self.op.instance_name)
self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
- def _LockInstancesNodes(self, primary_only=False):
+ def _LockInstancesNodes(self, primary_only=False,
+ level=locking.LEVEL_NODE):
"""Helper function to declare instances' nodes for locking.
This function should be called after locking one or more instances to lock
@type primary_only: boolean
@param primary_only: only lock primary nodes of locked instances
+ @param level: Which lock level to use for locking nodes
"""
- assert locking.LEVEL_NODE in self.recalculate_locks, \
+ assert level in self.recalculate_locks, \
"_LockInstancesNodes helper function called with no nodes to recalculate"
# TODO: check if we're really been called with the instance locks held
if not primary_only:
wanted_nodes.extend(instance.secondary_nodes)
- if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
- self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
- elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
- self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
+ if self.recalculate_locks[level] == constants.LOCKS_REPLACE:
+ self.needed_locks[level] = wanted_nodes
+ elif self.recalculate_locks[level] == constants.LOCKS_APPEND:
+ self.needed_locks[level].extend(wanted_nodes)
+ else:
+ raise errors.ProgrammerError("Unknown recalculation mode")
- del self.recalculate_locks[locking.LEVEL_NODE]
+ del self.recalculate_locks[level]
class NoHooksLU(LogicalUnit): # pylint: disable=W0223
"""Destroys the cluster.
"""
- master = self.cfg.GetMasterNode()
+ master_params = self.cfg.GetMasterNetworkParameters()
# Run post hooks on master node before it's removed
- _RunPostHook(self, master)
+ _RunPostHook(self, master_params.name)
- result = self.rpc.call_node_deactivate_master_ip(master)
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master role")
- return master
+ return master_params.name
def _VerifyCertificate(filename):
_ErrorIf(bool(missing), constants.CV_ENODENET, node,
"missing bridges: %s" % utils.CommaJoin(sorted(missing)))
+ def _VerifyNodeUserScripts(self, ninfo, nresult):
+ """Check the results of user scripts presence and executability on the node
+
+ @type ninfo: L{objects.Node}
+ @param ninfo: the node to check
+ @param nresult: the remote results for the node
+
+ """
+ node = ninfo.name
+
+ test = not constants.NV_USERSCRIPTS in nresult
+ self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
+ "did not return user scripts information")
+
+ broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
+ if not test:
+ self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
+ "user scripts not present or not executable: %s" %
+ utils.CommaJoin(sorted(broken_scripts)))
+
def _VerifyNodeNetwork(self, ninfo, nresult):
"""Check the node network connectivity results.
feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
+ user_scripts = []
+ if self.cfg.GetUseExternalMipScript():
+ user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
+
node_verify_param = {
constants.NV_FILELIST:
utils.UniqueSequence(filename
constants.NV_MASTERIP: (master_node, master_ip),
constants.NV_OSLIST: None,
constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
+ constants.NV_USERSCRIPTS: user_scripts,
}
if vg_name is not None:
nimg.call_ok = self._VerifyNode(node_i, nresult)
self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
self._VerifyNodeNetwork(node_i, nresult)
+ self._VerifyNodeUserScripts(node_i, nresult)
self._VerifyOob(node_i, nresult)
if nimg.vm_capable:
if self.op.instances:
self.wanted_names = _GetWantedInstances(self, self.op.instances)
self.needed_locks = {
- locking.LEVEL_NODE: [],
+ locking.LEVEL_NODE_RES: [],
locking.LEVEL_INSTANCE: self.wanted_names,
}
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
else:
self.wanted_names = None
self.needed_locks = {
- locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_NODE_RES: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
}
self.share_locks = _ShareAll()
def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE and self.wanted_names is not None:
- self._LockInstancesNodes(primary_only=True)
+ if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
+ self._LockInstancesNodes(primary_only=True, level=level)
def CheckPrereq(self):
"""Check prerequisites.
for idx, disk in enumerate(instance.disks):
per_node_disks[pnode].append((instance, idx, disk))
+ assert not (frozenset(per_node_disks.keys()) -
+ self.owned_locks(locking.LEVEL_NODE_RES)), \
+ "Not owning correct locks"
+ assert not self.owned_locks(locking.LEVEL_NODE)
+
changed = []
for node, dskl in per_node_disks.items():
newl = [v[2].Copy() for v in dskl]
"""
clustername = self.op.name
- ip = self.ip
+ new_ip = self.ip
# shutdown the master IP
- master = self.cfg.GetMasterNode()
- result = self.rpc.call_node_deactivate_master_ip(master)
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master role")
try:
cluster = self.cfg.GetClusterInfo()
cluster.cluster_name = clustername
- cluster.master_ip = ip
+ cluster.master_ip = new_ip
self.cfg.Update(cluster, feedback_fn)
# update the known hosts file
ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
node_list = self.cfg.GetOnlineNodeList()
try:
- node_list.remove(master)
+ node_list.remove(master_params.name)
except ValueError:
pass
_UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
finally:
- result = self.rpc.call_node_activate_master_ip(master)
+ master_params.ip = new_ip
+ result = self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
msg = result.fail_msg
if msg:
self.LogWarning("Could not re-enable the master role on"
if self.op.reserved_lvs is not None:
self.cluster.reserved_lvs = self.op.reserved_lvs
+ if self.op.use_external_mip_script is not None:
+ self.cluster.use_external_mip_script = self.op.use_external_mip_script
+
def helper_os(aname, mods, desc):
desc += " OS list"
lst = getattr(self.cluster, aname)
helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
if self.op.master_netdev:
- master = self.cfg.GetMasterNode()
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
feedback_fn("Shutting down master ip on the current netdev (%s)" %
self.cluster.master_netdev)
- result = self.rpc.call_node_deactivate_master_ip(master)
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master ip")
feedback_fn("Changing master_netdev from %s to %s" %
- (self.cluster.master_netdev, self.op.master_netdev))
+ (master_params.netdev, self.op.master_netdev))
self.cluster.master_netdev = self.op.master_netdev
if self.op.master_netmask:
- master = self.cfg.GetMasterNode()
+ master_params = self.cfg.GetMasterNetworkParameters()
feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
- result = self.rpc.call_node_change_master_netmask(master,
- self.op.master_netmask)
+ result = self.rpc.call_node_change_master_netmask(master_params.name,
+ master_params.netmask,
+ self.op.master_netmask,
+ master_params.ip,
+ master_params.netdev)
if result.fail_msg:
msg = "Could not change the master IP netmask: %s" % result.fail_msg
- self.LogWarning(msg)
feedback_fn(msg)
- else:
- self.cluster.master_netmask = self.op.master_netmask
+
+ self.cluster.master_netmask = self.op.master_netmask
self.cfg.Update(self.cluster, feedback_fn)
if self.op.master_netdev:
+ master_params = self.cfg.GetMasterNetworkParameters()
feedback_fn("Starting the master ip on the new master netdev (%s)" %
self.op.master_netdev)
- result = self.rpc.call_node_activate_master_ip(master)
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
if result.fail_msg:
self.LogWarning("Could not re-enable the master ip on"
" the master, please restart manually: %s",
# Files which should only be on master candidates
files_mc = set()
+
if not redist:
files_mc.add(constants.CLUSTER_CONF_FILE)
+ # FIXME: this should also be replicated but Ganeti doesn't support files_mc
+ # replication
+ files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
+
# Files which should only be on VM-capable nodes
files_vm = set(filename
for hv_name in cluster.enabled_hypervisors
"""Activate the master IP.
"""
- master = self.cfg.GetMasterNode()
- self.rpc.call_node_activate_master_ip(master)
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
class LUClusterDeactivateMasterIp(NoHooksLU):
"""Deactivate the master IP.
"""
- master = self.cfg.GetMasterNode()
- self.rpc.call_node_deactivate_master_ip(master)
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ self.rpc.call_node_deactivate_master_ip(master_params.name, master_params,
+ ems)
def _WaitForSync(lu, instance, disks=None, oneshot=False):
modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+ "Not owning BGL"
+
# Promote nodes to master candidate as needed
_AdjustCandidatePool(self, exceptions=[node.name])
self.context.RemoveNode(node.name)
def ExpandNames(self):
self.nq.ExpandNames(self)
+ def DeclareLocks(self, level):
+ self.nq.DeclareLocks(self, level)
+
def Exec(self, feedback_fn):
return self.nq.OldStyleQuery(self)
selected=self.op.output_fields)
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
+
if not self.op.nodes:
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
else:
selected=self.op.output_fields)
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
if self.op.nodes:
self.needed_locks[locking.LEVEL_NODE] = \
new_node = self.new_node
node = new_node.name
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+ "Not owning BGL"
+
# We adding a new node so we assume it's powered
new_node.powered = True
self.lock_all = self.op.auto_promote and self.might_demote
self.lock_instances = self.op.secondary_ip is not None
+ def _InstanceFilter(self, instance):
+ """Filter for getting affected instances.
+
+ """
+ return (instance.disk_template in constants.DTS_INT_MIRROR and
+ self.op.node_name in instance.all_nodes)
+
def ExpandNames(self):
if self.lock_all:
self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
if self.lock_instances:
- self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
-
- def DeclareLocks(self, level):
- # If we have locked all instances, before waiting to lock nodes, release
- # all the ones living on nodes unrelated to the current operation.
- if level == locking.LEVEL_NODE and self.lock_instances:
- self.affected_instances = []
- if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
- instances_keep = []
-
- # Build list of instances to release
- locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
- for instance_name, instance in self.cfg.GetMultiInstanceInfo(locked_i):
- if (instance.disk_template in constants.DTS_INT_MIRROR and
- self.op.node_name in instance.all_nodes):
- instances_keep.append(instance_name)
- self.affected_instances.append(instance)
-
- _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
-
- assert (set(self.owned_locks(locking.LEVEL_INSTANCE)) ==
- set(instances_keep))
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
+ frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
def BuildHooksEnv(self):
"""Build hooks env.
"""
node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
+ if self.lock_instances:
+ affected_instances = \
+ self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
+
+ # Verify instance locks
+ owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
+ wanted_instances = frozenset(affected_instances.keys())
+ if wanted_instances - owned_instances:
+ raise errors.OpPrereqError("Instances affected by changing node %s's"
+ " secondary IP address have changed since"
+ " locks were acquired, wanted '%s', have"
+ " '%s'; retry the operation" %
+ (self.op.node_name,
+ utils.CommaJoin(wanted_instances),
+ utils.CommaJoin(owned_instances)),
+ errors.ECODE_STATE)
+ else:
+ affected_instances = None
+
if (self.op.master_candidate is not None or
self.op.drained is not None or
self.op.offline is not None):
raise errors.OpPrereqError("Cannot change the secondary ip on a single"
" homed cluster", errors.ECODE_INVAL)
+ assert not (frozenset(affected_instances) -
+ self.owned_locks(locking.LEVEL_INSTANCE))
+
if node.offline:
- if self.affected_instances:
- raise errors.OpPrereqError("Cannot change secondary ip: offline"
- " node has instances (%s) configured"
- " to use it" % self.affected_instances)
+ if affected_instances:
+ raise errors.OpPrereqError("Cannot change secondary IP address:"
+ " offline node has instances (%s)"
+ " configured to use it" %
+ utils.CommaJoin(affected_instances.keys()))
else:
# On online nodes, check that no instances are running, and that
# the node has the new ip and we can reach it.
- for instance in self.affected_instances:
+ for instance in affected_instances.values():
_CheckInstanceDown(self, instance, "cannot change secondary ip")
_CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
"candidate_pool_size": cluster.candidate_pool_size,
"master_netdev": cluster.master_netdev,
"master_netmask": cluster.master_netmask,
+ "use_external_mip_script": cluster.use_external_mip_script,
"volume_group_name": cluster.volume_group_name,
"drbd_usermode_helper": cluster.drbd_usermode_helper,
"file_storage_dir": cluster.file_storage_dir,
# otherwise we need to lock all nodes for disk re-creation
primary_only = bool(self.op.nodes)
self._LockInstancesNodes(primary_only=primary_only)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
self.op.instance_name, errors.ECODE_INVAL)
# if we replace nodes *and* the old primary is offline, we don't
# check
- assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
+ assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
+ assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
if not (self.op.nodes and old_pnode.offline):
_CheckInstanceDown(self, instance, "cannot recreate disks")
"""
instance = self.instance
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
to_skip = []
mods = [] # keeps track of needed logical_id changes
def ExpandNames(self):
self._ExpandAndLockInstance()
self.needed_locks[locking.LEVEL_NODE] = []
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
" node %s: %s" %
(instance.name, instance.primary_node, msg))
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+ assert not (set(instance.all_nodes) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Not owning correct locks"
+
_RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
target_node = _ExpandNodeName(self.cfg, self.op.target_node)
self.op.target_node = target_node
self.needed_locks[locking.LEVEL_NODE] = [target_node]
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes(primary_only=True)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
self.LogInfo("Shutting down instance %s on source node %s",
instance.name, source_node)
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
result = self.rpc.call_instance_shutdown(source_node, instance,
self.op.shutdown_timeout)
msg = result.fail_msg
self.add_locks[locking.LEVEL_INSTANCE] = instance_name
if self.op.iallocator:
+ # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
+ # specifying a group on instance creation and then selecting nodes from
+ # that group
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
else:
self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
nodelist = [self.op.pnode]
self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
nodelist.append(self.op.snode)
self.needed_locks[locking.LEVEL_NODE] = nodelist
+ # Lock resources of instance's primary and secondary nodes (copy to
+ # prevent accidential modification)
+ self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
# in case of import lock the source node too
if self.op.mode == constants.INSTANCE_IMPORT:
instance = self.op.instance_name
pnode_name = self.pnode.name
+ assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Node locks differ from node resource locks"
+
ht_kind = self.op.hypervisor
if ht_kind in constants.HTS_REQ_PORT:
network_port = self.cfg.AllocatePort()
raise errors.OpExecError("There are some degraded disks for"
" this instance")
+ # Release all node resource locks
+ _ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
if self.op.mode == constants.INSTANCE_CREATE:
if not self.op.no_install:
raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
% self.op.mode)
+ assert not self.owned_locks(locking.LEVEL_NODE_RES)
+
if self.op.start:
iobj.admin_up = True
self.cfg.Update(iobj, feedback_fn)
REQ_BGL = False
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self._ExpandAndLockInstance()
def CheckPrereq(self):
def ExpandNames(self):
self._ExpandAndLockInstance()
self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
+ self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
instance = self.instance
disk = self.disk
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
if not disks_ok:
raise errors.OpExecError("Cannot activate block device to grow")
+ feedback_fn("Growing disk %s of instance '%s' by %s" %
+ (self.op.disk, instance.name,
+ utils.FormatUnit(self.op.amount, "h")))
+
# First run all grow ops in dry-run mode
for node in instance.all_nodes:
self.cfg.SetDiskID(disk, node)
disk.RecordGrow(self.op.amount)
self.cfg.Update(instance, feedback_fn)
+
+ # Changes have been recorded, release node lock
+ _ReleaseLocks(self, locking.LEVEL_NODE)
+
+ # Downgrade lock while waiting for sync
+ self.glm.downgrade(locking.LEVEL_INSTANCE)
+
if self.op.wait_for_sync:
disk_abort = not _WaitForSync(self, instance, disks=[disk])
if disk_abort:
" not supposed to be running because no wait for"
" sync mode was requested")
+ assert self.owned_locks(locking.LEVEL_NODE_RES)
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+
class LUInstanceQueryData(NoHooksLU):
"""Query runtime instance data.
projects / ganeti-local / blobdiff