_ErrorIf(test, self.ENODEHV, node,
"hypervisor %s verify failure: '%s'", hv_name, hv_result)
+ hvp_result = nresult.get(constants.NV_HVPARAMS, None)
+ if ninfo.vm_capable and isinstance(hvp_result, list):
+ for item, hv_name, hv_result in hvp_result:
+ _ErrorIf(True, self.ENODEHV, node,
+ "hypervisor %s parameter verify failure (source %s): %s",
+ hv_name, item, hv_result)
+
test = nresult.get(constants.NV_NODESETUP,
["Missing NODESETUP results"])
_ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
node_current)
for node, n_img in node_image.items():
- if (not node == node_current):
+ if node != node_current:
test = instance in n_img.instances
_ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
"instance should not run on node %s", node)
for idx, (success, status) in enumerate(disks)]
for nname, success, bdev_status, idx in diskdata:
- _ErrorIf(instanceconfig.admin_up and not success,
+ # the 'ghost node' construction in Exec() ensures that we have a
+ # node here
+ snode = node_image[nname]
+ bad_snode = snode.ghost or snode.offline
+ _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
self.EINSTANCEFAULTYDISK, instance,
"couldn't retrieve status for disk/%s on %s: %s",
idx, nname, bdev_status)
# WARNING: we currently take into account down instances as well
# as up ones, considering that even if they're down someone
# might want to start them even in the event of a node failure.
+ if n_img.offline:
+ # we're skipping offline nodes from the N+1 warning, since
+ # most likely we don't have good memory infromation from them;
+ # we already list instances living on such nodes, and that's
+ # enough warning
+ continue
for prinode, instances in n_img.sbp.items():
needed_mem = 0
for instance in instances:
return instdisk
+ def _VerifyHVP(self, hvp_data):
+ """Verifies locally the syntax of the hypervisor parameters.
+
+ """
+ for item, hv_name, hv_params in hvp_data:
+ msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
+ (item, hv_name))
+ try:
+ hv_class = hypervisor.GetHypervisor(hv_name)
+ utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
+ hv_class.CheckParameterSyntax(hv_params)
+ except errors.GenericError, err:
+ self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
+
+
def BuildHooksEnv(self):
"""Build hooks env.
local_checksums = utils.FingerprintFiles(file_names)
+ # Compute the set of hypervisor parameters
+ hvp_data = []
+ for hv_name in hypervisors:
+ hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
+ for os_name, os_hvp in cluster.os_hvp.items():
+ for hv_name, hv_params in os_hvp.items():
+ if not hv_params:
+ continue
+ full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
+ hvp_data.append(("os %s" % os_name, hv_name, full_params))
+ # TODO: collapse identical parameter values in a single one
+ for instance in instanceinfo.values():
+ if not instance.hvparams:
+ continue
+ hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
+ cluster.FillHV(instance)))
+ # and verify them locally
+ self._VerifyHVP(hvp_data)
+
feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
node_verify_param = {
constants.NV_FILELIST: file_names,
constants.NV_NODELIST: [node.name for node in nodeinfo
if not node.offline],
constants.NV_HYPERVISOR: hypervisors,
+ constants.NV_HVPARAMS: hvp_data,
constants.NV_NODENETTEST: [(node.name, node.primary_ip,
node.secondary_ip) for node in nodeinfo
if not node.offline],
self.ENODERPC, pnode, "instance %s, connection to"
" primary node failed", instance)
- if pnode_img.offline:
- inst_nodes_offline.append(pnode)
+ _ErrorIf(pnode_img.offline, self.EINSTANCEBADNODE, instance,
+ "instance lives on offline node %s", inst_config.primary_node)
# 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
# warn that the instance lives on offline nodes
_ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
- "instance lives on offline node(s) %s",
+ "instance has offline secondary node(s) %s",
utils.CommaJoin(inst_nodes_offline))
# ... or ghost/non-vm_capable nodes
for node in inst_config.all_nodes:
result = res_nodes, res_instances, res_missing = {}, [], {}
nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
- instances = [self.cfg.GetInstanceInfo(name)
- for name in self.cfg.GetInstanceList()]
+ instances = self.cfg.GetAllInstancesInfo().values()
nv_dict = {}
for inst in instances:
inst_lvs = {}
- if (not inst.admin_up or
- inst.disk_template not in constants.DTS_NET_MIRROR):
+ if not inst.admin_up:
continue
inst.MapLVsByNode(inst_lvs)
# transform { iname: {node: [vol,],},} to {(node, vol): iname}
if not nv_dict:
return result
- vg_names = self.rpc.call_vg_list(nodes)
- for node in nodes:
- vg_names[node].Raise("Cannot get list of VGs")
-
- for node in nodes:
- # node_volume
- node_res = self.rpc.call_lv_list([node],
- vg_names[node].payload.keys())[node]
+ node_lvs = self.rpc.call_lv_list(nodes, [])
+ for node, node_res in node_lvs.items():
if node_res.offline:
continue
msg = node_res.fail_msg
newl = [v[2].Copy() for v in dskl]
for dsk in newl:
self.cfg.SetDiskID(dsk, node)
- result = self.rpc.call_blockdev_getsizes(node, newl)
+ result = self.rpc.call_blockdev_getsize(node, newl)
if result.fail_msg:
- self.LogWarning("Failure in blockdev_getsizes call to node"
+ self.LogWarning("Failure in blockdev_getsize call to node"
" %s, ignoring", node)
continue
- if len(result.data) != len(dskl):
+ if len(result.payload) != len(dskl):
+ logging.warning("Invalid result from node %s: len(dksl)=%d,"
+ " result.payload=%s", node, len(dskl), result.payload)
self.LogWarning("Invalid result from node %s, ignoring node results",
node)
continue
- for ((instance, idx, disk), size) in zip(dskl, result.data):
+ for ((instance, idx, disk), size) in zip(dskl, result.payload):
if size is None:
self.LogWarning("Disk %d of instance %s did not return size"
" information, ignoring", idx, instance.name)
utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
+ # TODO: we need a more general way to handle resetting
+ # cluster-level parameters to default values
+ if self.new_ndparams["oob_program"] == "":
+ self.new_ndparams["oob_program"] = \
+ constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
+
if self.op.nicparams:
utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
"""Compute the list of OSes.
"""
- valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
+ valid_nodes = [node.name
+ for node in self.cfg.GetAllNodesInfo().values()
+ if not node.offline and node.vm_capable]
node_data = self.rpc.call_os_diagnose(valid_nodes)
pol = self._DiagnoseByOS(node_data)
output = []
# Gather data as requested
if query.NQ_LIVE in self.requested_data:
- node_data = lu.rpc.call_node_info(nodenames, lu.cfg.GetVGName(),
+ # filter out non-vm_capable nodes
+ toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
+
+ node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
lu.cfg.GetHypervisorType())
live_data = dict((name, nresult.payload)
for (name, nresult) in node_data.items()
"""Computes the list of instances and their attributes.
"""
+ cluster = lu.cfg.GetClusterInfo()
all_info = lu.cfg.GetAllInstancesInfo()
instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
instance_list = [all_info[name] for name in instance_names]
- nodes = frozenset([inst.primary_node for inst in instance_list])
+ nodes = frozenset(itertools.chain(*(inst.all_nodes
+ for inst in instance_list)))
hv_list = list(set([inst.hypervisor for inst in instance_list]))
bad_nodes = []
offline_nodes = []
+ wrongnode_inst = set()
# Gather data as requested
- if query.IQ_LIVE in self.requested_data:
+ if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
live_data = {}
node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
for name in nodes:
if result.fail_msg:
bad_nodes.append(name)
elif result.payload:
- live_data.update(result.payload)
+ for inst in result.payload:
+ if inst in all_info:
+ if all_info[inst].primary_node == name:
+ live_data.update(result.payload)
+ else:
+ wrongnode_inst.add(inst)
+ else:
+ # orphan instance; we don't list it here as we don't
+ # handle this case yet in the output of instance listing
+ logging.warning("Orphan instance '%s' found on node %s",
+ inst, name)
# else no instance is alive
else:
live_data = {}
else:
disk_usage = None
+ if query.IQ_CONSOLE in self.requested_data:
+ consinfo = {}
+ for inst in instance_list:
+ if inst.name in live_data:
+ # Instance is running
+ consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
+ else:
+ consinfo[inst.name] = None
+ assert set(consinfo.keys()) == set(instance_names)
+ else:
+ consinfo = None
+
return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
disk_usage, offline_nodes, bad_nodes,
- live_data)
+ live_data, wrongnode_inst, consinfo)
class LUQuery(NoHooksLU):
errors.ECODE_STATE)
if node.master_candidate and self.might_demote and not self.lock_all:
- assert not self.op.auto_promote, "auto-promote set but lock_all not"
+ assert not self.op.auto_promote, "auto_promote set but lock_all not"
# check if after removing the current node, we're missing master
# candidates
(mc_remaining, mc_should, _) = \
self.cfg.GetMasterCandidateStats(exceptions=[node.name])
if mc_remaining < mc_should:
raise errors.OpPrereqError("Not enough master candidates, please"
- " pass auto_promote to allow promotion",
- errors.ECODE_STATE)
+ " pass auto promote option to allow"
+ " promotion", errors.ECODE_STATE)
self.old_flags = old_flags = (node.master_candidate,
node.drained, node.offline)
# SyncSource, etc.)
# 1st pass, assemble on all nodes in secondary mode
- for inst_disk in disks:
+ for idx, inst_disk in enumerate(disks):
for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
if ignore_size:
node_disk = node_disk.Copy()
node_disk.UnsetSize()
lu.cfg.SetDiskID(node_disk, node)
- result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
+ result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
msg = result.fail_msg
if msg:
lu.proc.LogWarning("Could not prepare block device %s on node %s"
# FIXME: race condition on drbd migration to primary
# 2nd pass, do only the primary node
- for inst_disk in disks:
+ for idx, inst_disk in enumerate(disks):
dev_path = None
for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
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)
+ result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
msg = result.fail_msg
if msg:
lu.proc.LogWarning("Could not prepare block device %s on node %s"
"""
instance = self.instance
- _SafeShutdownInstanceDisks(self, instance)
+ if self.op.force:
+ _ShutdownInstanceDisks(self, instance)
+ else:
+ _SafeShutdownInstanceDisks(self, instance)
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
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)
+ instance.name, True, idx)
if result.fail_msg:
self.LogWarning("Can't assemble newly created disk %d: %s",
idx, result.fail_msg)
"""
node = instance.primary_node
+
+ for device in instance.disks:
+ lu.cfg.SetDiskID(device, node)
+
logging.info("Pause sync of instance %s disks", instance.name)
result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
try:
for idx, device in enumerate(instance.disks):
lu.LogInfo("* Wiping disk %d", idx)
- logging.info("Wiping disk %d for instance %s", idx, instance.name)
+ logging.info("Wiping disk %d for instance %s, node %s",
+ idx, instance.name, node)
# The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
# MAX_WIPE_CHUNK at max
return req_size_dict[disk_template]
+def _FilterVmNodes(lu, nodenames):
+ """Filters out non-vm_capable nodes from a list.
+
+ @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
+ @rtype: list
+ @return: the list of vm-capable nodes
+
+ """
+ vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
+ return [name for name in nodenames if name not in vm_nodes]
+
+
def _CheckHVParams(lu, nodenames, hvname, hvparams):
"""Hypervisor parameter validation.
@raise errors.OpPrereqError: if the parameters are not valid
"""
+ nodenames = _FilterVmNodes(lu, nodenames)
hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
hvname,
hvparams)
@raise errors.OpPrereqError: if the parameters are not valid
"""
+ nodenames = _FilterVmNodes(lu, nodenames)
result = lu.rpc.call_os_validate(required, nodenames, osname,
[constants.OS_VALIDATE_PARAMETERS],
osparams)
logging.debug("Connecting to console of %s on %s", instance.name, node)
- 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 = hyper.GetInstanceConsole(instance, hvparams, beparams)
+ return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
- assert console.instance == instance.name
- assert console.Validate()
- return console.ToDict()
+def _GetInstanceConsole(cluster, instance):
+ """Returns console information for an instance.
+
+ @type cluster: L{objects.Cluster}
+ @type instance: L{objects.Instance}
+ @rtype: dict
+
+ """
+ hyper = hypervisor.GetHypervisor(instance.hypervisor)
+ # 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 = hyper.GetInstanceConsole(instance, hvparams, beparams)
+
+ assert console.instance == instance.name
+ assert console.Validate()
+
+ return console.ToDict()
class LUInstanceReplaceDisks(LogicalUnit):
# Verify the cluster would not be left group-less.
if len(self.cfg.GetNodeGroupList()) == 1:
- raise errors.OpPrereqError("Group '%s' is the last group in the cluster,"
- " which cannot be left without at least one"
- " group" % self.op.group_name,
+ raise errors.OpPrereqError("Group '%s' is the only group,"
+ " cannot be removed" %
+ self.op.group_name,
errors.ECODE_STATE)
def BuildHooksEnv(self):
"i_pri_up_memory": i_p_up_mem,
}
pnr_dyn.update(node_results[nname])
-
- node_results[nname] = pnr_dyn
+ node_results[nname] = pnr_dyn
return node_results
projects / ganeti-local / blobdiff