from ganeti import query
from ganeti import qlang
from ganeti import opcodes
+from ganeti import ht
import ganeti.masterd.instance # pylint: disable-msg=W0611
self.rpc = rpc
# Dicts used to declare locking needs to mcpu
self.needed_locks = None
- self.acquired_locks = {}
self.share_locks = dict.fromkeys(locking.LEVELS, 0)
self.add_locks = {}
self.remove_locks = {}
# future we might want to have different behaviors depending on the value
# of self.recalculate_locks[locking.LEVEL_NODE]
wanted_nodes = []
- for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
+ for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
instance = self.context.cfg.GetInstanceInfo(instance_name)
wanted_nodes.append(instance.primary_node)
if not primary_only:
"""
if self.do_locking:
- names = lu.acquired_locks[lock_level]
+ names = lu.glm.list_owned(lock_level)
else:
names = all_names
# caller specified names and we must keep the same order
assert self.names
- assert not self.do_locking or lu.acquired_locks[lock_level]
+ assert not self.do_locking or lu.glm.is_owned(lock_level)
missing = set(self.wanted).difference(names)
if missing:
release = []
# Determine which locks to release
- for name in lu.acquired_locks[level]:
+ for name in lu.glm.list_owned(level):
if should_release(name):
release.append(name)
else:
retain.append(name)
- assert len(lu.acquired_locks[level]) == (len(retain) + len(release))
+ assert len(lu.glm.list_owned(level)) == (len(retain) + len(release))
# Release just some locks
lu.glm.release(level, names=release)
- lu.acquired_locks[level] = retain
assert frozenset(lu.glm.list_owned(level)) == frozenset(retain)
else:
# Release everything
lu.glm.release(level)
- del lu.acquired_locks[level]
- assert not lu.glm.list_owned(level), "No locks should be owned"
+ assert not lu.glm.is_owned(level), "No locks should be owned"
def _RunPostHook(lu, node_name):
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
memory, vcpus, nics, disk_template, disks,
- bep, hvp, hypervisor_name):
+ bep, hvp, hypervisor_name, tags):
"""Builds instance related env variables for hooks
This builds the hook environment from individual variables.
@param hvp: the hypervisor parameters for the instance
@type hypervisor_name: string
@param hypervisor_name: the hypervisor for the instance
+ @type tags: list
+ @param tags: list of instance tags as strings
@rtype: dict
@return: the hook environment for this instance
env["INSTANCE_DISK_COUNT"] = disk_count
+ if not tags:
+ tags = []
+
+ env["INSTANCE_TAGS"] = " ".join(tags)
+
for source, kind in [(bep, "BE"), (hvp, "HV")]:
for key, value in source.items():
env["INSTANCE_%s_%s" % (kind, key)] = value
'bep': bep,
'hvp': hvp,
'hypervisor_name': instance.hypervisor,
+ 'tags': instance.tags,
}
if override:
args.update(override)
iallocator = getattr(lu.op, iallocator_slot, None)
if node is not None and iallocator is not None:
- raise errors.OpPrereqError("Do not specify both, iallocator and node.",
+ raise errors.OpPrereqError("Do not specify both, iallocator and node",
errors.ECODE_INVAL)
elif node is None and iallocator is None:
default_iallocator = lu.cfg.GetDefaultIAllocator()
setattr(lu.op, iallocator_slot, default_iallocator)
else:
raise errors.OpPrereqError("No iallocator or node given and no"
- " cluster-wide default iallocator found."
- " Please specify either an iallocator or a"
+ " cluster-wide default iallocator found;"
+ " please specify either an iallocator or a"
" node, or set a cluster-wide default"
- " iallocator.")
+ " iallocator")
class LUClusterPostInit(LogicalUnit):
def _VerifyCertificate(filename):
- """Verifies a certificate for LUClusterVerify.
+ """Verifies a certificate for L{LUClusterVerifyConfig}.
@type filename: string
@param filename: Path to PEM file
cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
utils.ReadFile(filename))
except Exception, err: # pylint: disable-msg=W0703
- return (LUClusterVerify.ETYPE_ERROR,
+ return (LUClusterVerifyConfig.ETYPE_ERROR,
"Failed to load X509 certificate %s: %s" % (filename, err))
(errcode, msg) = \
if errcode is None:
return (None, fnamemsg)
elif errcode == utils.CERT_WARNING:
- return (LUClusterVerify.ETYPE_WARNING, fnamemsg)
+ return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
elif errcode == utils.CERT_ERROR:
- return (LUClusterVerify.ETYPE_ERROR, fnamemsg)
+ return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
-class LUClusterVerify(LogicalUnit):
- """Verifies the cluster status.
+def _GetAllHypervisorParameters(cluster, instances):
+ """Compute the set of all hypervisor parameters.
+
+ @type cluster: L{objects.Cluster}
+ @param cluster: the cluster object
+ @param instances: list of L{objects.Instance}
+ @param instances: additional instances from which to obtain parameters
+ @rtype: list of (origin, hypervisor, parameters)
+ @return: a list with all parameters found, indicating the hypervisor they
+ apply to, and the origin (can be "cluster", "os X", or "instance Y")
"""
- HPATH = "cluster-verify"
- HTYPE = constants.HTYPE_CLUSTER
- REQ_BGL = False
+ hvp_data = []
+
+ for hv_name in cluster.enabled_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 hv_params:
+ 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 instances:
+ if instance.hvparams:
+ hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
+ cluster.FillHV(instance)))
+
+ return hvp_data
+
+
+class _VerifyErrors(object):
+ """Mix-in for cluster/group verify LUs.
+
+ It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
+ self.op and self._feedback_fn to be available.)
+ """
TCLUSTER = "cluster"
TNODE = "node"
TINSTANCE = "instance"
ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
+ ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
+ ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
ETYPE_ERROR = "ERROR"
ETYPE_WARNING = "WARNING"
+ 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.
+
+ """
+ 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: # This is a mix-in. pylint: disable-msg=E1101
+ 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) # Mix-in. pylint: disable-msg=E1101
+
+ 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) # pylint: disable-msg=E1101
+ 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
+
+
+class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
+ """Verifies the cluster config.
+
+ """
+ REQ_BGL = True
+
+ 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 ExpandNames(self):
+ # Information can be safely retrieved as the BGL is acquired in exclusive
+ # mode
+ self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
+ self.all_node_info = self.cfg.GetAllNodesInfo()
+ self.all_inst_info = self.cfg.GetAllInstancesInfo()
+ self.needed_locks = {}
+
+ def Exec(self, feedback_fn):
+ """Verify integrity of cluster, performing various test on nodes.
+
+ """
+ self.bad = False
+ self._feedback_fn = feedback_fn
+
+ feedback_fn("* Verifying cluster config")
+
+ for msg in self.cfg.VerifyConfig():
+ self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
+
+ feedback_fn("* Verifying cluster certificate files")
+
+ for cert_filename in constants.ALL_CERT_FILES:
+ (errcode, msg) = _VerifyCertificate(cert_filename)
+ self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
+
+ feedback_fn("* Verifying hypervisor parameters")
+
+ self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
+ self.all_inst_info.values()))
+
+ feedback_fn("* Verifying all nodes belong to an existing group")
+
+ # We do this verification here because, should this bogus circumstance
+ # occur, it would never be caught by VerifyGroup, which only acts on
+ # nodes/instances reachable from existing node groups.
+
+ dangling_nodes = set(node.name for node in self.all_node_info.values()
+ if node.group not in self.all_group_info)
+
+ dangling_instances = {}
+ no_node_instances = []
+
+ for inst in self.all_inst_info.values():
+ if inst.primary_node in dangling_nodes:
+ dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
+ elif inst.primary_node not in self.all_node_info:
+ no_node_instances.append(inst.name)
+
+ pretty_dangling = [
+ "%s (%s)" %
+ (node.name,
+ utils.CommaJoin(dangling_instances.get(node.name,
+ ["no instances"])))
+ for node in dangling_nodes]
+
+ self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
+ "the following nodes (and their instances) belong to a non"
+ " existing group: %s", utils.CommaJoin(pretty_dangling))
+
+ self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
+ "the following instances have a non-existing primary-node:"
+ " %s", utils.CommaJoin(no_node_instances))
+
+ return (not self.bad, [g.name for g in self.all_group_info.values()])
+
+
+class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
+ """Verifies the status of a node group.
+
+ """
+ HPATH = "cluster-verify"
+ HTYPE = constants.HTYPE_CLUSTER
+ REQ_BGL = False
+
_HOOKS_INDENT_RE = re.compile("^", re.M)
class NodeImage(object):
self.oslist = {}
def ExpandNames(self):
+ # This raises errors.OpPrereqError on its own:
+ self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+ # Get instances in node group; this is unsafe and needs verification later
+ inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
+
self.needed_locks = {
- locking.LEVEL_NODE: locking.ALL_SET,
- locking.LEVEL_INSTANCE: locking.ALL_SET,
- }
+ locking.LEVEL_INSTANCE: inst_names,
+ locking.LEVEL_NODEGROUP: [self.group_uuid],
+ locking.LEVEL_NODE: [],
+ }
+
self.share_locks = dict.fromkeys(locking.LEVELS, 1)
- def _Error(self, ecode, item, msg, *args, **kwargs):
- """Format an error message.
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ # Get members of node group; this is unsafe and needs verification later
+ nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
- Based on the opcode's error_codes parameter, either format a
- parseable error code, or a simpler error string.
+ all_inst_info = self.cfg.GetAllInstancesInfo()
- This must be called only from Exec and functions called from Exec.
+ # In Exec(), we warn about mirrored instances that have primary and
+ # secondary living in separate node groups. To fully verify that
+ # volumes for these instances are healthy, we will need to do an
+ # extra call to their secondaries. We ensure here those nodes will
+ # be locked.
+ for inst in self.glm.list_owned(locking.LEVEL_INSTANCE):
+ # Important: access only the instances whose lock is owned
+ if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
+ nodes.update(all_inst_info[inst].secondary_nodes)
- """
- 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)
+ self.needed_locks[locking.LEVEL_NODE] = nodes
- def _ErrorIf(self, cond, *args, **kwargs):
- """Log an error message if the passed condition is True.
+ def CheckPrereq(self):
+ group_nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
+ group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
- """
- 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
+ unlocked_nodes = \
+ group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+
+ unlocked_instances = \
+ group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
+
+ if unlocked_nodes:
+ raise errors.OpPrereqError("Missing lock for nodes: %s" %
+ utils.CommaJoin(unlocked_nodes))
+
+ if unlocked_instances:
+ raise errors.OpPrereqError("Missing lock for instances: %s" %
+ utils.CommaJoin(unlocked_instances))
+
+ self.all_node_info = self.cfg.GetAllNodesInfo()
+ self.all_inst_info = self.cfg.GetAllInstancesInfo()
+
+ self.my_node_names = utils.NiceSort(group_nodes)
+ self.my_inst_names = utils.NiceSort(group_instances)
+
+ self.my_node_info = dict((name, self.all_node_info[name])
+ for name in self.my_node_names)
+
+ self.my_inst_info = dict((name, self.all_inst_info[name])
+ for name in self.my_inst_names)
+
+ # We detect here the nodes that will need the extra RPC calls for verifying
+ # split LV volumes; they should be locked.
+ extra_lv_nodes = set()
+
+ for inst in self.my_inst_info.values():
+ if inst.disk_template in constants.DTS_INT_MIRROR:
+ group = self.my_node_info[inst.primary_node].group
+ for nname in inst.secondary_nodes:
+ if self.all_node_info[nname].group != group:
+ extra_lv_nodes.add(nname)
+
+ unlocked_lv_nodes = \
+ extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+
+ if unlocked_lv_nodes:
+ raise errors.OpPrereqError("these nodes could be locked: %s" %
+ utils.CommaJoin(unlocked_lv_nodes))
+ self.extra_lv_nodes = list(extra_lv_nodes)
def _VerifyNode(self, ninfo, nresult):
"""Perform some basic validation on data returned from a node.
ntime_diff)
def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
- """Check the node time.
+ """Check the node LVM results.
@type ninfo: L{objects.Node}
@param ninfo: the node to check
_ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
" '%s' of VG '%s'", pvname, owner_vg)
+ def _VerifyNodeBridges(self, ninfo, nresult, bridges):
+ """Check the node bridges.
+
+ @type ninfo: L{objects.Node}
+ @param ninfo: the node to check
+ @param nresult: the remote results for the node
+ @param bridges: the expected list of bridges
+
+ """
+ if not bridges:
+ return
+
+ node = ninfo.name
+ _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+
+ missing = nresult.get(constants.NV_BRIDGES, None)
+ test = not isinstance(missing, list)
+ _ErrorIf(test, self.ENODENET, node,
+ "did not return valid bridge information")
+ if not test:
+ _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
+ utils.CommaJoin(sorted(missing)))
+
def _VerifyNodeNetwork(self, ninfo, nresult):
- """Check the node time.
+ """Check the node network connectivity results.
@type ninfo: L{objects.Node}
@param ninfo: the node to check
"instance not running on its primary node %s",
node_current)
- for node, n_img in node_image.items():
- if node != node_current:
- test = instance in n_img.instances
- _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
- "instance should not run on node %s", node)
-
diskdata = [(nname, success, status, idx)
for (nname, disks) in diskstatus.items()
for idx, (success, status) in enumerate(disks)]
self._ErrorIf(test, self.ENODEORPHANLV, node,
"volume %s is unknown", volume)
- def _VerifyOrphanInstances(self, instancelist, node_image):
- """Verify the list of running instances.
-
- This checks what instances are running but unknown to the cluster.
-
- """
- for node, n_img in node_image.items():
- for o_inst in n_img.instances:
- 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_image, instance_cfg):
"""Verify N+1 Memory Resilience.
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.
the output be logged in the verify output and the verification to fail.
"""
- cfg = self.cfg
-
env = {
- "CLUSTER_TAGS": " ".join(cfg.GetClusterInfo().GetTags())
+ "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
}
env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
- for node in cfg.GetAllNodesInfo().values())
+ for node in self.my_node_info.values())
return env
"""Build hooks nodes.
"""
- return ([], self.cfg.GetNodeList())
+ assert self.my_node_names, ("Node list not gathered,"
+ " has CheckPrereq been executed?")
+ return ([], self.my_node_names)
def Exec(self, feedback_fn):
- """Verify integrity of cluster, performing various test on nodes.
+ """Verify integrity of the node group, performing various test on nodes.
"""
# This method has too many local variables. pylint: disable-msg=R0914
_ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
verbose = self.op.verbose
self._feedback_fn = feedback_fn
- feedback_fn("* Verifying global settings")
- for msg in self.cfg.VerifyConfig():
- _ErrorIf(True, self.ECLUSTERCFG, None, msg)
-
- # Check the cluster certificates
- for cert_filename in constants.ALL_CERT_FILES:
- (errcode, msg) = _VerifyCertificate(cert_filename)
- _ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
vg_name = self.cfg.GetVGName()
drbd_helper = self.cfg.GetDRBDHelper()
- hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
cluster = self.cfg.GetClusterInfo()
- nodelist = utils.NiceSort(self.cfg.GetNodeList())
- nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
- nodeinfo_byname = dict(zip(nodelist, nodeinfo))
- instancelist = utils.NiceSort(self.cfg.GetInstanceList())
- instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
- for iname in instancelist)
groupinfo = self.cfg.GetAllNodeGroupsInfo()
+ hypervisors = cluster.enabled_hypervisors
+ node_data_list = [self.my_node_info[name] for name in self.my_node_names]
+
i_non_redundant = [] # Non redundant instances
i_non_a_balanced = [] # Non auto-balanced instances
n_offline = 0 # Count of offline nodes
master_node = self.master_node = self.cfg.GetMasterNode()
master_ip = self.cfg.GetMasterIP()
- # 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(self.my_node_names))
+
+ # We will make nodes contact all nodes in their group, and one node from
+ # every other group.
+ # TODO: should it be a *random* node, different every time?
+ online_nodes = [node.name for node in node_data_list if not node.offline]
+ other_group_nodes = {}
+
+ for name in sorted(self.all_node_info):
+ node = self.all_node_info[name]
+ if (node.group not in other_group_nodes
+ and node.group != self.group_uuid
+ and not node.offline):
+ other_group_nodes[node.group] = node.name
- feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
node_verify_param = {
constants.NV_FILELIST:
utils.UniqueSequence(filename
for files in filemap
for filename in files),
- constants.NV_NODELIST: [node.name for node in nodeinfo
- if not node.offline],
+ constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
constants.NV_HYPERVISOR: hypervisors,
- constants.NV_HVPARAMS: hvp_data,
- constants.NV_NODENETTEST: [(node.name, node.primary_ip,
- node.secondary_ip) for node in nodeinfo
+ constants.NV_HVPARAMS:
+ _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
+ constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
+ for node in node_data_list
if not node.offline],
constants.NV_INSTANCELIST: hypervisors,
constants.NV_VERSION: None,
if drbd_helper:
node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
+ # bridge checks
+ # FIXME: this needs to be changed per node-group, not cluster-wide
+ bridges = set()
+ default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
+ if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+ bridges.add(default_nicpp[constants.NIC_LINK])
+ for instance in self.my_inst_info.values():
+ for nic in instance.nics:
+ full_nic = cluster.SimpleFillNIC(nic.nicparams)
+ if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
+ bridges.add(full_nic[constants.NIC_LINK])
+
+ if bridges:
+ node_verify_param[constants.NV_BRIDGES] = list(bridges)
+
# Build our expected cluster state
node_image = dict((node.name, self.NodeImage(offline=node.offline,
name=node.name,
vm_capable=node.vm_capable))
- for node in nodeinfo)
+ for node in node_data_list)
# Gather OOB paths
oob_paths = []
- for node in nodeinfo:
+ for node in self.all_node_info.values():
path = _SupportsOob(self.cfg, node)
if path and path not in oob_paths:
oob_paths.append(path)
if oob_paths:
node_verify_param[constants.NV_OOB_PATHS] = oob_paths
- for instance in instancelist:
- inst_config = instanceinfo[instance]
+ for instance in self.my_inst_names:
+ inst_config = self.my_inst_info[instance]
for nname in inst_config.all_nodes:
if nname not in node_image:
- # ghost node
gnode = self.NodeImage(name=nname)
- gnode.ghost = True
+ gnode.ghost = (nname not in self.all_node_info)
node_image[nname] = gnode
inst_config.MapLVsByNode(node_vol_should)
# time before and after executing the request, we can at least have a time
# window.
nvinfo_starttime = time.time()
- all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
+ all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
+ node_verify_param,
self.cfg.GetClusterName())
nvinfo_endtime = time.time()
+ if self.extra_lv_nodes and vg_name is not None:
+ extra_lv_nvinfo = \
+ self.rpc.call_node_verify(self.extra_lv_nodes,
+ {constants.NV_LVLIST: vg_name},
+ self.cfg.GetClusterName())
+ else:
+ extra_lv_nvinfo = {}
+
all_drbd_map = self.cfg.ComputeDRBDMap()
- feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
- instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
+ feedback_fn("* Gathering disk information (%s nodes)" %
+ len(self.my_node_names))
+ instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
+ self.my_inst_info)
feedback_fn("* Verifying configuration file consistency")
- self._VerifyFiles(_ErrorIf, nodeinfo, master_node, all_nvinfo, filemap)
+
+ # If not all nodes are being checked, we need to make sure the master node
+ # and a non-checked vm_capable node are in the list.
+ absent_nodes = set(self.all_node_info).difference(self.my_node_info)
+ if absent_nodes:
+ vf_nvinfo = all_nvinfo.copy()
+ vf_node_info = list(self.my_node_info.values())
+ additional_nodes = []
+ if master_node not in self.my_node_info:
+ additional_nodes.append(master_node)
+ vf_node_info.append(self.all_node_info[master_node])
+ # Add the first vm_capable node we find which is not included
+ for node in absent_nodes:
+ nodeinfo = self.all_node_info[node]
+ if nodeinfo.vm_capable and not nodeinfo.offline:
+ additional_nodes.append(node)
+ vf_node_info.append(self.all_node_info[node])
+ break
+ key = constants.NV_FILELIST
+ vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
+ {key: node_verify_param[key]},
+ self.cfg.GetClusterName()))
+ else:
+ vf_nvinfo = all_nvinfo
+ vf_node_info = self.my_node_info.values()
+
+ self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
feedback_fn("* Verifying node status")
refos_img = None
- for node_i in nodeinfo:
+ for node_i in node_data_list:
node = node_i.name
nimg = node_image[node]
if nimg.vm_capable:
self._VerifyNodeLVM(node_i, nresult, vg_name)
- self._VerifyNodeDrbd(node_i, nresult, instanceinfo, drbd_helper,
+ self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
all_drbd_map)
self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
self._UpdateNodeInstances(node_i, nresult, nimg)
self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
self._UpdateNodeOS(node_i, nresult, nimg)
+
if not nimg.os_fail:
if refos_img is None:
refos_img = nimg
self._VerifyNodeOS(node_i, nimg, refos_img)
+ self._VerifyNodeBridges(node_i, nresult, bridges)
+
+ # Check whether all running instancies are primary for the node. (This
+ # can no longer be done from _VerifyInstance below, since some of the
+ # wrong instances could be from other node groups.)
+ non_primary_inst = set(nimg.instances).difference(nimg.pinst)
+
+ for inst in non_primary_inst:
+ test = inst in self.all_inst_info
+ _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
+ "instance should not run on node %s", node_i.name)
+ _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
+ "node is running unknown instance %s", inst)
+
+ for node, result in extra_lv_nvinfo.items():
+ self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
+ node_image[node], vg_name)
feedback_fn("* Verifying instance status")
- for instance in instancelist:
+ for instance in self.my_inst_names:
if verbose:
feedback_fn("* Verifying instance %s" % instance)
- inst_config = instanceinfo[instance]
+ inst_config = self.my_inst_info[instance]
self._VerifyInstance(instance, inst_config, node_image,
instdisk[instance])
inst_nodes_offline = []
instance_groups = {}
for node in instance_nodes:
- instance_groups.setdefault(nodeinfo_byname[node].group,
+ instance_groups.setdefault(self.all_node_info[node].group,
[]).append(node)
pretty_list = [
feedback_fn("* Verifying orphan volumes")
reserved = utils.FieldSet(*cluster.reserved_lvs)
- self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
- feedback_fn("* Verifying orphan instances")
- self._VerifyOrphanInstances(instancelist, node_image)
+ # We will get spurious "unknown volume" warnings if any node of this group
+ # is secondary for an instance whose primary is in another group. To avoid
+ # them, we find these instances and add their volumes to node_vol_should.
+ for inst in self.all_inst_info.values():
+ for secondary in inst.secondary_nodes:
+ if (secondary in self.my_node_info
+ and inst.name not in self.my_inst_info):
+ inst.MapLVsByNode(node_vol_should)
+ break
+
+ self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
feedback_fn("* Verifying N+1 Memory redundancy")
- self._VerifyNPlusOneMemory(node_image, instanceinfo)
+ self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
feedback_fn("* Other Notes")
if i_non_redundant:
def ExpandNames(self):
if self.op.instances:
- self.wanted_names = []
- for name in self.op.instances:
- full_name = _ExpandInstanceName(self.cfg, name)
- self.wanted_names.append(full_name)
+ self.wanted_names = _GetWantedInstances(self, self.op.instances)
self.needed_locks = {
locking.LEVEL_NODE: [],
locking.LEVEL_INSTANCE: self.wanted_names,
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 DeclareLocks(self, level):
if level == locking.LEVEL_NODE and self.wanted_names is not None:
"""
if self.wanted_names is None:
- self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+ self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
in self.wanted_names]
" drbd-based instances exist",
errors.ECODE_INVAL)
- node_list = self.acquired_locks[locking.LEVEL_NODE]
+ node_list = self.glm.list_owned(locking.LEVEL_NODE)
# if vg_name not None, checks given volume group on all nodes
if self.op.vg_name:
# if we're moving instances to routed, check that they have an ip
target_mode = params_filled[constants.NIC_MODE]
if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
- nic_errors.append("Instance %s, nic/%d: routed nick with no ip" %
- (instance.name, nic_idx))
+ nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
+ " address" % (instance.name, nic_idx))
if nic_errors:
raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
"\n".join(nic_errors))
REG_BGL = False
_SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
+ def ExpandNames(self):
+ """Gather locks we need.
+
+ """
+ if self.op.node_names:
+ self.op.node_names = _GetWantedNodes(self, self.op.node_names)
+ lock_names = self.op.node_names
+ else:
+ lock_names = locking.ALL_SET
+
+ self.needed_locks = {
+ locking.LEVEL_NODE: lock_names,
+ }
+
def CheckPrereq(self):
"""Check prerequisites.
" not marked offline") % node_name,
errors.ECODE_STATE)
- def ExpandNames(self):
- """Gather locks we need.
-
- """
- if self.op.node_names:
- self.op.node_names = [_ExpandNodeName(self.cfg, name)
- for name in self.op.node_names]
- lock_names = self.op.node_names
- else:
- lock_names = locking.ALL_SET
-
- self.needed_locks = {
- locking.LEVEL_NODE: lock_names,
- }
-
def Exec(self, feedback_fn):
"""Execute OOB and return result if we expect any.
master_node = self.master_node
ret = []
- for idx, node in enumerate(self.nodes):
+ for idx, node in enumerate(utils.NiceSort(self.nodes,
+ key=lambda node: node.name)):
node_entry = [(constants.RS_NORMAL, node.name)]
ret.append(node_entry)
"""
# Locking is not used
- assert not (lu.acquired_locks or self.do_locking or self.use_locking)
+ assert not (compat.any(lu.glm.is_owned(level)
+ for level in locking.LEVELS
+ if level != locking.LEVEL_CLUSTER) or
+ self.do_locking or self.use_locking)
valid_nodes = [node.name
for node in lu.cfg.GetAllNodesInfo().values()
"""Computes the list of nodes and their attributes.
"""
- nodenames = self.acquired_locks[locking.LEVEL_NODE]
+ nodenames = self.glm.list_owned(locking.LEVEL_NODE)
volumes = self.rpc.call_node_volumes(nodenames)
ilist = [self.cfg.GetInstanceInfo(iname) for iname
"""Computes the list of nodes and their attributes.
"""
- self.nodes = self.acquired_locks[locking.LEVEL_NODE]
+ self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
# Always get name to sort by
if constants.SF_NAME in self.op.output_fields:
instances_keep = []
# Build list of instances to release
- for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
+ for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
instance = self.context.cfg.GetInstanceInfo(instance_name)
if (instance.disk_template in constants.DTS_INT_MIRROR and
self.op.node_name in instance.all_nodes):
_ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
- assert (set(self.acquired_locks.get(locking.LEVEL_INSTANCE, [])) ==
+ assert (set(self.glm.list_owned(locking.LEVEL_INSTANCE)) ==
set(instances_keep))
def BuildHooksEnv(self):
instance = self.instance
force = self.op.force
- self.cfg.MarkInstanceUp(instance.name)
+ if not self.op.no_remember:
+ self.cfg.MarkInstanceUp(instance.name)
if self.primary_offline:
assert self.op.ignore_offline_nodes
node_current = instance.primary_node
timeout = self.op.timeout
- self.cfg.MarkInstanceDown(instance.name)
+ if not self.op.no_remember:
+ self.cfg.MarkInstanceDown(instance.name)
if self.primary_offline:
assert self.op.ignore_offline_nodes
HTYPE = constants.HTYPE_INSTANCE
REQ_BGL = False
+ def CheckArguments(self):
+ # normalise the disk list
+ self.op.disks = sorted(frozenset(self.op.disks))
+
def ExpandNames(self):
self._ExpandAndLockInstance()
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ if self.op.nodes:
+ self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
+ self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
+ else:
+ self.needed_locks[locking.LEVEL_NODE] = []
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODE:
+ # if we replace the nodes, we only need to lock the old primary,
+ # otherwise we need to lock all nodes for disk re-creation
+ primary_only = bool(self.op.nodes)
+ self._LockInstancesNodes(primary_only=primary_only)
def BuildHooksEnv(self):
"""Build hooks env.
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 self.op.nodes:
+ if len(self.op.nodes) != len(instance.all_nodes):
+ raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
+ " %d replacement nodes were specified" %
+ (instance.name, len(instance.all_nodes),
+ len(self.op.nodes)),
+ errors.ECODE_INVAL)
+ assert instance.disk_template != constants.DT_DRBD8 or \
+ len(self.op.nodes) == 2
+ assert instance.disk_template != constants.DT_PLAIN or \
+ len(self.op.nodes) == 1
+ primary_node = self.op.nodes[0]
+ else:
+ primary_node = instance.primary_node
+ _CheckNodeOnline(self, primary_node)
if instance.disk_template == constants.DT_DISKLESS:
raise errors.OpPrereqError("Instance '%s' has no disks" %
self.op.instance_name, errors.ECODE_INVAL)
- _CheckInstanceDown(self, instance, "cannot recreate disks")
+ # 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]
+ old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
+ if not (self.op.nodes and old_pnode.offline):
+ _CheckInstanceDown(self, instance, "cannot recreate disks")
if not self.op.disks:
self.op.disks = range(len(instance.disks))
if idx >= len(instance.disks):
raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
errors.ECODE_INVAL)
-
+ if self.op.disks != range(len(instance.disks)) and self.op.nodes:
+ raise errors.OpPrereqError("Can't recreate disks partially and"
+ " change the nodes at the same time",
+ errors.ECODE_INVAL)
self.instance = instance
def Exec(self, feedback_fn):
"""Recreate the disks.
"""
+ # change primary node, if needed
+ if self.op.nodes:
+ self.instance.primary_node = self.op.nodes[0]
+ self.LogWarning("Changing the instance's nodes, you will have to"
+ " remove any disks left on the older nodes manually")
+
to_skip = []
- for idx, _ in enumerate(self.instance.disks):
+ 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
+ # update secondaries for disks, if needed
+ if self.op.nodes:
+ if disk.dev_type == constants.LD_DRBD8:
+ # need to update the nodes
+ assert len(self.op.nodes) == 2
+ logical_id = list(disk.logical_id)
+ logical_id[0] = self.op.nodes[0]
+ logical_id[1] = self.op.nodes[1]
+ disk.logical_id = tuple(logical_id)
+
+ if self.op.nodes:
+ self.cfg.Update(self.instance, feedback_fn)
_CreateDisks(self, self.instance, to_skip=to_skip)
new_name = self.op.new_name
if self.op.name_check:
hostname = netutils.GetHostname(name=new_name)
- self.LogInfo("Resolved given name '%s' to '%s'", new_name,
- hostname.name)
+ if hostname != new_name:
+ self.LogInfo("Resolved given name '%s' to '%s'", new_name,
+ hostname.name)
if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
" same as given hostname '%s'") %
REQ_BGL = False
def CheckArguments(self):
- _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
+ pass
def ExpandNames(self):
self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
- self.needed_locks = {}
-
- # Create tasklets for migrating instances for all instances on this node
- names = []
- tasklets = []
-
- self.lock_all_nodes = False
-
- for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
- logging.debug("Migrating instance %s", inst.name)
- names.append(inst.name)
-
- tasklets.append(TLMigrateInstance(self, inst.name, cleanup=False))
-
- if inst.disk_template in constants.DTS_EXT_MIRROR:
- # We need to lock all nodes, as the iallocator will choose the
- # destination nodes afterwards
- self.lock_all_nodes = True
-
- self.tasklets = tasklets
-
- # Declare node locks
- if self.lock_all_nodes:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
- else:
- self.needed_locks[locking.LEVEL_NODE] = [self.op.node_name]
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
-
- # Declare instance locks
- self.needed_locks[locking.LEVEL_INSTANCE] = names
-
- def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE and not self.lock_all_nodes:
- self._LockInstancesNodes()
+ self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ self.needed_locks = {
+ locking.LEVEL_NODE: [self.op.node_name],
+ }
def BuildHooksEnv(self):
"""Build hooks env.
nl = [self.cfg.GetMasterNode()]
return (nl, nl)
+ def CheckPrereq(self):
+ pass
+
+ def Exec(self, feedback_fn):
+ # Prepare jobs for migration instances
+ jobs = [
+ [opcodes.OpInstanceMigrate(instance_name=inst.name,
+ mode=self.op.mode,
+ live=self.op.live,
+ iallocator=self.op.iallocator,
+ target_node=self.op.target_node)]
+ for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
+ ]
+
+ # TODO: Run iallocator in this opcode and pass correct placement options to
+ # OpInstanceMigrate. Since other jobs can modify the cluster between
+ # running the iallocator and the actual migration, a good consistency model
+ # will have to be found.
+
+ assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+ frozenset([self.op.node_name]))
+
+ return ResultWithJobs(jobs)
+
class TLMigrateInstance(Tasklet):
"""Tasklet class for instance migration.
# self.target_node is already populated, either directly or by the
# iallocator run
target_node = self.target_node
+ if self.target_node == instance.primary_node:
+ raise errors.OpPrereqError("Cannot migrate instance %s"
+ " to its primary (%s)" %
+ (instance.name, instance.primary_node))
if len(self.lu.tasklets) == 1:
# It is safe to release locks only when we're the only tasklet
# in the LU
- _ReleaseLocks(self, locking.LEVEL_NODE,
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE,
keep=[instance.primary_node, self.target_node])
else:
mode=constants.IALLOCATOR_MODE_ALLOC,
name=self.op.instance_name,
disk_template=self.op.disk_template,
- tags=[],
+ tags=self.op.tags,
os=self.op.os_type,
vcpus=self.be_full[constants.BE_VCPUS],
- mem_size=self.be_full[constants.BE_MEMORY],
+ memory=self.be_full[constants.BE_MEMORY],
disks=self.disks,
nics=nics,
hypervisor=self.op.hypervisor,
bep=self.be_full,
hvp=self.hv_full,
hypervisor_name=self.op.hypervisor,
+ tags=self.op.tags,
))
return env
src_path = self.op.src_path
if src_node is None:
- locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
+ locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
exp_list = self.rpc.call_export_list(locked_nodes)
found = False
for node in exp_list:
nics.append(ndict)
self.op.nics = nics
+ if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
+ self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
+
if (self.op.hypervisor is None and
einfo.has_option(constants.INISECT_INS, "hypervisor")):
self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
+
if einfo.has_section(constants.INISECT_HYP):
# use the export parameters but do not override the ones
# specified by the user
",".join(enabled_hvs)),
errors.ECODE_STATE)
+ # Check tag validity
+ for tag in self.op.tags:
+ objects.TaggableObject.ValidateTag(tag)
+
# check hypervisor parameter syntax (locally)
utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
osparams=self.op.osparams,
)
+ if self.op.tags:
+ for tag in self.op.tags:
+ iobj.AddTag(tag)
+
if self.adopt_disks:
if self.op.disk_template == constants.DT_PLAIN:
# rename LVs to the newly-generated names; we need to construct
# Lock member nodes of all locked groups
self.needed_locks[locking.LEVEL_NODE] = [node_name
- for group_uuid in self.acquired_locks[locking.LEVEL_NODEGROUP]
+ for group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
for node_name in self.cfg.GetNodeGroup(group_uuid).members]
else:
self._LockInstancesNodes()
"""Check prerequisites.
"""
- assert (locking.LEVEL_NODEGROUP in self.acquired_locks or
+ assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
self.op.iallocator is None)
- if locking.LEVEL_NODEGROUP in self.acquired_locks:
+ owned_groups = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ if owned_groups:
groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
- prevgroups = self.acquired_locks[locking.LEVEL_NODEGROUP]
- if prevgroups != groups:
+ if owned_groups != groups:
raise errors.OpExecError("Node groups used by instance '%s' changed"
" since lock was acquired, current list is %r,"
" used to be '%s'" %
(self.op.instance_name,
utils.CommaJoin(groups),
- utils.CommaJoin(prevgroups)))
+ utils.CommaJoin(owned_groups)))
return LogicalUnit.CheckPrereq(self)
if remote_node is None:
self.remote_node_info = None
else:
- assert remote_node in self.lu.acquired_locks[locking.LEVEL_NODE], \
+ assert remote_node in self.lu.glm.list_owned(locking.LEVEL_NODE), \
"Remote node '%s' is not locked" % remote_node
self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
def Exec(self, feedback_fn):
+ instances = []
+ for node in self.op.nodes:
+ instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
+ if not instances:
+ return []
+
if self.op.remote_node is not None:
- instances = []
- for node in self.op.nodes:
- instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
result = []
for i in instances:
if i.primary_node == self.op.remote_node:
if not disks_ok:
raise errors.OpExecError("Cannot activate block device to grow")
+ # First run all grow ops in dry-run mode
for node in instance.all_nodes:
self.cfg.SetDiskID(disk, node)
- result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
+ result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, True)
+ result.Raise("Grow request failed to node %s" % node)
+
+ # We know that (as far as we can test) operations across different
+ # nodes will succeed, time to run it for real
+ for node in instance.all_nodes:
+ self.cfg.SetDiskID(disk, node)
+ result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, False)
result.Raise("Grow request failed to node %s" % node)
# TODO: Rewrite code to work properly
"""
if self.wanted_names is None:
assert self.op.use_locking, "Locking was not used"
- self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+ self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
self.wanted_instances = [self.cfg.GetInstanceInfo(name)
for name in self.wanted_names]
self.be_inst = i_bedict # the new dict (without defaults)
else:
self.be_new = self.be_inst = {}
+ be_old = cluster.FillBE(instance)
# osparams processing
if self.op.osparams:
self.warn = []
- if constants.BE_MEMORY in self.op.beparams and not self.op.force:
+ if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
+ be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
mem_check_list = [pnode]
if be_new[constants.BE_AUTO_BALANCE]:
# either we changed auto_balance to yes or it was from before
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)
+ nres.Raise("Can't get info from secondary node %s" % node,
+ prereq=True, ecode=errors.ECODE_STATE)
+ if not isinstance(nres.payload.get('memory_free', None), int):
+ raise errors.OpPrereqError("Secondary node %s didn't return free"
+ " memory information" % node,
+ errors.ECODE_STATE)
elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
- self.warn.append("Not enough memory to failover instance to"
- " secondary node %s" % node)
+ raise errors.OpPrereqError("This change will prevent the instance"
+ " from failover to its secondary node"
+ " %s, due to not enough memory" % node,
+ errors.ECODE_STATE)
# NIC processing
self.nic_pnew = {}
self.cfg.Update(instance, feedback_fn)
# disks are created, waiting for sync
- disk_abort = not _WaitForSync(self, instance)
+ disk_abort = not _WaitForSync(self, instance,
+ oneshot=not self.op.wait_for_sync)
if disk_abort:
raise errors.OpExecError("There are some degraded disks for"
" this instance, please cleanup manually")
that node.
"""
- self.nodes = self.acquired_locks[locking.LEVEL_NODE]
+ self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
rpcresult = self.rpc.call_export_list(self.nodes)
result = {}
for node in rpcresult:
fqdn_warn = True
instance_name = self.op.instance_name
- locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
+ locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
exportlist = self.rpc.call_export_list(locked_nodes)
found = False
for node in exportlist:
# We want to lock all the affected nodes and groups. We have readily
# available the list of nodes, and the *destination* group. To gather the
- # list of "source" groups, we need to fetch node information.
- self.node_data = self.cfg.GetAllNodesInfo()
- affected_groups = set(self.node_data[node].group for node in self.op.nodes)
- affected_groups.add(self.group_uuid)
-
+ # list of "source" groups, we need to fetch node information later on.
self.needed_locks = {
- locking.LEVEL_NODEGROUP: list(affected_groups),
+ locking.LEVEL_NODEGROUP: set([self.group_uuid]),
locking.LEVEL_NODE: self.op.nodes,
}
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
+
+ # Try to get all affected nodes' groups without having the group or node
+ # lock yet. Needs verification later in the code flow.
+ groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
+
+ self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
+
def CheckPrereq(self):
"""Check prerequisites.
"""
+ assert self.needed_locks[locking.LEVEL_NODEGROUP]
+ assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+ frozenset(self.op.nodes))
+
+ expected_locks = (set([self.group_uuid]) |
+ self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
+ actual_locks = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ if actual_locks != expected_locks:
+ raise errors.OpExecError("Nodes changed groups since locks were acquired,"
+ " current groups are '%s', used to be '%s'" %
+ (utils.CommaJoin(expected_locks),
+ utils.CommaJoin(actual_locks)))
+
+ self.node_data = self.cfg.GetAllNodesInfo()
self.group = self.cfg.GetNodeGroup(self.group_uuid)
instance_data = self.cfg.GetAllInstancesInfo()
for node in self.op.nodes:
self.node_data[node].group = self.group_uuid
+ # FIXME: Depends on side-effects of modifying the result of
+ # C{cfg.GetAllNodesInfo}
+
self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
@staticmethod
"""
# pylint: disable-msg=R0902
# lots of instance attributes
- _ALLO_KEYS = [
- "name", "mem_size", "disks", "disk_template",
- "os", "tags", "nics", "vcpus", "hypervisor",
- ]
- _RELO_KEYS = [
- "name", "relocate_from",
- ]
- _EVAC_KEYS = [
- "evac_nodes",
- ]
def __init__(self, cfg, rpc, mode, **kwargs):
self.cfg = cfg
self.in_text = self.out_text = self.in_data = self.out_data = None
# init all input fields so that pylint is happy
self.mode = mode
- self.mem_size = self.disks = self.disk_template = None
+ self.memory = self.disks = self.disk_template = None
self.os = self.tags = self.nics = self.vcpus = None
self.hypervisor = None
self.relocate_from = None
self.name = None
self.evac_nodes = None
+ self.instances = None
+ self.evac_mode = None
+ self.target_groups = []
# computed fields
self.required_nodes = None
# init result fields
self.success = self.info = self.result = None
- if self.mode == constants.IALLOCATOR_MODE_ALLOC:
- keyset = self._ALLO_KEYS
- fn = self._AddNewInstance
- elif self.mode == constants.IALLOCATOR_MODE_RELOC:
- keyset = self._RELO_KEYS
- fn = self._AddRelocateInstance
- elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
- keyset = self._EVAC_KEYS
- fn = self._AddEvacuateNodes
- else:
+
+ try:
+ (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
+ except KeyError:
raise errors.ProgrammerError("Unknown mode '%s' passed to the"
" IAllocator" % self.mode)
+
+ keyset = [n for (n, _) in keydata]
+
for key in kwargs:
if key not in keyset:
raise errors.ProgrammerError("Invalid input parameter '%s' to"
if key not in kwargs:
raise errors.ProgrammerError("Missing input parameter '%s' to"
" IAllocator" % key)
- self._BuildInputData(fn)
+ self._BuildInputData(compat.partial(fn, self), keydata)
def _ComputeClusterData(self):
"""Compute the generic allocator input data.
hypervisor_name = self.hypervisor
elif self.mode == constants.IALLOCATOR_MODE_RELOC:
hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
- elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
+ else:
hypervisor_name = cluster_info.enabled_hypervisors[0]
node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
"""Compute node groups data.
"""
- ng = {}
- for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
- ng[guuid] = {
- "name": gdata.name,
- "alloc_policy": gdata.alloc_policy,
- }
+ ng = dict((guuid, {
+ "name": gdata.name,
+ "alloc_policy": gdata.alloc_policy,
+ })
+ for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
+
return ng
@staticmethod
@returns: a dict of name: (node dict, node config)
"""
- node_results = {}
- for ninfo in node_cfg.values():
- # fill in static (config-based) values
- pnr = {
- "tags": list(ninfo.GetTags()),
- "primary_ip": ninfo.primary_ip,
- "secondary_ip": ninfo.secondary_ip,
- "offline": ninfo.offline,
- "drained": ninfo.drained,
- "master_candidate": ninfo.master_candidate,
- "group": ninfo.group,
- "master_capable": ninfo.master_capable,
- "vm_capable": ninfo.vm_capable,
- }
-
- node_results[ninfo.name] = pnr
+ # fill in static (config-based) values
+ node_results = dict((ninfo.name, {
+ "tags": list(ninfo.GetTags()),
+ "primary_ip": ninfo.primary_ip,
+ "secondary_ip": ninfo.secondary_ip,
+ "offline": ninfo.offline,
+ "drained": ninfo.drained,
+ "master_candidate": ninfo.master_candidate,
+ "group": ninfo.group,
+ "master_capable": ninfo.master_capable,
+ "vm_capable": ninfo.vm_capable,
+ })
+ for ninfo in node_cfg.values())
return node_results
nic_data = []
for nic in iinfo.nics:
filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
- nic_dict = {"mac": nic.mac,
- "ip": nic.ip,
- "mode": filled_params[constants.NIC_MODE],
- "link": filled_params[constants.NIC_LINK],
- }
+ 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)
self.required_nodes = 2
else:
self.required_nodes = 1
+
request = {
"name": self.name,
"disk_template": self.disk_template,
"tags": self.tags,
"os": self.os,
"vcpus": self.vcpus,
- "memory": self.mem_size,
+ "memory": self.memory,
"disks": self.disks,
"disk_space_total": disk_space,
"nics": self.nics,
"required_nodes": self.required_nodes,
+ "hypervisor": self.hypervisor,
}
+
return request
def _AddRelocateInstance(self):
}
return request
- def _BuildInputData(self, fn):
+ def _AddNodeEvacuate(self):
+ """Get data for node-evacuate requests.
+
+ """
+ return {
+ "instances": self.instances,
+ "evac_mode": self.evac_mode,
+ }
+
+ def _AddChangeGroup(self):
+ """Get data for node-evacuate requests.
+
+ """
+ return {
+ "instances": self.instances,
+ "target_groups": self.target_groups,
+ }
+
+ def _BuildInputData(self, fn, keydata):
"""Build input data structures.
"""
request = fn()
request["type"] = self.mode
+ for keyname, keytype in keydata:
+ if keyname not in request:
+ raise errors.ProgrammerError("Request parameter %s is missing" %
+ keyname)
+ val = request[keyname]
+ if not keytype(val):
+ raise errors.ProgrammerError("Request parameter %s doesn't pass"
+ " validation, value %s, expected"
+ " type %s" % (keyname, val, keytype))
self.in_data["request"] = request
self.in_text = serializer.Dump(self.in_data)
+ _STRING_LIST = ht.TListOf(ht.TString)
+ _JOBSET_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
+ # pylint: disable-msg=E1101
+ # Class '...' has no 'OP_ID' member
+ "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
+ opcodes.OpInstanceMigrate.OP_ID,
+ opcodes.OpInstanceReplaceDisks.OP_ID])
+ })))
+ _MODE_DATA = {
+ constants.IALLOCATOR_MODE_ALLOC:
+ (_AddNewInstance,
+ [
+ ("name", ht.TString),
+ ("memory", ht.TInt),
+ ("disks", ht.TListOf(ht.TDict)),
+ ("disk_template", ht.TString),
+ ("os", ht.TString),
+ ("tags", _STRING_LIST),
+ ("nics", ht.TListOf(ht.TDict)),
+ ("vcpus", ht.TInt),
+ ("hypervisor", ht.TString),
+ ], ht.TList),
+ constants.IALLOCATOR_MODE_RELOC:
+ (_AddRelocateInstance,
+ [("name", ht.TString), ("relocate_from", _STRING_LIST)],
+ ht.TList),
+ constants.IALLOCATOR_MODE_MEVAC:
+ (_AddEvacuateNodes, [("evac_nodes", _STRING_LIST)],
+ ht.TListOf(ht.TAnd(ht.TIsLength(2), _STRING_LIST))),
+ constants.IALLOCATOR_MODE_NODE_EVAC:
+ (_AddNodeEvacuate, [
+ ("instances", _STRING_LIST),
+ ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
+ ], _JOBSET_LIST),
+ constants.IALLOCATOR_MODE_CHG_GROUP:
+ (_AddChangeGroup, [
+ ("instances", _STRING_LIST),
+ ("target_groups", _STRING_LIST),
+ ], _JOBSET_LIST),
+ }
+
def Run(self, name, validate=True, call_fn=None):
"""Run an instance allocator and return the results.
" missing key '%s'" % key)
setattr(self, key, rdict[key])
- if not isinstance(rdict["result"], list):
- raise errors.OpExecError("Can't parse iallocator results: 'result' key"
- " is not a list")
-
- if self.mode == constants.IALLOCATOR_MODE_RELOC:
- assert self.relocate_from is not None
- assert self.required_nodes == 1
+ if not self._result_check(self.result):
+ raise errors.OpExecError("Iallocator returned invalid result,"
+ " expected %s, got %s" %
+ (self._result_check, self.result),
+ errors.ECODE_INVAL)
+ if self.mode in (constants.IALLOCATOR_MODE_RELOC,
+ constants.IALLOCATOR_MODE_MEVAC):
node2group = dict((name, ndata["group"])
for (name, ndata) in self.in_data["nodes"].items())
fn = compat.partial(self._NodesToGroups, node2group,
self.in_data["nodegroups"])
- request_groups = fn(self.relocate_from)
- result_groups = fn(rdict["result"])
-
- if result_groups != request_groups:
- raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
- " differ from original groups (%s)" %
- (utils.CommaJoin(result_groups),
- utils.CommaJoin(request_groups)))
+ if self.mode == constants.IALLOCATOR_MODE_RELOC:
+ assert self.relocate_from is not None
+ assert self.required_nodes == 1
+
+ request_groups = fn(self.relocate_from)
+ result_groups = fn(rdict["result"])
+
+ if result_groups != request_groups:
+ raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
+ " differ from original groups (%s)" %
+ (utils.CommaJoin(result_groups),
+ utils.CommaJoin(request_groups)))
+ elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
+ request_groups = fn(self.evac_nodes)
+ for (instance_name, secnode) in self.result:
+ result_groups = fn([secnode])
+ if result_groups != request_groups:
+ raise errors.OpExecError("Iallocator returned new secondary node"
+ " '%s' (group '%s') for instance '%s'"
+ " which is not in original group '%s'" %
+ (secnode, utils.CommaJoin(result_groups),
+ instance_name,
+ utils.CommaJoin(request_groups)))
+ else:
+ raise errors.ProgrammerError("Unhandled mode '%s'" % self.mode)
self.out_data = rdict
"""
if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
- for attr in ["mem_size", "disks", "disk_template",
+ for attr in ["memory", "disks", "disk_template",
"os", "tags", "nics", "vcpus"]:
if not hasattr(self.op, attr):
raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
errors.ECODE_INVAL)
for row in self.op.disks:
if (not isinstance(row, dict) or
- "size" not in row or
- not isinstance(row["size"], int) or
- "mode" not in row or
- row["mode"] not in ['r', 'w']):
+ constants.IDISK_SIZE not in row or
+ not isinstance(row[constants.IDISK_SIZE], int) or
+ constants.IDISK_MODE not in row or
+ row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
raise errors.OpPrereqError("Invalid contents of the 'disks'"
" parameter", errors.ECODE_INVAL)
if self.op.hypervisor is None:
if not hasattr(self.op, "evac_nodes"):
raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
" opcode input", errors.ECODE_INVAL)
+ elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
+ constants.IALLOCATOR_MODE_NODE_EVAC):
+ if not self.op.instances:
+ raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
+ self.op.instances = _GetWantedInstances(self, self.op.instances)
else:
raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
self.op.mode, errors.ECODE_INVAL)
ial = IAllocator(self.cfg, self.rpc,
mode=self.op.mode,
name=self.op.name,
- mem_size=self.op.mem_size,
+ memory=self.op.memory,
disks=self.op.disks,
disk_template=self.op.disk_template,
os=self.op.os,
ial = IAllocator(self.cfg, self.rpc,
mode=self.op.mode,
evac_nodes=self.op.evac_nodes)
+ elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
+ ial = IAllocator(self.cfg, self.rpc,
+ mode=self.op.mode,
+ instances=self.op.instances,
+ target_groups=self.op.target_groups)
+ elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
+ ial = IAllocator(self.cfg, self.rpc,
+ mode=self.op.mode,
+ instances=self.op.instances,
+ evac_mode=self.op.evac_mode)
else:
raise errors.ProgrammerError("Uncatched mode %s in"
" LUTestAllocator.Exec", self.op.mode)
projects / ganeti-local / blobdiff