=self.dynamic_fields,
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=self._FIELDS_DYNAMIC,
selected=self.op.output_fields)
# Lock all nodes, in shared mode
+ # Temporary removal of locks, should be reverted later
+ # TODO: reintroduce locks when they are lighter-weight
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ #self.share_locks[locking.LEVEL_NODE] = 1
+ #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
def CheckPrereq(self):
"""Check prerequisites.
def _DiagnoseByOS(node_list, rlist):
"""Remaps a per-node return list into an a per-os per-node dictionary
- Args:
- node_list: a list with the names of all nodes
- rlist: a map with node names as keys and OS objects as values
+ @param node_list: a list with the names of all nodes
+ @param rlist: a map with node names as keys and OS objects as values
- Returns:
- map: a map with osnames as keys and as value another map, with
- nodes as
- keys and list of OS objects as values
- e.g. {"debian-etch": {"node1": [<object>,...],
- "node2": [<object>,]}
- }
+ @rtype: dict
+ @return: a dictionary with osnames as keys and as value another map, with
+ nodes as keys and tuples of (path, status, diagnose) as values, eg::
+
+ {"debian-etch": {"node1": [(/usr/lib/..., True, ""),
+ (/srv/..., False, "invalid api")],
+ "node2": [(/srv/..., True, "")]}
+ }
"""
all_os = {}
- for node_name, nr in rlist.iteritems():
- if not nr:
+ # we build here the list of nodes that didn't fail the RPC (at RPC
+ # level), so that nodes with a non-responding node daemon don't
+ # make all OSes invalid
+ good_nodes = [node_name for node_name in rlist
+ if not rlist[node_name].fail_msg]
+ for node_name, nr in rlist.items():
+ if nr.fail_msg or not nr.payload:
continue
- for os_obj in nr:
- if os_obj.name not in all_os:
+ for name, path, status, diagnose in nr.payload:
+ if name not in all_os:
# build a list of nodes for this os containing empty lists
# for each node in node_list
- all_os[os_obj.name] = {}
- for nname in node_list:
- all_os[os_obj.name][nname] = []
- all_os[os_obj.name][node_name].append(os_obj)
+ all_os[name] = {}
+ for nname in good_nodes:
+ all_os[name][nname] = []
+ all_os[name][node_name].append((path, status, diagnose))
return all_os
def Exec(self, feedback_fn):
"""Compute the list of OSes.
"""
- node_list = self.acquired_locks[locking.LEVEL_NODE]
- node_data = rpc.call_os_diagnose(node_list)
- if node_data == False:
- raise errors.OpExecError("Can't gather the list of OSes")
- pol = self._DiagnoseByOS(node_list, node_data)
+ valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
+ node_data = self.rpc.call_os_diagnose(valid_nodes)
+ pol = self._DiagnoseByOS(valid_nodes, node_data)
output = []
- for os_name, os_data in pol.iteritems():
+ for os_name, os_data in pol.items():
row = []
for field in self.op.output_fields:
if field == "name":
val = os_name
elif field == "valid":
- val = utils.all([osl and osl[0] for osl in os_data.values()])
+ val = utils.all([osl and osl[0][1] for osl in os_data.values()])
elif field == "node_status":
+ # this is just a copy of the dict
val = {}
- for node_name, nos_list in os_data.iteritems():
- val[node_name] = [(v.status, v.path) for v in nos_list]
+ for node_name, nos_list in os_data.items():
+ val[node_name] = nos_list
else:
raise errors.ParameterError(field)
row.append(val)
"NODE_NAME": self.op.node_name,
}
all_nodes = self.cfg.GetNodeList()
- all_nodes.remove(self.op.node_name)
+ if self.op.node_name in all_nodes:
+ all_nodes.remove(self.op.node_name)
return env, all_nodes, all_nodes
def CheckPrereq(self):
- it does not have primary or secondary instances
- it's not the master
- Any errors are signalled by raising errors.OpPrereqError.
+ Any errors are signaled by raising errors.OpPrereqError.
"""
node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
instance_list = self.cfg.GetInstanceList()
- masternode = self.sstore.GetMasterNode()
+ masternode = self.cfg.GetMasterNode()
if node.name == masternode:
raise errors.OpPrereqError("Node is the master node,"
" you need to failover first.")
for instance_name in instance_list:
instance = self.cfg.GetInstanceInfo(instance_name)
- if node.name == instance.primary_node:
- raise errors.OpPrereqError("Instance %s still running on the node,"
- " please remove first." % instance_name)
- if node.name in instance.secondary_nodes:
- raise errors.OpPrereqError("Instance %s has node as a secondary,"
+ if node.name in instance.all_nodes:
+ raise errors.OpPrereqError("Instance %s is still running on the node,"
" please remove first." % instance_name)
self.op.node_name = node.name
self.node = node
"""
node = self.node
- logger.Info("stopping the node daemon and removing configs from node %s" %
- node.name)
+ logging.info("Stopping the node daemon and removing configs from node %s",
+ node.name)
self.context.RemoveNode(node.name)
- rpc.call_node_leave_cluster(node.name)
+ # Run post hooks on the node before it's removed
+ hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
+ try:
+ h_results = hm.RunPhase(constants.HOOKS_PHASE_POST, [node.name])
+ except:
+ self.LogWarning("Errors occurred running hooks on %s" % node.name)
+
+ result = self.rpc.call_node_leave_cluster(node.name)
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Errors encountered on the remote node while leaving"
+ " the cluster: %s", msg)
+
+ # Promote nodes to master candidate as needed
+ _AdjustCandidatePool(self)
class LUQueryNodes(NoHooksLU):
"""Logical unit for querying nodes.
"""
- _OP_REQP = ["output_fields", "names"]
+ _OP_REQP = ["output_fields", "names", "use_locking"]
REQ_BGL = False
+ _SIMPLE_FIELDS = ["name", "serial_no", "ctime", "mtime", "uuid",
+ "master_candidate", "offline", "drained"]
+
+ _FIELDS_DYNAMIC = utils.FieldSet(
+ "dtotal", "dfree",
+ "mtotal", "mnode", "mfree",
+ "bootid",
+ "ctotal", "cnodes", "csockets",
+ )
+
+ _FIELDS_STATIC = utils.FieldSet(*[
+ "pinst_cnt", "sinst_cnt",
+ "pinst_list", "sinst_list",
+ "pip", "sip", "tags",
+ "master",
+ "role"] + _SIMPLE_FIELDS
+ )
+
def ExpandNames(self):
- self.dynamic_fields = frozenset([
- "dtotal", "dfree",
- "mtotal", "mnode", "mfree",
- "bootid",
- "ctotal",
- ])
-
- _CheckOutputFields(static=["name", "pinst_cnt", "sinst_cnt",
- "pinst_list", "sinst_list",
- "pip", "sip", "tags"],
- dynamic=self.dynamic_fields,
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=self._FIELDS_DYNAMIC,
selected=self.op.output_fields)
self.needed_locks = {}
self.share_locks[locking.LEVEL_NODE] = 1
- # TODO: we could lock nodes only if the user asked for dynamic fields. For
- # that we need atomic ways to get info for a group of nodes from the
- # config, though.
- if not self.op.names:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+ if self.op.names:
+ self.wanted = _GetWantedNodes(self, self.op.names)
else:
- self.needed_locks[locking.LEVEL_NODE] = \
- _GetWantedNodes(self, self.op.names)
+ self.wanted = locking.ALL_SET
+
+ self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
+ self.do_locking = self.do_node_query and self.op.use_locking
+ if self.do_locking:
+ # if we don't request only static fields, we need to lock the nodes
+ self.needed_locks[locking.LEVEL_NODE] = self.wanted
+
def CheckPrereq(self):
"""Check prerequisites.
"""
- # This of course is valid only if we locked the nodes
- self.wanted = self.acquired_locks[locking.LEVEL_NODE]
+ # The validation of the node list is done in the _GetWantedNodes,
+ # if non empty, and if empty, there's no validation to do
+ pass
def Exec(self, feedback_fn):
"""Computes the list of nodes and their attributes.
"""
- nodenames = self.wanted
- nodelist = [self.cfg.GetNodeInfo(name) for name in nodenames]
+ all_info = self.cfg.GetAllNodesInfo()
+ if self.do_locking:
+ nodenames = self.acquired_locks[locking.LEVEL_NODE]
+ elif self.wanted != locking.ALL_SET:
+ nodenames = self.wanted
+ missing = set(nodenames).difference(all_info.keys())
+ if missing:
+ raise errors.OpExecError(
+ "Some nodes were removed before retrieving their data: %s" % missing)
+ else:
+ nodenames = all_info.keys()
+
+ nodenames = utils.NiceSort(nodenames)
+ nodelist = [all_info[name] for name in nodenames]
# begin data gathering
- if self.dynamic_fields.intersection(self.op.output_fields):
+ if self.do_node_query:
live_data = {}
- node_data = rpc.call_node_info(nodenames, self.cfg.GetVGName())
+ node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
+ self.cfg.GetHypervisorType())
for name in nodenames:
- nodeinfo = node_data.get(name, None)
- if nodeinfo:
+ nodeinfo = node_data[name]
+ if not nodeinfo.fail_msg and nodeinfo.payload:
+ nodeinfo = nodeinfo.payload
+ fn = utils.TryConvert
live_data[name] = {
- "mtotal": utils.TryConvert(int, nodeinfo['memory_total']),
- "mnode": utils.TryConvert(int, nodeinfo['memory_dom0']),
- "mfree": utils.TryConvert(int, nodeinfo['memory_free']),
- "dtotal": utils.TryConvert(int, nodeinfo['vg_size']),
- "dfree": utils.TryConvert(int, nodeinfo['vg_free']),
- "ctotal": utils.TryConvert(int, nodeinfo['cpu_total']),
- "bootid": nodeinfo['bootid'],
+ "mtotal": fn(int, nodeinfo.get('memory_total', None)),
+ "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
+ "mfree": fn(int, nodeinfo.get('memory_free', None)),
+ "dtotal": fn(int, nodeinfo.get('vg_size', None)),
+ "dfree": fn(int, nodeinfo.get('vg_free', None)),
+ "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
+ "bootid": nodeinfo.get('bootid', None),
+ "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
+ "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
}
else:
live_data[name] = {}
if secnode in node_to_secondary:
node_to_secondary[secnode].add(inst.name)
+ master_node = self.cfg.GetMasterNode()
+
# end data gathering
output = []
for node in nodelist:
node_output = []
for field in self.op.output_fields:
- if field == "name":
- val = node.name
+ if field in self._SIMPLE_FIELDS:
+ val = getattr(node, field)
elif field == "pinst_list":
val = list(node_to_primary[node.name])
elif field == "sinst_list":
val = node.secondary_ip
elif field == "tags":
val = list(node.GetTags())
- elif field in self.dynamic_fields:
+ elif field == "master":
+ val = node.name == master_node
+ elif self._FIELDS_DYNAMIC.Matches(field):
val = live_data[node.name].get(field, None)
+ elif field == "role":
+ if node.name == master_node:
+ val = "M"
+ elif node.master_candidate:
+ val = "C"
+ elif node.drained:
+ val = "D"
+ elif node.offline:
+ val = "O"
+ else:
+ val = "R"
else:
raise errors.ParameterError(field)
node_output.append(val)
"""
_OP_REQP = ["nodes", "output_fields"]
REQ_BGL = False
+ _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
+ _FIELDS_STATIC = utils.FieldSet("node")
def ExpandNames(self):
- _CheckOutputFields(static=["node"],
- dynamic=["phys", "vg", "name", "size", "instance"],
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=self._FIELDS_DYNAMIC,
selected=self.op.output_fields)
self.needed_locks = {}
"""
nodenames = self.nodes
- volumes = rpc.call_node_volumes(nodenames)
+ volumes = self.rpc.call_node_volumes(nodenames)
ilist = [self.cfg.GetInstanceInfo(iname) for iname
in self.cfg.GetInstanceList()]
output = []
for node in nodenames:
- if node not in volumes or not volumes[node]:
+ nresult = volumes[node]
+ if nresult.offline:
+ continue
+ msg = nresult.fail_msg
+ if msg:
+ self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
continue
- node_vols = volumes[node][:]
+ node_vols = nresult.payload[:]
node_vols.sort(key=lambda vol: vol['dev'])
for vol in node_vols:
return output
+class LUQueryNodeStorage(NoHooksLU):
+ """Logical unit for getting information on storage units on node(s).
+
+ """
+ _OP_REQP = ["nodes", "storage_type", "output_fields"]
+ REQ_BGL = False
+ _FIELDS_STATIC = utils.FieldSet("node")
+
+ def ExpandNames(self):
+ storage_type = self.op.storage_type
+
+ if storage_type not in constants.VALID_STORAGE_FIELDS:
+ raise errors.OpPrereqError("Unknown storage type: %s" % storage_type)
+
+ dynamic_fields = constants.VALID_STORAGE_FIELDS[storage_type]
+
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=utils.FieldSet(*dynamic_fields),
+ selected=self.op.output_fields)
+
+ self.needed_locks = {}
+ self.share_locks[locking.LEVEL_NODE] = 1
+
+ if self.op.nodes:
+ self.needed_locks[locking.LEVEL_NODE] = \
+ _GetWantedNodes(self, self.op.nodes)
+ else:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the fields required are valid output fields.
+
+ """
+ self.op.name = getattr(self.op, "name", None)
+
+ self.nodes = self.acquired_locks[locking.LEVEL_NODE]
+
+ def Exec(self, feedback_fn):
+ """Computes the list of nodes and their attributes.
+
+ """
+ # Always get name to sort by
+ if constants.SF_NAME in self.op.output_fields:
+ fields = self.op.output_fields[:]
+ else:
+ fields = [constants.SF_NAME] + self.op.output_fields
+
+ # Never ask for node as it's only known to the LU
+ while "node" in fields:
+ fields.remove("node")
+
+ field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
+ name_idx = field_idx[constants.SF_NAME]
+
+ st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
+ data = self.rpc.call_storage_list(self.nodes,
+ self.op.storage_type, st_args,
+ self.op.name, fields)
+
+ result = []
+
+ for node in utils.NiceSort(self.nodes):
+ nresult = data[node]
+ if nresult.offline:
+ continue
+
+ msg = nresult.fail_msg
+ if msg:
+ self.LogWarning("Can't get storage data from node %s: %s", node, msg)
+ continue
+
+ rows = dict([(row[name_idx], row) for row in nresult.payload])
+
+ for name in utils.NiceSort(rows.keys()):
+ row = rows[name]
+
+ out = []
+
+ for field in self.op.output_fields:
+ if field == "node":
+ val = node
+ elif field in field_idx:
+ val = row[field_idx[field]]
+ else:
+ raise errors.ParameterError(field)
+
+ out.append(val)
+
+ result.append(out)
+
+ return result
+
+
+class LUModifyNodeStorage(NoHooksLU):
+ """Logical unit for modifying a storage volume on a node.
+
+ """
+ _OP_REQP = ["node_name", "storage_type", "name", "changes"]
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if node_name is None:
+ raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
+
+ self.op.node_name = node_name
+
+ storage_type = self.op.storage_type
+ if storage_type not in constants.VALID_STORAGE_FIELDS:
+ raise errors.OpPrereqError("Unknown storage type: %s" % storage_type)
+
+ def ExpandNames(self):
+ self.needed_locks = {
+ locking.LEVEL_NODE: self.op.node_name,
+ }
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ """
+ storage_type = self.op.storage_type
+
+ try:
+ modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
+ except KeyError:
+ raise errors.OpPrereqError("Storage units of type '%s' can not be"
+ " modified" % storage_type)
+
+ diff = set(self.op.changes.keys()) - modifiable
+ if diff:
+ raise errors.OpPrereqError("The following fields can not be modified for"
+ " storage units of type '%s': %r" %
+ (storage_type, list(diff)))
+
+ def Exec(self, feedback_fn):
+ """Computes the list of nodes and their attributes.
+
+ """
+ st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
+ result = self.rpc.call_storage_modify(self.op.node_name,
+ self.op.storage_type, st_args,
+ self.op.name, self.op.changes)
+ result.Raise("Failed to modify storage unit '%s' on %s" %
+ (self.op.name, self.op.node_name))
+
+
class LUAddNode(LogicalUnit):
"""Logical unit for adding node to the cluster.
- it is resolvable
- its parameters (single/dual homed) matches the cluster
- Any errors are signalled by raising errors.OpPrereqError.
+ Any errors are signaled by raising errors.OpPrereqError.
"""
node_name = self.op.node_name
# check that the type of the node (single versus dual homed) is the
# same as for the master
- myself = cfg.GetNodeInfo(self.sstore.GetMasterNode())
+ myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
master_singlehomed = myself.secondary_ip == myself.primary_ip
newbie_singlehomed = secondary_ip == primary_ip
if master_singlehomed != newbie_singlehomed:
raise errors.OpPrereqError("The master has a private ip but the"
" new node doesn't have one")
- # checks reachablity
+ # checks reachability
if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
raise errors.OpPrereqError("Node not reachable by ping")
raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
" based ping to noded port")
- self.new_node = objects.Node(name=node,
- primary_ip=primary_ip,
- secondary_ip=secondary_ip)
+ cp_size = self.cfg.GetClusterInfo().candidate_pool_size
+ if self.op.readd:
+ exceptions = [node]
+ else:
+ exceptions = []
+ mc_now, mc_max = self.cfg.GetMasterCandidateStats(exceptions)
+ # the new node will increase mc_max with one, so:
+ mc_max = min(mc_max + 1, cp_size)
+ self.master_candidate = mc_now < mc_max
+
+ if self.op.readd:
+ self.new_node = self.cfg.GetNodeInfo(node)
+ assert self.new_node is not None, "Can't retrieve locked node %s" % node
+ else:
+ self.new_node = objects.Node(name=node,
+ primary_ip=primary_ip,
+ secondary_ip=secondary_ip,
+ master_candidate=self.master_candidate,
+ offline=False, drained=False)
def Exec(self, feedback_fn):
"""Adds the new node to the cluster.
new_node = self.new_node
node = new_node.name
+ # for re-adds, reset the offline/drained/master-candidate flags;
+ # we need to reset here, otherwise offline would prevent RPC calls
+ # later in the procedure; this also means that if the re-add
+ # fails, we are left with a non-offlined, broken node
+ if self.op.readd:
+ new_node.drained = new_node.offline = False
+ self.LogInfo("Readding a node, the offline/drained flags were reset")
+ # if we demote the node, we do cleanup later in the procedure
+ new_node.master_candidate = self.master_candidate
+
+ # notify the user about any possible mc promotion
+ if new_node.master_candidate:
+ self.LogInfo("Node will be a master candidate")
+
# check connectivity
- result = rpc.call_version([node])[node]
- if result:
- if constants.PROTOCOL_VERSION == result:
- logger.Info("communication to node %s fine, sw version %s match" %
- (node, result))
- else:
- raise errors.OpExecError("Version mismatch master version %s,"
- " node version %s" %
- (constants.PROTOCOL_VERSION, result))
+ result = self.rpc.call_version([node])[node]
+ result.Raise("Can't get version information from node %s" % node)
+ if constants.PROTOCOL_VERSION == result.payload:
+ logging.info("Communication to node %s fine, sw version %s match",
+ node, result.payload)
else:
- raise errors.OpExecError("Cannot get version from the new node")
+ raise errors.OpExecError("Version mismatch master version %s,"
+ " node version %s" %
+ (constants.PROTOCOL_VERSION, result.payload))
# setup ssh on node
- logger.Info("copy ssh key to node %s" % node)
+ logging.info("Copy ssh key to node %s", node)
priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
keyarray = []
keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
priv_key, pub_key]
for i in keyfiles:
- f = open(i, 'r')
- try:
- keyarray.append(f.read())
- finally:
- f.close()
-
- result = rpc.call_node_add(node, keyarray[0], keyarray[1], keyarray[2],
- keyarray[3], keyarray[4], keyarray[5])
+ keyarray.append(utils.ReadFile(i))
- if not result:
- raise errors.OpExecError("Cannot transfer ssh keys to the new node")
+ result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
+ keyarray[2],
+ keyarray[3], keyarray[4], keyarray[5])
+ result.Raise("Cannot transfer ssh keys to the new node")
# Add node to our /etc/hosts, and add key to known_hosts
- utils.AddHostToEtcHosts(new_node.name)
+ if self.cfg.GetClusterInfo().modify_etc_hosts:
+ utils.AddHostToEtcHosts(new_node.name)
if new_node.secondary_ip != new_node.primary_ip:
- if not rpc.call_node_tcp_ping(new_node.name,
- constants.LOCALHOST_IP_ADDRESS,
- new_node.secondary_ip,
- constants.DEFAULT_NODED_PORT,
- 10, False):
+ result = self.rpc.call_node_has_ip_address(new_node.name,
+ new_node.secondary_ip)
+ result.Raise("Failure checking secondary ip on node %s" % new_node.name,
+ prereq=True)
+ if not result.payload:
raise errors.OpExecError("Node claims it doesn't have the secondary ip"
" you gave (%s). Please fix and re-run this"
" command." % new_node.secondary_ip)
- node_verify_list = [self.sstore.GetMasterNode()]
+ node_verify_list = [self.cfg.GetMasterNode()]
node_verify_param = {
- 'nodelist': [node],
+ constants.NV_NODELIST: [node],
# TODO: do a node-net-test as well?
}
- result = rpc.call_node_verify(node_verify_list, node_verify_param)
+ result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
+ self.cfg.GetClusterName())
for verifier in node_verify_list:
- if not result[verifier]:
- raise errors.OpExecError("Cannot communicate with %s's node daemon"
- " for remote verification" % verifier)
- if result[verifier]['nodelist']:
- for failed in result[verifier]['nodelist']:
+ result[verifier].Raise("Cannot communicate with node %s" % verifier)
+ nl_payload = result[verifier].payload[constants.NV_NODELIST]
+ if nl_payload:
+ for failed in nl_payload:
feedback_fn("ssh/hostname verification failed %s -> %s" %
- (verifier, result[verifier]['nodelist'][failed]))
+ (verifier, nl_payload[failed]))
raise errors.OpExecError("ssh/hostname verification failed.")
- # Distribute updated /etc/hosts and known_hosts to all nodes,
- # including the node just added
- myself = self.cfg.GetNodeInfo(self.sstore.GetMasterNode())
- dist_nodes = self.cfg.GetNodeList()
- if not self.op.readd:
- dist_nodes.append(node)
- if myself.name in dist_nodes:
- dist_nodes.remove(myself.name)
-
- logger.Debug("Copying hosts and known_hosts to all nodes")
- for fname in (constants.ETC_HOSTS, constants.SSH_KNOWN_HOSTS_FILE):
- result = rpc.call_upload_file(dist_nodes, fname)
- for to_node in dist_nodes:
- if not result[to_node]:
- logger.Error("copy of file %s to node %s failed" %
- (fname, to_node))
-
- to_copy = self.sstore.GetFileList()
- if self.sstore.GetHypervisorType() == constants.HT_XEN_HVM31:
- to_copy.append(constants.VNC_PASSWORD_FILE)
- for fname in to_copy:
- result = rpc.call_upload_file([node], fname)
- if not result[node]:
- logger.Error("could not copy file %s to node %s" % (fname, node))
-
if self.op.readd:
+ _RedistributeAncillaryFiles(self)
self.context.ReaddNode(new_node)
+ # make sure we redistribute the config
+ self.cfg.Update(new_node)
+ # and make sure the new node will not have old files around
+ if not new_node.master_candidate:
+ result = self.rpc.call_node_demote_from_mc(new_node.name)
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Node failed to demote itself from master"
+ " candidate status: %s" % msg)
else:
+ _RedistributeAncillaryFiles(self, additional_nodes=[node])
self.context.AddNode(new_node)
+class LUSetNodeParams(LogicalUnit):
+ """Modifies the parameters of a node.
+
+ """
+ HPATH = "node-modify"
+ HTYPE = constants.HTYPE_NODE
+ _OP_REQP = ["node_name"]
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if node_name is None:
+ raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
+ self.op.node_name = node_name
+ _CheckBooleanOpField(self.op, 'master_candidate')
+ _CheckBooleanOpField(self.op, 'offline')
+ _CheckBooleanOpField(self.op, 'drained')
+ all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
+ if all_mods.count(None) == 3:
+ raise errors.OpPrereqError("Please pass at least one modification")
+ if all_mods.count(True) > 1:
+ raise errors.OpPrereqError("Can't set the node into more than one"
+ " state at the same time")
+
+ def ExpandNames(self):
+ self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master node.
+
+ """
+ env = {
+ "OP_TARGET": self.op.node_name,
+ "MASTER_CANDIDATE": str(self.op.master_candidate),
+ "OFFLINE": str(self.op.offline),
+ "DRAINED": str(self.op.drained),
+ }
+ nl = [self.cfg.GetMasterNode(),
+ self.op.node_name]
+ return env, nl, nl
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This only checks the instance list against the existing names.
+
+ """
+ node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
+
+ if (self.op.master_candidate is not None or
+ self.op.drained is not None or
+ self.op.offline is not None):
+ # we can't change the master's node flags
+ if self.op.node_name == self.cfg.GetMasterNode():
+ raise errors.OpPrereqError("The master role can be changed"
+ " only via masterfailover")
+
+ if ((self.op.master_candidate == False or self.op.offline == True or
+ self.op.drained == True) and node.master_candidate):
+ cp_size = self.cfg.GetClusterInfo().candidate_pool_size
+ num_candidates, _ = self.cfg.GetMasterCandidateStats()
+ if num_candidates <= cp_size:
+ msg = ("Not enough master candidates (desired"
+ " %d, new value will be %d)" % (cp_size, num_candidates-1))
+ if self.op.force:
+ self.LogWarning(msg)
+ else:
+ raise errors.OpPrereqError(msg)
+
+ if (self.op.master_candidate == True and
+ ((node.offline and not self.op.offline == False) or
+ (node.drained and not self.op.drained == False))):
+ raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
+ " to master_candidate" % node.name)
+
+ return
+
+ def Exec(self, feedback_fn):
+ """Modifies a node.
+
+ """
+ node = self.node
+
+ result = []
+ changed_mc = False
+
+ if self.op.offline is not None:
+ node.offline = self.op.offline
+ result.append(("offline", str(self.op.offline)))
+ if self.op.offline == True:
+ if node.master_candidate:
+ node.master_candidate = False
+ changed_mc = True
+ result.append(("master_candidate", "auto-demotion due to offline"))
+ if node.drained:
+ node.drained = False
+ result.append(("drained", "clear drained status due to offline"))
+
+ if self.op.master_candidate is not None:
+ node.master_candidate = self.op.master_candidate
+ changed_mc = True
+ result.append(("master_candidate", str(self.op.master_candidate)))
+ if self.op.master_candidate == False:
+ rrc = self.rpc.call_node_demote_from_mc(node.name)
+ msg = rrc.fail_msg
+ if msg:
+ self.LogWarning("Node failed to demote itself: %s" % msg)
+
+ if self.op.drained is not None:
+ node.drained = self.op.drained
+ result.append(("drained", str(self.op.drained)))
+ if self.op.drained == True:
+ if node.master_candidate:
+ node.master_candidate = False
+ changed_mc = True
+ result.append(("master_candidate", "auto-demotion due to drain"))
+ rrc = self.rpc.call_node_demote_from_mc(node.name)
+ msg = rrc.fail_msg
+ if msg:
+ self.LogWarning("Node failed to demote itself: %s" % msg)
+ if node.offline:
+ node.offline = False
+ result.append(("offline", "clear offline status due to drain"))
+
+ # this will trigger configuration file update, if needed
+ self.cfg.Update(node)
+ # this will trigger job queue propagation or cleanup
+ if changed_mc:
+ self.context.ReaddNode(node)
+
+ return result
+
+
+class LUPowercycleNode(NoHooksLU):
+ """Powercycles a node.
+
+ """
+ _OP_REQP = ["node_name", "force"]
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if node_name is None:
+ raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
+ self.op.node_name = node_name
+ if node_name == self.cfg.GetMasterNode() and not self.op.force:
+ raise errors.OpPrereqError("The node is the master and the force"
+ " parameter was not set")
+
+ def ExpandNames(self):
+ """Locking for PowercycleNode.
+
+ This is a last-resort option and shouldn't block on other
+ jobs. Therefore, we grab no locks.
+
+ """
+ self.needed_locks = {}
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This LU has no prereqs.
+
+ """
+ pass
+
+ def Exec(self, feedback_fn):
+ """Reboots a node.
+
+ """
+ result = self.rpc.call_node_powercycle(self.op.node_name,
+ self.cfg.GetHypervisorType())
+ result.Raise("Failed to schedule the reboot")
+ return result.payload
+
+
class LUQueryClusterInfo(NoHooksLU):
"""Query cluster configuration.
"""
_OP_REQP = []
- REQ_MASTER = False
REQ_BGL = False
def ExpandNames(self):
"""Return cluster config.
"""
+ cluster = self.cfg.GetClusterInfo()
result = {
- "name": self.sstore.GetClusterName(),
"software_version": constants.RELEASE_VERSION,
"protocol_version": constants.PROTOCOL_VERSION,
"config_version": constants.CONFIG_VERSION,
- "os_api_version": constants.OS_API_VERSION,
+ "os_api_version": max(constants.OS_API_VERSIONS),
"export_version": constants.EXPORT_VERSION,
- "master": self.sstore.GetMasterNode(),
"architecture": (platform.architecture()[0], platform.machine()),
- "hypervisor_type": self.sstore.GetHypervisorType(),
+ "name": cluster.cluster_name,
+ "master": cluster.master_node,
+ "default_hypervisor": cluster.enabled_hypervisors[0],
+ "enabled_hypervisors": cluster.enabled_hypervisors,
+ "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
+ for hypervisor_name in cluster.enabled_hypervisors]),
+ "beparams": cluster.beparams,
+ "nicparams": cluster.nicparams,
+ "candidate_pool_size": cluster.candidate_pool_size,
+ "master_netdev": cluster.master_netdev,
+ "volume_group_name": cluster.volume_group_name,
+ "file_storage_dir": cluster.file_storage_dir,
+ "ctime": cluster.ctime,
+ "mtime": cluster.mtime,
+ "uuid": cluster.uuid,
+ "tags": list(cluster.GetTags()),
}
return result
-class LUDumpClusterConfig(NoHooksLU):
- """Return a text-representation of the cluster-config.
+class LUQueryConfigValues(NoHooksLU):
+ """Return configuration values.
"""
_OP_REQP = []
REQ_BGL = False
+ _FIELDS_DYNAMIC = utils.FieldSet()
+ _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
+ "watcher_pause")
def ExpandNames(self):
self.needed_locks = {}
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=self._FIELDS_DYNAMIC,
+ selected=self.op.output_fields)
+
def CheckPrereq(self):
"""No prerequisites.
"""Dump a representation of the cluster config to the standard output.
"""
- return self.cfg.DumpConfig()
+ values = []
+ for field in self.op.output_fields:
+ if field == "cluster_name":
+ entry = self.cfg.GetClusterName()
+ elif field == "master_node":
+ entry = self.cfg.GetMasterNode()
+ elif field == "drain_flag":
+ entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
+ elif field == "watcher_pause":
+ return utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
+ else:
+ raise errors.ParameterError(field)
+ values.append(entry)
+ return values
class LUActivateInstanceDisks(NoHooksLU):
self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, self.instance.primary_node)
+ if not hasattr(self.op, "ignore_size"):
+ self.op.ignore_size = False
def Exec(self, feedback_fn):
"""Activate the disks.
"""
- disks_ok, disks_info = _AssembleInstanceDisks(self.instance, self.cfg)
+ disks_ok, disks_info = \
+ _AssembleInstanceDisks(self, self.instance,
+ ignore_size=self.op.ignore_size)
if not disks_ok:
raise errors.OpExecError("Cannot activate block devices")
return disks_info
-def _AssembleInstanceDisks(instance, cfg, ignore_secondaries=False):
+def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False,
+ ignore_size=False):
"""Prepare the block devices for an instance.
This sets up the block devices on all nodes.
- Args:
- instance: a ganeti.objects.Instance object
- ignore_secondaries: if true, errors on secondary nodes won't result
- in an error return from the function
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance for whose disks we assemble
+ @type ignore_secondaries: boolean
+ @param ignore_secondaries: if true, errors on secondary nodes
+ won't result in an error return from the function
+ @type ignore_size: boolean
+ @param ignore_size: if true, the current known size of the disk
+ will not be used during the disk activation, useful for cases
+ when the size is wrong
+ @return: False if the operation failed, otherwise a list of
+ (host, instance_visible_name, node_visible_name)
+ with the mapping from node devices to instance devices
- Returns:
- false if the operation failed
- list of (host, instance_visible_name, node_visible_name) if the operation
- suceeded with the mapping from node devices to instance devices
"""
device_info = []
disks_ok = True
# 1st pass, assemble on all nodes in secondary mode
for inst_disk in instance.disks:
for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
- cfg.SetDiskID(node_disk, node)
- result = rpc.call_blockdev_assemble(node, node_disk, iname, False)
- if not result:
- logger.Error("could not prepare block device %s on node %s"
- " (is_primary=False, pass=1)" % (inst_disk.iv_name, node))
+ if ignore_size:
+ node_disk = node_disk.Copy()
+ node_disk.UnsetSize()
+ lu.cfg.SetDiskID(node_disk, node)
+ result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
+ msg = result.fail_msg
+ if msg:
+ lu.proc.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=False, pass=1): %s",
+ inst_disk.iv_name, node, msg)
if not ignore_secondaries:
disks_ok = False
for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
if node != instance.primary_node:
continue
- cfg.SetDiskID(node_disk, node)
- result = rpc.call_blockdev_assemble(node, node_disk, iname, True)
- if not result:
- logger.Error("could not prepare block device %s on node %s"
- " (is_primary=True, pass=2)" % (inst_disk.iv_name, node))
+ if ignore_size:
+ node_disk = node_disk.Copy()
+ node_disk.UnsetSize()
+ lu.cfg.SetDiskID(node_disk, node)
+ result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
+ msg = result.fail_msg
+ if msg:
+ lu.proc.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=True, pass=2): %s",
+ inst_disk.iv_name, node, msg)
disks_ok = False
- device_info.append((instance.primary_node, inst_disk.iv_name, result))
+ device_info.append((instance.primary_node, inst_disk.iv_name,
+ result.payload))
# leave the disks configured for the primary node
# this is a workaround that would be fixed better by
# improving the logical/physical id handling
for disk in instance.disks:
- cfg.SetDiskID(disk, instance.primary_node)
+ lu.cfg.SetDiskID(disk, instance.primary_node)
return disks_ok, device_info
-def _StartInstanceDisks(cfg, instance, force):
+def _StartInstanceDisks(lu, instance, force):
"""Start the disks of an instance.
"""
- disks_ok, dummy = _AssembleInstanceDisks(instance, cfg,
+ disks_ok, _ = _AssembleInstanceDisks(lu, instance,
ignore_secondaries=force)
if not disks_ok:
- _ShutdownInstanceDisks(instance, cfg)
+ _ShutdownInstanceDisks(lu, instance)
if force is not None and not force:
- logger.Error("If the message above refers to a secondary node,"
- " you can retry the operation using '--force'.")
+ lu.proc.LogWarning("", hint="If the message above refers to a"
+ " secondary node,"
+ " you can retry the operation using '--force'.")
raise errors.OpExecError("Disk consistency error")
"""
instance = self.instance
- _SafeShutdownInstanceDisks(instance, self.cfg)
+ _SafeShutdownInstanceDisks(self, instance)
-def _SafeShutdownInstanceDisks(instance, cfg):
+def _SafeShutdownInstanceDisks(lu, instance):
"""Shutdown block devices of an instance.
This function checks if an instance is running, before calling
_ShutdownInstanceDisks.
"""
- ins_l = rpc.call_instance_list([instance.primary_node])
- ins_l = ins_l[instance.primary_node]
- if not type(ins_l) is list:
- raise errors.OpExecError("Can't contact node '%s'" %
- instance.primary_node)
+ pnode = instance.primary_node
+ ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
+ ins_l.Raise("Can't contact node %s" % pnode)
- if instance.name in ins_l:
+ if instance.name in ins_l.payload:
raise errors.OpExecError("Instance is running, can't shutdown"
" block devices.")
- _ShutdownInstanceDisks(instance, cfg)
+ _ShutdownInstanceDisks(lu, instance)
-def _ShutdownInstanceDisks(instance, cfg, ignore_primary=False):
+def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
"""Shutdown block devices of an instance.
This does the shutdown on all nodes of the instance.
ignored.
"""
- result = True
+ all_result = True
for disk in instance.disks:
for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
- cfg.SetDiskID(top_disk, node)
- if not rpc.call_blockdev_shutdown(node, top_disk):
- logger.Error("could not shutdown block device %s on node %s" %
- (disk.iv_name, node))
+ lu.cfg.SetDiskID(top_disk, node)
+ result = lu.rpc.call_blockdev_shutdown(node, top_disk)
+ msg = result.fail_msg
+ if msg:
+ lu.LogWarning("Could not shutdown block device %s on node %s: %s",
+ disk.iv_name, node, msg)
if not ignore_primary or node != instance.primary_node:
- result = False
- return result
+ all_result = False
+ return all_result
-def _CheckNodeFreeMemory(cfg, node, reason, requested):
+def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
"""Checks if a node has enough free memory.
This function check if a given node has the needed amount of free
information from the node, this function raise an OpPrereqError
exception.
- Args:
- - cfg: a ConfigWriter instance
- - node: the node name
- - reason: string to use in the error message
- - requested: the amount of memory in MiB
+ @type lu: C{LogicalUnit}
+ @param lu: a logical unit from which we get configuration data
+ @type node: C{str}
+ @param node: the node to check
+ @type reason: C{str}
+ @param reason: string to use in the error message
+ @type requested: C{int}
+ @param requested: the amount of memory in MiB to check for
+ @type hypervisor_name: C{str}
+ @param hypervisor_name: the hypervisor to ask for memory stats
+ @raise errors.OpPrereqError: if the node doesn't have enough memory, or
+ we cannot check the node
"""
- nodeinfo = rpc.call_node_info([node], cfg.GetVGName())
- if not nodeinfo or not isinstance(nodeinfo, dict):
- raise errors.OpPrereqError("Could not contact node %s for resource"
- " information" % (node,))
-
- free_mem = nodeinfo[node].get('memory_free')
+ nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
+ nodeinfo[node].Raise("Can't get data from node %s" % node, prereq=True)
+ free_mem = nodeinfo[node].payload.get('memory_free', None)
if not isinstance(free_mem, int):
raise errors.OpPrereqError("Can't compute free memory on node %s, result"
- " was '%s'" % (node, free_mem))
+ " was '%s'" % (node, free_mem))
if requested > free_mem:
raise errors.OpPrereqError("Not enough memory on node %s for %s:"
- " needed %s MiB, available %s MiB" %
- (node, reason, requested, free_mem))
+ " needed %s MiB, available %s MiB" %
+ (node, reason, requested, free_mem))
class LUStartupInstance(LogicalUnit):
def ExpandNames(self):
self._ExpandAndLockInstance()
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- self._LockInstancesNodes()
def BuildHooksEnv(self):
"""Build hooks env.
env = {
"FORCE": self.op.force,
}
- env.update(_BuildInstanceHookEnvByObject(self.instance))
- nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
- list(self.instance.secondary_nodes))
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
def CheckPrereq(self):
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
- # check bridges existance
- _CheckInstanceBridgesExist(instance)
-
- _CheckNodeFreeMemory(self.cfg, instance.primary_node,
- "starting instance %s" % instance.name,
- instance.memory)
+ # extra beparams
+ self.beparams = getattr(self.op, "beparams", {})
+ if self.beparams:
+ if not isinstance(self.beparams, dict):
+ raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
+ " dict" % (type(self.beparams), ))
+ # fill the beparams dict
+ utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
+ self.op.beparams = self.beparams
+
+ # extra hvparams
+ self.hvparams = getattr(self.op, "hvparams", {})
+ if self.hvparams:
+ if not isinstance(self.hvparams, dict):
+ raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
+ " dict" % (type(self.hvparams), ))
+
+ # check hypervisor parameter syntax (locally)
+ cluster = self.cfg.GetClusterInfo()
+ utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
+ filled_hvp = objects.FillDict(cluster.hvparams[instance.hypervisor],
+ instance.hvparams)
+ filled_hvp.update(self.hvparams)
+ hv_type = hypervisor.GetHypervisor(instance.hypervisor)
+ hv_type.CheckParameterSyntax(filled_hvp)
+ _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
+ self.op.hvparams = self.hvparams
+
+ _CheckNodeOnline(self, instance.primary_node)
+
+ bep = self.cfg.GetClusterInfo().FillBE(instance)
+ # check bridges existence
+ _CheckInstanceBridgesExist(self, instance)
+
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node,
+ prereq=True)
+ if not remote_info.payload: # not running already
+ _CheckNodeFreeMemory(self, instance.primary_node,
+ "starting instance %s" % instance.name,
+ bep[constants.BE_MEMORY], instance.hypervisor)
def Exec(self, feedback_fn):
"""Start the instance.
"""
instance = self.instance
force = self.op.force
- extra_args = getattr(self.op, "extra_args", "")
self.cfg.MarkInstanceUp(instance.name)
node_current = instance.primary_node
- _StartInstanceDisks(self.cfg, instance, force)
+ _StartInstanceDisks(self, instance, force)
- if not rpc.call_instance_start(node_current, instance, extra_args):
- _ShutdownInstanceDisks(instance, self.cfg)
- raise errors.OpExecError("Could not start instance")
+ result = self.rpc.call_instance_start(node_current, instance,
+ self.hvparams, self.beparams)
+ msg = result.fail_msg
+ if msg:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance: %s" % msg)
class LURebootInstance(LogicalUnit):
constants.INSTANCE_REBOOT_HARD,
constants.INSTANCE_REBOOT_FULL))
self._ExpandAndLockInstance()
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- primary_only = not constants.INSTANCE_REBOOT_FULL
- self._LockInstancesNodes(primary_only=primary_only)
def BuildHooksEnv(self):
"""Build hooks env.
"""
env = {
"IGNORE_SECONDARIES": self.op.ignore_secondaries,
+ "REBOOT_TYPE": self.op.reboot_type,
}
- env.update(_BuildInstanceHookEnvByObject(self.instance))
- nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
- list(self.instance.secondary_nodes))
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
def CheckPrereq(self):
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
- # check bridges existance
- _CheckInstanceBridgesExist(instance)
+ _CheckNodeOnline(self, instance.primary_node)
+
+ # check bridges existence
+ _CheckInstanceBridgesExist(self, instance)
def Exec(self, feedback_fn):
"""Reboot the instance.
instance = self.instance
ignore_secondaries = self.op.ignore_secondaries
reboot_type = self.op.reboot_type
- extra_args = getattr(self.op, "extra_args", "")
node_current = instance.primary_node
if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
constants.INSTANCE_REBOOT_HARD]:
- if not rpc.call_instance_reboot(node_current, instance,
- reboot_type, extra_args):
- raise errors.OpExecError("Could not reboot instance")
+ for disk in instance.disks:
+ self.cfg.SetDiskID(disk, node_current)
+ result = self.rpc.call_instance_reboot(node_current, instance,
+ reboot_type)
+ result.Raise("Could not reboot instance")
else:
- if not rpc.call_instance_shutdown(node_current, instance):
- raise errors.OpExecError("could not shutdown instance for full reboot")
- _ShutdownInstanceDisks(instance, self.cfg)
- _StartInstanceDisks(self.cfg, instance, ignore_secondaries)
- if not rpc.call_instance_start(node_current, instance, extra_args):
- _ShutdownInstanceDisks(instance, self.cfg)
- raise errors.OpExecError("Could not start instance for full reboot")
+ result = self.rpc.call_instance_shutdown(node_current, instance)
+ result.Raise("Could not shutdown instance for full reboot")
+ _ShutdownInstanceDisks(self, instance)
+ _StartInstanceDisks(self, instance, ignore_secondaries)
+ result = self.rpc.call_instance_start(node_current, instance, None, None)
+ msg = result.fail_msg
+ if msg:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance for"
+ " full reboot: %s" % msg)
self.cfg.MarkInstanceUp(instance.name)
def ExpandNames(self):
self._ExpandAndLockInstance()
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- self._LockInstancesNodes()
def BuildHooksEnv(self):
"""Build hooks env.
This runs on master, primary and secondary nodes of the instance.
"""
- env = _BuildInstanceHookEnvByObject(self.instance)
- nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
- list(self.instance.secondary_nodes))
+ env = _BuildInstanceHookEnvByObject(self, self.instance)
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
def CheckPrereq(self):
self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, self.instance.primary_node)
def Exec(self, feedback_fn):
"""Shutdown the instance.
instance = self.instance
node_current = instance.primary_node
self.cfg.MarkInstanceDown(instance.name)
- if not rpc.call_instance_shutdown(node_current, instance):
- logger.Error("could not shutdown instance")
+ result = self.rpc.call_instance_shutdown(node_current, instance)
+ msg = result.fail_msg
+ if msg:
+ self.proc.LogWarning("Could not shutdown instance: %s" % msg)
- _ShutdownInstanceDisks(instance, self.cfg)
+ _ShutdownInstanceDisks(self, instance)
class LUReinstallInstance(LogicalUnit):
def ExpandNames(self):
self._ExpandAndLockInstance()
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- self._LockInstancesNodes()
def BuildHooksEnv(self):
"""Build hooks env.
This runs on master, primary and secondary nodes of the instance.
"""
- env = _BuildInstanceHookEnvByObject(self.instance)
- nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
- list(self.instance.secondary_nodes))
+ env = _BuildInstanceHookEnvByObject(self, self.instance)
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
def CheckPrereq(self):
instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, instance.primary_node)
if instance.disk_template == constants.DT_DISKLESS:
raise errors.OpPrereqError("Instance '%s' has no disks" %
self.op.instance_name)
- if instance.status != "down":
+ if instance.admin_up:
raise errors.OpPrereqError("Instance '%s' is marked to be up" %
self.op.instance_name)
- remote_info = rpc.call_instance_info(instance.primary_node, instance.name)
- if remote_info:
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node,
+ prereq=True)
+ if remote_info.payload:
raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
(self.op.instance_name,
instance.primary_node))
if pnode is None:
raise errors.OpPrereqError("Primary node '%s' is unknown" %
self.op.pnode)
- os_obj = rpc.call_os_get(pnode.name, self.op.os_type)
- if not os_obj:
- raise errors.OpPrereqError("OS '%s' not in supported OS list for"
- " primary node" % self.op.os_type)
+ result = self.rpc.call_os_get(pnode.name, self.op.os_type)
+ result.Raise("OS '%s' not in supported OS list for primary node %s" %
+ (self.op.os_type, pnode.name), prereq=True)
self.instance = instance
if self.op.os_type is not None:
feedback_fn("Changing OS to '%s'..." % self.op.os_type)
inst.os = self.op.os_type
- self.cfg.AddInstance(inst)
+ self.cfg.Update(inst)
- _StartInstanceDisks(self.cfg, inst, None)
+ _StartInstanceDisks(self, inst, None)
try:
feedback_fn("Running the instance OS create scripts...")
- if not rpc.call_instance_os_add(inst.primary_node, inst, "sda", "sdb"):
- raise errors.OpExecError("Could not install OS for instance %s"
- " on node %s" %
- (inst.name, inst.primary_node))
+ result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
+ result.Raise("Could not install OS for instance %s on node %s" %
+ (inst.name, inst.primary_node))
finally:
- _ShutdownInstanceDisks(inst, self.cfg)
+ _ShutdownInstanceDisks(self, inst)
+
+
+class LURecreateInstanceDisks(LogicalUnit):
+ """Recreate an instance's missing disks.
+
+ """
+ HPATH = "instance-recreate-disks"
+ HTYPE = constants.HTYPE_INSTANCE
+ _OP_REQP = ["instance_name", "disks"]
+ REQ_BGL = False
+
+ def CheckArguments(self):
+ """Check the arguments.
+
+ """
+ if not isinstance(self.op.disks, list):
+ raise errors.OpPrereqError("Invalid disks parameter")
+ for item in self.op.disks:
+ if (not isinstance(item, int) or
+ item < 0):
+ raise errors.OpPrereqError("Invalid disk specification '%s'" %
+ str(item))
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on master, primary and secondary nodes of the instance.
+
+ """
+ env = _BuildInstanceHookEnvByObject(self, self.instance)
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
+ return env, nl, nl
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster and is not running.
+
+ """
+ instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, instance.primary_node)
+
+ if instance.disk_template == constants.DT_DISKLESS:
+ raise errors.OpPrereqError("Instance '%s' has no disks" %
+ self.op.instance_name)
+ if instance.admin_up:
+ raise errors.OpPrereqError("Instance '%s' is marked to be up" %
+ self.op.instance_name)
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node,
+ prereq=True)
+ if remote_info.payload:
+ raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
+ (self.op.instance_name,
+ instance.primary_node))
+
+ if not self.op.disks:
+ self.op.disks = range(len(instance.disks))
+ else:
+ for idx in self.op.disks:
+ if idx >= len(instance.disks):
+ raise errors.OpPrereqError("Invalid disk index passed '%s'" % idx)
+
+ self.instance = instance
+
+ def Exec(self, feedback_fn):
+ """Recreate the disks.
+
+ """
+ to_skip = []
+ for idx, disk in enumerate(self.instance.disks):
+ if idx not in self.op.disks: # disk idx has not been passed in
+ to_skip.append(idx)
+ continue
+
+ _CreateDisks(self, self.instance, to_skip=to_skip)
class LURenameInstance(LogicalUnit):
This runs on master, primary and secondary nodes of the instance.
"""
- env = _BuildInstanceHookEnvByObject(self.instance)
+ env = _BuildInstanceHookEnvByObject(self, self.instance)
env["INSTANCE_NEW_NAME"] = self.op.new_name
- nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
- list(self.instance.secondary_nodes))
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
def CheckPrereq(self):
if instance is None:
raise errors.OpPrereqError("Instance '%s' not known" %
self.op.instance_name)
- if instance.status != "down":
+ _CheckNodeOnline(self, instance.primary_node)
+
+ if instance.admin_up:
raise errors.OpPrereqError("Instance '%s' is marked to be up" %
self.op.instance_name)
- remote_info = rpc.call_instance_info(instance.primary_node, instance.name)
- if remote_info:
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node,
+ prereq=True)
+ if remote_info.payload:
raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
(self.op.instance_name,
instance.primary_node))
self.cfg.RenameInstance(inst.name, self.op.new_name)
# Change the instance lock. This is definitely safe while we hold the BGL
- self.context.glm.remove(locking.LEVEL_INSTANCE, inst.name)
+ self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
# re-read the instance from the configuration after rename
if inst.disk_template == constants.DT_FILE:
new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
- result = rpc.call_file_storage_dir_rename(inst.primary_node,
- old_file_storage_dir,
- new_file_storage_dir)
-
- if not result:
- raise errors.OpExecError("Could not connect to node '%s' to rename"
- " directory '%s' to '%s' (but the instance"
- " has been renamed in Ganeti)" % (
- inst.primary_node, old_file_storage_dir,
- new_file_storage_dir))
-
- if not result[0]:
- raise errors.OpExecError("Could not rename directory '%s' to '%s'"
- " (but the instance has been renamed in"
- " Ganeti)" % (old_file_storage_dir,
- new_file_storage_dir))
-
- _StartInstanceDisks(self.cfg, inst, None)
+ result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
+ old_file_storage_dir,
+ new_file_storage_dir)
+ result.Raise("Could not rename on node %s directory '%s' to '%s'"
+ " (but the instance has been renamed in Ganeti)" %
+ (inst.primary_node, old_file_storage_dir,
+ new_file_storage_dir))
+
+ _StartInstanceDisks(self, inst, None)
try:
- if not rpc.call_instance_run_rename(inst.primary_node, inst, old_name,
- "sda", "sdb"):
+ result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
+ old_name)
+ msg = result.fail_msg
+ if msg:
msg = ("Could not run OS rename script for instance %s on node %s"
- " (but the instance has been renamed in Ganeti)" %
- (inst.name, inst.primary_node))
- logger.Error(msg)
+ " (but the instance has been renamed in Ganeti): %s" %
+ (inst.name, inst.primary_node, msg))
+ self.proc.LogWarning(msg)
finally:
- _ShutdownInstanceDisks(inst, self.cfg)
+ _ShutdownInstanceDisks(self, inst)
class LURemoveInstance(LogicalUnit):
HPATH = "instance-remove"
HTYPE = constants.HTYPE_INSTANCE
_OP_REQP = ["instance_name", "ignore_failures"]
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
def BuildHooksEnv(self):
"""Build hooks env.
This runs on master, primary and secondary nodes of the instance.
"""
- env = _BuildInstanceHookEnvByObject(self.instance)
- nl = [self.sstore.GetMasterNode()]
+ env = _BuildInstanceHookEnvByObject(self, self.instance)
+ nl = [self.cfg.GetMasterNode()]
return env, nl, nl
def CheckPrereq(self):
This checks that the instance is in the cluster.
"""
- instance = self.cfg.GetInstanceInfo(
- self.cfg.ExpandInstanceName(self.op.instance_name))
- if instance is None:
- raise errors.OpPrereqError("Instance '%s' not known" %
- self.op.instance_name)
- self.instance = instance
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
def Exec(self, feedback_fn):
"""Remove the instance.
"""
instance = self.instance
- logger.Info("shutting down instance %s on node %s" %
- (instance.name, instance.primary_node))
+ logging.info("Shutting down instance %s on node %s",
+ instance.name, instance.primary_node)
- if not rpc.call_instance_shutdown(instance.primary_node, instance):
+ result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
+ msg = result.fail_msg
+ if msg:
if self.op.ignore_failures:
- feedback_fn("Warning: can't shutdown instance")
+ feedback_fn("Warning: can't shutdown instance: %s" % msg)
else:
- raise errors.OpExecError("Could not shutdown instance %s on node %s" %
- (instance.name, instance.primary_node))
+ raise errors.OpExecError("Could not shutdown instance %s on"
+ " node %s: %s" %
+ (instance.name, instance.primary_node, msg))
- logger.Info("removing block devices for instance %s" % instance.name)
+ logging.info("Removing block devices for instance %s", instance.name)
- if not _RemoveDisks(instance, self.cfg):
+ if not _RemoveDisks(self, instance):
if self.op.ignore_failures:
feedback_fn("Warning: can't remove instance's disks")
else:
raise errors.OpExecError("Can't remove instance's disks")
- logger.Info("removing instance %s out of cluster config" % instance.name)
+ logging.info("Removing instance %s out of cluster config", instance.name)
self.cfg.RemoveInstance(instance.name)
- # Remove the new instance from the Ganeti Lock Manager
- self.context.glm.remove(locking.LEVEL_INSTANCE, instance.name)
+ self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
class LUQueryInstances(NoHooksLU):
"""Logical unit for querying instances.
"""
- _OP_REQP = ["output_fields", "names"]
+ _OP_REQP = ["output_fields", "names", "use_locking"]
REQ_BGL = False
+ _SIMPLE_FIELDS = ["name", "os", "network_port", "hypervisor",
+ "serial_no", "ctime", "mtime", "uuid"]
+ _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
+ "admin_state",
+ "disk_template", "ip", "mac", "bridge",
+ "nic_mode", "nic_link",
+ "sda_size", "sdb_size", "vcpus", "tags",
+ "network_port", "beparams",
+ r"(disk)\.(size)/([0-9]+)",
+ r"(disk)\.(sizes)", "disk_usage",
+ r"(nic)\.(mac|ip|mode|link)/([0-9]+)",
+ r"(nic)\.(bridge)/([0-9]+)",
+ r"(nic)\.(macs|ips|modes|links|bridges)",
+ r"(disk|nic)\.(count)",
+ "hvparams",
+ ] + _SIMPLE_FIELDS +
+ ["hv/%s" % name
+ for name in constants.HVS_PARAMETERS] +
+ ["be/%s" % name
+ for name in constants.BES_PARAMETERS])
+ _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
+
def ExpandNames(self):
- self.dynamic_fields = frozenset(["oper_state", "oper_ram", "status"])
- _CheckOutputFields(static=["name", "os", "pnode", "snodes",
- "admin_state", "admin_ram",
- "disk_template", "ip", "mac", "bridge",
- "sda_size", "sdb_size", "vcpus", "tags",
- "auto_balance",
- "network_port", "kernel_path", "initrd_path",
- "hvm_boot_order", "hvm_acpi", "hvm_pae",
- "hvm_cdrom_image_path", "hvm_nic_type",
- "hvm_disk_type", "vnc_bind_address"],
- dynamic=self.dynamic_fields,
+ _CheckOutputFields(static=self._FIELDS_STATIC,
+ dynamic=self._FIELDS_DYNAMIC,
selected=self.op.output_fields)
self.needed_locks = {}
self.share_locks[locking.LEVEL_INSTANCE] = 1
self.share_locks[locking.LEVEL_NODE] = 1
- # TODO: we could lock instances (and nodes) only if the user asked for
- # dynamic fields. For that we need atomic ways to get info for a group of
- # instances from the config, though.
- if not self.op.names:
- self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+ if self.op.names:
+ self.wanted = _GetWantedInstances(self, self.op.names)
else:
- self.needed_locks[locking.LEVEL_INSTANCE] = \
- _GetWantedInstances(self, self.op.names)
+ self.wanted = locking.ALL_SET
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
+ self.do_locking = self.do_node_query and self.op.use_locking
+ if self.do_locking:
+ self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
def DeclareLocks(self, level):
- # TODO: locking of nodes could be avoided when not querying them
- if level == locking.LEVEL_NODE:
+ if level == locking.LEVEL_NODE and self.do_locking:
self._LockInstancesNodes()
def CheckPrereq(self):
"""Check prerequisites.
"""
- # This of course is valid only if we locked the instances
- self.wanted = self.acquired_locks[locking.LEVEL_INSTANCE]
+ pass
def Exec(self, feedback_fn):
"""Computes the list of nodes and their attributes.
"""
- instance_names = self.wanted
- instance_list = [self.cfg.GetInstanceInfo(iname) for iname
- in instance_names]
+ all_info = self.cfg.GetAllInstancesInfo()
+ if self.wanted == locking.ALL_SET:
+ # caller didn't specify instance names, so ordering is not important
+ if self.do_locking:
+ instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+ else:
+ instance_names = all_info.keys()
+ instance_names = utils.NiceSort(instance_names)
+ else:
+ # caller did specify names, so we must keep the ordering
+ if self.do_locking:
+ tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
+ else:
+ tgt_set = all_info.keys()
+ missing = set(self.wanted).difference(tgt_set)
+ if missing:
+ raise errors.OpExecError("Some instances were removed before"
+ " retrieving their data: %s" % missing)
+ instance_names = self.wanted
+
+ instance_list = [all_info[iname] for iname in instance_names]
# begin data gathering
nodes = frozenset([inst.primary_node for inst in instance_list])
+ hv_list = list(set([inst.hypervisor for inst in instance_list]))
bad_nodes = []
- if self.dynamic_fields.intersection(self.op.output_fields):
+ off_nodes = []
+ if self.do_node_query:
live_data = {}
- node_data = rpc.call_all_instances_info(nodes)
+ node_data = self.rpc.call_all_instances_info(nodes, hv_list)
for name in nodes:
result = node_data[name]
- if result:
- live_data.update(result)
- elif result == False:
+ if result.offline:
+ # offline nodes will be in both lists
+ off_nodes.append(name)
+ if result.fail_msg:
bad_nodes.append(name)
- # else no instance is alive
+ else:
+ if result.payload:
+ live_data.update(result.payload)
+ # else no instance is alive
else:
live_data = dict([(name, {}) for name in instance_names])
# end data gathering
+ HVPREFIX = "hv/"
+ BEPREFIX = "be/"
output = []
+ cluster = self.cfg.GetClusterInfo()
for instance in instance_list:
iout = []
+ i_hv = cluster.FillHV(instance)
+ i_be = cluster.FillBE(instance)
+ i_nicp = [objects.FillDict(cluster.nicparams[constants.PP_DEFAULT],
+ nic.nicparams) for nic in instance.nics]
for field in self.op.output_fields:
- if field == "name":
- val = instance.name
- elif field == "os":
- val = instance.os
+ st_match = self._FIELDS_STATIC.Matches(field)
+ if field in self._SIMPLE_FIELDS:
+ val = getattr(instance, field)
elif field == "pnode":
val = instance.primary_node
elif field == "snodes":
val = list(instance.secondary_nodes)
elif field == "admin_state":
- val = (instance.status != "down")
+ val = instance.admin_up
elif field == "oper_state":
if instance.primary_node in bad_nodes:
val = None
else:
val = bool(live_data.get(instance.name))
elif field == "status":
- if instance.primary_node in bad_nodes:
+ if instance.primary_node in off_nodes:
+ val = "ERROR_nodeoffline"
+ elif instance.primary_node in bad_nodes:
val = "ERROR_nodedown"
else:
running = bool(live_data.get(instance.name))
if running:
- if instance.status != "down":
+ if instance.admin_up:
val = "running"
else:
val = "ERROR_up"
else:
- if instance.status != "down":
+ if instance.admin_up:
val = "ERROR_down"
else:
val = "ADMIN_down"
- elif field == "admin_ram":
- val = instance.memory
elif field == "oper_ram":
if instance.primary_node in bad_nodes:
val = None
val = live_data[instance.name].get("memory", "?")
else:
val = "-"
+ elif field == "vcpus":
+ val = i_be[constants.BE_VCPUS]
elif field == "disk_template":
val = instance.disk_template
elif field == "ip":
- val = instance.nics[0].ip
+ if instance.nics:
+ val = instance.nics[0].ip
+ else:
+ val = None
+ elif field == "nic_mode":
+ if instance.nics:
+ val = i_nicp[0][constants.NIC_MODE]
+ else:
+ val = None
+ elif field == "nic_link":
+ if instance.nics:
+ val = i_nicp[0][constants.NIC_LINK]
+ else:
+ val = None
elif field == "bridge":
- val = instance.nics[0].bridge
+ if (instance.nics and
+ i_nicp[0][constants.NIC_MODE] == constants.NIC_MODE_BRIDGED):
+ val = i_nicp[0][constants.NIC_LINK]
+ else:
+ val = None
elif field == "mac":
- val = instance.nics[0].mac
+ if instance.nics:
+ val = instance.nics[0].mac
+ else:
+ val = None
elif field == "sda_size" or field == "sdb_size":
- disk = instance.FindDisk(field[:3])
- if disk is None:
+ idx = ord(field[2]) - ord('a')
+ try:
+ val = instance.FindDisk(idx).size
+ except errors.OpPrereqError:
val = None
- else:
- val = disk.size
- elif field == "vcpus":
- val = instance.vcpus
+ elif field == "disk_usage": # total disk usage per node
+ disk_sizes = [{'size': disk.size} for disk in instance.disks]
+ val = _ComputeDiskSize(instance.disk_template, disk_sizes)
elif field == "tags":
val = list(instance.GetTags())
- elif field in ("network_port", "kernel_path", "initrd_path",
- "hvm_boot_order", "hvm_acpi", "hvm_pae",
- "hvm_cdrom_image_path", "hvm_nic_type",
- "hvm_disk_type", "vnc_bind_address"):
- val = getattr(instance, field, None)
- if val is not None:
- pass
- elif field in ("hvm_nic_type", "hvm_disk_type",
- "kernel_path", "initrd_path"):
- val = "default"
+ elif field == "hvparams":
+ val = i_hv
+ elif (field.startswith(HVPREFIX) and
+ field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
+ val = i_hv.get(field[len(HVPREFIX):], None)
+ elif field == "beparams":
+ val = i_be
+ elif (field.startswith(BEPREFIX) and
+ field[len(BEPREFIX):] in constants.BES_PARAMETERS):
+ val = i_be.get(field[len(BEPREFIX):], None)
+ elif st_match and st_match.groups():
+ # matches a variable list
+ st_groups = st_match.groups()
+ if st_groups and st_groups[0] == "disk":
+ if st_groups[1] == "count":
+ val = len(instance.disks)
+ elif st_groups[1] == "sizes":
+ val = [disk.size for disk in instance.disks]
+ elif st_groups[1] == "size":
+ try:
+ val = instance.FindDisk(st_groups[2]).size
+ except errors.OpPrereqError:
+ val = None
+ else:
+ assert False, "Unhandled disk parameter"
+ elif st_groups[0] == "nic":
+ if st_groups[1] == "count":
+ val = len(instance.nics)
+ elif st_groups[1] == "macs":
+ val = [nic.mac for nic in instance.nics]
+ elif st_groups[1] == "ips":
+ val = [nic.ip for nic in instance.nics]
+ elif st_groups[1] == "modes":
+ val = [nicp[constants.NIC_MODE] for nicp in i_nicp]
+ elif st_groups[1] == "links":
+ val = [nicp[constants.NIC_LINK] for nicp in i_nicp]
+ elif st_groups[1] == "bridges":
+ val = []
+ for nicp in i_nicp:
+ if nicp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+ val.append(nicp[constants.NIC_LINK])
+ else:
+ val.append(None)
+ else:
+ # index-based item
+ nic_idx = int(st_groups[2])
+ if nic_idx >= len(instance.nics):
+ val = None
+ else:
+ if st_groups[1] == "mac":
+ val = instance.nics[nic_idx].mac
+ elif st_groups[1] == "ip":
+ val = instance.nics[nic_idx].ip
+ elif st_groups[1] == "mode":
+ val = i_nicp[nic_idx][constants.NIC_MODE]
+ elif st_groups[1] == "link":
+ val = i_nicp[nic_idx][constants.NIC_LINK]
+ elif st_groups[1] == "bridge":
+ nic_mode = i_nicp[nic_idx][constants.NIC_MODE]
+ if nic_mode == constants.NIC_MODE_BRIDGED:
+ val = i_nicp[nic_idx][constants.NIC_LINK]
+ else:
+ val = None
+ else:
+ assert False, "Unhandled NIC parameter"
else:
- val = "-"
+ assert False, ("Declared but unhandled variable parameter '%s'" %
+ field)
else:
- raise errors.ParameterError(field)
+ assert False, "Declared but unhandled parameter '%s'" % field
iout.append(val)
output.append(iout)
env = {
"IGNORE_CONSISTENCY": self.op.ignore_consistency,
}
- env.update(_BuildInstanceHookEnvByObject(self.instance))
- nl = [self.sstore.GetMasterNode()] + list(self.instance.secondary_nodes)
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
return env, nl, nl
def CheckPrereq(self):
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ bep = self.cfg.GetClusterInfo().FillBE(instance)
if instance.disk_template not in constants.DTS_NET_MIRROR:
raise errors.OpPrereqError("Instance's disk layout is not"
" network mirrored, cannot failover.")
"a mirrored disk template")
target_node = secondary_nodes[0]
- # check memory requirements on the secondary node
- _CheckNodeFreeMemory(self.cfg, target_node, "failing over instance %s" %
- instance.name, instance.memory)
+ _CheckNodeOnline(self, target_node)
+ _CheckNodeNotDrained(self, target_node)
+ if instance.admin_up:
+ # check memory requirements on the secondary node
+ _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
+ instance.name, bep[constants.BE_MEMORY],
+ instance.hypervisor)
+ else:
+ self.LogInfo("Not checking memory on the secondary node as"
+ " instance will not be started")
# check bridge existance
- brlist = [nic.bridge for nic in instance.nics]
- if not rpc.call_bridges_exist(target_node, brlist):
- raise errors.OpPrereqError("One or more target bridges %s does not"
- " exist on destination node '%s'" %
- (brlist, target_node))
+ _CheckInstanceBridgesExist(self, instance, node=target_node)
def Exec(self, feedback_fn):
"""Failover an instance.
feedback_fn("* checking disk consistency between source and target")
for dev in instance.disks:
# for drbd, these are drbd over lvm
- if not _CheckDiskConsistency(self.cfg, dev, target_node, False):
- if instance.status == "up" and not self.op.ignore_consistency:
+ if not _CheckDiskConsistency(self, dev, target_node, False):
+ if instance.admin_up and not self.op.ignore_consistency:
raise errors.OpExecError("Disk %s is degraded on target node,"
" aborting failover." % dev.iv_name)
feedback_fn("* shutting down instance on source node")
- logger.Info("Shutting down instance %s on node %s" %
- (instance.name, source_node))
+ logging.info("Shutting down instance %s on node %s",
+ instance.name, source_node)
- if not rpc.call_instance_shutdown(source_node, instance):
+ result = self.rpc.call_instance_shutdown(source_node, instance)
+ msg = result.fail_msg
+ if msg:
if self.op.ignore_consistency:
- logger.Error("Could not shutdown instance %s on node %s. Proceeding"
- " anyway. Please make sure node %s is down" %
- (instance.name, source_node, source_node))
+ self.proc.LogWarning("Could not shutdown instance %s on node %s."
+ " Proceeding anyway. Please make sure node"
+ " %s is down. Error details: %s",
+ instance.name, source_node, source_node, msg)
else:
- raise errors.OpExecError("Could not shutdown instance %s on node %s" %
- (instance.name, source_node))
+ raise errors.OpExecError("Could not shutdown instance %s on"
+ " node %s: %s" %
+ (instance.name, source_node, msg))
feedback_fn("* deactivating the instance's disks on source node")
- if not _ShutdownInstanceDisks(instance, self.cfg, ignore_primary=True):
+ if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
raise errors.OpExecError("Can't shut down the instance's disks.")
instance.primary_node = target_node
self.cfg.Update(instance)
# Only start the instance if it's marked as up
- if instance.status == "up":
+ if instance.admin_up:
feedback_fn("* activating the instance's disks on target node")
- logger.Info("Starting instance %s on node %s" %
- (instance.name, target_node))
+ logging.info("Starting instance %s on node %s",
+ instance.name, target_node)
- disks_ok, dummy = _AssembleInstanceDisks(instance, self.cfg,
+ disks_ok, _ = _AssembleInstanceDisks(self, instance,
ignore_secondaries=True)
if not disks_ok:
- _ShutdownInstanceDisks(instance, self.cfg)
+ _ShutdownInstanceDisks(self, instance)
raise errors.OpExecError("Can't activate the instance's disks")
feedback_fn("* starting the instance on the target node")
- if not rpc.call_instance_start(target_node, instance, None):
- _ShutdownInstanceDisks(instance, self.cfg)
- raise errors.OpExecError("Could not start instance %s on node %s." %
- (instance.name, target_node))
+ result = self.rpc.call_instance_start(target_node, instance, None, None)
+ msg = result.fail_msg
+ if msg:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance %s on node %s: %s" %
+ (instance.name, target_node, msg))
-def _CreateBlockDevOnPrimary(cfg, node, instance, device, info):
- """Create a tree of block devices on the primary node.
+class LUMigrateInstance(LogicalUnit):
+ """Migrate an instance.
- This always creates all devices.
+ This is migration without shutting down, compared to the failover,
+ which is done with shutdown.
"""
- if device.children:
- for child in device.children:
- if not _CreateBlockDevOnPrimary(cfg, node, instance, child, info):
- return False
-
- cfg.SetDiskID(device, node)
- new_id = rpc.call_blockdev_create(node, device, device.size,
- instance.name, True, info)
- if not new_id:
- return False
- if device.physical_id is None:
- device.physical_id = new_id
- return True
+ HPATH = "instance-migrate"
+ HTYPE = constants.HTYPE_INSTANCE
+ _OP_REQP = ["instance_name", "live", "cleanup"]
+
+ REQ_BGL = False
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
-def _CreateBlockDevOnSecondary(cfg, node, instance, device, force, info):
- """Create a tree of block devices on a secondary node.
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
- If this device type has to be created on secondaries, create it and
- all its children.
+ self._migrater = TLMigrateInstance(self, self.op.instance_name,
+ self.op.live, self.op.cleanup)
+ self.tasklets = [self._migrater]
- If not, just recurse to children keeping the same 'force' value.
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
- """
- if device.CreateOnSecondary():
- force = True
- if device.children:
- for child in device.children:
- if not _CreateBlockDevOnSecondary(cfg, node, instance,
- child, force, info):
- return False
+ def BuildHooksEnv(self):
+ """Build hooks env.
- if not force:
- return True
- cfg.SetDiskID(device, node)
- new_id = rpc.call_blockdev_create(node, device, device.size,
- instance.name, False, info)
- if not new_id:
- return False
- if device.physical_id is None:
- device.physical_id = new_id
- return True
+ This runs on master, primary and secondary nodes of the instance.
+ """
+ instance = self._migrater.instance
+ env = _BuildInstanceHookEnvByObject(self, instance)
+ env["MIGRATE_LIVE"] = self.op.live
+ env["MIGRATE_CLEANUP"] = self.op.cleanup
+ nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
+ return env, nl, nl
-def _GenerateUniqueNames(cfg, exts):
- """Generate a suitable LV name.
- This will generate a logical volume name for the given instance.
+class LUMoveInstance(LogicalUnit):
+ """Move an instance by data-copying.
"""
- results = []
- for val in exts:
- new_id = cfg.GenerateUniqueID()
- results.append("%s%s" % (new_id, val))
- return results
+ HPATH = "instance-move"
+ HTYPE = constants.HTYPE_INSTANCE
+ _OP_REQP = ["instance_name", "target_node"]
+ REQ_BGL = False
+ def ExpandNames(self):
+ self._ExpandAndLockInstance()
+ target_node = self.cfg.ExpandNodeName(self.op.target_node)
+ if target_node is None:
+ raise errors.OpPrereqError("Node '%s' not known" %
+ self.op.target_node)
+ self.op.target_node = target_node
+ self.needed_locks[locking.LEVEL_NODE] = [target_node]
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
-def _GenerateDRBD8Branch(cfg, primary, secondary, size, names, iv_name):
- """Generate a drbd8 device complete with its children.
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes(primary_only=True)
- """
- port = cfg.AllocatePort()
- vgname = cfg.GetVGName()
- dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
- logical_id=(vgname, names[0]))
- dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
- logical_id=(vgname, names[1]))
- drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
- logical_id = (primary, secondary, port),
- children = [dev_data, dev_meta],
- iv_name=iv_name)
- return drbd_dev
+ def BuildHooksEnv(self):
+ """Build hooks env.
+ This runs on master, primary and secondary nodes of the instance.
-def _GenerateDiskTemplate(cfg, template_name,
- instance_name, primary_node,
- secondary_nodes, disk_sz, swap_sz,
- file_storage_dir, file_driver):
- """Generate the entire disk layout for a given template type.
+ """
+ env = {
+ "TARGET_NODE": self.op.target_node,
+ }
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+ nl = [self.cfg.GetMasterNode()] + [self.instance.primary_node,
+ self.op.target_node]
+ return env, nl, nl
- """
- #TODO: compute space requirements
+ def CheckPrereq(self):
+ """Check prerequisites.
- vgname = cfg.GetVGName()
- if template_name == constants.DT_DISKLESS:
- disks = []
- elif template_name == constants.DT_PLAIN:
- if len(secondary_nodes) != 0:
- raise errors.ProgrammerError("Wrong template configuration")
+ This checks that the instance is in the cluster.
- names = _GenerateUniqueNames(cfg, [".sda", ".sdb"])
- sda_dev = objects.Disk(dev_type=constants.LD_LV, size=disk_sz,
- logical_id=(vgname, names[0]),
- iv_name = "sda")
- sdb_dev = objects.Disk(dev_type=constants.LD_LV, size=swap_sz,
- logical_id=(vgname, names[1]),
- iv_name = "sdb")
- disks = [sda_dev, sdb_dev]
- elif template_name == constants.DT_DRBD8:
- if len(secondary_nodes) != 1:
- raise errors.ProgrammerError("Wrong template configuration")
- remote_node = secondary_nodes[0]
- names = _GenerateUniqueNames(cfg, [".sda_data", ".sda_meta",
- ".sdb_data", ".sdb_meta"])
- drbd_sda_dev = _GenerateDRBD8Branch(cfg, primary_node, remote_node,
- disk_sz, names[0:2], "sda")
- drbd_sdb_dev = _GenerateDRBD8Branch(cfg, primary_node, remote_node,
- swap_sz, names[2:4], "sdb")
- disks = [drbd_sda_dev, drbd_sdb_dev]
- elif template_name == constants.DT_FILE:
- if len(secondary_nodes) != 0:
- raise errors.ProgrammerError("Wrong template configuration")
+ """
+ self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.op.instance_name
- file_sda_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk_sz,
- iv_name="sda", logical_id=(file_driver,
- "%s/sda" % file_storage_dir))
- file_sdb_dev = objects.Disk(dev_type=constants.LD_FILE, size=swap_sz,
- iv_name="sdb", logical_id=(file_driver,
- "%s/sdb" % file_storage_dir))
- disks = [file_sda_dev, file_sdb_dev]
- else:
- raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
- return disks
+ node = self.cfg.GetNodeInfo(self.op.target_node)
+ assert node is not None, \
+ "Cannot retrieve locked node %s" % self.op.target_node
+ self.target_node = target_node = node.name
-def _GetInstanceInfoText(instance):
- """Compute that text that should be added to the disk's metadata.
+ if target_node == instance.primary_node:
+ raise errors.OpPrereqError("Instance %s is already on the node %s" %
+ (instance.name, target_node))
- """
- return "originstname+%s" % instance.name
+ bep = self.cfg.GetClusterInfo().FillBE(instance)
+ for idx, dsk in enumerate(instance.disks):
+ if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
+ raise errors.OpPrereqError("Instance disk %d has a complex layout,"
+ " cannot copy")
-def _CreateDisks(cfg, instance):
- """Create all disks for an instance.
+ _CheckNodeOnline(self, target_node)
+ _CheckNodeNotDrained(self, target_node)
- This abstracts away some work from AddInstance.
+ if instance.admin_up:
+ # check memory requirements on the secondary node
+ _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
+ instance.name, bep[constants.BE_MEMORY],
+ instance.hypervisor)
+ else:
+ self.LogInfo("Not checking memory on the secondary node as"
+ " instance will not be started")
- Args:
- instance: the instance object
+ # check bridge existance
+ _CheckInstanceBridgesExist(self, instance, node=target_node)
- Returns:
- True or False showing the success of the creation process
+ def Exec(self, feedback_fn):
+ """Move an instance.
- """
- info = _GetInstanceInfoText(instance)
+ The move is done by shutting it down on its present node, copying
+ the data over (slow) and starting it on the new node.
- if instance.disk_template == constants.DT_FILE:
- file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
- result = rpc.call_file_storage_dir_create(instance.primary_node,
- file_storage_dir)
+ """
+ instance = self.instance
- if not result:
- logger.Error("Could not connect to node '%s'" % instance.primary_node)
- return False
+ source_node = instance.primary_node
+ target_node = self.target_node
- if not result[0]:
- logger.Error("failed to create directory '%s'" % file_storage_dir)
- return False
+ self.LogInfo("Shutting down instance %s on source node %s",
+ instance.name, source_node)
- for device in instance.disks:
- logger.Info("creating volume %s for instance %s" %
- (device.iv_name, instance.name))
- #HARDCODE
- for secondary_node in instance.secondary_nodes:
- if not _CreateBlockDevOnSecondary(cfg, secondary_node, instance,
- device, False, info):
- logger.Error("failed to create volume %s (%s) on secondary node %s!" %
- (device.iv_name, device, secondary_node))
- return False
- #HARDCODE
- if not _CreateBlockDevOnPrimary(cfg, instance.primary_node,
- instance, device, info):
- logger.Error("failed to create volume %s on primary!" %
- device.iv_name)
- return False
+ result = self.rpc.call_instance_shutdown(source_node, instance)
+ msg = result.fail_msg
+ if msg:
+ if self.op.ignore_consistency:
+ self.proc.LogWarning("Could not shutdown instance %s on node %s."
+ " Proceeding anyway. Please make sure node"
+ " %s is down. Error details: %s",
+ instance.name, source_node, source_node, msg)
+ else:
+ raise errors.OpExecError("Could not shutdown instance %s on"
+ " node %s: %s" %
+ (instance.name, source_node, msg))
+
+ # create the target disks
+ try:
+ _CreateDisks(self, instance, target_node=target_node)
+ except errors.OpExecError:
+ self.LogWarning("Device creation failed, reverting...")
+ try:
+ _RemoveDisks(self, instance, target_node=target_node)
+ finally:
+ self.cfg.ReleaseDRBDMinors(instance.name)
+ raise
+
+ cluster_name = self.cfg.GetClusterInfo().cluster_name
+
+ errs = []
+ # activate, get path, copy the data over
+ for idx, disk in enumerate(instance.disks):
+ self.LogInfo("Copying data for disk %d", idx)
+ result = self.rpc.call_blockdev_assemble(target_node, disk,
+ instance.name, True)
+ if result.fail_msg:
+ self.LogWarning("Can't assemble newly created disk %d: %s",
+ idx, result.fail_msg)
+ errs.append(result.fail_msg)
+ break
+ dev_path = result.payload
+ result = self.rpc.call_blockdev_export(source_node, disk,
+ target_node, dev_path,
+ cluster_name)
+ if result.fail_msg:
+ self.LogWarning("Can't copy data over for disk %d: %s",
+ idx, result.fail_msg)
+ errs.append(result.fail_msg)
+ break
+
+ if errs:
+ self.LogWarning("Some disks failed to copy, aborting")
+ try:
+ _RemoveDisks(self, instance, target_node=target_node)
+ finally:
+ self.cfg.ReleaseDRBDMinors(instance.name)
+ raise errors.OpExecError("Errors during disk copy: %s" %
+ (",".join(errs),))
- return True
+ instance.primary_node = target_node
+ self.cfg.Update(instance)
+ self.LogInfo("Removing the disks on the original node")
+ _RemoveDisks(self, instance, target_node=source_node)
-def _RemoveDisks(instance, cfg):
- """Remove all disks for an instance.
+ # Only start the instance if it's marked as up
+ if instance.admin_up:
+ self.LogInfo("Starting instance %s on node %s",
+ instance.name, target_node)
- This abstracts away some work from `AddInstance()` and
- `RemoveInstance()`. Note that in case some of the devices couldn't
- be removed, the removal will continue with the other ones (compare
- with `_CreateDisks()`).
+ disks_ok, _ = _AssembleInstanceDisks(self, instance,
+ ignore_secondaries=True)
+ if not disks_ok:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Can't activate the instance's disks")
+
+ result = self.rpc.call_instance_start(target_node, instance, None, None)
+ msg = result.fail_msg
+ if msg:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance %s on node %s: %s" %
+ (instance.name, target_node, msg))
- Args:
- instance: the instance object
- Returns:
- True or False showing the success of the removal proces
+class LUMigrateNode(LogicalUnit):
+ """Migrate all instances from a node.
"""
- logger.Info("removing block devices for instance %s" % instance.name)
+ HPATH = "node-migrate"
+ HTYPE = constants.HTYPE_NODE
+ _OP_REQP = ["node_name", "live"]
+ REQ_BGL = False
- result = True
- for device in instance.disks:
- for node, disk in device.ComputeNodeTree(instance.primary_node):
- cfg.SetDiskID(disk, node)
- if not rpc.call_blockdev_remove(node, disk):
- logger.Error("could not remove block device %s on node %s,"
- " continuing anyway" %
- (device.iv_name, node))
- result = False
+ def ExpandNames(self):
+ self.op.node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if self.op.node_name is None:
+ raise errors.OpPrereqError("Node '%s' not known" % self.op.node_name)
- if instance.disk_template == constants.DT_FILE:
- file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
- if not rpc.call_file_storage_dir_remove(instance.primary_node,
- file_storage_dir):
- logger.Error("could not remove directory '%s'" % file_storage_dir)
- result = False
+ self.needed_locks = {
+ locking.LEVEL_NODE: [self.op.node_name],
+ }
- return result
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ # Create tasklets for migrating instances for all instances on this node
+ names = []
+ tasklets = []
-def _ComputeDiskSize(disk_template, disk_size, swap_size):
- """Compute disk size requirements in the volume group
+ for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
+ logging.debug("Migrating instance %s", inst.name)
+ names.append(inst.name)
- This is currently hard-coded for the two-drive layout.
+ tasklets.append(TLMigrateInstance(self, inst.name, self.op.live, False))
- """
- # Required free disk space as a function of disk and swap space
- req_size_dict = {
- constants.DT_DISKLESS: None,
- constants.DT_PLAIN: disk_size + swap_size,
- # 256 MB are added for drbd metadata, 128MB for each drbd device
- constants.DT_DRBD8: disk_size + swap_size + 256,
+ self.tasklets = tasklets
+
+ # Declare instance locks
+ self.needed_locks[locking.LEVEL_INSTANCE] = names
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, the primary and all the secondaries.
+
+ """
+ env = {
+ "NODE_NAME": self.op.node_name,
+ }
+
+ nl = [self.cfg.GetMasterNode()]
+
+ return (env, nl, nl)
+
+
+class TLMigrateInstance(Tasklet):
+ def __init__(self, lu, instance_name, live, cleanup):
+ """Initializes this class.
+
+ """
+ Tasklet.__init__(self, lu)
+
+ # Parameters
+ self.instance_name = instance_name
+ self.live = live
+ self.cleanup = cleanup
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ instance = self.cfg.GetInstanceInfo(
+ self.cfg.ExpandInstanceName(self.instance_name))
+ if instance is None:
+ raise errors.OpPrereqError("Instance '%s' not known" %
+ self.instance_name)
+
+ if instance.disk_template != constants.DT_DRBD8:
+ raise errors.OpPrereqError("Instance's disk layout is not"
+ " drbd8, cannot migrate.")
+
+ secondary_nodes = instance.secondary_nodes
+ if not secondary_nodes:
+ raise errors.ConfigurationError("No secondary node but using"
+ " drbd8 disk template")
+
+ i_be = self.cfg.GetClusterInfo().FillBE(instance)
+
+ target_node = secondary_nodes[0]
+ # check memory requirements on the secondary node
+ _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
+ instance.name, i_be[constants.BE_MEMORY],
+ instance.hypervisor)
+
+ # check bridge existance
+ _CheckInstanceBridgesExist(self, instance, node=target_node)
+
+ if not self.cleanup:
+ _CheckNodeNotDrained(self, target_node)
+ result = self.rpc.call_instance_migratable(instance.primary_node,
+ instance)
+ result.Raise("Can't migrate, please use failover", prereq=True)
+
+ self.instance = instance
+
+ def _WaitUntilSync(self):
+ """Poll with custom rpc for disk sync.
+
+ This uses our own step-based rpc call.
+
+ """
+ self.feedback_fn("* wait until resync is done")
+ all_done = False
+ while not all_done:
+ all_done = True
+ result = self.rpc.call_drbd_wait_sync(self.all_nodes,
+ self.nodes_ip,
+ self.instance.disks)
+ min_percent = 100
+ for node, nres in result.items():
+ nres.Raise("Cannot resync disks on node %s" % node)
+ node_done, node_percent = nres.payload
+ all_done = all_done and node_done
+ if node_percent is not None:
+ min_percent = min(min_percent, node_percent)
+ if not all_done:
+ if min_percent < 100:
+ self.feedback_fn(" - progress: %.1f%%" % min_percent)
+ time.sleep(2)
+
+ def _EnsureSecondary(self, node):
+ """Demote a node to secondary.
+
+ """
+ self.feedback_fn("* switching node %s to secondary mode" % node)
+
+ for dev in self.instance.disks:
+ self.cfg.SetDiskID(dev, node)
+
+ result = self.rpc.call_blockdev_close(node, self.instance.name,
+ self.instance.disks)
+ result.Raise("Cannot change disk to secondary on node %s" % node)
+
+ def _GoStandalone(self):
+ """Disconnect from the network.
+
+ """
+ self.feedback_fn("* changing into standalone mode")
+ result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
+ self.instance.disks)
+ for node, nres in result.items():
+ nres.Raise("Cannot disconnect disks node %s" % node)
+
+ def _GoReconnect(self, multimaster):
+ """Reconnect to the network.
+
+ """
+ if multimaster:
+ msg = "dual-master"
+ else:
+ msg = "single-master"
+ self.feedback_fn("* changing disks into %s mode" % msg)
+ result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
+ self.instance.disks,
+ self.instance.name, multimaster)
+ for node, nres in result.items():
+ nres.Raise("Cannot change disks config on node %s" % node)
+
+ def _ExecCleanup(self):
+ """Try to cleanup after a failed migration.
+
+ The cleanup is done by:
+ - check that the instance is running only on one node
+ (and update the config if needed)
+ - change disks on its secondary node to secondary
+ - wait until disks are fully synchronized
+ - disconnect from the network
+ - change disks into single-master mode
+ - wait again until disks are fully synchronized
+
+ """
+ instance = self.instance
+ target_node = self.target_node
+ source_node = self.source_node
+
+ # check running on only one node
+ self.feedback_fn("* checking where the instance actually runs"
+ " (if this hangs, the hypervisor might be in"
+ " a bad state)")
+ ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
+ for node, result in ins_l.items():
+ result.Raise("Can't contact node %s" % node)
+
+ runningon_source = instance.name in ins_l[source_node].payload
+ runningon_target = instance.name in ins_l[target_node].payload
+
+ if runningon_source and runningon_target:
+ raise errors.OpExecError("Instance seems to be running on two nodes,"
+ " or the hypervisor is confused. You will have"
+ " to ensure manually that it runs only on one"
+ " and restart this operation.")
+
+ if not (runningon_source or runningon_target):
+ raise errors.OpExecError("Instance does not seem to be running at all."
+ " In this case, it's safer to repair by"
+ " running 'gnt-instance stop' to ensure disk"
+ " shutdown, and then restarting it.")
+
+ if runningon_target:
+ # the migration has actually succeeded, we need to update the config
+ self.feedback_fn("* instance running on secondary node (%s),"
+ " updating config" % target_node)
+ instance.primary_node = target_node
+ self.cfg.Update(instance)
+ demoted_node = source_node
+ else:
+ self.feedback_fn("* instance confirmed to be running on its"
+ " primary node (%s)" % source_node)
+ demoted_node = target_node
+
+ self._EnsureSecondary(demoted_node)
+ try:
+ self._WaitUntilSync()
+ except errors.OpExecError:
+ # we ignore here errors, since if the device is standalone, it
+ # won't be able to sync
+ pass
+ self._GoStandalone()
+ self._GoReconnect(False)
+ self._WaitUntilSync()
+
+ self.feedback_fn("* done")
+
+ def _RevertDiskStatus(self):
+ """Try to revert the disk status after a failed migration.
+
+ """
+ target_node = self.target_node
+ try:
+ self._EnsureSecondary(target_node)
+ self._GoStandalone()
+ self._GoReconnect(False)
+ self._WaitUntilSync()
+ except errors.OpExecError, err:
+ self.lu.LogWarning("Migration failed and I can't reconnect the"
+ " drives: error '%s'\n"
+ "Please look and recover the instance status" %
+ str(err))
+
+ def _AbortMigration(self):
+ """Call the hypervisor code to abort a started migration.
+
+ """
+ instance = self.instance
+ target_node = self.target_node
+ migration_info = self.migration_info
+
+ abort_result = self.rpc.call_finalize_migration(target_node,
+ instance,
+ migration_info,
+ False)
+ abort_msg = abort_result.fail_msg
+ if abort_msg:
+ logging.error("Aborting migration failed on target node %s: %s" %
+ (target_node, abort_msg))
+ # Don't raise an exception here, as we stil have to try to revert the
+ # disk status, even if this step failed.
+
+ def _ExecMigration(self):
+ """Migrate an instance.
+
+ The migrate is done by:
+ - change the disks into dual-master mode
+ - wait until disks are fully synchronized again
+ - migrate the instance
+ - change disks on the new secondary node (the old primary) to secondary
+ - wait until disks are fully synchronized
+ - change disks into single-master mode
+
+ """
+ instance = self.instance
+ target_node = self.target_node
+ source_node = self.source_node
+
+ self.feedback_fn("* checking disk consistency between source and target")
+ for dev in instance.disks:
+ if not _CheckDiskConsistency(self, dev, target_node, False):
+ raise errors.OpExecError("Disk %s is degraded or not fully"
+ " synchronized on target node,"
+ " aborting migrate." % dev.iv_name)
+
+ # First get the migration information from the remote node
+ result = self.rpc.call_migration_info(source_node, instance)
+ msg = result.fail_msg
+ if msg:
+ log_err = ("Failed fetching source migration information from %s: %s" %
+ (source_node, msg))
+ logging.error(log_err)
+ raise errors.OpExecError(log_err)
+
+ self.migration_info = migration_info = result.payload
+
+ # Then switch the disks to master/master mode
+ self._EnsureSecondary(target_node)
+ self._GoStandalone()
+ self._GoReconnect(True)
+ self._WaitUntilSync()
+
+ self.feedback_fn("* preparing %s to accept the instance" % target_node)
+ result = self.rpc.call_accept_instance(target_node,
+ instance,
+ migration_info,
+ self.nodes_ip[target_node])
+
+ msg = result.fail_msg
+ if msg:
+ logging.error("Instance pre-migration failed, trying to revert"
+ " disk status: %s", msg)
+ self._AbortMigration()
+ self._RevertDiskStatus()
+ raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
+ (instance.name, msg))
+
+ self.feedback_fn("* migrating instance to %s" % target_node)
+ time.sleep(10)
+ result = self.rpc.call_instance_migrate(source_node, instance,
+ self.nodes_ip[target_node],
+ self.live)
+ msg = result.fail_msg
+ if msg:
+ logging.error("Instance migration failed, trying to revert"
+ " disk status: %s", msg)
+ self._AbortMigration()
+ self._RevertDiskStatus()
+ raise errors.OpExecError("Could not migrate instance %s: %s" %
+ (instance.name, msg))
+ time.sleep(10)
+
+ instance.primary_node = target_node
+ # distribute new instance config to the other nodes
+ self.cfg.Update(instance)
+
+ result = self.rpc.call_finalize_migration(target_node,
+ instance,
+ migration_info,
+ True)
+ msg = result.fail_msg
+ if msg:
+ logging.error("Instance migration succeeded, but finalization failed:"
+ " %s" % msg)
+ raise errors.OpExecError("Could not finalize instance migration: %s" %
+ msg)
+
+ self._EnsureSecondary(source_node)
+ self._WaitUntilSync()
+ self._GoStandalone()
+ self._GoReconnect(False)
+ self._WaitUntilSync()
+
+ self.feedback_fn("* done")
+
+ def Exec(self, feedback_fn):
+ """Perform the migration.
+
+ """
+ feedback_fn("Migrating instance %s" % self.instance.name)
+
+ self.feedback_fn = feedback_fn
+
+ self.source_node = self.instance.primary_node
+ self.target_node = self.instance.secondary_nodes[0]
+ self.all_nodes = [self.source_node, self.target_node]
+ self.nodes_ip = {
+ self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
+ self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
+ }
+
+ if self.cleanup:
+ return self._ExecCleanup()
+ else:
+ return self._ExecMigration()
+
+
+def _CreateBlockDev(lu, node, instance, device, force_create,
+ info, force_open):
+ """Create a tree of block devices on a given node.
+
+ If this device type has to be created on secondaries, create it and
+ all its children.
+
+ If not, just recurse to children keeping the same 'force' value.
+
+ @param lu: the lu on whose behalf we execute
+ @param node: the node on which to create the device
+ @type instance: L{objects.Instance}
+ @param instance: the instance which owns the device
+ @type device: L{objects.Disk}
+ @param device: the device to create
+ @type force_create: boolean
+ @param force_create: whether to force creation of this device; this
+ will be change to True whenever we find a device which has
+ CreateOnSecondary() attribute
+ @param info: the extra 'metadata' we should attach to the device
+ (this will be represented as a LVM tag)
+ @type force_open: boolean
+ @param force_open: this parameter will be passes to the
+ L{backend.BlockdevCreate} function where it specifies
+ whether we run on primary or not, and it affects both
+ the child assembly and the device own Open() execution
+
+ """
+ if device.CreateOnSecondary():
+ force_create = True
+
+ if device.children:
+ for child in device.children:
+ _CreateBlockDev(lu, node, instance, child, force_create,
+ info, force_open)
+
+ if not force_create:
+ return
+
+ _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
+
+
+def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
+ """Create a single block device on a given node.
+
+ This will not recurse over children of the device, so they must be
+ created in advance.
+
+ @param lu: the lu on whose behalf we execute
+ @param node: the node on which to create the device
+ @type instance: L{objects.Instance}
+ @param instance: the instance which owns the device
+ @type device: L{objects.Disk}
+ @param device: the device to create
+ @param info: the extra 'metadata' we should attach to the device
+ (this will be represented as a LVM tag)
+ @type force_open: boolean
+ @param force_open: this parameter will be passes to the
+ L{backend.BlockdevCreate} function where it specifies
+ whether we run on primary or not, and it affects both
+ the child assembly and the device own Open() execution
+
+ """
+ lu.cfg.SetDiskID(device, node)
+ result = lu.rpc.call_blockdev_create(node, device, device.size,
+ instance.name, force_open, info)
+ result.Raise("Can't create block device %s on"
+ " node %s for instance %s" % (device, node, instance.name))
+ if device.physical_id is None:
+ device.physical_id = result.payload
+
+
+def _GenerateUniqueNames(lu, exts):
+ """Generate a suitable LV name.
+
+ This will generate a logical volume name for the given instance.
+
+ """
+ results = []
+ for val in exts:
+ new_id = lu.cfg.GenerateUniqueID()
+ results.append("%s%s" % (new_id, val))
+ return results
+
+
+def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
+ p_minor, s_minor):
+ """Generate a drbd8 device complete with its children.
+
+ """
+ port = lu.cfg.AllocatePort()
+ vgname = lu.cfg.GetVGName()
+ shared_secret = lu.cfg.GenerateDRBDSecret()
+ dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
+ logical_id=(vgname, names[0]))
+ dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
+ logical_id=(vgname, names[1]))
+ drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
+ logical_id=(primary, secondary, port,
+ p_minor, s_minor,
+ shared_secret),
+ children=[dev_data, dev_meta],
+ iv_name=iv_name)
+ return drbd_dev
+
+
+def _GenerateDiskTemplate(lu, template_name,
+ instance_name, primary_node,
+ secondary_nodes, disk_info,
+ file_storage_dir, file_driver,
+ base_index):
+ """Generate the entire disk layout for a given template type.
+
+ """
+ #TODO: compute space requirements
+
+ vgname = lu.cfg.GetVGName()
+ disk_count = len(disk_info)
+ disks = []
+ if template_name == constants.DT_DISKLESS:
+ pass
+ elif template_name == constants.DT_PLAIN:
+ if len(secondary_nodes) != 0:
+ raise errors.ProgrammerError("Wrong template configuration")
+
+ names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
+ for i in range(disk_count)])
+ for idx, disk in enumerate(disk_info):
+ disk_index = idx + base_index
+ disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
+ logical_id=(vgname, names[idx]),
+ iv_name="disk/%d" % disk_index,
+ mode=disk["mode"])
+ disks.append(disk_dev)
+ elif template_name == constants.DT_DRBD8:
+ if len(secondary_nodes) != 1:
+ raise errors.ProgrammerError("Wrong template configuration")
+ remote_node = secondary_nodes[0]
+ minors = lu.cfg.AllocateDRBDMinor(
+ [primary_node, remote_node] * len(disk_info), instance_name)
+
+ names = []
+ for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
+ for i in range(disk_count)]):
+ names.append(lv_prefix + "_data")
+ names.append(lv_prefix + "_meta")
+ for idx, disk in enumerate(disk_info):
+ disk_index = idx + base_index
+ disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
+ disk["size"], names[idx*2:idx*2+2],
+ "disk/%d" % disk_index,
+ minors[idx*2], minors[idx*2+1])
+ disk_dev.mode = disk["mode"]
+ disks.append(disk_dev)
+ elif template_name == constants.DT_FILE:
+ if len(secondary_nodes) != 0:
+ raise errors.ProgrammerError("Wrong template configuration")
+
+ for idx, disk in enumerate(disk_info):
+ disk_index = idx + base_index
+ disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
+ iv_name="disk/%d" % disk_index,
+ logical_id=(file_driver,
+ "%s/disk%d" % (file_storage_dir,
+ disk_index)),
+ mode=disk["mode"])
+ disks.append(disk_dev)
+ else:
+ raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
+ return disks
+
+
+def _GetInstanceInfoText(instance):
+ """Compute that text that should be added to the disk's metadata.
+
+ """
+ return "originstname+%s" % instance.name
+
+
+def _CreateDisks(lu, instance, to_skip=None, target_node=None):
+ """Create all disks for an instance.
+
+ This abstracts away some work from AddInstance.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance whose disks we should create
+ @type to_skip: list
+ @param to_skip: list of indices to skip
+ @type target_node: string
+ @param target_node: if passed, overrides the target node for creation
+ @rtype: boolean
+ @return: the success of the creation
+
+ """
+ info = _GetInstanceInfoText(instance)
+ if target_node is None:
+ pnode = instance.primary_node
+ all_nodes = instance.all_nodes
+ else:
+ pnode = target_node
+ all_nodes = [pnode]
+
+ if instance.disk_template == constants.DT_FILE:
+ file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
+ result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
+
+ result.Raise("Failed to create directory '%s' on"
+ " node %s" % (file_storage_dir, pnode))
+
+ # Note: this needs to be kept in sync with adding of disks in
+ # LUSetInstanceParams
+ for idx, device in enumerate(instance.disks):
+ if to_skip and idx in to_skip:
+ continue
+ logging.info("Creating volume %s for instance %s",
+ device.iv_name, instance.name)
+ #HARDCODE
+ for node in all_nodes:
+ f_create = node == pnode
+ _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
+
+
+def _RemoveDisks(lu, instance, target_node=None):
+ """Remove all disks for an instance.
+
+ This abstracts away some work from `AddInstance()` and
+ `RemoveInstance()`. Note that in case some of the devices couldn't
+ be removed, the removal will continue with the other ones (compare
+ with `_CreateDisks()`).
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit on whose behalf we execute
+ @type instance: L{objects.Instance}
+ @param instance: the instance whose disks we should remove
+ @type target_node: string
+ @param target_node: used to override the node on which to remove the disks
+ @rtype: boolean
+ @return: the success of the removal
+
+ """
+ logging.info("Removing block devices for instance %s", instance.name)
+
+ all_result = True
+ for device in instance.disks:
+ if target_node:
+ edata = [(target_node, device)]
+ else:
+ edata = device.ComputeNodeTree(instance.primary_node)
+ for node, disk in edata:
+ lu.cfg.SetDiskID(disk, node)
+ msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
+ if msg:
+ lu.LogWarning("Could not remove block device %s on node %s,"
+ " continuing anyway: %s", device.iv_name, node, msg)
+ all_result = False
+
+ if instance.disk_template == constants.DT_FILE:
+ file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
+ if target_node:
+ tgt = target_node
+ else:
+ tgt = instance.primary_node
+ result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
+ if result.fail_msg:
+ lu.LogWarning("Could not remove directory '%s' on node %s: %s",
+ file_storage_dir, instance.primary_node, result.fail_msg)
+ all_result = False
+
+ return all_result
+
+
+def _ComputeDiskSize(disk_template, disks):
+ """Compute disk size requirements in the volume group
+
+ """
+ # Required free disk space as a function of disk and swap space
+ req_size_dict = {
+ constants.DT_DISKLESS: None,
+ constants.DT_PLAIN: sum(d["size"] for d in disks),
+ # 128 MB are added for drbd metadata for each disk
+ constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
constants.DT_FILE: None,
}
return req_size_dict[disk_template]
+def _CheckHVParams(lu, nodenames, hvname, hvparams):
+ """Hypervisor parameter validation.
+
+ This function abstract the hypervisor parameter validation to be
+ used in both instance create and instance modify.
+
+ @type lu: L{LogicalUnit}
+ @param lu: the logical unit for which we check
+ @type nodenames: list
+ @param nodenames: the list of nodes on which we should check
+ @type hvname: string
+ @param hvname: the name of the hypervisor we should use
+ @type hvparams: dict
+ @param hvparams: the parameters which we need to check
+ @raise errors.OpPrereqError: if the parameters are not valid
+
+ """
+ hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
+ hvname,
+ hvparams)
+ for node in nodenames:
+ info = hvinfo[node]
+ if info.offline:
+ continue
+ info.Raise("Hypervisor parameter validation failed on node %s" % node)
+
+
class LUCreateInstance(LogicalUnit):
"""Create an instance.
"""
HPATH = "instance-add"
HTYPE = constants.HTYPE_INSTANCE
- _OP_REQP = ["instance_name", "mem_size", "disk_size",
- "disk_template", "swap_size", "mode", "start", "vcpus",
- "wait_for_sync", "ip_check", "mac"]
+ _OP_REQP = ["instance_name", "disks", "disk_template",
+ "mode", "start",
+ "wait_for_sync", "ip_check", "nics",
+ "hvparams", "beparams"]
+ REQ_BGL = False
+
+ def _ExpandNode(self, node):
+ """Expands and checks one node name.
+
+ """
+ node_full = self.cfg.ExpandNodeName(node)
+ if node_full is None:
+ raise errors.OpPrereqError("Unknown node %s" % node)
+ return node_full
+
+ def ExpandNames(self):
+ """ExpandNames for CreateInstance.
+
+ Figure out the right locks for instance creation.
+
+ """
+ self.needed_locks = {}
+
+ # set optional parameters to none if they don't exist
+ for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
+ if not hasattr(self.op, attr):
+ setattr(self.op, attr, None)
+
+ # cheap checks, mostly valid constants given
+
+ # verify creation mode
+ if self.op.mode not in (constants.INSTANCE_CREATE,
+ constants.INSTANCE_IMPORT):
+ raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
+ self.op.mode)
+
+ # disk template and mirror node verification
+ if self.op.disk_template not in constants.DISK_TEMPLATES:
+ raise errors.OpPrereqError("Invalid disk template name")
+
+ if self.op.hypervisor is None:
+ self.op.hypervisor = self.cfg.GetHypervisorType()
+
+ cluster = self.cfg.GetClusterInfo()
+ enabled_hvs = cluster.enabled_hypervisors
+ if self.op.hypervisor not in enabled_hvs:
+ raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
+ " cluster (%s)" % (self.op.hypervisor,
+ ",".join(enabled_hvs)))
+
+ # check hypervisor parameter syntax (locally)
+ utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
+ filled_hvp = objects.FillDict(cluster.hvparams[self.op.hypervisor],
+ self.op.hvparams)
+ hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
+ hv_type.CheckParameterSyntax(filled_hvp)
+ self.hv_full = filled_hvp
+
+ # fill and remember the beparams dict
+ utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
+ self.be_full = objects.FillDict(cluster.beparams[constants.PP_DEFAULT],
+ self.op.beparams)
+
+ #### instance parameters check
+
+ # instance name verification
+ hostname1 = utils.HostInfo(self.op.instance_name)
+ self.op.instance_name = instance_name = hostname1.name
+
+ # this is just a preventive check, but someone might still add this
+ # instance in the meantime, and creation will fail at lock-add time
+ if instance_name in self.cfg.GetInstanceList():
+ raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
+ instance_name)
+
+ self.add_locks[locking.LEVEL_INSTANCE] = instance_name
+
+ # NIC buildup
+ self.nics = []
+ for idx, nic in enumerate(self.op.nics):
+ nic_mode_req = nic.get("mode", None)
+ nic_mode = nic_mode_req
+ if nic_mode is None:
+ nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
+
+ # in routed mode, for the first nic, the default ip is 'auto'
+ if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
+ default_ip_mode = constants.VALUE_AUTO
+ else:
+ default_ip_mode = constants.VALUE_NONE
+
+ # ip validity checks
+ ip = nic.get("ip", default_ip_mode)
+ if ip is None or ip.lower() == constants.VALUE_NONE:
+ nic_ip = None
+ elif ip.lower() == constants.VALUE_AUTO:
+ nic_ip = hostname1.ip
+ else:
+ if not utils.IsValidIP(ip):
+ raise errors.OpPrereqError("Given IP address '%s' doesn't look"
+ " like a valid IP" % ip)
+ nic_ip = ip
+
+ # TODO: check the ip for uniqueness !!
+ if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
+ raise errors.OpPrereqError("Routed nic mode requires an ip address")
+
+ # MAC address verification
+ mac = nic.get("mac", constants.VALUE_AUTO)
+ if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ if not utils.IsValidMac(mac.lower()):
+ raise errors.OpPrereqError("Invalid MAC address specified: %s" %
+ mac)
+ else:
+ # or validate/reserve the current one
+ if self.cfg.IsMacInUse(mac):
+ raise errors.OpPrereqError("MAC address %s already in use"
+ " in cluster" % mac)
+
+ # bridge verification
+ bridge = nic.get("bridge", None)
+ link = nic.get("link", None)
+ if bridge and link:
+ raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
+ " at the same time")
+ elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
+ raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic")
+ elif bridge:
+ link = bridge
+
+ nicparams = {}
+ if nic_mode_req:
+ nicparams[constants.NIC_MODE] = nic_mode_req
+ if link:
+ nicparams[constants.NIC_LINK] = link
+
+ check_params = objects.FillDict(cluster.nicparams[constants.PP_DEFAULT],
+ nicparams)
+ objects.NIC.CheckParameterSyntax(check_params)
+ self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
+
+ # disk checks/pre-build
+ self.disks = []
+ for disk in self.op.disks:
+ mode = disk.get("mode", constants.DISK_RDWR)
+ if mode not in constants.DISK_ACCESS_SET:
+ raise errors.OpPrereqError("Invalid disk access mode '%s'" %
+ mode)
+ size = disk.get("size", None)
+ if size is None:
+ raise errors.OpPrereqError("Missing disk size")
+ try:
+ size = int(size)
+ except ValueError:
+ raise errors.OpPrereqError("Invalid disk size '%s'" % size)
+ self.disks.append({"size": size, "mode": mode})
+
+ # used in CheckPrereq for ip ping check
+ self.check_ip = hostname1.ip
+
+ # file storage checks
+ if (self.op.file_driver and
+ not self.op.file_driver in constants.FILE_DRIVER):
+ raise errors.OpPrereqError("Invalid file driver name '%s'" %
+ self.op.file_driver)
+
+ if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
+ raise errors.OpPrereqError("File storage directory path not absolute")
+
+ ### Node/iallocator related checks
+ if [self.op.iallocator, self.op.pnode].count(None) != 1:
+ raise errors.OpPrereqError("One and only one of iallocator and primary"
+ " node must be given")
+
+ if self.op.iallocator:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ else:
+ self.op.pnode = self._ExpandNode(self.op.pnode)
+ nodelist = [self.op.pnode]
+ if self.op.snode is not None:
+ self.op.snode = self._ExpandNode(self.op.snode)
+ nodelist.append(self.op.snode)
+ self.needed_locks[locking.LEVEL_NODE] = nodelist
+
+ # in case of import lock the source node too
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ src_node = getattr(self.op, "src_node", None)
+ src_path = getattr(self.op, "src_path", None)
+
+ if src_path is None:
+ self.op.src_path = src_path = self.op.instance_name
+
+ if src_node is None:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.op.src_node = None
+ if os.path.isabs(src_path):
+ raise errors.OpPrereqError("Importing an instance from an absolute"
+ " path requires a source node option.")
+ else:
+ self.op.src_node = src_node = self._ExpandNode(src_node)
+ if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
+ self.needed_locks[locking.LEVEL_NODE].append(src_node)
+ if not os.path.isabs(src_path):
+ self.op.src_path = src_path = \
+ os.path.join(constants.EXPORT_DIR, src_path)
+
+ else: # INSTANCE_CREATE
+ if getattr(self.op, "os_type", None) is None:
+ raise errors.OpPrereqError("No guest OS specified")
def _RunAllocator(self):
"""Run the allocator based on input opcode.
"""
- disks = [{"size": self.op.disk_size, "mode": "w"},
- {"size": self.op.swap_size, "mode": "w"}]
- nics = [{"mac": self.op.mac, "ip": getattr(self.op, "ip", None),
- "bridge": self.op.bridge}]
- ial = IAllocator(self.cfg, self.sstore,
+ nics = [n.ToDict() for n in self.nics]
+ ial = IAllocator(self.cfg, self.rpc,
mode=constants.IALLOCATOR_MODE_ALLOC,
name=self.op.instance_name,
disk_template=self.op.disk_template,
tags=[],
os=self.op.os_type,
- vcpus=self.op.vcpus,
- mem_size=self.op.mem_size,
- disks=disks,
+ vcpus=self.be_full[constants.BE_VCPUS],
+ mem_size=self.be_full[constants.BE_MEMORY],
+ disks=self.disks,
nics=nics,
+ hypervisor=self.op.hypervisor,
)
ial.Run(self.op.iallocator)
if len(ial.nodes) != ial.required_nodes:
raise errors.OpPrereqError("iallocator '%s' returned invalid number"
" of nodes (%s), required %s" %
- (len(ial.nodes), ial.required_nodes))
+ (self.op.iallocator, len(ial.nodes),
+ ial.required_nodes))
self.op.pnode = ial.nodes[0]
- logger.ToStdout("Selected nodes for the instance: %s" %
- (", ".join(ial.nodes),))
- logger.Info("Selected nodes for instance %s via iallocator %s: %s" %
- (self.op.instance_name, self.op.iallocator, ial.nodes))
+ self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
+ self.op.instance_name, self.op.iallocator,
+ ", ".join(ial.nodes))
if ial.required_nodes == 2:
self.op.snode = ial.nodes[1]
"""
env = {
- "INSTANCE_DISK_TEMPLATE": self.op.disk_template,
- "INSTANCE_DISK_SIZE": self.op.disk_size,
- "INSTANCE_SWAP_SIZE": self.op.swap_size,
- "INSTANCE_ADD_MODE": self.op.mode,
+ "ADD_MODE": self.op.mode,
}
if self.op.mode == constants.INSTANCE_IMPORT:
- env["INSTANCE_SRC_NODE"] = self.op.src_node
- env["INSTANCE_SRC_PATH"] = self.op.src_path
- env["INSTANCE_SRC_IMAGE"] = self.src_image
+ env["SRC_NODE"] = self.op.src_node
+ env["SRC_PATH"] = self.op.src_path
+ env["SRC_IMAGES"] = self.src_images
- env.update(_BuildInstanceHookEnv(name=self.op.instance_name,
+ env.update(_BuildInstanceHookEnv(
+ name=self.op.instance_name,
primary_node=self.op.pnode,
secondary_nodes=self.secondaries,
- status=self.instance_status,
+ status=self.op.start,
os_type=self.op.os_type,
- memory=self.op.mem_size,
- vcpus=self.op.vcpus,
- nics=[(self.inst_ip, self.op.bridge, self.op.mac)],
+ memory=self.be_full[constants.BE_MEMORY],
+ vcpus=self.be_full[constants.BE_VCPUS],
+ nics=_NICListToTuple(self, self.nics),
+ disk_template=self.op.disk_template,
+ disks=[(d["size"], d["mode"]) for d in self.disks],
+ bep=self.be_full,
+ hvp=self.hv_full,
+ hypervisor_name=self.op.hypervisor,
))
- nl = ([self.sstore.GetMasterNode(), self.op.pnode] +
+ nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
self.secondaries)
return env, nl, nl
"""Check prerequisites.
"""
- # set optional parameters to none if they don't exist
- for attr in ["kernel_path", "initrd_path", "hvm_boot_order", "pnode",
- "iallocator", "hvm_acpi", "hvm_pae", "hvm_cdrom_image_path",
- "hvm_nic_type", "hvm_disk_type", "vnc_bind_address"]:
- if not hasattr(self.op, attr):
- setattr(self.op, attr, None)
-
- if self.op.mode not in (constants.INSTANCE_CREATE,
- constants.INSTANCE_IMPORT):
- raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
- self.op.mode)
-
- if (not self.cfg.GetVGName() and
- self.op.disk_template not in constants.DTS_NOT_LVM):
- raise errors.OpPrereqError("Cluster does not support lvm-based"
- " instances")
-
- if self.op.mode == constants.INSTANCE_IMPORT:
- src_node = getattr(self.op, "src_node", None)
- src_path = getattr(self.op, "src_path", None)
- if src_node is None or src_path is None:
- raise errors.OpPrereqError("Importing an instance requires source"
- " node and path options")
- src_node_full = self.cfg.ExpandNodeName(src_node)
- if src_node_full is None:
- raise errors.OpPrereqError("Unknown source node '%s'" % src_node)
- self.op.src_node = src_node = src_node_full
-
- if not os.path.isabs(src_path):
- raise errors.OpPrereqError("The source path must be absolute")
-
- export_info = rpc.call_export_info(src_node, src_path)
-
- if not export_info:
- raise errors.OpPrereqError("No export found in dir %s" % src_path)
+ if (not self.cfg.GetVGName() and
+ self.op.disk_template not in constants.DTS_NOT_LVM):
+ raise errors.OpPrereqError("Cluster does not support lvm-based"
+ " instances")
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ src_node = self.op.src_node
+ src_path = self.op.src_path
+
+ if src_node is None:
+ locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
+ exp_list = self.rpc.call_export_list(locked_nodes)
+ found = False
+ for node in exp_list:
+ if exp_list[node].fail_msg:
+ continue
+ if src_path in exp_list[node].payload:
+ found = True
+ self.op.src_node = src_node = node
+ self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
+ src_path)
+ break
+ if not found:
+ raise errors.OpPrereqError("No export found for relative path %s" %
+ src_path)
+
+ _CheckNodeOnline(self, src_node)
+ result = self.rpc.call_export_info(src_node, src_path)
+ result.Raise("No export or invalid export found in dir %s" % src_path)
+
+ export_info = objects.SerializableConfigParser.Loads(str(result.payload))
if not export_info.has_section(constants.INISECT_EXP):
raise errors.ProgrammerError("Corrupted export config")
raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
(ei_version, constants.EXPORT_VERSION))
- if int(export_info.get(constants.INISECT_INS, 'disk_count')) > 1:
- raise errors.OpPrereqError("Can't import instance with more than"
- " one data disk")
+ # Check that the new instance doesn't have less disks than the export
+ instance_disks = len(self.disks)
+ export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
+ if instance_disks < export_disks:
+ raise errors.OpPrereqError("Not enough disks to import."
+ " (instance: %d, export: %d)" %
+ (instance_disks, export_disks))
- # FIXME: are the old os-es, disk sizes, etc. useful?
self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
- diskimage = os.path.join(src_path, export_info.get(constants.INISECT_INS,
- 'disk0_dump'))
- self.src_image = diskimage
- else: # INSTANCE_CREATE
- if getattr(self.op, "os_type", None) is None:
- raise errors.OpPrereqError("No guest OS specified")
-
- #### instance parameters check
-
- # disk template and mirror node verification
- if self.op.disk_template not in constants.DISK_TEMPLATES:
- raise errors.OpPrereqError("Invalid disk template name")
-
- # instance name verification
- hostname1 = utils.HostInfo(self.op.instance_name)
+ disk_images = []
+ for idx in range(export_disks):
+ option = 'disk%d_dump' % idx
+ if export_info.has_option(constants.INISECT_INS, option):
+ # FIXME: are the old os-es, disk sizes, etc. useful?
+ export_name = export_info.get(constants.INISECT_INS, option)
+ image = os.path.join(src_path, export_name)
+ disk_images.append(image)
+ else:
+ disk_images.append(False)
- self.op.instance_name = instance_name = hostname1.name
- instance_list = self.cfg.GetInstanceList()
- if instance_name in instance_list:
- raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
- instance_name)
+ self.src_images = disk_images
- # ip validity checks
- ip = getattr(self.op, "ip", None)
- if ip is None or ip.lower() == "none":
- inst_ip = None
- elif ip.lower() == "auto":
- inst_ip = hostname1.ip
- else:
- if not utils.IsValidIP(ip):
- raise errors.OpPrereqError("given IP address '%s' doesn't look"
- " like a valid IP" % ip)
- inst_ip = ip
- self.inst_ip = self.op.ip = inst_ip
+ old_name = export_info.get(constants.INISECT_INS, 'name')
+ # FIXME: int() here could throw a ValueError on broken exports
+ exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
+ if self.op.instance_name == old_name:
+ for idx, nic in enumerate(self.nics):
+ if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
+ nic_mac_ini = 'nic%d_mac' % idx
+ nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
+ # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
+ # ip ping checks (we use the same ip that was resolved in ExpandNames)
if self.op.start and not self.op.ip_check:
raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
" adding an instance in start mode")
if self.op.ip_check:
- if utils.TcpPing(hostname1.ip, constants.DEFAULT_NODED_PORT):
+ if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
raise errors.OpPrereqError("IP %s of instance %s already in use" %
- (hostname1.ip, instance_name))
-
- # MAC address verification
- if self.op.mac != "auto":
- if not utils.IsValidMac(self.op.mac.lower()):
- raise errors.OpPrereqError("invalid MAC address specified: %s" %
- self.op.mac)
-
- # bridge verification
- bridge = getattr(self.op, "bridge", None)
- if bridge is None:
- self.op.bridge = self.cfg.GetDefBridge()
- else:
- self.op.bridge = bridge
-
- # boot order verification
- if self.op.hvm_boot_order is not None:
- if len(self.op.hvm_boot_order.strip("acdn")) != 0:
- raise errors.OpPrereqError("invalid boot order specified,"
- " must be one or more of [acdn]")
- # file storage checks
- if (self.op.file_driver and
- not self.op.file_driver in constants.FILE_DRIVER):
- raise errors.OpPrereqError("Invalid file driver name '%s'" %
- self.op.file_driver)
+ (self.check_ip, self.op.instance_name))
+
+ #### mac address generation
+ # By generating here the mac address both the allocator and the hooks get
+ # the real final mac address rather than the 'auto' or 'generate' value.
+ # There is a race condition between the generation and the instance object
+ # creation, which means that we know the mac is valid now, but we're not
+ # sure it will be when we actually add the instance. If things go bad
+ # adding the instance will abort because of a duplicate mac, and the
+ # creation job will fail.
+ for nic in self.nics:
+ if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ nic.mac = self.cfg.GenerateMAC()
- if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
- raise errors.OpPrereqError("File storage directory not a relative"
- " path")
#### allocator run
- if [self.op.iallocator, self.op.pnode].count(None) != 1:
- raise errors.OpPrereqError("One and only one of iallocator and primary"
- " node must be given")
-
if self.op.iallocator is not None:
self._RunAllocator()
#### node related checks
# check primary node
- pnode = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.pnode))
- if pnode is None:
- raise errors.OpPrereqError("Primary node '%s' is unknown" %
- self.op.pnode)
- self.op.pnode = pnode.name
- self.pnode = pnode
+ self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
+ assert self.pnode is not None, \
+ "Cannot retrieve locked node %s" % self.op.pnode
+ if pnode.offline:
+ raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
+ pnode.name)
+ if pnode.drained:
+ raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
+ pnode.name)
+
self.secondaries = []
# mirror node verification
if self.op.disk_template in constants.DTS_NET_MIRROR:
- if getattr(self.op, "snode", None) is None:
+ if self.op.snode is None:
raise errors.OpPrereqError("The networked disk templates need"
" a mirror node")
-
- snode_name = self.cfg.ExpandNodeName(self.op.snode)
- if snode_name is None:
- raise errors.OpPrereqError("Unknown secondary node '%s'" %
- self.op.snode)
- elif snode_name == pnode.name:
+ if self.op.snode == pnode.name:
raise errors.OpPrereqError("The secondary node cannot be"
" the primary node.")
- self.secondaries.append(snode_name)
+ _CheckNodeOnline(self, self.op.snode)
+ _CheckNodeNotDrained(self, self.op.snode)
+ self.secondaries.append(self.op.snode)
+
+ nodenames = [pnode.name] + self.secondaries
req_size = _ComputeDiskSize(self.op.disk_template,
- self.op.disk_size, self.op.swap_size)
+ self.disks)
# Check lv size requirements
if req_size is not None:
- nodenames = [pnode.name] + self.secondaries
- nodeinfo = rpc.call_node_info(nodenames, self.cfg.GetVGName())
+ nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
+ self.op.hypervisor)
for node in nodenames:
- info = nodeinfo.get(node, None)
- if not info:
- raise errors.OpPrereqError("Cannot get current information"
- " from node '%s'" % node)
+ info = nodeinfo[node]
+ info.Raise("Cannot get current information from node %s" % node)
+ info = info.payload
vg_free = info.get('vg_free', None)
if not isinstance(vg_free, int):
raise errors.OpPrereqError("Can't compute free disk space on"
" node %s" % node)
- if req_size > info['vg_free']:
+ if req_size > vg_free:
raise errors.OpPrereqError("Not enough disk space on target node %s."
" %d MB available, %d MB required" %
- (node, info['vg_free'], req_size))
-
- # os verification
- os_obj = rpc.call_os_get(pnode.name, self.op.os_type)
- if not os_obj:
- raise errors.OpPrereqError("OS '%s' not in supported os list for"
- " primary node" % self.op.os_type)
+ (node, vg_free, req_size))
- if self.op.kernel_path == constants.VALUE_NONE:
- raise errors.OpPrereqError("Can't set instance kernel to none")
+ _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
+ # os verification
+ result = self.rpc.call_os_get(pnode.name, self.op.os_type)
+ result.Raise("OS '%s' not in supported os list for primary node %s" %
+ (self.op.os_type, pnode.name), prereq=True)
- # bridge check on primary node
- if not rpc.call_bridges_exist(self.pnode.name, [self.op.bridge]):
- raise errors.OpPrereqError("target bridge '%s' does not exist on"
- " destination node '%s'" %
- (self.op.bridge, pnode.name))
+ _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
# memory check on primary node
if self.op.start:
- _CheckNodeFreeMemory(self.cfg, self.pnode.name,
+ _CheckNodeFreeMemory(self, self.pnode.name,
"creating instance %s" % self.op.instance_name,
- self.op.mem_size)
-
- # hvm_cdrom_image_path verification
- if self.op.hvm_cdrom_image_path is not None:
- if not os.path.isabs(self.op.hvm_cdrom_image_path):
- raise errors.OpPrereqError("The path to the HVM CDROM image must"
- " be an absolute path or None, not %s" %
- self.op.hvm_cdrom_image_path)
- if not os.path.isfile(self.op.hvm_cdrom_image_path):
- raise errors.OpPrereqError("The HVM CDROM image must either be a"
- " regular file or a symlink pointing to"
- " an existing regular file, not %s" %
- self.op.hvm_cdrom_image_path)
-
- # vnc_bind_address verification
- if self.op.vnc_bind_address is not None:
- if not utils.IsValidIP(self.op.vnc_bind_address):
- raise errors.OpPrereqError("given VNC bind address '%s' doesn't look"
- " like a valid IP address" %
- self.op.vnc_bind_address)
-
- # Xen HVM device type checks
- if self.sstore.GetHypervisorType() == constants.HT_XEN_HVM31:
- if self.op.hvm_nic_type not in constants.HT_HVM_VALID_NIC_TYPES:
- raise errors.OpPrereqError("Invalid NIC type %s specified for Xen HVM"
- " hypervisor" % self.op.hvm_nic_type)
- if self.op.hvm_disk_type not in constants.HT_HVM_VALID_DISK_TYPES:
- raise errors.OpPrereqError("Invalid disk type %s specified for Xen HVM"
- " hypervisor" % self.op.hvm_disk_type)
+ self.be_full[constants.BE_MEMORY],
+ self.op.hypervisor)
- if self.op.start:
- self.instance_status = 'up'
- else:
- self.instance_status = 'down'
+ self.dry_run_result = list(nodenames)
def Exec(self, feedback_fn):
"""Create and add the instance to the cluster.
instance = self.op.instance_name
pnode_name = self.pnode.name
- if self.op.mac == "auto":
- mac_address = self.cfg.GenerateMAC()
- else:
- mac_address = self.op.mac
-
- nic = objects.NIC(bridge=self.op.bridge, mac=mac_address)
- if self.inst_ip is not None:
- nic.ip = self.inst_ip
-
- ht_kind = self.sstore.GetHypervisorType()
+ ht_kind = self.op.hypervisor
if ht_kind in constants.HTS_REQ_PORT:
network_port = self.cfg.AllocatePort()
else:
network_port = None
- if self.op.vnc_bind_address is None:
- self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
+ ##if self.op.vnc_bind_address is None:
+ ## self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
# this is needed because os.path.join does not accept None arguments
if self.op.file_storage_dir is None:
# build the full file storage dir path
file_storage_dir = os.path.normpath(os.path.join(
- self.sstore.GetFileStorageDir(),
+ self.cfg.GetFileStorageDir(),
string_file_storage_dir, instance))
- disks = _GenerateDiskTemplate(self.cfg,
+ disks = _GenerateDiskTemplate(self,
self.op.disk_template,
instance, pnode_name,
- self.secondaries, self.op.disk_size,
- self.op.swap_size,
+ self.secondaries,
+ self.disks,
file_storage_dir,
- self.op.file_driver)
+ self.op.file_driver,
+ 0)
iobj = objects.Instance(name=instance, os=self.op.os_type,
primary_node=pnode_name,
- memory=self.op.mem_size,
- vcpus=self.op.vcpus,
- nics=[nic], disks=disks,
+ nics=self.nics, disks=disks,
disk_template=self.op.disk_template,
- status=self.instance_status,
+ admin_up=False,
network_port=network_port,
- kernel_path=self.op.kernel_path,
- initrd_path=self.op.initrd_path,
- hvm_boot_order=self.op.hvm_boot_order,
- hvm_acpi=self.op.hvm_acpi,
- hvm_pae=self.op.hvm_pae,
- hvm_cdrom_image_path=self.op.hvm_cdrom_image_path,
- vnc_bind_address=self.op.vnc_bind_address,
- hvm_nic_type=self.op.hvm_nic_type,
- hvm_disk_type=self.op.hvm_disk_type,
+ beparams=self.op.beparams,
+ hvparams=self.op.hvparams,
+ hypervisor=self.op.hypervisor,
)
feedback_fn("* creating instance disks...")
- if not _CreateDisks(self.cfg, iobj):
- _RemoveDisks(iobj, self.cfg)
- raise errors.OpExecError("Device creation failed, reverting...")
+ try:
+ _CreateDisks(self, iobj)
+ except errors.OpExecError:
+ self.LogWarning("Device creation failed, reverting...")
+ try:
+ _RemoveDisks(self, iobj)
+ finally:
+ self.cfg.ReleaseDRBDMinors(instance)
+ raise
feedback_fn("adding instance %s to cluster config" % instance)
self.cfg.AddInstance(iobj)
- # Add the new instance to the Ganeti Lock Manager
- self.context.glm.add(locking.LEVEL_INSTANCE, instance)
+ # Declare that we don't want to remove the instance lock anymore, as we've
+ # added the instance to the config
+ del self.remove_locks[locking.LEVEL_INSTANCE]
+ # Unlock all the nodes
+ if self.op.mode == constants.INSTANCE_IMPORT:
+ nodes_keep = [self.op.src_node]
+ nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
+ if node != self.op.src_node]
+ self.context.glm.release(locking.LEVEL_NODE, nodes_release)
+ self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
+ else:
+ self.context.glm.release(locking.LEVEL_NODE)
+ del self.acquired_locks[locking.LEVEL_NODE]
if self.op.wait_for_sync:
- disk_abort = not _WaitForSync(self.cfg, iobj, self.proc)
+ disk_abort = not _WaitForSync(self, iobj)
elif iobj.disk_template in constants.DTS_NET_MIRROR:
# make sure the disks are not degraded (still sync-ing is ok)
time.sleep(15)
feedback_fn("* checking mirrors status")
- disk_abort = not _WaitForSync(self.cfg, iobj, self.proc, oneshot=True)
+ disk_abort = not _WaitForSync(self, iobj, oneshot=True)
else:
disk_abort = False
if disk_abort:
- _RemoveDisks(iobj, self.cfg)
+ _RemoveDisks(self, iobj)
self.cfg.RemoveInstance(iobj.name)
- # Remove the new instance from the Ganeti Lock Manager
- self.context.glm.remove(locking.LEVEL_INSTANCE, iobj.name)
+ # Make sure the instance lock gets removed
+ self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
raise errors.OpExecError("There are some degraded disks for"
" this instance")
if iobj.disk_template != constants.DT_DISKLESS:
if self.op.mode == constants.INSTANCE_CREATE:
feedback_fn("* running the instance OS create scripts...")
- if not rpc.call_instance_os_add(pnode_name, iobj, "sda", "sdb"):
- raise errors.OpExecError("could not add os for instance %s"
- " on node %s" %
- (instance, pnode_name))
+ result = self.rpc.call_instance_os_add(pnode_name, iobj, False)
+ result.Raise("Could not add os for instance %s"
+ " on node %s" % (instance, pnode_name))
elif self.op.mode == constants.INSTANCE_IMPORT:
feedback_fn("* running the instance OS import scripts...")
src_node = self.op.src_node
- src_image = self.src_image
- if not rpc.call_instance_os_import(pnode_name, iobj, "sda", "sdb",
- src_node, src_image):
- raise errors.OpExecError("Could not import os for instance"
- " %s on node %s" %
- (instance, pnode_name))
+ src_images = self.src_images
+ cluster_name = self.cfg.GetClusterName()
+ import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
+ src_node, src_images,
+ cluster_name)
+ msg = import_result.fail_msg
+ if msg:
+ self.LogWarning("Error while importing the disk images for instance"
+ " %s on node %s: %s" % (instance, pnode_name, msg))
else:
# also checked in the prereq part
raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
% self.op.mode)
if self.op.start:
- logger.Info("starting instance %s on node %s" % (instance, pnode_name))
+ iobj.admin_up = True
+ self.cfg.Update(iobj)
+ logging.info("Starting instance %s on node %s", instance, pnode_name)
feedback_fn("* starting instance...")
- if not rpc.call_instance_start(pnode_name, iobj, None):
- raise errors.OpExecError("Could not start instance")
+ result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
+ result.Raise("Could not start instance")
+
+ return list(iobj.all_nodes)
class LUConnectConsole(NoHooksLU):
self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, self.instance.primary_node)
def Exec(self, feedback_fn):
"""Connect to the console of an instance
instance = self.instance
node = instance.primary_node
- node_insts = rpc.call_instance_list([node])[node]
- if node_insts is False:
- raise errors.OpExecError("Can't connect to node %s." % node)
+ node_insts = self.rpc.call_instance_list([node],
+ [instance.hypervisor])[node]
+ node_insts.Raise("Can't get node information from %s" % node)
- if instance.name not in node_insts:
+ if instance.name not in node_insts.payload:
raise errors.OpExecError("Instance %s is not running." % instance.name)
- logger.Debug("connecting to console of %s on %s" % (instance.name, node))
+ logging.debug("Connecting to console of %s on %s", instance.name, node)
- hyper = hypervisor.GetHypervisor()
- console_cmd = hyper.GetShellCommandForConsole(instance)
+ hyper = hypervisor.GetHypervisor(instance.hypervisor)
+ cluster = self.cfg.GetClusterInfo()
+ # beparams and hvparams are passed separately, to avoid editing the
+ # instance and then saving the defaults in the instance itself.
+ hvparams = cluster.FillHV(instance)
+ beparams = cluster.FillBE(instance)
+ console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
# build ssh cmdline
return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
_OP_REQP = ["instance_name", "mode", "disks"]
REQ_BGL = False
+ def CheckArguments(self):
+ if not hasattr(self.op, "remote_node"):
+ self.op.remote_node = None
+ if not hasattr(self.op, "iallocator"):
+ self.op.iallocator = None
+
+ TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
+ self.op.iallocator)
+
def ExpandNames(self):
self._ExpandAndLockInstance()
+ if self.op.iallocator is not None:
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+ elif self.op.remote_node is not None:
+ remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
+ if remote_node is None:
+ raise errors.OpPrereqError("Node '%s' not known" %
+ self.op.remote_node)
+
+ self.op.remote_node = remote_node
+
+ # Warning: do not remove the locking of the new secondary here
+ # unless DRBD8.AddChildren is changed to work in parallel;
+ # currently it doesn't since parallel invocations of
+ # FindUnusedMinor will conflict
+ self.needed_locks[locking.LEVEL_NODE] = [remote_node]
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
+ else:
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
+ self.op.iallocator, self.op.remote_node,
+ self.op.disks)
+
+ self.tasklets = [self.replacer]
+
+ def DeclareLocks(self, level):
+ # If we're not already locking all nodes in the set we have to declare the
+ # instance's primary/secondary nodes.
+ if (level == locking.LEVEL_NODE and
+ self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
+ self._LockInstancesNodes()
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, the primary and all the secondaries.
+
+ """
+ instance = self.replacer.instance
+ env = {
+ "MODE": self.op.mode,
+ "NEW_SECONDARY": self.op.remote_node,
+ "OLD_SECONDARY": instance.secondary_nodes[0],
+ }
+ env.update(_BuildInstanceHookEnvByObject(self, instance))
+ nl = [
+ self.cfg.GetMasterNode(),
+ instance.primary_node,
+ ]
+ if self.op.remote_node is not None:
+ nl.append(self.op.remote_node)
+ return env, nl, nl
+
+
+class LUEvacuateNode(LogicalUnit):
+ """Relocate the secondary instances from a node.
+
+ """
+ HPATH = "node-evacuate"
+ HTYPE = constants.HTYPE_NODE
+ _OP_REQP = ["node_name"]
+ REQ_BGL = False
+
+ def CheckArguments(self):
if not hasattr(self.op, "remote_node"):
self.op.remote_node = None
+ if not hasattr(self.op, "iallocator"):
+ self.op.iallocator = None
- ia_name = getattr(self.op, "iallocator", None)
- if ia_name is not None:
- if self.op.remote_node is not None:
- raise errors.OpPrereqError("Give either the iallocator or the new"
- " secondary, not both")
+ TLReplaceDisks.CheckArguments(constants.REPLACE_DISK_CHG,
+ self.op.remote_node,
+ self.op.iallocator)
+
+ def ExpandNames(self):
+ self.op.node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if self.op.node_name is None:
+ raise errors.OpPrereqError("Node '%s' not known" % self.op.node_name)
+
+ self.needed_locks = {}
+
+ # Declare node locks
+ if self.op.iallocator is not None:
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
elif self.op.remote_node is not None:
remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
if remote_node is None:
raise errors.OpPrereqError("Node '%s' not known" %
self.op.remote_node)
+
self.op.remote_node = remote_node
+
+ # Warning: do not remove the locking of the new secondary here
+ # unless DRBD8.AddChildren is changed to work in parallel;
+ # currently it doesn't since parallel invocations of
+ # FindUnusedMinor will conflict
self.needed_locks[locking.LEVEL_NODE] = [remote_node]
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
else:
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ raise errors.OpPrereqError("Invalid parameters")
+
+ # Create tasklets for replacing disks for all secondary instances on this
+ # node
+ names = []
+ tasklets = []
+
+ for inst in _GetNodeSecondaryInstances(self.cfg, self.op.node_name):
+ logging.debug("Replacing disks for instance %s", inst.name)
+ names.append(inst.name)
+
+ replacer = TLReplaceDisks(self, inst.name, constants.REPLACE_DISK_CHG,
+ self.op.iallocator, self.op.remote_node, [])
+ tasklets.append(replacer)
+
+ self.tasklets = tasklets
+ self.instance_names = names
+
+ # Declare instance locks
+ self.needed_locks[locking.LEVEL_INSTANCE] = self.instance_names
def DeclareLocks(self, level):
# If we're not already locking all nodes in the set we have to declare the
self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
self._LockInstancesNodes()
- def _RunAllocator(self):
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ This runs on the master, the primary and all the secondaries.
+
+ """
+ env = {
+ "NODE_NAME": self.op.node_name,
+ }
+
+ nl = [self.cfg.GetMasterNode()]
+
+ if self.op.remote_node is not None:
+ env["NEW_SECONDARY"] = self.op.remote_node
+ nl.append(self.op.remote_node)
+
+ return (env, nl, nl)
+
+
+class TLReplaceDisks(Tasklet):
+ """Replaces disks for an instance.
+
+ Note: Locking is not within the scope of this class.
+
+ """
+ def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
+ disks):
+ """Initializes this class.
+
+ """
+ Tasklet.__init__(self, lu)
+
+ # Parameters
+ self.instance_name = instance_name
+ self.mode = mode
+ self.iallocator_name = iallocator_name
+ self.remote_node = remote_node
+ self.disks = disks
+
+ # Runtime data
+ self.instance = None
+ self.new_node = None
+ self.target_node = None
+ self.other_node = None
+ self.remote_node_info = None
+ self.node_secondary_ip = None
+
+ @staticmethod
+ def CheckArguments(mode, remote_node, iallocator):
+ """Helper function for users of this class.
+
+ """
+ # check for valid parameter combination
+ if mode == constants.REPLACE_DISK_CHG:
+ if remote_node is None and iallocator is None:
+ raise errors.OpPrereqError("When changing the secondary either an"
+ " iallocator script must be used or the"
+ " new node given")
+
+ if remote_node is not None and iallocator is not None:
+ raise errors.OpPrereqError("Give either the iallocator or the new"
+ " secondary, not both")
+
+ elif remote_node is not None or iallocator is not None:
+ # Not replacing the secondary
+ raise errors.OpPrereqError("The iallocator and new node options can"
+ " only be used when changing the"
+ " secondary node")
+
+ @staticmethod
+ def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
"""Compute a new secondary node using an IAllocator.
"""
- ial = IAllocator(self.cfg, self.sstore,
+ ial = IAllocator(lu.cfg, lu.rpc,
mode=constants.IALLOCATOR_MODE_RELOC,
- name=self.op.instance_name,
- relocate_from=[self.sec_node])
+ name=instance_name,
+ relocate_from=relocate_from)
- ial.Run(self.op.iallocator)
+ ial.Run(iallocator_name)
if not ial.success:
- raise errors.OpPrereqError("Can't compute nodes using"
- " iallocator '%s': %s" % (self.op.iallocator,
- ial.info))
+ raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
+ " %s" % (iallocator_name, ial.info))
+
if len(ial.nodes) != ial.required_nodes:
raise errors.OpPrereqError("iallocator '%s' returned invalid number"
" of nodes (%s), required %s" %
(len(ial.nodes), ial.required_nodes))
- self.op.remote_node = ial.nodes[0]
- logger.ToStdout("Selected new secondary for the instance: %s" %
- self.op.remote_node)
- def BuildHooksEnv(self):
- """Build hooks env.
+ remote_node_name = ial.nodes[0]
+
+ lu.LogInfo("Selected new secondary for instance '%s': %s",
+ instance_name, remote_node_name)
+
+ return remote_node_name
+
+ def _FindFaultyDisks(self, node_name):
+ return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
+ node_name, True)
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ This checks that the instance is in the cluster.
+
+ """
+ self.instance = self.cfg.GetInstanceInfo(self.instance_name)
+ assert self.instance is not None, \
+ "Cannot retrieve locked instance %s" % self.instance_name
+
+ if self.instance.disk_template != constants.DT_DRBD8:
+ raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
+ " instances")
+
+ if len(self.instance.secondary_nodes) != 1:
+ raise errors.OpPrereqError("The instance has a strange layout,"
+ " expected one secondary but found %d" %
+ len(self.instance.secondary_nodes))
+
+ secondary_node = self.instance.secondary_nodes[0]
+
+ if self.iallocator_name is None:
+ remote_node = self.remote_node
+ else:
+ remote_node = self._RunAllocator(self.lu, self.iallocator_name,
+ self.instance.name, secondary_node)
+
+ if remote_node is not None:
+ self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
+ assert self.remote_node_info is not None, \
+ "Cannot retrieve locked node %s" % remote_node
+ else:
+ self.remote_node_info = None
+
+ if remote_node == self.instance.primary_node:
+ raise errors.OpPrereqError("The specified node is the primary node of"
+ " the instance.")
+
+ if remote_node == secondary_node:
+ raise errors.OpPrereqError("The specified node is already the"
+ " secondary node of the instance.")
+
+ if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
+ constants.REPLACE_DISK_CHG):
+ raise errors.OpPrereqError("Cannot specify disks to be replaced")
+
+ if self.mode == constants.REPLACE_DISK_AUTO:
+ faulty_primary = self._FindFaultyDisks(self.instance.primary_node)
+ faulty_secondary = self._FindFaultyDisks(secondary_node)
+
+ if faulty_primary and faulty_secondary:
+ raise errors.OpPrereqError("Instance %s has faulty disks on more than"
+ " one node and can not be repaired"
+ " automatically" % self.instance_name)
+
+ if faulty_primary:
+ self.disks = faulty_primary
+ self.target_node = self.instance.primary_node
+ self.other_node = secondary_node
+ check_nodes = [self.target_node, self.other_node]
+ elif faulty_secondary:
+ self.disks = faulty_secondary
+ self.target_node = secondary_node
+ self.other_node = self.instance.primary_node
+ check_nodes = [self.target_node, self.other_node]
+ else:
+ self.disks = []
+ check_nodes = []
+
+ else:
+ # Non-automatic modes
+ if self.mode == constants.REPLACE_DISK_PRI:
+ self.target_node = self.instance.primary_node
+ self.other_node = secondary_node
+ check_nodes = [self.target_node, self.other_node]
+
+ elif self.mode == constants.REPLACE_DISK_SEC:
+ self.target_node = secondary_node
+ self.other_node = self.instance.primary_node
+ check_nodes = [self.target_node, self.other_node]
+
+ elif self.mode == constants.REPLACE_DISK_CHG:
+ self.new_node = remote_node
+ self.other_node = self.instance.primary_node
+ self.target_node = secondary_node
+ check_nodes = [self.new_node, self.other_node]
+
+ _CheckNodeNotDrained(self.lu, remote_node)
+
+ else:
+ raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
+ self.mode)
+
+ # If not specified all disks should be replaced
+ if not self.disks:
+ self.disks = range(len(self.instance.disks))
+
+ for node in check_nodes:
+ _CheckNodeOnline(self.lu, node)
+
+ # Check whether disks are valid
+ for disk_idx in self.disks:
+ self.instance.FindDisk(disk_idx)
+
+ # Get secondary node IP addresses
+ node_2nd_ip = {}
+
+ for node_name in [self.target_node, self.other_node, self.new_node]:
+ if node_name is not None:
+ node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
+
+ self.node_secondary_ip = node_2nd_ip
+
+ def Exec(self, feedback_fn):
+ """Execute disk replacement.
+
+ This dispatches the disk replacement to the appropriate handler.
+
+ """
+ if not self.disks:
+ feedback_fn("No disks need replacement")
+ return
+
+ feedback_fn("Replacing disk(s) %s for %s" %
+ (", ".join([str(i) for i in self.disks]), self.instance.name))
+
+ activate_disks = (not self.instance.admin_up)
+
+ # Activate the instance disks if we're replacing them on a down instance
+ if activate_disks:
+ _StartInstanceDisks(self.lu, self.instance, True)
+
+ try:
+ # Should we replace the secondary node?
+ if self.new_node is not None:
+ return self._ExecDrbd8Secondary()
+ else:
+ return self._ExecDrbd8DiskOnly()
+
+ finally:
+ # Deactivate the instance disks if we're replacing them on a down instance
+ if activate_disks:
+ _SafeShutdownInstanceDisks(self.lu, self.instance)
+
+ def _CheckVolumeGroup(self, nodes):
+ self.lu.LogInfo("Checking volume groups")
+
+ vgname = self.cfg.GetVGName()
+
+ # Make sure volume group exists on all involved nodes
+ results = self.rpc.call_vg_list(nodes)
+ if not results:
+ raise errors.OpExecError("Can't list volume groups on the nodes")
+
+ for node in nodes:
+ res = results[node]
+ res.Raise("Error checking node %s" % node)
+ if vgname not in res.payload:
+ raise errors.OpExecError("Volume group '%s' not found on node %s" %
+ (vgname, node))
+
+ def _CheckDisksExistence(self, nodes):
+ # Check disk existence
+ for idx, dev in enumerate(self.instance.disks):
+ if idx not in self.disks:
+ continue
+
+ for node in nodes:
+ self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
+ self.cfg.SetDiskID(dev, node)
+
+ result = self.rpc.call_blockdev_find(node, dev)
+
+ msg = result.fail_msg
+ if msg or not result.payload:
+ if not msg:
+ msg = "disk not found"
+ raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
+ (idx, node, msg))
+
+ def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
+ for idx, dev in enumerate(self.instance.disks):
+ if idx not in self.disks:
+ continue
+
+ self.lu.LogInfo("Checking disk/%d consistency on node %s" %
+ (idx, node_name))
+
+ if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
+ ldisk=ldisk):
+ raise errors.OpExecError("Node %s has degraded storage, unsafe to"
+ " replace disks for instance %s" %
+ (node_name, self.instance.name))
+
+ def _CreateNewStorage(self, node_name):
+ vgname = self.cfg.GetVGName()
+ iv_names = {}
+
+ for idx, dev in enumerate(self.instance.disks):
+ if idx not in self.disks:
+ continue
+
+ self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
+
+ self.cfg.SetDiskID(dev, node_name)
+
+ lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
+ names = _GenerateUniqueNames(self.lu, lv_names)
+
+ lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
+ logical_id=(vgname, names[0]))
+ lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
+ logical_id=(vgname, names[1]))
+
+ new_lvs = [lv_data, lv_meta]
+ old_lvs = dev.children
+ iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
+
+ # we pass force_create=True to force the LVM creation
+ for new_lv in new_lvs:
+ _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
+ _GetInstanceInfoText(self.instance), False)
- This runs on the master, the primary and all the secondaries.
+ return iv_names
- """
- env = {
- "MODE": self.op.mode,
- "NEW_SECONDARY": self.op.remote_node,
- "OLD_SECONDARY": self.instance.secondary_nodes[0],
- }
- env.update(_BuildInstanceHookEnvByObject(self.instance))
- nl = [
- self.sstore.GetMasterNode(),
- self.instance.primary_node,
- ]
- if self.op.remote_node is not None:
- nl.append(self.op.remote_node)
- return env, nl, nl
+ def _CheckDevices(self, node_name, iv_names):
+ for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
+ self.cfg.SetDiskID(dev, node_name)
- def CheckPrereq(self):
- """Check prerequisites.
+ result = self.rpc.call_blockdev_find(node_name, dev)
- This checks that the instance is in the cluster.
+ msg = result.fail_msg
+ if msg or not result.payload:
+ if not msg:
+ msg = "disk not found"
+ raise errors.OpExecError("Can't find DRBD device %s: %s" %
+ (name, msg))
- """
- instance = self.cfg.GetInstanceInfo(self.op.instance_name)
- assert instance is not None, \
- "Cannot retrieve locked instance %s" % self.op.instance_name
- self.instance = instance
+ if result.payload.is_degraded:
+ raise errors.OpExecError("DRBD device %s is degraded!" % name)
- if instance.disk_template not in constants.DTS_NET_MIRROR:
- raise errors.OpPrereqError("Instance's disk layout is not"
- " network mirrored.")
+ def _RemoveOldStorage(self, node_name, iv_names):
+ for name, (dev, old_lvs, _) in iv_names.iteritems():
+ self.lu.LogInfo("Remove logical volumes for %s" % name)
- if len(instance.secondary_nodes) != 1:
- raise errors.OpPrereqError("The instance has a strange layout,"
- " expected one secondary but found %d" %
- len(instance.secondary_nodes))
+ for lv in old_lvs:
+ self.cfg.SetDiskID(lv, node_name)
- self.sec_node = instance.secondary_nodes[0]
+ msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
+ if msg:
+ self.lu.LogWarning("Can't remove old LV: %s" % msg,
+ hint="remove unused LVs manually")
- ia_name = getattr(self.op, "iallocator", None)
- if ia_name is not None:
- self._RunAllocator()
+ def _ExecDrbd8DiskOnly(self):
+ """Replace a disk on the primary or secondary for DRBD 8.
- remote_node = self.op.remote_node
- if remote_node is not None:
- self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
- assert self.remote_node_info is not None, \
- "Cannot retrieve locked node %s" % remote_node
- else:
- self.remote_node_info = None
- if remote_node == instance.primary_node:
- raise errors.OpPrereqError("The specified node is the primary node of"
- " the instance.")
- elif remote_node == self.sec_node:
- if self.op.mode == constants.REPLACE_DISK_SEC:
- # this is for DRBD8, where we can't execute the same mode of
- # replacement as for drbd7 (no different port allocated)
- raise errors.OpPrereqError("Same secondary given, cannot execute"
- " replacement")
- if instance.disk_template == constants.DT_DRBD8:
- if (self.op.mode == constants.REPLACE_DISK_ALL and
- remote_node is not None):
- # switch to replace secondary mode
- self.op.mode = constants.REPLACE_DISK_SEC
-
- if self.op.mode == constants.REPLACE_DISK_ALL:
- raise errors.OpPrereqError("Template 'drbd' only allows primary or"
- " secondary disk replacement, not"
- " both at once")
- elif self.op.mode == constants.REPLACE_DISK_PRI:
- if remote_node is not None:
- raise errors.OpPrereqError("Template 'drbd' does not allow changing"
- " the secondary while doing a primary"
- " node disk replacement")
- self.tgt_node = instance.primary_node
- self.oth_node = instance.secondary_nodes[0]
- elif self.op.mode == constants.REPLACE_DISK_SEC:
- self.new_node = remote_node # this can be None, in which case
- # we don't change the secondary
- self.tgt_node = instance.secondary_nodes[0]
- self.oth_node = instance.primary_node
- else:
- raise errors.ProgrammerError("Unhandled disk replace mode")
+ The algorithm for replace is quite complicated:
- for name in self.op.disks:
- if instance.FindDisk(name) is None:
- raise errors.OpPrereqError("Disk '%s' not found for instance '%s'" %
- (name, instance.name))
+ 1. for each disk to be replaced:
- def _ExecD8DiskOnly(self, feedback_fn):
- """Replace a disk on the primary or secondary for dbrd8.
+ 1. create new LVs on the target node with unique names
+ 1. detach old LVs from the drbd device
+ 1. rename old LVs to name_replaced.<time_t>
+ 1. rename new LVs to old LVs
+ 1. attach the new LVs (with the old names now) to the drbd device
- The algorithm for replace is quite complicated:
- - for each disk to be replaced:
- - create new LVs on the target node with unique names
- - detach old LVs from the drbd device
- - rename old LVs to name_replaced.<time_t>
- - rename new LVs to old LVs
- - attach the new LVs (with the old names now) to the drbd device
- - wait for sync across all devices
- - for each modified disk:
- - remove old LVs (which have the name name_replaces.<time_t>)
+ 1. wait for sync across all devices
+
+ 1. for each modified disk:
+
+ 1. remove old LVs (which have the name name_replaces.<time_t>)
Failures are not very well handled.
"""
steps_total = 6
- warning, info = (self.proc.LogWarning, self.proc.LogInfo)
- instance = self.instance
- iv_names = {}
- vgname = self.cfg.GetVGName()
- # start of work
- cfg = self.cfg
- tgt_node = self.tgt_node
- oth_node = self.oth_node
# Step: check device activation
- self.proc.LogStep(1, steps_total, "check device existence")
- info("checking volume groups")
- my_vg = cfg.GetVGName()
- results = rpc.call_vg_list([oth_node, tgt_node])
- if not results:
- raise errors.OpExecError("Can't list volume groups on the nodes")
- for node in oth_node, tgt_node:
- res = results.get(node, False)
- if not res or my_vg not in res:
- raise errors.OpExecError("Volume group '%s' not found on %s" %
- (my_vg, node))
- for dev in instance.disks:
- if not dev.iv_name in self.op.disks:
- continue
- for node in tgt_node, oth_node:
- info("checking %s on %s" % (dev.iv_name, node))
- cfg.SetDiskID(dev, node)
- if not rpc.call_blockdev_find(node, dev):
- raise errors.OpExecError("Can't find device %s on node %s" %
- (dev.iv_name, node))
+ self.lu.LogStep(1, steps_total, "Check device existence")
+ self._CheckDisksExistence([self.other_node, self.target_node])
+ self._CheckVolumeGroup([self.target_node, self.other_node])
# Step: check other node consistency
- self.proc.LogStep(2, steps_total, "check peer consistency")
- for dev in instance.disks:
- if not dev.iv_name in self.op.disks:
- continue
- info("checking %s consistency on %s" % (dev.iv_name, oth_node))
- if not _CheckDiskConsistency(self.cfg, dev, oth_node,
- oth_node==instance.primary_node):
- raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
- " to replace disks on this node (%s)" %
- (oth_node, tgt_node))
+ self.lu.LogStep(2, steps_total, "Check peer consistency")
+ self._CheckDisksConsistency(self.other_node,
+ self.other_node == self.instance.primary_node,
+ False)
# Step: create new storage
- self.proc.LogStep(3, steps_total, "allocate new storage")
- for dev in instance.disks:
- if not dev.iv_name in self.op.disks:
- continue
- size = dev.size
- cfg.SetDiskID(dev, tgt_node)
- lv_names = [".%s_%s" % (dev.iv_name, suf) for suf in ["data", "meta"]]
- names = _GenerateUniqueNames(cfg, lv_names)
- lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
- logical_id=(vgname, names[0]))
- lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
- logical_id=(vgname, names[1]))
- new_lvs = [lv_data, lv_meta]
- old_lvs = dev.children
- iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
- info("creating new local storage on %s for %s" %
- (tgt_node, dev.iv_name))
- # since we *always* want to create this LV, we use the
- # _Create...OnPrimary (which forces the creation), even if we
- # are talking about the secondary node
- for new_lv in new_lvs:
- if not _CreateBlockDevOnPrimary(cfg, tgt_node, instance, new_lv,
- _GetInstanceInfoText(instance)):
- raise errors.OpExecError("Failed to create new LV named '%s' on"
- " node '%s'" %
- (new_lv.logical_id[1], tgt_node))
+ self.lu.LogStep(3, steps_total, "Allocate new storage")
+ iv_names = self._CreateNewStorage(self.target_node)
# Step: for each lv, detach+rename*2+attach
- self.proc.LogStep(4, steps_total, "change drbd configuration")
+ self.lu.LogStep(4, steps_total, "Changing drbd configuration")
for dev, old_lvs, new_lvs in iv_names.itervalues():
- info("detaching %s drbd from local storage" % dev.iv_name)
- if not rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs):
- raise errors.OpExecError("Can't detach drbd from local storage on node"
- " %s for device %s" % (tgt_node, dev.iv_name))
+ self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
+
+ result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
+ old_lvs)
+ result.Raise("Can't detach drbd from local storage on node"
+ " %s for device %s" % (self.target_node, dev.iv_name))
#dev.children = []
#cfg.Update(instance)
temp_suffix = int(time.time())
ren_fn = lambda d, suff: (d.physical_id[0],
d.physical_id[1] + "_replaced-%s" % suff)
- # build the rename list based on what LVs exist on the node
- rlist = []
+
+ # Build the rename list based on what LVs exist on the node
+ rename_old_to_new = []
for to_ren in old_lvs:
- find_res = rpc.call_blockdev_find(tgt_node, to_ren)
- if find_res is not None: # device exists
- rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
-
- info("renaming the old LVs on the target node")
- if not rpc.call_blockdev_rename(tgt_node, rlist):
- raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
- # now we rename the new LVs to the old LVs
- info("renaming the new LVs on the target node")
- rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
- if not rpc.call_blockdev_rename(tgt_node, rlist):
- raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
+ result = self.rpc.call_blockdev_find(self.target_node, to_ren)
+ if not result.fail_msg and result.payload:
+ # device exists
+ rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
+
+ self.lu.LogInfo("Renaming the old LVs on the target node")
+ result = self.rpc.call_blockdev_rename(self.target_node,
+ rename_old_to_new)
+ result.Raise("Can't rename old LVs on node %s" % self.target_node)
+
+ # Now we rename the new LVs to the old LVs
+ self.lu.LogInfo("Renaming the new LVs on the target node")
+ rename_new_to_old = [(new, old.physical_id)
+ for old, new in zip(old_lvs, new_lvs)]
+ result = self.rpc.call_blockdev_rename(self.target_node,
+ rename_new_to_old)
+ result.Raise("Can't rename new LVs on node %s" % self.target_node)
for old, new in zip(old_lvs, new_lvs):
new.logical_id = old.logical_id
- cfg.SetDiskID(new, tgt_node)
+ self.cfg.SetDiskID(new, self.target_node)
for disk in old_lvs:
disk.logical_id = ren_fn(disk, temp_suffix)
- cfg.SetDiskID(disk, tgt_node)
-
- # now that the new lvs have the old name, we can add them to the device
- info("adding new mirror component on %s" % tgt_node)
- if not rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs):
+ self.cfg.SetDiskID(disk, self.target_node)
+
+ # Now that the new lvs have the old name, we can add them to the device
+ self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
+ result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
+ new_lvs)
+ msg = result.fail_msg
+ if msg:
for new_lv in new_lvs:
- if not rpc.call_blockdev_remove(tgt_node, new_lv):
- warning("Can't rollback device %s", hint="manually cleanup unused"
- " logical volumes")
- raise errors.OpExecError("Can't add local storage to drbd")
+ msg2 = self.rpc.call_blockdev_remove(self.target_node,
+ new_lv).fail_msg
+ if msg2:
+ self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
+ hint=("cleanup manually the unused logical"
+ "volumes"))
+ raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
dev.children = new_lvs
- cfg.Update(instance)
- # Step: wait for sync
+ self.cfg.Update(self.instance)
- # this can fail as the old devices are degraded and _WaitForSync
- # does a combined result over all disks, so we don't check its
- # return value
- self.proc.LogStep(5, steps_total, "sync devices")
- _WaitForSync(cfg, instance, self.proc, unlock=True)
+ # Wait for sync
+ # This can fail as the old devices are degraded and _WaitForSync
+ # does a combined result over all disks, so we don't check its return value
+ self.lu.LogStep(5, steps_total, "Sync devices")
+ _WaitForSync(self.lu, self.instance, unlock=True)
- # so check manually all the devices
- for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
- cfg.SetDiskID(dev, instance.primary_node)
- is_degr = rpc.call_blockdev_find(instance.primary_node, dev)[5]
- if is_degr:
- raise errors.OpExecError("DRBD device %s is degraded!" % name)
+ # Check all devices manually
+ self._CheckDevices(self.instance.primary_node, iv_names)
# Step: remove old storage
- self.proc.LogStep(6, steps_total, "removing old storage")
- for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
- info("remove logical volumes for %s" % name)
- for lv in old_lvs:
- cfg.SetDiskID(lv, tgt_node)
- if not rpc.call_blockdev_remove(tgt_node, lv):
- warning("Can't remove old LV", hint="manually remove unused LVs")
- continue
+ self.lu.LogStep(6, steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
- def _ExecD8Secondary(self, feedback_fn):
- """Replace the secondary node for drbd8.
+ def _ExecDrbd8Secondary(self):
+ """Replace the secondary node for DRBD 8.
The algorithm for replace is quite complicated:
- for all disks of the instance:
"""
steps_total = 6
- warning, info = (self.proc.LogWarning, self.proc.LogInfo)
- instance = self.instance
- iv_names = {}
- vgname = self.cfg.GetVGName()
- # start of work
- cfg = self.cfg
- old_node = self.tgt_node
- new_node = self.new_node
- pri_node = instance.primary_node
# Step: check device activation
- self.proc.LogStep(1, steps_total, "check device existence")
- info("checking volume groups")
- my_vg = cfg.GetVGName()
- results = rpc.call_vg_list([pri_node, new_node])
- if not results:
- raise errors.OpExecError("Can't list volume groups on the nodes")
- for node in pri_node, new_node:
- res = results.get(node, False)
- if not res or my_vg not in res:
- raise errors.OpExecError("Volume group '%s' not found on %s" %
- (my_vg, node))
- for dev in instance.disks:
- if not dev.iv_name in self.op.disks:
- continue
- info("checking %s on %s" % (dev.iv_name, pri_node))
- cfg.SetDiskID(dev, pri_node)
- if not rpc.call_blockdev_find(pri_node, dev):
- raise errors.OpExecError("Can't find device %s on node %s" %
- (dev.iv_name, pri_node))
+ self.lu.LogStep(1, steps_total, "Check device existence")
+ self._CheckDisksExistence([self.instance.primary_node])
+ self._CheckVolumeGroup([self.instance.primary_node])
# Step: check other node consistency
- self.proc.LogStep(2, steps_total, "check peer consistency")
- for dev in instance.disks:
- if not dev.iv_name in self.op.disks:
- continue
- info("checking %s consistency on %s" % (dev.iv_name, pri_node))
- if not _CheckDiskConsistency(self.cfg, dev, pri_node, True, ldisk=True):
- raise errors.OpExecError("Primary node (%s) has degraded storage,"
- " unsafe to replace the secondary" %
- pri_node)
+ self.lu.LogStep(2, steps_total, "Check peer consistency")
+ self._CheckDisksConsistency(self.instance.primary_node, True, True)
# Step: create new storage
- self.proc.LogStep(3, steps_total, "allocate new storage")
- for dev in instance.disks:
- size = dev.size
- info("adding new local storage on %s for %s" % (new_node, dev.iv_name))
- # since we *always* want to create this LV, we use the
- # _Create...OnPrimary (which forces the creation), even if we
- # are talking about the secondary node
+ self.lu.LogStep(3, steps_total, "Allocate new storage")
+ for idx, dev in enumerate(self.instance.disks):
+ self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
+ (self.new_node, idx))
+ # we pass force_create=True to force LVM creation
for new_lv in dev.children:
- if not _CreateBlockDevOnPrimary(cfg, new_node, instance, new_lv,
- _GetInstanceInfoText(instance)):
- raise errors.OpExecError("Failed to create new LV named '%s' on"
- " node '%s'" %
- (new_lv.logical_id[1], new_node))
-
- iv_names[dev.iv_name] = (dev, dev.children)
+ _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
+ _GetInstanceInfoText(self.instance), False)
+
+ # Step 4: dbrd minors and drbd setups changes
+ # after this, we must manually remove the drbd minors on both the
+ # error and the success paths
+ self.lu.LogStep(4, steps_total, "Changing drbd configuration")
+ minors = self.cfg.AllocateDRBDMinor([self.new_node
+ for dev in self.instance.disks],
+ self.instance.name)
+ logging.debug("Allocated minors %r" % (minors,))
- self.proc.LogStep(4, steps_total, "changing drbd configuration")
- for dev in instance.disks:
- size = dev.size
- info("activating a new drbd on %s for %s" % (new_node, dev.iv_name))
- # create new devices on new_node
- new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
- logical_id=(pri_node, new_node,
- dev.logical_id[2]),
- children=dev.children)
- if not _CreateBlockDevOnSecondary(cfg, new_node, instance,
- new_drbd, False,
- _GetInstanceInfoText(instance)):
- raise errors.OpExecError("Failed to create new DRBD on"
- " node '%s'" % new_node)
-
- for dev in instance.disks:
- # we have new devices, shutdown the drbd on the old secondary
- info("shutting down drbd for %s on old node" % dev.iv_name)
- cfg.SetDiskID(dev, old_node)
- if not rpc.call_blockdev_shutdown(old_node, dev):
- warning("Failed to shutdown drbd for %s on old node" % dev.iv_name,
- hint="Please cleanup this device manually as soon as possible")
-
- info("detaching primary drbds from the network (=> standalone)")
- done = 0
- for dev in instance.disks:
- cfg.SetDiskID(dev, pri_node)
- # set the physical (unique in bdev terms) id to None, meaning
- # detach from network
- dev.physical_id = (None,) * len(dev.physical_id)
- # and 'find' the device, which will 'fix' it to match the
- # standalone state
- if rpc.call_blockdev_find(pri_node, dev):
- done += 1
+ iv_names = {}
+ for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
+ self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
+ (self.new_node, idx))
+ # create new devices on new_node; note that we create two IDs:
+ # one without port, so the drbd will be activated without
+ # networking information on the new node at this stage, and one
+ # with network, for the latter activation in step 4
+ (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
+ if self.instance.primary_node == o_node1:
+ p_minor = o_minor1
else:
- warning("Failed to detach drbd %s from network, unusual case" %
- dev.iv_name)
+ p_minor = o_minor2
- if not done:
- # no detaches succeeded (very unlikely)
- raise errors.OpExecError("Can't detach at least one DRBD from old node")
+ new_alone_id = (self.instance.primary_node, self.new_node, None,
+ p_minor, new_minor, o_secret)
+ new_net_id = (self.instance.primary_node, self.new_node, o_port,
+ p_minor, new_minor, o_secret)
+
+ iv_names[idx] = (dev, dev.children, new_net_id)
+ logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
+ new_net_id)
+ new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
+ logical_id=new_alone_id,
+ children=dev.children,
+ size=dev.size)
+ try:
+ _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
+ _GetInstanceInfoText(self.instance), False)
+ except errors.GenericError:
+ self.cfg.ReleaseDRBDMinors(self.instance.name)
+ raise
+
+ # We have new devices, shutdown the drbd on the old secondary
+ for idx, dev in enumerate(self.instance.disks):
+ self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
+ self.cfg.SetDiskID(dev, self.target_node)
+ msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
+ if msg:
+ self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
+ "node: %s" % (idx, msg),
+ hint=("Please cleanup this device manually as"
+ " soon as possible"))
+
+ self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
+ result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
+ self.node_secondary_ip,
+ self.instance.disks)\
+ [self.instance.primary_node]
+
+ msg = result.fail_msg
+ if msg:
+ # detaches didn't succeed (unlikely)
+ self.cfg.ReleaseDRBDMinors(self.instance.name)
+ raise errors.OpExecError("Can't detach the disks from the network on"
+ " old node: %s" % (msg,))
# if we managed to detach at least one, we update all the disks of
# the instance to point to the new secondary
- info("updating instance configuration")
- for dev in instance.disks:
- dev.logical_id = (pri_node, new_node) + dev.logical_id[2:]
- cfg.SetDiskID(dev, pri_node)
- cfg.Update(instance)
+ self.lu.LogInfo("Updating instance configuration")
+ for dev, _, new_logical_id in iv_names.itervalues():
+ dev.logical_id = new_logical_id
+ self.cfg.SetDiskID(dev, self.instance.primary_node)
+
+ self.cfg.Update(self.instance)
# and now perform the drbd attach
- info("attaching primary drbds to new secondary (standalone => connected)")
- failures = []
- for dev in instance.disks:
- info("attaching primary drbd for %s to new secondary node" % dev.iv_name)
- # since the attach is smart, it's enough to 'find' the device,
- # it will automatically activate the network, if the physical_id
- # is correct
- cfg.SetDiskID(dev, pri_node)
- if not rpc.call_blockdev_find(pri_node, dev):
- warning("can't attach drbd %s to new secondary!" % dev.iv_name,
- "please do a gnt-instance info to see the status of disks")
-
- # this can fail as the old devices are degraded and _WaitForSync
- # does a combined result over all disks, so we don't check its
- # return value
- self.proc.LogStep(5, steps_total, "sync devices")
- _WaitForSync(cfg, instance, self.proc, unlock=True)
-
- # so check manually all the devices
- for name, (dev, old_lvs) in iv_names.iteritems():
- cfg.SetDiskID(dev, pri_node)
- is_degr = rpc.call_blockdev_find(pri_node, dev)[5]
- if is_degr:
- raise errors.OpExecError("DRBD device %s is degraded!" % name)
+ self.lu.LogInfo("Attaching primary drbds to new secondary"
+ " (standalone => connected)")
+ result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
+ self.new_node],
+ self.node_secondary_ip,
+ self.instance.disks,
+ self.instance.name,
+ False)
+ for to_node, to_result in result.items():
+ msg = to_result.fail_msg
+ if msg:
+ self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
+ to_node, msg,
+ hint=("please do a gnt-instance info to see the"
+ " status of disks"))
+
+ # Wait for sync
+ # This can fail as the old devices are degraded and _WaitForSync
+ # does a combined result over all disks, so we don't check its return value
+ self.lu.LogStep(5, steps_total, "Sync devices")
+ _WaitForSync(self.lu, self.instance, unlock=True)
+
+ # Check all devices manually
+ self._CheckDevices(self.instance.primary_node, iv_names)
- self.proc.LogStep(6, steps_total, "removing old storage")
- for name, (dev, old_lvs) in iv_names.iteritems():
- info("remove logical volumes for %s" % name)
- for lv in old_lvs:
- cfg.SetDiskID(lv, old_node)
- if not rpc.call_blockdev_remove(old_node, lv):
- warning("Can't remove LV on old secondary",
- hint="Cleanup stale volumes by hand")
+ # Step: remove old storage
+ self.lu.LogStep(6, steps_total, "Removing old storage")
+ self._RemoveOldStorage(self.target_node, iv_names)
- def Exec(self, feedback_fn):
- """Execute disk replacement.
- This dispatches the disk replacement to the appropriate handler.
+class LURepairNodeStorage(NoHooksLU):
+ """Repairs the volume group on a node.
- """
- instance = self.instance
+ """
+ _OP_REQP = ["node_name"]
+ REQ_BGL = False
- # Activate the instance disks if we're replacing them on a down instance
- if instance.status == "down":
- _StartInstanceDisks(self.cfg, instance, True)
+ def CheckArguments(self):
+ node_name = self.cfg.ExpandNodeName(self.op.node_name)
+ if node_name is None:
+ raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
- if instance.disk_template == constants.DT_DRBD8:
- if self.op.remote_node is None:
- fn = self._ExecD8DiskOnly
- else:
- fn = self._ExecD8Secondary
- else:
- raise errors.ProgrammerError("Unhandled disk replacement case")
+ self.op.node_name = node_name
+
+ def ExpandNames(self):
+ self.needed_locks = {
+ locking.LEVEL_NODE: [self.op.node_name],
+ }
+
+ def _CheckFaultyDisks(self, instance, node_name):
+ if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
+ node_name, True):
+ raise errors.OpPrereqError("Instance '%s' has faulty disks on"
+ " node '%s'" % (instance.name, node_name))
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ """
+ storage_type = self.op.storage_type
- ret = fn(feedback_fn)
+ if (constants.SO_FIX_CONSISTENCY not in
+ constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
+ raise errors.OpPrereqError("Storage units of type '%s' can not be"
+ " repaired" % storage_type)
- # Deactivate the instance disks if we're replacing them on a down instance
- if instance.status == "down":
- _SafeShutdownInstanceDisks(instance, self.cfg)
+ # Check whether any instance on this node has faulty disks
+ for inst in _GetNodeInstances(self.cfg, self.op.node_name):
+ check_nodes = set(inst.all_nodes)
+ check_nodes.discard(self.op.node_name)
+ for inst_node_name in check_nodes:
+ self._CheckFaultyDisks(inst, inst_node_name)
- return ret
+ def Exec(self, feedback_fn):
+ feedback_fn("Repairing storage unit '%s' on %s ..." %
+ (self.op.name, self.op.node_name))
+
+ st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
+ result = self.rpc.call_storage_execute(self.op.node_name,
+ self.op.storage_type, st_args,
+ self.op.name,
+ constants.SO_FIX_CONSISTENCY)
+ result.Raise("Failed to repair storage unit '%s' on %s" %
+ (self.op.name, self.op.node_name))
class LUGrowDisk(LogicalUnit):
"""
HPATH = "disk-grow"
HTYPE = constants.HTYPE_INSTANCE
- _OP_REQP = ["instance_name", "disk", "amount"]
+ _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
REQ_BGL = False
def ExpandNames(self):
"DISK": self.op.disk,
"AMOUNT": self.op.amount,
}
- env.update(_BuildInstanceHookEnvByObject(self.instance))
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
nl = [
- self.sstore.GetMasterNode(),
+ self.cfg.GetMasterNode(),
self.instance.primary_node,
]
return env, nl, nl
instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ nodenames = list(instance.all_nodes)
+ for node in nodenames:
+ _CheckNodeOnline(self, node)
+
self.instance = instance
raise errors.OpPrereqError("Instance's disk layout does not support"
" growing.")
- if instance.FindDisk(self.op.disk) is None:
- raise errors.OpPrereqError("Disk '%s' not found for instance '%s'" %
- (self.op.disk, instance.name))
+ self.disk = instance.FindDisk(self.op.disk)
- nodenames = [instance.primary_node] + list(instance.secondary_nodes)
- nodeinfo = rpc.call_node_info(nodenames, self.cfg.GetVGName())
+ nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
+ instance.hypervisor)
for node in nodenames:
- info = nodeinfo.get(node, None)
- if not info:
- raise errors.OpPrereqError("Cannot get current information"
- " from node '%s'" % node)
- vg_free = info.get('vg_free', None)
+ info = nodeinfo[node]
+ info.Raise("Cannot get current information from node %s" % node)
+ vg_free = info.payload.get('vg_free', None)
if not isinstance(vg_free, int):
raise errors.OpPrereqError("Can't compute free disk space on"
" node %s" % node)
- if self.op.amount > info['vg_free']:
+ if self.op.amount > vg_free:
raise errors.OpPrereqError("Not enough disk space on target node %s:"
" %d MiB available, %d MiB required" %
- (node, info['vg_free'], self.op.amount))
+ (node, vg_free, self.op.amount))
def Exec(self, feedback_fn):
"""Execute disk grow.
"""
instance = self.instance
- disk = instance.FindDisk(self.op.disk)
- for node in (instance.secondary_nodes + (instance.primary_node,)):
+ disk = self.disk
+ for node in instance.all_nodes:
self.cfg.SetDiskID(disk, node)
- result = rpc.call_blockdev_grow(node, disk, self.op.amount)
- if not result or not isinstance(result, (list, tuple)) or len(result) != 2:
- raise errors.OpExecError("grow request failed to node %s" % node)
- elif not result[0]:
- raise errors.OpExecError("grow request failed to node %s: %s" %
- (node, result[1]))
+ result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
+ result.Raise("Grow request failed to node %s" % node)
disk.RecordGrow(self.op.amount)
self.cfg.Update(instance)
- return
+ if self.op.wait_for_sync:
+ disk_abort = not _WaitForSync(self, instance)
+ if disk_abort:
+ self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
+ " status.\nPlease check the instance.")
class LUQueryInstanceData(NoHooksLU):
"""Query runtime instance data.
"""
- _OP_REQP = ["instances"]
+ _OP_REQP = ["instances", "static"]
REQ_BGL = False
+
def ExpandNames(self):
self.needed_locks = {}
- self.share_locks = dict(((i, 1) for i in locking.LEVELS))
+ self.share_locks = dict.fromkeys(locking.LEVELS, 1)
if not isinstance(self.op.instances, list):
raise errors.OpPrereqError("Invalid argument type 'instances'")
for name in self.op.instances:
full_name = self.cfg.ExpandInstanceName(name)
if full_name is None:
- raise errors.OpPrereqError("Instance '%s' not known" %
- self.op.instance_name)
+ raise errors.OpPrereqError("Instance '%s' not known" % name)
self.wanted_names.append(full_name)
self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
else:
in self.wanted_names]
return
+ def _ComputeBlockdevStatus(self, node, instance_name, dev):
+ """Returns the status of a block device
+
+ """
+ if self.op.static or not node:
+ return None
+
+ self.cfg.SetDiskID(dev, node)
+
+ result = self.rpc.call_blockdev_find(node, dev)
+ if result.offline:
+ return None
+
+ result.Raise("Can't compute disk status for %s" % instance_name)
+
+ status = result.payload
+ if status is None:
+ return None
+
+ return (status.dev_path, status.major, status.minor,
+ status.sync_percent, status.estimated_time,
+ status.is_degraded, status.ldisk_status)
+
def _ComputeDiskStatus(self, instance, snode, dev):
"""Compute block device status.
"""
- self.cfg.SetDiskID(dev, instance.primary_node)
- dev_pstatus = rpc.call_blockdev_find(instance.primary_node, dev)
if dev.dev_type in constants.LDS_DRBD:
# we change the snode then (otherwise we use the one passed in)
if dev.logical_id[0] == instance.primary_node:
else:
snode = dev.logical_id[0]
- if snode:
- self.cfg.SetDiskID(dev, snode)
- dev_sstatus = rpc.call_blockdev_find(snode, dev)
- else:
- dev_sstatus = None
+ dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
+ instance.name, dev)
+ dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
if dev.children:
dev_children = [self._ComputeDiskStatus(instance, snode, child)
"pstatus": dev_pstatus,
"sstatus": dev_sstatus,
"children": dev_children,
+ "mode": dev.mode,
+ "size": dev.size,
}
return data
def Exec(self, feedback_fn):
"""Gather and return data"""
result = {}
+
+ cluster = self.cfg.GetClusterInfo()
+
for instance in self.wanted_instances:
- remote_info = rpc.call_instance_info(instance.primary_node,
- instance.name)
- if remote_info and "state" in remote_info:
- remote_state = "up"
- else:
- remote_state = "down"
- if instance.status == "down":
- config_state = "down"
+ if not self.op.static:
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise("Error checking node %s" % instance.primary_node)
+ remote_info = remote_info.payload
+ if remote_info and "state" in remote_info:
+ remote_state = "up"
+ else:
+ remote_state = "down"
else:
+ remote_state = None
+ if instance.admin_up:
config_state = "up"
+ else:
+ config_state = "down"
disks = [self._ComputeDiskStatus(instance, None, device)
for device in instance.disks]
"pnode": instance.primary_node,
"snodes": instance.secondary_nodes,
"os": instance.os,
- "memory": instance.memory,
- "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
+ # this happens to be the same format used for hooks
+ "nics": _NICListToTuple(self, instance.nics),
"disks": disks,
- "vcpus": instance.vcpus,
+ "hypervisor": instance.hypervisor,
+ "network_port": instance.network_port,
+ "hv_instance": instance.hvparams,
+ "hv_actual": cluster.FillHV(instance),
+ "be_instance": instance.beparams,
+ "be_actual": cluster.FillBE(instance),
+ "serial_no": instance.serial_no,
+ "mtime": instance.mtime,
+ "ctime": instance.ctime,
+ "uuid": instance.uuid,
}
- htkind = self.sstore.GetHypervisorType()
- if htkind == constants.HT_XEN_PVM30:
- idict["kernel_path"] = instance.kernel_path
- idict["initrd_path"] = instance.initrd_path
-
- if htkind == constants.HT_XEN_HVM31:
- idict["hvm_boot_order"] = instance.hvm_boot_order
- idict["hvm_acpi"] = instance.hvm_acpi
- idict["hvm_pae"] = instance.hvm_pae
- idict["hvm_cdrom_image_path"] = instance.hvm_cdrom_image_path
- idict["hvm_nic_type"] = instance.hvm_nic_type
- idict["hvm_disk_type"] = instance.hvm_disk_type
-
- if htkind in constants.HTS_REQ_PORT:
- if instance.vnc_bind_address is None:
- vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
- else:
- vnc_bind_address = instance.vnc_bind_address
- if instance.network_port is None:
- vnc_console_port = None
- elif vnc_bind_address == constants.BIND_ADDRESS_GLOBAL:
- vnc_console_port = "%s:%s" % (instance.primary_node,
- instance.network_port)
- elif vnc_bind_address == constants.LOCALHOST_IP_ADDRESS:
- vnc_console_port = "%s:%s on node %s" % (vnc_bind_address,
- instance.network_port,
- instance.primary_node)
- else:
- vnc_console_port = "%s:%s" % (instance.vnc_bind_address,
- instance.network_port)
- idict["vnc_console_port"] = vnc_console_port
- idict["vnc_bind_address"] = vnc_bind_address
- idict["network_port"] = instance.network_port
-
result[instance.name] = idict
return result
_OP_REQP = ["instance_name"]
REQ_BGL = False
+ def CheckArguments(self):
+ if not hasattr(self.op, 'nics'):
+ self.op.nics = []
+ if not hasattr(self.op, 'disks'):
+ self.op.disks = []
+ if not hasattr(self.op, 'beparams'):
+ self.op.beparams = {}
+ if not hasattr(self.op, 'hvparams'):
+ self.op.hvparams = {}
+ self.op.force = getattr(self.op, "force", False)
+ if not (self.op.nics or self.op.disks or
+ self.op.hvparams or self.op.beparams):
+ raise errors.OpPrereqError("No changes submitted")
+
+ # Disk validation
+ disk_addremove = 0
+ for disk_op, disk_dict in self.op.disks:
+ if disk_op == constants.DDM_REMOVE:
+ disk_addremove += 1
+ continue
+ elif disk_op == constants.DDM_ADD:
+ disk_addremove += 1
+ else:
+ if not isinstance(disk_op, int):
+ raise errors.OpPrereqError("Invalid disk index")
+ if not isinstance(disk_dict, dict):
+ msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
+ raise errors.OpPrereqError(msg)
+
+ if disk_op == constants.DDM_ADD:
+ mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
+ if mode not in constants.DISK_ACCESS_SET:
+ raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
+ size = disk_dict.get('size', None)
+ if size is None:
+ raise errors.OpPrereqError("Required disk parameter size missing")
+ try:
+ size = int(size)
+ except ValueError, err:
+ raise errors.OpPrereqError("Invalid disk size parameter: %s" %
+ str(err))
+ disk_dict['size'] = size
+ else:
+ # modification of disk
+ if 'size' in disk_dict:
+ raise errors.OpPrereqError("Disk size change not possible, use"
+ " grow-disk")
+
+ if disk_addremove > 1:
+ raise errors.OpPrereqError("Only one disk add or remove operation"
+ " supported at a time")
+
+ # NIC validation
+ nic_addremove = 0
+ for nic_op, nic_dict in self.op.nics:
+ if nic_op == constants.DDM_REMOVE:
+ nic_addremove += 1
+ continue
+ elif nic_op == constants.DDM_ADD:
+ nic_addremove += 1
+ else:
+ if not isinstance(nic_op, int):
+ raise errors.OpPrereqError("Invalid nic index")
+ if not isinstance(nic_dict, dict):
+ msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
+ raise errors.OpPrereqError(msg)
+
+ # nic_dict should be a dict
+ nic_ip = nic_dict.get('ip', None)
+ if nic_ip is not None:
+ if nic_ip.lower() == constants.VALUE_NONE:
+ nic_dict['ip'] = None
+ else:
+ if not utils.IsValidIP(nic_ip):
+ raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
+
+ nic_bridge = nic_dict.get('bridge', None)
+ nic_link = nic_dict.get('link', None)
+ if nic_bridge and nic_link:
+ raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
+ " at the same time")
+ elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
+ nic_dict['bridge'] = None
+ elif nic_link and nic_link.lower() == constants.VALUE_NONE:
+ nic_dict['link'] = None
+
+ if nic_op == constants.DDM_ADD:
+ nic_mac = nic_dict.get('mac', None)
+ if nic_mac is None:
+ nic_dict['mac'] = constants.VALUE_AUTO
+
+ if 'mac' in nic_dict:
+ nic_mac = nic_dict['mac']
+ if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ if not utils.IsValidMac(nic_mac):
+ raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
+ if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
+ raise errors.OpPrereqError("'auto' is not a valid MAC address when"
+ " modifying an existing nic")
+
+ if nic_addremove > 1:
+ raise errors.OpPrereqError("Only one NIC add or remove operation"
+ " supported at a time")
+
def ExpandNames(self):
self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODE] = []
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ self._LockInstancesNodes()
def BuildHooksEnv(self):
"""Build hooks env.
"""
args = dict()
- if self.mem:
- args['memory'] = self.mem
- if self.vcpus:
- args['vcpus'] = self.vcpus
- if self.do_ip or self.do_bridge or self.mac:
- if self.do_ip:
- ip = self.ip
- else:
- ip = self.instance.nics[0].ip
- if self.bridge:
- bridge = self.bridge
- else:
- bridge = self.instance.nics[0].bridge
- if self.mac:
- mac = self.mac
- else:
- mac = self.instance.nics[0].mac
- args['nics'] = [(ip, bridge, mac)]
- env = _BuildInstanceHookEnvByObject(self.instance, override=args)
- nl = [self.sstore.GetMasterNode(),
- self.instance.primary_node] + list(self.instance.secondary_nodes)
+ if constants.BE_MEMORY in self.be_new:
+ args['memory'] = self.be_new[constants.BE_MEMORY]
+ if constants.BE_VCPUS in self.be_new:
+ args['vcpus'] = self.be_new[constants.BE_VCPUS]
+ # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
+ # information at all.
+ if self.op.nics:
+ args['nics'] = []
+ nic_override = dict(self.op.nics)
+ c_nicparams = self.cluster.nicparams[constants.PP_DEFAULT]
+ for idx, nic in enumerate(self.instance.nics):
+ if idx in nic_override:
+ this_nic_override = nic_override[idx]
+ else:
+ this_nic_override = {}
+ if 'ip' in this_nic_override:
+ ip = this_nic_override['ip']
+ else:
+ ip = nic.ip
+ if 'mac' in this_nic_override:
+ mac = this_nic_override['mac']
+ else:
+ mac = nic.mac
+ if idx in self.nic_pnew:
+ nicparams = self.nic_pnew[idx]
+ else:
+ nicparams = objects.FillDict(c_nicparams, nic.nicparams)
+ mode = nicparams[constants.NIC_MODE]
+ link = nicparams[constants.NIC_LINK]
+ args['nics'].append((ip, mac, mode, link))
+ if constants.DDM_ADD in nic_override:
+ ip = nic_override[constants.DDM_ADD].get('ip', None)
+ mac = nic_override[constants.DDM_ADD]['mac']
+ nicparams = self.nic_pnew[constants.DDM_ADD]
+ mode = nicparams[constants.NIC_MODE]
+ link = nicparams[constants.NIC_LINK]
+ args['nics'].append((ip, mac, mode, link))
+ elif constants.DDM_REMOVE in nic_override:
+ del args['nics'][-1]
+
+ env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
+ nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
return env, nl, nl
+ def _GetUpdatedParams(self, old_params, update_dict,
+ default_values, parameter_types):
+ """Return the new params dict for the given params.
+
+ @type old_params: dict
+ @param old_params: old parameters
+ @type update_dict: dict
+ @param update_dict: dict containing new parameter values,
+ or constants.VALUE_DEFAULT to reset the
+ parameter to its default value
+ @type default_values: dict
+ @param default_values: default values for the filled parameters
+ @type parameter_types: dict
+ @param parameter_types: dict mapping target dict keys to types
+ in constants.ENFORCEABLE_TYPES
+ @rtype: (dict, dict)
+ @return: (new_parameters, filled_parameters)
+
+ """
+ params_copy = copy.deepcopy(old_params)
+ for key, val in update_dict.iteritems():
+ if val == constants.VALUE_DEFAULT:
+ try:
+ del params_copy[key]
+ except KeyError:
+ pass
+ else:
+ params_copy[key] = val
+ utils.ForceDictType(params_copy, parameter_types)
+ params_filled = objects.FillDict(default_values, params_copy)
+ return (params_copy, params_filled)
+
def CheckPrereq(self):
"""Check prerequisites.
This only checks the instance list against the existing names.
"""
- # FIXME: all the parameters could be checked before, in ExpandNames, or in
- # a separate CheckArguments function, if we implement one, so the operation
- # can be aborted without waiting for any lock, should it have an error...
- self.mem = getattr(self.op, "mem", None)
- self.vcpus = getattr(self.op, "vcpus", None)
- self.ip = getattr(self.op, "ip", None)
- self.mac = getattr(self.op, "mac", None)
- self.bridge = getattr(self.op, "bridge", None)
- self.kernel_path = getattr(self.op, "kernel_path", None)
- self.initrd_path = getattr(self.op, "initrd_path", None)
- self.hvm_boot_order = getattr(self.op, "hvm_boot_order", None)
- self.hvm_acpi = getattr(self.op, "hvm_acpi", None)
- self.hvm_pae = getattr(self.op, "hvm_pae", None)
- self.hvm_nic_type = getattr(self.op, "hvm_nic_type", None)
- self.hvm_disk_type = getattr(self.op, "hvm_disk_type", None)
- self.hvm_cdrom_image_path = getattr(self.op, "hvm_cdrom_image_path", None)
- self.vnc_bind_address = getattr(self.op, "vnc_bind_address", None)
- self.force = getattr(self.op, "force", None)
- all_parms = [self.mem, self.vcpus, self.ip, self.bridge, self.mac,
- self.kernel_path, self.initrd_path, self.hvm_boot_order,
- self.hvm_acpi, self.hvm_pae, self.hvm_cdrom_image_path,
- self.vnc_bind_address, self.hvm_nic_type, self.hvm_disk_type]
- if all_parms.count(None) == len(all_parms):
- raise errors.OpPrereqError("No changes submitted")
- if self.mem is not None:
- try:
- self.mem = int(self.mem)
- except ValueError, err:
- raise errors.OpPrereqError("Invalid memory size: %s" % str(err))
- if self.vcpus is not None:
- try:
- self.vcpus = int(self.vcpus)
- except ValueError, err:
- raise errors.OpPrereqError("Invalid vcpus number: %s" % str(err))
- if self.ip is not None:
- self.do_ip = True
- if self.ip.lower() == "none":
- self.ip = None
- else:
- if not utils.IsValidIP(self.ip):
- raise errors.OpPrereqError("Invalid IP address '%s'." % self.ip)
- else:
- self.do_ip = False
- self.do_bridge = (self.bridge is not None)
- if self.mac is not None:
- if self.cfg.IsMacInUse(self.mac):
- raise errors.OpPrereqError('MAC address %s already in use in cluster' %
- self.mac)
- if not utils.IsValidMac(self.mac):
- raise errors.OpPrereqError('Invalid MAC address %s' % self.mac)
-
- if self.kernel_path is not None:
- self.do_kernel_path = True
- if self.kernel_path == constants.VALUE_NONE:
- raise errors.OpPrereqError("Can't set instance to no kernel")
-
- if self.kernel_path != constants.VALUE_DEFAULT:
- if not os.path.isabs(self.kernel_path):
- raise errors.OpPrereqError("The kernel path must be an absolute"
- " filename")
- else:
- self.do_kernel_path = False
-
- if self.initrd_path is not None:
- self.do_initrd_path = True
- if self.initrd_path not in (constants.VALUE_NONE,
- constants.VALUE_DEFAULT):
- if not os.path.isabs(self.initrd_path):
- raise errors.OpPrereqError("The initrd path must be an absolute"
- " filename")
- else:
- self.do_initrd_path = False
-
- # boot order verification
- if self.hvm_boot_order is not None:
- if self.hvm_boot_order != constants.VALUE_DEFAULT:
- if len(self.hvm_boot_order.strip("acdn")) != 0:
- raise errors.OpPrereqError("invalid boot order specified,"
- " must be one or more of [acdn]"
- " or 'default'")
-
- # hvm_cdrom_image_path verification
- if self.op.hvm_cdrom_image_path is not None:
- if not (os.path.isabs(self.op.hvm_cdrom_image_path) or
- self.op.hvm_cdrom_image_path.lower() == "none"):
- raise errors.OpPrereqError("The path to the HVM CDROM image must"
- " be an absolute path or None, not %s" %
- self.op.hvm_cdrom_image_path)
- if not (os.path.isfile(self.op.hvm_cdrom_image_path) or
- self.op.hvm_cdrom_image_path.lower() == "none"):
- raise errors.OpPrereqError("The HVM CDROM image must either be a"
- " regular file or a symlink pointing to"
- " an existing regular file, not %s" %
- self.op.hvm_cdrom_image_path)
-
- # vnc_bind_address verification
- if self.op.vnc_bind_address is not None:
- if not utils.IsValidIP(self.op.vnc_bind_address):
- raise errors.OpPrereqError("given VNC bind address '%s' doesn't look"
- " like a valid IP address" %
- self.op.vnc_bind_address)
+ self.force = self.op.force
+
+ # checking the new params on the primary/secondary nodes
instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+ cluster = self.cluster = self.cfg.GetClusterInfo()
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ pnode = instance.primary_node
+ nodelist = list(instance.all_nodes)
+
+ # hvparams processing
+ if self.op.hvparams:
+ i_hvdict, hv_new = self._GetUpdatedParams(
+ instance.hvparams, self.op.hvparams,
+ cluster.hvparams[instance.hypervisor],
+ constants.HVS_PARAMETER_TYPES)
+ # local check
+ hypervisor.GetHypervisor(
+ instance.hypervisor).CheckParameterSyntax(hv_new)
+ _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
+ self.hv_new = hv_new # the new actual values
+ self.hv_inst = i_hvdict # the new dict (without defaults)
+ else:
+ self.hv_new = self.hv_inst = {}
+
+ # beparams processing
+ if self.op.beparams:
+ i_bedict, be_new = self._GetUpdatedParams(
+ instance.beparams, self.op.beparams,
+ cluster.beparams[constants.PP_DEFAULT],
+ constants.BES_PARAMETER_TYPES)
+ self.be_new = be_new # the new actual values
+ self.be_inst = i_bedict # the new dict (without defaults)
+ else:
+ self.be_new = self.be_inst = {}
+
self.warn = []
- if self.mem is not None and not self.force:
- pnode = self.instance.primary_node
- nodelist = [pnode]
- nodelist.extend(instance.secondary_nodes)
- instance_info = rpc.call_instance_info(pnode, instance.name)
- nodeinfo = rpc.call_node_info(nodelist, self.cfg.GetVGName())
-
- if pnode not in nodeinfo or not isinstance(nodeinfo[pnode], dict):
+
+ if constants.BE_MEMORY in self.op.beparams and not self.force:
+ mem_check_list = [pnode]
+ if be_new[constants.BE_AUTO_BALANCE]:
+ # either we changed auto_balance to yes or it was from before
+ mem_check_list.extend(instance.secondary_nodes)
+ instance_info = self.rpc.call_instance_info(pnode, instance.name,
+ instance.hypervisor)
+ nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
+ instance.hypervisor)
+ pninfo = nodeinfo[pnode]
+ msg = pninfo.fail_msg
+ if msg:
# Assume the primary node is unreachable and go ahead
- self.warn.append("Can't get info from primary node %s" % pnode)
+ self.warn.append("Can't get info from primary node %s: %s" %
+ (pnode, msg))
+ elif not isinstance(pninfo.payload.get('memory_free', None), int):
+ self.warn.append("Node data from primary node %s doesn't contain"
+ " free memory information" % pnode)
+ elif instance_info.fail_msg:
+ self.warn.append("Can't get instance runtime information: %s" %
+ instance_info.fail_msg)
else:
- if instance_info:
- current_mem = instance_info['memory']
+ if instance_info.payload:
+ current_mem = int(instance_info.payload['memory'])
else:
# Assume instance not running
# (there is a slight race condition here, but it's not very probable,
# and we have no other way to check)
current_mem = 0
- miss_mem = self.mem - current_mem - nodeinfo[pnode]['memory_free']
+ miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
+ pninfo.payload['memory_free'])
if miss_mem > 0:
raise errors.OpPrereqError("This change will prevent the instance"
" from starting, due to %d MB of memory"
" missing on its primary node" % miss_mem)
- for node in instance.secondary_nodes:
- if node not in nodeinfo or not isinstance(nodeinfo[node], dict):
- self.warn.append("Can't get info from secondary node %s" % node)
- elif self.mem > nodeinfo[node]['memory_free']:
- self.warn.append("Not enough memory to failover instance to secondary"
- " node %s" % node)
-
- # Xen HVM device type checks
- if self.sstore.GetHypervisorType() == constants.HT_XEN_HVM31:
- if self.op.hvm_nic_type is not None:
- if self.op.hvm_nic_type not in constants.HT_HVM_VALID_NIC_TYPES:
- raise errors.OpPrereqError("Invalid NIC type %s specified for Xen"
- " HVM hypervisor" % self.op.hvm_nic_type)
- if self.op.hvm_disk_type is not None:
- if self.op.hvm_disk_type not in constants.HT_HVM_VALID_DISK_TYPES:
- raise errors.OpPrereqError("Invalid disk type %s specified for Xen"
- " HVM hypervisor" % self.op.hvm_disk_type)
+ if be_new[constants.BE_AUTO_BALANCE]:
+ for node, nres in nodeinfo.items():
+ if node not in instance.secondary_nodes:
+ continue
+ msg = nres.fail_msg
+ if msg:
+ self.warn.append("Can't get info from secondary node %s: %s" %
+ (node, msg))
+ elif not isinstance(nres.payload.get('memory_free', None), int):
+ self.warn.append("Secondary node %s didn't return free"
+ " memory information" % node)
+ elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
+ self.warn.append("Not enough memory to failover instance to"
+ " secondary node %s" % node)
+
+ # NIC processing
+ self.nic_pnew = {}
+ self.nic_pinst = {}
+ for nic_op, nic_dict in self.op.nics:
+ if nic_op == constants.DDM_REMOVE:
+ if not instance.nics:
+ raise errors.OpPrereqError("Instance has no NICs, cannot remove")
+ continue
+ if nic_op != constants.DDM_ADD:
+ # an existing nic
+ if nic_op < 0 or nic_op >= len(instance.nics):
+ raise errors.OpPrereqError("Invalid NIC index %s, valid values"
+ " are 0 to %d" %
+ (nic_op, len(instance.nics)))
+ old_nic_params = instance.nics[nic_op].nicparams
+ old_nic_ip = instance.nics[nic_op].ip
+ else:
+ old_nic_params = {}
+ old_nic_ip = None
+
+ update_params_dict = dict([(key, nic_dict[key])
+ for key in constants.NICS_PARAMETERS
+ if key in nic_dict])
+
+ if 'bridge' in nic_dict:
+ update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
+
+ new_nic_params, new_filled_nic_params = \
+ self._GetUpdatedParams(old_nic_params, update_params_dict,
+ cluster.nicparams[constants.PP_DEFAULT],
+ constants.NICS_PARAMETER_TYPES)
+ objects.NIC.CheckParameterSyntax(new_filled_nic_params)
+ self.nic_pinst[nic_op] = new_nic_params
+ self.nic_pnew[nic_op] = new_filled_nic_params
+ new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
+
+ if new_nic_mode == constants.NIC_MODE_BRIDGED:
+ nic_bridge = new_filled_nic_params[constants.NIC_LINK]
+ msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
+ if msg:
+ msg = "Error checking bridges on node %s: %s" % (pnode, msg)
+ if self.force:
+ self.warn.append(msg)
+ else:
+ raise errors.OpPrereqError(msg)
+ if new_nic_mode == constants.NIC_MODE_ROUTED:
+ if 'ip' in nic_dict:
+ nic_ip = nic_dict['ip']
+ else:
+ nic_ip = old_nic_ip
+ if nic_ip is None:
+ raise errors.OpPrereqError('Cannot set the nic ip to None'
+ ' on a routed nic')
+ if 'mac' in nic_dict:
+ nic_mac = nic_dict['mac']
+ if nic_mac is None:
+ raise errors.OpPrereqError('Cannot set the nic mac to None')
+ elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
+ # otherwise generate the mac
+ nic_dict['mac'] = self.cfg.GenerateMAC()
+ else:
+ # or validate/reserve the current one
+ if self.cfg.IsMacInUse(nic_mac):
+ raise errors.OpPrereqError("MAC address %s already in use"
+ " in cluster" % nic_mac)
+
+ # DISK processing
+ if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
+ raise errors.OpPrereqError("Disk operations not supported for"
+ " diskless instances")
+ for disk_op, disk_dict in self.op.disks:
+ if disk_op == constants.DDM_REMOVE:
+ if len(instance.disks) == 1:
+ raise errors.OpPrereqError("Cannot remove the last disk of"
+ " an instance")
+ ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
+ ins_l = ins_l[pnode]
+ msg = ins_l.fail_msg
+ if msg:
+ raise errors.OpPrereqError("Can't contact node %s: %s" %
+ (pnode, msg))
+ if instance.name in ins_l.payload:
+ raise errors.OpPrereqError("Instance is running, can't remove"
+ " disks.")
+
+ if (disk_op == constants.DDM_ADD and
+ len(instance.nics) >= constants.MAX_DISKS):
+ raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
+ " add more" % constants.MAX_DISKS)
+ if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
+ # an existing disk
+ if disk_op < 0 or disk_op >= len(instance.disks):
+ raise errors.OpPrereqError("Invalid disk index %s, valid values"
+ " are 0 to %d" %
+ (disk_op, len(instance.disks)))
return
"""Modifies an instance.
All parameters take effect only at the next restart of the instance.
+
"""
# Process here the warnings from CheckPrereq, as we don't have a
# feedback_fn there.
result = []
instance = self.instance
- if self.mem:
- instance.memory = self.mem
- result.append(("mem", self.mem))
- if self.vcpus:
- instance.vcpus = self.vcpus
- result.append(("vcpus", self.vcpus))
- if self.do_ip:
- instance.nics[0].ip = self.ip
- result.append(("ip", self.ip))
- if self.bridge:
- instance.nics[0].bridge = self.bridge
- result.append(("bridge", self.bridge))
- if self.mac:
- instance.nics[0].mac = self.mac
- result.append(("mac", self.mac))
- if self.do_kernel_path:
- instance.kernel_path = self.kernel_path
- result.append(("kernel_path", self.kernel_path))
- if self.do_initrd_path:
- instance.initrd_path = self.initrd_path
- result.append(("initrd_path", self.initrd_path))
- if self.hvm_boot_order:
- if self.hvm_boot_order == constants.VALUE_DEFAULT:
- instance.hvm_boot_order = None
+ cluster = self.cluster
+ # disk changes
+ for disk_op, disk_dict in self.op.disks:
+ if disk_op == constants.DDM_REMOVE:
+ # remove the last disk
+ device = instance.disks.pop()
+ device_idx = len(instance.disks)
+ for node, disk in device.ComputeNodeTree(instance.primary_node):
+ self.cfg.SetDiskID(disk, node)
+ msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
+ if msg:
+ self.LogWarning("Could not remove disk/%d on node %s: %s,"
+ " continuing anyway", device_idx, node, msg)
+ result.append(("disk/%d" % device_idx, "remove"))
+ elif disk_op == constants.DDM_ADD:
+ # add a new disk
+ if instance.disk_template == constants.DT_FILE:
+ file_driver, file_path = instance.disks[0].logical_id
+ file_path = os.path.dirname(file_path)
+ else:
+ file_driver = file_path = None
+ disk_idx_base = len(instance.disks)
+ new_disk = _GenerateDiskTemplate(self,
+ instance.disk_template,
+ instance.name, instance.primary_node,
+ instance.secondary_nodes,
+ [disk_dict],
+ file_path,
+ file_driver,
+ disk_idx_base)[0]
+ instance.disks.append(new_disk)
+ info = _GetInstanceInfoText(instance)
+
+ logging.info("Creating volume %s for instance %s",
+ new_disk.iv_name, instance.name)
+ # Note: this needs to be kept in sync with _CreateDisks
+ #HARDCODE
+ for node in instance.all_nodes:
+ f_create = node == instance.primary_node
+ try:
+ _CreateBlockDev(self, node, instance, new_disk,
+ f_create, info, f_create)
+ except errors.OpExecError, err:
+ self.LogWarning("Failed to create volume %s (%s) on"
+ " node %s: %s",
+ new_disk.iv_name, new_disk, node, err)
+ result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
+ (new_disk.size, new_disk.mode)))
else:
- instance.hvm_boot_order = self.hvm_boot_order
- result.append(("hvm_boot_order", self.hvm_boot_order))
- if self.hvm_acpi is not None:
- instance.hvm_acpi = self.hvm_acpi
- result.append(("hvm_acpi", self.hvm_acpi))
- if self.hvm_pae is not None:
- instance.hvm_pae = self.hvm_pae
- result.append(("hvm_pae", self.hvm_pae))
- if self.hvm_nic_type is not None:
- instance.hvm_nic_type = self.hvm_nic_type
- result.append(("hvm_nic_type", self.hvm_nic_type))
- if self.hvm_disk_type is not None:
- instance.hvm_disk_type = self.hvm_disk_type
- result.append(("hvm_disk_type", self.hvm_disk_type))
- if self.hvm_cdrom_image_path:
- if self.hvm_cdrom_image_path == constants.VALUE_NONE:
- instance.hvm_cdrom_image_path = None
+ # change a given disk
+ instance.disks[disk_op].mode = disk_dict['mode']
+ result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
+ # NIC changes
+ for nic_op, nic_dict in self.op.nics:
+ if nic_op == constants.DDM_REMOVE:
+ # remove the last nic
+ del instance.nics[-1]
+ result.append(("nic.%d" % len(instance.nics), "remove"))
+ elif nic_op == constants.DDM_ADD:
+ # mac and bridge should be set, by now
+ mac = nic_dict['mac']
+ ip = nic_dict.get('ip', None)
+ nicparams = self.nic_pinst[constants.DDM_ADD]
+ new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
+ instance.nics.append(new_nic)
+ result.append(("nic.%d" % (len(instance.nics) - 1),
+ "add:mac=%s,ip=%s,mode=%s,link=%s" %
+ (new_nic.mac, new_nic.ip,
+ self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
+ self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
+ )))
else:
- instance.hvm_cdrom_image_path = self.hvm_cdrom_image_path
- result.append(("hvm_cdrom_image_path", self.hvm_cdrom_image_path))
- if self.vnc_bind_address:
- instance.vnc_bind_address = self.vnc_bind_address
- result.append(("vnc_bind_address", self.vnc_bind_address))
+ for key in 'mac', 'ip':
+ if key in nic_dict:
+ setattr(instance.nics[nic_op], key, nic_dict[key])
+ if nic_op in self.nic_pnew:
+ instance.nics[nic_op].nicparams = self.nic_pnew[nic_op]
+ for key, val in nic_dict.iteritems():
+ result.append(("nic.%s/%d" % (key, nic_op), val))
+
+ # hvparams changes
+ if self.op.hvparams:
+ instance.hvparams = self.hv_inst
+ for key, val in self.op.hvparams.iteritems():
+ result.append(("hv/%s" % key, val))
+
+ # beparams changes
+ if self.op.beparams:
+ instance.beparams = self.be_inst
+ for key, val in self.op.beparams.iteritems():
+ result.append(("be/%s" % key, val))
self.cfg.Update(instance)
def Exec(self, feedback_fn):
"""Compute the list of all the exported system images.
- Returns:
- a dictionary with the structure node->(export-list)
- where export-list is a list of the instances exported on
- that node.
+ @rtype: dict
+ @return: a dictionary with the structure node->(export-list)
+ where export-list is a list of the instances exported on
+ that node.
"""
- return rpc.call_export_list(self.nodes)
+ rpcresult = self.rpc.call_export_list(self.nodes)
+ result = {}
+ for node in rpcresult:
+ if rpcresult[node].fail_msg:
+ result[node] = False
+ else:
+ result[node] = rpcresult[node].payload
+
+ return result
class LUExportInstance(LogicalUnit):
# remove it from its current node. In the future we could fix this by:
# - making a tasklet to search (share-lock all), then create the new one,
# then one to remove, after
- # - removing the removal operation altoghether
+ # - removing the removal operation altogether
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
def DeclareLocks(self, level):
"EXPORT_NODE": self.op.target_node,
"EXPORT_DO_SHUTDOWN": self.op.shutdown,
}
- env.update(_BuildInstanceHookEnvByObject(self.instance))
- nl = [self.sstore.GetMasterNode(), self.instance.primary_node,
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+ nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
self.op.target_node]
return env, nl, nl
self.instance = self.cfg.GetInstanceInfo(instance_name)
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckNodeOnline(self, self.instance.primary_node)
self.dst_node = self.cfg.GetNodeInfo(
self.cfg.ExpandNodeName(self.op.target_node))
- assert self.dst_node is not None, \
- "Cannot retrieve locked node %s" % self.op.target_node
+ if self.dst_node is None:
+ # This is wrong node name, not a non-locked node
+ raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
+ _CheckNodeOnline(self, self.dst_node.name)
+ _CheckNodeNotDrained(self, self.dst_node.name)
# instance disk type verification
for disk in self.instance.disks:
instance = self.instance
dst_node = self.dst_node
src_node = instance.primary_node
+
if self.op.shutdown:
# shutdown the instance, but not the disks
- if not rpc.call_instance_shutdown(src_node, instance):
- raise errors.OpExecError("Could not shutdown instance %s on node %s" %
- (instance.name, src_node))
+ feedback_fn("Shutting down instance %s" % instance.name)
+ result = self.rpc.call_instance_shutdown(src_node, instance)
+ result.Raise("Could not shutdown instance %s on"
+ " node %s" % (instance.name, src_node))
vgname = self.cfg.GetVGName()
snap_disks = []
- try:
- for disk in instance.disks:
- if disk.iv_name == "sda":
- # new_dev_name will be a snapshot of an lvm leaf of the one we passed
- new_dev_name = rpc.call_blockdev_snapshot(src_node, disk)
+ # set the disks ID correctly since call_instance_start needs the
+ # correct drbd minor to create the symlinks
+ for disk in instance.disks:
+ self.cfg.SetDiskID(disk, src_node)
- if not new_dev_name:
- logger.Error("could not snapshot block device %s on node %s" %
- (disk.logical_id[1], src_node))
- else:
- new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
- logical_id=(vgname, new_dev_name),
- physical_id=(vgname, new_dev_name),
- iv_name=disk.iv_name)
- snap_disks.append(new_dev)
+ # per-disk results
+ dresults = []
+ try:
+ for idx, disk in enumerate(instance.disks):
+ feedback_fn("Creating a snapshot of disk/%s on node %s" %
+ (idx, src_node))
+
+ # result.payload will be a snapshot of an lvm leaf of the one we passed
+ result = self.rpc.call_blockdev_snapshot(src_node, disk)
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Could not snapshot disk/%s on node %s: %s",
+ idx, src_node, msg)
+ snap_disks.append(False)
+ else:
+ disk_id = (vgname, result.payload)
+ new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
+ logical_id=disk_id, physical_id=disk_id,
+ iv_name=disk.iv_name)
+ snap_disks.append(new_dev)
finally:
- if self.op.shutdown and instance.status == "up":
- if not rpc.call_instance_start(src_node, instance, None):
- _ShutdownInstanceDisks(instance, self.cfg)
- raise errors.OpExecError("Could not start instance")
+ if self.op.shutdown and instance.admin_up:
+ feedback_fn("Starting instance %s" % instance.name)
+ result = self.rpc.call_instance_start(src_node, instance, None, None)
+ msg = result.fail_msg
+ if msg:
+ _ShutdownInstanceDisks(self, instance)
+ raise errors.OpExecError("Could not start instance: %s" % msg)
# TODO: check for size
- for dev in snap_disks:
- if not rpc.call_snapshot_export(src_node, dev, dst_node.name, instance):
- logger.Error("could not export block device %s from node %s to node %s"
- % (dev.logical_id[1], src_node, dst_node.name))
- if not rpc.call_blockdev_remove(src_node, dev):
- logger.Error("could not remove snapshot block device %s from node %s" %
- (dev.logical_id[1], src_node))
+ cluster_name = self.cfg.GetClusterName()
+ for idx, dev in enumerate(snap_disks):
+ feedback_fn("Exporting snapshot %s from %s to %s" %
+ (idx, src_node, dst_node.name))
+ if dev:
+ result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
+ instance, cluster_name, idx)
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Could not export disk/%s from node %s to"
+ " node %s: %s", idx, src_node, dst_node.name, msg)
+ dresults.append(False)
+ else:
+ dresults.append(True)
+ msg = self.rpc.call_blockdev_remove(src_node, dev).fail_msg
+ if msg:
+ self.LogWarning("Could not remove snapshot for disk/%d from node"
+ " %s: %s", idx, src_node, msg)
+ else:
+ dresults.append(False)
- if not rpc.call_finalize_export(dst_node.name, instance, snap_disks):
- logger.Error("could not finalize export for instance %s on node %s" %
- (instance.name, dst_node.name))
+ feedback_fn("Finalizing export on %s" % dst_node.name)
+ result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
+ fin_resu = True
+ msg = result.fail_msg
+ if msg:
+ self.LogWarning("Could not finalize export for instance %s"
+ " on node %s: %s", instance.name, dst_node.name, msg)
+ fin_resu = False
nodelist = self.cfg.GetNodeList()
nodelist.remove(dst_node.name)
# on one-node clusters nodelist will be empty after the removal
# if we proceed the backup would be removed because OpQueryExports
# substitutes an empty list with the full cluster node list.
+ iname = instance.name
if nodelist:
- exportlist = rpc.call_export_list(nodelist)
+ feedback_fn("Removing old exports for instance %s" % iname)
+ exportlist = self.rpc.call_export_list(nodelist)
for node in exportlist:
- if instance.name in exportlist[node]:
- if not rpc.call_export_remove(node, instance.name):
- logger.Error("could not remove older export for instance %s"
- " on node %s" % (instance.name, node))
+ if exportlist[node].fail_msg:
+ continue
+ if iname in exportlist[node].payload:
+ msg = self.rpc.call_export_remove(node, iname).fail_msg
+ if msg:
+ self.LogWarning("Could not remove older export for instance %s"
+ " on node %s: %s", iname, node, msg)
+ return fin_resu, dresults
class LURemoveExport(NoHooksLU):
"""
_OP_REQP = ["instance_name"]
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.needed_locks = {}
+ # We need all nodes to be locked in order for RemoveExport to work, but we
+ # don't need to lock the instance itself, as nothing will happen to it (and
+ # we can remove exports also for a removed instance)
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
def CheckPrereq(self):
"""Check prerequisites.
fqdn_warn = True
instance_name = self.op.instance_name
- exportlist = rpc.call_export_list(self.cfg.GetNodeList())
+ locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
+ exportlist = self.rpc.call_export_list(locked_nodes)
found = False
for node in exportlist:
- if instance_name in exportlist[node]:
+ msg = exportlist[node].fail_msg
+ if msg:
+ self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
+ continue
+ if instance_name in exportlist[node].payload:
found = True
- if not rpc.call_export_remove(node, instance_name):
- logger.Error("could not remove export for instance %s"
- " on node %s" % (instance_name, node))
+ result = self.rpc.call_export_remove(node, instance_name)
+ msg = result.fail_msg
+ if msg:
+ logging.error("Could not remove export for instance %s"
+ " on node %s: %s", instance_name, node, msg)
if fqdn_warn and not found:
feedback_fn("Export not found. If trying to remove an export belonging"
This is an abstract class which is the parent of all the other tags LUs.
"""
- def CheckPrereq(self):
- """Check prerequisites.
- """
- if self.op.kind == constants.TAG_CLUSTER:
- self.target = self.cfg.GetClusterInfo()
- elif self.op.kind == constants.TAG_NODE:
+ def ExpandNames(self):
+ self.needed_locks = {}
+ if self.op.kind == constants.TAG_NODE:
name = self.cfg.ExpandNodeName(self.op.name)
if name is None:
raise errors.OpPrereqError("Invalid node name (%s)" %
(self.op.name,))
self.op.name = name
- self.target = self.cfg.GetNodeInfo(name)
+ self.needed_locks[locking.LEVEL_NODE] = name
elif self.op.kind == constants.TAG_INSTANCE:
name = self.cfg.ExpandInstanceName(self.op.name)
if name is None:
raise errors.OpPrereqError("Invalid instance name (%s)" %
(self.op.name,))
self.op.name = name
- self.target = self.cfg.GetInstanceInfo(name)
+ self.needed_locks[locking.LEVEL_INSTANCE] = name
+
+ def CheckPrereq(self):
+ """Check prerequisites.
+
+ """
+ if self.op.kind == constants.TAG_CLUSTER:
+ self.target = self.cfg.GetClusterInfo()
+ elif self.op.kind == constants.TAG_NODE:
+ self.target = self.cfg.GetNodeInfo(self.op.name)
+ elif self.op.kind == constants.TAG_INSTANCE:
+ self.target = self.cfg.GetInstanceInfo(self.op.name)
else:
raise errors.OpPrereqError("Wrong tag type requested (%s)" %
str(self.op.kind))
"""
_OP_REQP = ["kind", "name"]
+ REQ_BGL = False
def Exec(self, feedback_fn):
"""Returns the tag list.
"""
_OP_REQP = ["pattern"]
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.needed_locks = {}
def CheckPrereq(self):
"""Check prerequisites.
"""
cfg = self.cfg
tgts = [("/cluster", cfg.GetClusterInfo())]
- ilist = [cfg.GetInstanceInfo(name) for name in cfg.GetInstanceList()]
+ ilist = cfg.GetAllInstancesInfo().values()
tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
- nlist = [cfg.GetNodeInfo(name) for name in cfg.GetNodeList()]
+ nlist = cfg.GetAllNodesInfo().values()
tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
results = []
for path, target in tgts:
"""
_OP_REQP = ["kind", "name", "tags"]
+ REQ_BGL = False
def CheckPrereq(self):
"""Check prerequisites.
"""
_OP_REQP = ["kind", "name", "tags"]
+ REQ_BGL = False
def CheckPrereq(self):
"""Check prerequisites.
if not utils.TestDelay(self.op.duration):
raise errors.OpExecError("Error during master delay test")
if self.op.on_nodes:
- result = rpc.call_test_delay(self.op.on_nodes, self.op.duration)
- if not result:
- raise errors.OpExecError("Complete failure from rpc call")
+ result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
for node, node_result in result.items():
- if not node_result:
- raise errors.OpExecError("Failure during rpc call to node %s,"
- " result: %s" % (node, node_result))
+ node_result.Raise("Failure during rpc call to node %s" % node)
class IAllocator(object):
"""IAllocator framework.
An IAllocator instance has three sets of attributes:
- - cfg/sstore that are needed to query the cluster
+ - cfg that is needed to query the cluster
- input data (all members of the _KEYS class attribute are required)
- four buffer attributes (in|out_data|text), that represent the
input (to the external script) in text and data structure format,
"""
_ALLO_KEYS = [
"mem_size", "disks", "disk_template",
- "os", "tags", "nics", "vcpus",
+ "os", "tags", "nics", "vcpus", "hypervisor",
]
_RELO_KEYS = [
"relocate_from",
]
- def __init__(self, cfg, sstore, mode, name, **kwargs):
+ def __init__(self, cfg, rpc, mode, name, **kwargs):
self.cfg = cfg
- self.sstore = sstore
+ self.rpc = rpc
# init buffer variables
self.in_text = self.out_text = self.in_data = self.out_data = None
# init all input fields so that pylint is happy
self.name = name
self.mem_size = self.disks = self.disk_template = None
self.os = self.tags = self.nics = self.vcpus = None
+ self.hypervisor = None
self.relocate_from = None
# computed fields
self.required_nodes = None
"""
cfg = self.cfg
+ cluster_info = cfg.GetClusterInfo()
# cluster data
data = {
- "version": 1,
- "cluster_name": self.sstore.GetClusterName(),
- "cluster_tags": list(cfg.GetClusterInfo().GetTags()),
- "hypervisor_type": self.sstore.GetHypervisorType(),
+ "version": constants.IALLOCATOR_VERSION,
+ "cluster_name": cfg.GetClusterName(),
+ "cluster_tags": list(cluster_info.GetTags()),
+ "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
# we don't have job IDs
}
-
- i_list = [cfg.GetInstanceInfo(iname) for iname in cfg.GetInstanceList()]
+ iinfo = cfg.GetAllInstancesInfo().values()
+ i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
# node data
node_results = {}
node_list = cfg.GetNodeList()
- node_data = rpc.call_node_info(node_list, cfg.GetVGName())
- for nname in node_list:
+
+ if self.mode == constants.IALLOCATOR_MODE_ALLOC:
+ hypervisor_name = self.hypervisor
+ elif self.mode == constants.IALLOCATOR_MODE_RELOC:
+ hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
+
+ node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
+ hypervisor_name)
+ node_iinfo = \
+ self.rpc.call_all_instances_info(node_list,
+ cluster_info.enabled_hypervisors)
+ for nname, nresult in node_data.items():
+ # first fill in static (config-based) values
ninfo = cfg.GetNodeInfo(nname)
- if nname not in node_data or not isinstance(node_data[nname], dict):
- raise errors.OpExecError("Can't get data for node %s" % nname)
- remote_info = node_data[nname]
- for attr in ['memory_total', 'memory_free', 'memory_dom0',
- 'vg_size', 'vg_free', 'cpu_total']:
- if attr not in remote_info:
- raise errors.OpExecError("Node '%s' didn't return attribute '%s'" %
- (nname, attr))
- try:
- remote_info[attr] = int(remote_info[attr])
- except ValueError, err:
- raise errors.OpExecError("Node '%s' returned invalid value for '%s':"
- " %s" % (nname, attr, str(err)))
- # compute memory used by primary instances
- i_p_mem = i_p_up_mem = 0
- for iinfo in i_list:
- if iinfo.primary_node == nname:
- i_p_mem += iinfo.memory
- if iinfo.status == "up":
- i_p_up_mem += iinfo.memory
-
- # compute memory used by instances
pnr = {
"tags": list(ninfo.GetTags()),
- "total_memory": remote_info['memory_total'],
- "reserved_memory": remote_info['memory_dom0'],
- "free_memory": remote_info['memory_free'],
- "i_pri_memory": i_p_mem,
- "i_pri_up_memory": i_p_up_mem,
- "total_disk": remote_info['vg_size'],
- "free_disk": remote_info['vg_free'],
"primary_ip": ninfo.primary_ip,
"secondary_ip": ninfo.secondary_ip,
- "total_cpus": remote_info['cpu_total'],
+ "offline": ninfo.offline,
+ "drained": ninfo.drained,
+ "master_candidate": ninfo.master_candidate,
}
+
+ if not (ninfo.offline or ninfo.drained):
+ nresult.Raise("Can't get data for node %s" % nname)
+ node_iinfo[nname].Raise("Can't get node instance info from node %s" %
+ nname)
+ remote_info = nresult.payload
+
+ for attr in ['memory_total', 'memory_free', 'memory_dom0',
+ 'vg_size', 'vg_free', 'cpu_total']:
+ if attr not in remote_info:
+ raise errors.OpExecError("Node '%s' didn't return attribute"
+ " '%s'" % (nname, attr))
+ if not isinstance(remote_info[attr], int):
+ raise errors.OpExecError("Node '%s' returned invalid value"
+ " for '%s': %s" %
+ (nname, attr, remote_info[attr]))
+ # compute memory used by primary instances
+ i_p_mem = i_p_up_mem = 0
+ for iinfo, beinfo in i_list:
+ if iinfo.primary_node == nname:
+ i_p_mem += beinfo[constants.BE_MEMORY]
+ if iinfo.name not in node_iinfo[nname].payload:
+ i_used_mem = 0
+ else:
+ i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
+ i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
+ remote_info['memory_free'] -= max(0, i_mem_diff)
+
+ if iinfo.admin_up:
+ i_p_up_mem += beinfo[constants.BE_MEMORY]
+
+ # compute memory used by instances
+ pnr_dyn = {
+ "total_memory": remote_info['memory_total'],
+ "reserved_memory": remote_info['memory_dom0'],
+ "free_memory": remote_info['memory_free'],
+ "total_disk": remote_info['vg_size'],
+ "free_disk": remote_info['vg_free'],
+ "total_cpus": remote_info['cpu_total'],
+ "i_pri_memory": i_p_mem,
+ "i_pri_up_memory": i_p_up_mem,
+ }
+ pnr.update(pnr_dyn)
+
node_results[nname] = pnr
data["nodes"] = node_results
# instance data
instance_data = {}
- for iinfo in i_list:
- nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
- for n in iinfo.nics]
+ for iinfo, beinfo in i_list:
+ nic_data = []
+ for nic in iinfo.nics:
+ filled_params = objects.FillDict(
+ cluster_info.nicparams[constants.PP_DEFAULT],
+ nic.nicparams)
+ nic_dict = {"mac": nic.mac,
+ "ip": nic.ip,
+ "mode": filled_params[constants.NIC_MODE],
+ "link": filled_params[constants.NIC_LINK],
+ }
+ if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+ nic_dict["bridge"] = filled_params[constants.NIC_LINK]
+ nic_data.append(nic_dict)
pir = {
"tags": list(iinfo.GetTags()),
- "should_run": iinfo.status == "up",
- "vcpus": iinfo.vcpus,
- "memory": iinfo.memory,
+ "admin_up": iinfo.admin_up,
+ "vcpus": beinfo[constants.BE_VCPUS],
+ "memory": beinfo[constants.BE_MEMORY],
"os": iinfo.os,
"nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
"nics": nic_data,
- "disks": [{"size": dsk.size, "mode": "w"} for dsk in iinfo.disks],
+ "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
"disk_template": iinfo.disk_template,
+ "hypervisor": iinfo.hypervisor,
}
+ pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
+ pir["disks"])
instance_data[iinfo.name] = pir
data["instances"] = instance_data
"""
data = self.in_data
- if len(self.disks) != 2:
- raise errors.OpExecError("Only two-disk configurations supported")
- disk_space = _ComputeDiskSize(self.disk_template,
- self.disks[0]["size"], self.disks[1]["size"])
+ disk_space = _ComputeDiskSize(self.disk_template, self.disks)
if self.disk_template in constants.DTS_NET_MIRROR:
self.required_nodes = 2
raise errors.OpPrereqError("Instance has not exactly one secondary node")
self.required_nodes = 1
-
- disk_space = _ComputeDiskSize(instance.disk_template,
- instance.disks[0].size,
- instance.disks[1].size)
+ disk_sizes = [{'size': disk.size} for disk in instance.disks]
+ disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
request = {
"type": "relocate",
self.in_text = serializer.Dump(self.in_data)
- def Run(self, name, validate=True, call_fn=rpc.call_iallocator_runner):
+ def Run(self, name, validate=True, call_fn=None):
"""Run an instance allocator and return the results.
"""
- data = self.in_text
-
- result = call_fn(self.sstore.GetMasterNode(), name, self.in_text)
-
- if not isinstance(result, (list, tuple)) or len(result) != 4:
- raise errors.OpExecError("Invalid result from master iallocator runner")
+ if call_fn is None:
+ call_fn = self.rpc.call_iallocator_runner
- rcode, stdout, stderr, fail = result
+ result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
+ result.Raise("Failure while running the iallocator script")
- if rcode == constants.IARUN_NOTFOUND:
- raise errors.OpExecError("Can't find allocator '%s'" % name)
- elif rcode == constants.IARUN_FAILURE:
- raise errors.OpExecError("Instance allocator call failed: %s,"
- " output: %s" % (fail, stdout+stderr))
- self.out_text = stdout
+ self.out_text = result.payload
if validate:
self._ValidateResult()
" 'nics' parameter")
if not isinstance(self.op.disks, list):
raise errors.OpPrereqError("Invalid parameter 'disks'")
- if len(self.op.disks) != 2:
- raise errors.OpPrereqError("Only two-disk configurations supported")
for row in self.op.disks:
if (not isinstance(row, dict) or
"size" not in row or
row["mode"] not in ['r', 'w']):
raise errors.OpPrereqError("Invalid contents of the"
" 'disks' parameter")
+ if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
+ self.op.hypervisor = self.cfg.GetHypervisorType()
elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
if not hasattr(self.op, "name"):
raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
"""
if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
- ial = IAllocator(self.cfg, self.sstore,
+ ial = IAllocator(self.cfg, self.rpc,
mode=self.op.mode,
name=self.op.name,
mem_size=self.op.mem_size,
tags=self.op.tags,
nics=self.op.nics,
vcpus=self.op.vcpus,
+ hypervisor=self.op.hypervisor,
)
else:
- ial = IAllocator(self.cfg, self.sstore,
+ ial = IAllocator(self.cfg, self.rpc,
mode=self.op.mode,
name=self.op.name,
relocate_from=list(self.relocate_from),
projects / ganeti-local / blobdiff