"""Module implementing the master-side code."""
-# pylint: disable-msg=W0201,C0302
+# pylint: disable=W0201,C0302
# W0201 since most LU attributes are defined in CheckPrereq or similar
# functions
from ganeti import opcodes
from ganeti import ht
-import ganeti.masterd.instance # pylint: disable-msg=W0611
+import ganeti.masterd.instance # pylint: disable=W0611
class ResultWithJobs:
self.op = op
self.cfg = context.cfg
self.glm = context.glm
+ # readability alias
+ self.owned_locks = context.glm.list_owned
self.context = context
self.rpc = rpc
# Dicts used to declare locking needs to mcpu
# Used to force good behavior when calling helper functions
self.recalculate_locks = {}
# logging
- self.Log = processor.Log # pylint: disable-msg=C0103
- self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
- self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
- self.LogStep = processor.LogStep # pylint: disable-msg=C0103
+ self.Log = processor.Log # pylint: disable=C0103
+ self.LogWarning = processor.LogWarning # pylint: disable=C0103
+ self.LogInfo = processor.LogInfo # pylint: disable=C0103
+ self.LogStep = processor.LogStep # pylint: disable=C0103
# support for dry-run
self.dry_run_result = None
# support for generic debug attribute
"""
# API must be kept, thus we ignore the unused argument and could
# be a function warnings
- # pylint: disable-msg=W0613,R0201
+ # pylint: disable=W0613,R0201
return lu_result
def _ExpandAndLockInstance(self):
# 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.glm.list_owned(locking.LEVEL_INSTANCE):
- instance = self.context.cfg.GetInstanceInfo(instance_name)
+ locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
+ for _, instance in self.cfg.GetMultiInstanceInfo(locked_i):
wanted_nodes.append(instance.primary_node)
if not primary_only:
wanted_nodes.extend(instance.secondary_nodes)
del self.recalculate_locks[locking.LEVEL_NODE]
-class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
+class NoHooksLU(LogicalUnit): # pylint: disable=W0223
"""Simple LU which runs no hooks.
This LU is intended as a parent for other LogicalUnits which will
"""
if self.do_locking:
- names = lu.glm.list_owned(lock_level)
+ names = lu.owned_locks(lock_level)
else:
names = all_names
return dict.fromkeys(locking.LEVELS, 1)
+def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
+ """Checks if the owned node groups are still correct for an instance.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: The cluster configuration
+ @type instance_name: string
+ @param instance_name: Instance name
+ @type owned_groups: set or frozenset
+ @param owned_groups: List of currently owned node groups
+
+ """
+ inst_groups = cfg.GetInstanceNodeGroups(instance_name)
+
+ if not owned_groups.issuperset(inst_groups):
+ raise errors.OpPrereqError("Instance %s's node groups changed since"
+ " locks were acquired, current groups are"
+ " are '%s', owning groups '%s'; retry the"
+ " operation" %
+ (instance_name,
+ utils.CommaJoin(inst_groups),
+ utils.CommaJoin(owned_groups)),
+ errors.ECODE_STATE)
+
+ return inst_groups
+
+
+def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
+ """Checks if the instances in a node group are still correct.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: The cluster configuration
+ @type group_uuid: string
+ @param group_uuid: Node group UUID
+ @type owned_instances: set or frozenset
+ @param owned_instances: List of currently owned instances
+
+ """
+ wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
+ if owned_instances != wanted_instances:
+ raise errors.OpPrereqError("Instances in node group '%s' changed since"
+ " locks were acquired, wanted '%s', have '%s';"
+ " retry the operation" %
+ (group_uuid,
+ utils.CommaJoin(wanted_instances),
+ utils.CommaJoin(owned_instances)),
+ errors.ECODE_STATE)
+
+ return wanted_instances
+
+
def _SupportsOob(cfg, node):
"""Tells if node supports OOB.
release = []
# Determine which locks to release
- for name in lu.glm.list_owned(level):
+ for name in lu.owned_locks(level):
if should_release(name):
release.append(name)
else:
retain.append(name)
- assert len(lu.glm.list_owned(level)) == (len(retain) + len(release))
+ assert len(lu.owned_locks(level)) == (len(retain) + len(release))
# Release just some locks
lu.glm.release(level, names=release)
- assert frozenset(lu.glm.list_owned(level)) == frozenset(retain)
+ assert frozenset(lu.owned_locks(level)) == frozenset(retain)
else:
# Release everything
lu.glm.release(level)
try:
hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
except:
- # pylint: disable-msg=W0702
+ # pylint: disable=W0702
lu.LogWarning("Errors occurred running hooks on %s" % node_name)
}
if override:
args.update(override)
- return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
+ return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
def _AdjustCandidatePool(lu, exceptions):
" iallocator")
+def _GetDefaultIAllocator(cfg, iallocator):
+ """Decides on which iallocator to use.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: Cluster configuration object
+ @type iallocator: string or None
+ @param iallocator: Iallocator specified in opcode
+ @rtype: string
+ @return: Iallocator name
+
+ """
+ if not iallocator:
+ # Use default iallocator
+ iallocator = cfg.GetDefaultIAllocator()
+
+ if not iallocator:
+ raise errors.OpPrereqError("No iallocator was specified, neither in the"
+ " opcode nor as a cluster-wide default",
+ errors.ECODE_INVAL)
+
+ return iallocator
+
+
class LUClusterPostInit(LogicalUnit):
"""Logical unit for running hooks after cluster initialization.
# Run post hooks on master node before it's removed
_RunPostHook(self, master)
- result = self.rpc.call_node_stop_master(master, False)
+ result = self.rpc.call_node_deactivate_master_ip(master)
result.Raise("Could not disable the master role")
return master
try:
cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
utils.ReadFile(filename))
- except Exception, err: # pylint: disable-msg=W0703
+ except Exception, err: # pylint: disable=W0703
return (LUClusterVerifyConfig.ETYPE_ERROR,
"Failed to load X509 certificate %s: %s" % (filename, err))
if args:
msg = msg % args
# then format the whole message
- if self.op.error_codes: # This is a mix-in. pylint: disable-msg=E1101
+ if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
else:
if item:
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
+ self._feedback_fn(" - %s" % msg) # Mix-in. pylint: disable=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
+ or self.op.debug_simulate_errors) # pylint: disable=E1101
if cond:
self._Error(*args, **kwargs)
# do not mark the operation as failed for WARN cases only
self.bad = self.bad or cond
+class LUClusterVerify(NoHooksLU):
+ """Submits all jobs necessary to verify the cluster.
+
+ """
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.needed_locks = {}
+
+ def Exec(self, feedback_fn):
+ jobs = []
+
+ if self.op.group_name:
+ groups = [self.op.group_name]
+ depends_fn = lambda: None
+ else:
+ groups = self.cfg.GetNodeGroupList()
+
+ # Verify global configuration
+ jobs.append([opcodes.OpClusterVerifyConfig()])
+
+ # Always depend on global verification
+ depends_fn = lambda: [(-len(jobs), [])]
+
+ jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
+ depends=depends_fn())]
+ for group in groups)
+
+ # Fix up all parameters
+ for op in itertools.chain(*jobs): # pylint: disable=W0142
+ op.debug_simulate_errors = self.op.debug_simulate_errors
+ op.verbose = self.op.verbose
+ op.error_codes = self.op.error_codes
+ try:
+ op.skip_checks = self.op.skip_checks
+ except AttributeError:
+ assert not isinstance(op, opcodes.OpClusterVerifyGroup)
+
+ return ResultWithJobs(jobs)
+
+
class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
"""Verifies the cluster config.
def ExpandNames(self):
# Information can be safely retrieved as the BGL is acquired in exclusive
# mode
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
self.all_node_info = self.cfg.GetAllNodesInfo()
self.all_inst_info = self.cfg.GetAllInstancesInfo()
"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()])
+ return not self.bad
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
# 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):
+ for inst in self.owned_locks(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)
self.needed_locks[locking.LEVEL_NODE] = nodes
def CheckPrereq(self):
- group_nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
+ assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+ self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
+
+ group_nodes = set(self.group_info.members)
group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
unlocked_nodes = \
- group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+ group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
unlocked_instances = \
- group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
+ group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
if unlocked_nodes:
raise errors.OpPrereqError("Missing lock for nodes: %s" %
extra_lv_nodes.add(nname)
unlocked_lv_nodes = \
- extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+ extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
if unlocked_lv_nodes:
raise errors.OpPrereqError("these nodes could be locked: %s" %
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
# main result, nresult should be a non-empty dict
test = not nresult or not isinstance(nresult, dict)
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
ntime = nresult.get(constants.NV_TIME, None)
try:
return
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
# checks vg existence and size > 20G
vglist = nresult.get(constants.NV_VGLIST, None)
return
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
missing = nresult.get(constants.NV_BRIDGES, None)
test = not isinstance(missing, list)
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
test = constants.NV_NODELIST not in nresult
_ErrorIf(test, self.ENODESSH, node,
available on the instance's node.
"""
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
node_current = instanceconfig.primary_node
node_vol_should = {}
@classmethod
def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
- (files_all, files_all_opt, files_mc, files_vm)):
+ (files_all, files_opt, files_mc, files_vm)):
"""Verifies file checksums collected from all nodes.
@param errorif: Callback for reporting errors
@param all_nvinfo: RPC results
"""
- node_names = frozenset(node.name for node in nodeinfo if not node.offline)
+ # Define functions determining which nodes to consider for a file
+ files2nodefn = [
+ (files_all, None),
+ (files_mc, lambda node: (node.master_candidate or
+ node.name == master_node)),
+ (files_vm, lambda node: node.vm_capable),
+ ]
- assert master_node in node_names
- assert (len(files_all | files_all_opt | files_mc | files_vm) ==
- sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
- "Found file listed in more than one file list"
+ # Build mapping from filename to list of nodes which should have the file
+ nodefiles = {}
+ for (files, fn) in files2nodefn:
+ if fn is None:
+ filenodes = nodeinfo
+ else:
+ filenodes = filter(fn, nodeinfo)
+ nodefiles.update((filename,
+ frozenset(map(operator.attrgetter("name"), filenodes)))
+ for filename in files)
- # Define functions determining which nodes to consider for a file
- file2nodefn = dict([(filename, fn)
- for (files, fn) in [(files_all, None),
- (files_all_opt, None),
- (files_mc, lambda node: (node.master_candidate or
- node.name == master_node)),
- (files_vm, lambda node: node.vm_capable)]
- for filename in files])
+ assert set(nodefiles) == (files_all | files_mc | files_vm)
- fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
+ fileinfo = dict((filename, {}) for filename in nodefiles)
+ ignore_nodes = set()
for node in nodeinfo:
if node.offline:
+ ignore_nodes.add(node.name)
continue
nresult = all_nvinfo[node.name]
errorif(test, cls.ENODEFILECHECK, node.name,
"Node did not return file checksum data")
if test:
+ ignore_nodes.add(node.name)
continue
+ # Build per-checksum mapping from filename to nodes having it
for (filename, checksum) in node_files.items():
- # Check if the file should be considered for a node
- fn = file2nodefn[filename]
- if fn is None or fn(node):
- fileinfo[filename].setdefault(checksum, set()).add(node.name)
+ assert filename in nodefiles
+ fileinfo[filename].setdefault(checksum, set()).add(node.name)
for (filename, checksums) in fileinfo.items():
assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
# Nodes having the file
with_file = frozenset(node_name
for nodes in fileinfo[filename].values()
- for node_name in nodes)
+ for node_name in nodes) - ignore_nodes
+
+ expected_nodes = nodefiles[filename] - ignore_nodes
# Nodes missing file
- missing_file = node_names - with_file
+ missing_file = expected_nodes - with_file
- if filename in files_all_opt:
+ if filename in files_opt:
# All or no nodes
- errorif(missing_file and missing_file != node_names,
+ errorif(missing_file and missing_file != expected_nodes,
cls.ECLUSTERFILECHECK, None,
"File %s is optional, but it must exist on all or no"
" nodes (not found on %s)",
filename, utils.CommaJoin(utils.NiceSort(missing_file)))
else:
+ # Non-optional files
errorif(missing_file, cls.ECLUSTERFILECHECK, None,
"File %s is missing from node(s) %s", filename,
utils.CommaJoin(utils.NiceSort(missing_file)))
+ # Warn if a node has a file it shouldn't
+ unexpected = with_file - expected_nodes
+ errorif(unexpected,
+ cls.ECLUSTERFILECHECK, None,
+ "File %s should not exist on node(s) %s",
+ filename, utils.CommaJoin(utils.NiceSort(unexpected)))
+
# See if there are multiple versions of the file
test = len(checksums) > 1
if test:
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
if drbd_helper:
helper_result = nresult.get(constants.NV_DRBDHELPER, None)
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
remote_os = nresult.get(constants.NV_OSLIST, None)
test = (not isinstance(remote_os, list) or
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
nimg.lvm_fail = True
lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
"""
node = ninfo.name
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
# try to read free memory (from the hypervisor)
hv_info = nresult.get(constants.NV_HVINFO, None)
list of tuples (success, payload)
"""
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
node_disks = {}
node_disks_devonly = {}
return instdisk
+ @staticmethod
+ def _SshNodeSelector(group_uuid, all_nodes):
+ """Create endless iterators for all potential SSH check hosts.
+
+ """
+ nodes = [node for node in all_nodes
+ if (node.group != group_uuid and
+ not node.offline)]
+ keyfunc = operator.attrgetter("group")
+
+ return map(itertools.cycle,
+ [sorted(map(operator.attrgetter("name"), names))
+ for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
+ keyfunc)])
+
+ @classmethod
+ def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
+ """Choose which nodes should talk to which other nodes.
+
+ We will make nodes contact all nodes in their group, and one node from
+ every other group.
+
+ @warning: This algorithm has a known issue if one node group is much
+ smaller than others (e.g. just one node). In such a case all other
+ nodes will talk to the single node.
+
+ """
+ online_nodes = sorted(node.name for node in group_nodes if not node.offline)
+ sel = cls._SshNodeSelector(group_uuid, all_nodes)
+
+ return (online_nodes,
+ dict((name, sorted([i.next() for i in sel]))
+ for name in online_nodes))
+
def BuildHooksEnv(self):
"""Build hooks env.
"""Verify integrity of the node group, performing various test on nodes.
"""
- # This method has too many local variables. pylint: disable-msg=R0914
+ # This method has too many local variables. pylint: disable=R0914
+ feedback_fn("* Verifying group '%s'" % self.group_info.name)
if not self.my_node_names:
# empty node group
return True
self.bad = False
- _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+ _ErrorIf = self._ErrorIf # pylint: disable=C0103
verbose = self.op.verbose
self._feedback_fn = feedback_fn
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
-
node_verify_param = {
constants.NV_FILELIST:
utils.UniqueSequence(filename
for files in filemap
for filename in files),
- constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
+ constants.NV_NODELIST:
+ self._SelectSshCheckNodes(node_data_list, self.group_uuid,
+ self.all_node_info.values()),
constants.NV_HYPERVISOR: hypervisors,
constants.NV_HVPARAMS:
_GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
self._ErrorIf(test, self.ENODEHOOKS, node_name,
"Communication failure in hooks execution: %s", msg)
if res.offline or msg:
- # No need to investigate payload if node is offline or gave an error.
- # override manually lu_result here as _ErrorIf only
- # overrides self.bad
- lu_result = 1
+ # No need to investigate payload if node is offline or gave
+ # an error.
continue
for script, hkr, output in res.payload:
test = hkr == constants.HKR_FAIL
if test:
output = self._HOOKS_INDENT_RE.sub(" ", output)
feedback_fn("%s" % output)
- lu_result = 0
+ lu_result = False
return lu_result
}
def Exec(self, feedback_fn):
- group_names = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
# Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
# going via the node before it's locked, requiring verification
# later on
[group_uuid
- for instance_name in
- self.glm.list_owned(locking.LEVEL_INSTANCE)
- for group_uuid in
- self.cfg.GetInstanceNodeGroups(instance_name)])
+ for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
+ for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
elif level == locking.LEVEL_NODE:
# This will only lock the nodes in the group to be verified which contain
self._LockInstancesNodes()
# Lock all nodes in group to be verified
- assert self.group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
def CheckPrereq(self):
- owned_instances = frozenset(self.glm.list_owned(locking.LEVEL_INSTANCE))
- owned_groups = frozenset(self.glm.list_owned(locking.LEVEL_NODEGROUP))
- owned_nodes = frozenset(self.glm.list_owned(locking.LEVEL_NODE))
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
assert self.group_uuid in owned_groups
# Check if locked instances are still correct
- wanted_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
- if owned_instances != wanted_instances:
- raise errors.OpPrereqError("Instances in node group %s changed since"
- " locks were acquired, wanted %s, have %s;"
- " retry the operation" %
- (self.op.group_name,
- utils.CommaJoin(wanted_instances),
- utils.CommaJoin(owned_instances)),
- errors.ECODE_STATE)
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
# Get instance information
- self.instances = dict((name, self.cfg.GetInstanceInfo(name))
- for name in owned_instances)
+ self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
# Check if node groups for locked instances are still correct
for (instance_name, inst) in self.instances.items():
- assert self.group_uuid in self.cfg.GetInstanceNodeGroups(instance_name), \
- "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
assert owned_nodes.issuperset(inst.all_nodes), \
"Instance %s's nodes changed while we kept the lock" % instance_name
- inst_groups = self.cfg.GetInstanceNodeGroups(instance_name)
- if not owned_groups.issuperset(inst_groups):
- raise errors.OpPrereqError("Instance %s's node groups changed since"
- " locks were acquired, current groups are"
- " are '%s', owning groups '%s'; retry the"
- " operation" %
- (instance_name,
- utils.CommaJoin(inst_groups),
- utils.CommaJoin(owned_groups)),
- errors.ECODE_STATE)
+ inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
+ owned_groups)
+
+ assert self.group_uuid in inst_groups, \
+ "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
def Exec(self, feedback_fn):
"""Verify integrity of cluster disks.
if inst.admin_up])
if nv_dict:
- nodes = utils.NiceSort(set(self.glm.list_owned(locking.LEVEL_NODE)) &
+ nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
set(self.cfg.GetVmCapableNodeList()))
node_lvs = self.rpc.call_lv_list(nodes, [])
"""
if self.wanted_names is None:
- self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
+ self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
- self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
- in self.wanted_names]
+ self.wanted_instances = \
+ map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
def _EnsureChildSizes(self, disk):
"""Ensure children of the disk have the needed disk size.
# shutdown the master IP
master = self.cfg.GetMasterNode()
- result = self.rpc.call_node_stop_master(master, False)
+ result = self.rpc.call_node_deactivate_master_ip(master)
result.Raise("Could not disable the master role")
try:
pass
_UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
finally:
- result = self.rpc.call_node_start_master(master, False, False)
+ result = self.rpc.call_node_activate_master_ip(master)
msg = result.fail_msg
if msg:
self.LogWarning("Could not re-enable the master role on"
" drbd-based instances exist",
errors.ECODE_INVAL)
- node_list = self.glm.list_owned(locking.LEVEL_NODE)
+ node_list = self.owned_locks(locking.LEVEL_NODE)
# if vg_name not None, checks given volume group on all nodes
if self.op.vg_name:
master = self.cfg.GetMasterNode()
feedback_fn("Shutting down master ip on the current netdev (%s)" %
self.cluster.master_netdev)
- result = self.rpc.call_node_stop_master(master, False)
+ result = self.rpc.call_node_deactivate_master_ip(master)
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))
if self.op.master_netdev:
feedback_fn("Starting the master ip on the new master netdev (%s)" %
self.op.master_netdev)
- result = self.rpc.call_node_start_master(master, False, False)
+ result = self.rpc.call_node_activate_master_ip(master)
if result.fail_msg:
self.LogWarning("Could not re-enable the master ip on"
" the master, please restart manually: %s",
constants.SSH_KNOWN_HOSTS_FILE,
constants.CONFD_HMAC_KEY,
constants.CLUSTER_DOMAIN_SECRET_FILE,
+ constants.RAPI_USERS_FILE,
])
if not redist:
files_all.update(constants.ALL_CERT_FILES)
files_all.update(ssconf.SimpleStore().GetFileList())
+ else:
+ # we need to ship at least the RAPI certificate
+ files_all.add(constants.RAPI_CERT_FILE)
if cluster.modify_etc_hosts:
files_all.add(constants.ETC_HOSTS)
- # Files which must either exist on all nodes or on none
- files_all_opt = set([
+ # Files which are optional, these must:
+ # - be present in one other category as well
+ # - either exist or not exist on all nodes of that category (mc, vm all)
+ files_opt = set([
constants.RAPI_USERS_FILE,
])
# Files which should only be on VM-capable nodes
files_vm = set(filename
for hv_name in cluster.enabled_hypervisors
- for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
+
+ files_opt |= set(filename
+ for hv_name in cluster.enabled_hypervisors
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
- # Filenames must be unique
- assert (len(files_all | files_all_opt | files_mc | files_vm) ==
- sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
+ # Filenames in each category must be unique
+ all_files_set = files_all | files_mc | files_vm
+ assert (len(all_files_set) ==
+ sum(map(len, [files_all, files_mc, files_vm]))), \
"Found file listed in more than one file list"
- return (files_all, files_all_opt, files_mc, files_vm)
+ # Optional files must be present in one other category
+ assert all_files_set.issuperset(files_opt), \
+ "Optional file not in a different required list"
+
+ return (files_all, files_opt, files_mc, files_vm)
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
nodelist.remove(master_info.name)
# Gather file lists
- (files_all, files_all_opt, files_mc, files_vm) = \
+ (files_all, _, files_mc, files_vm) = \
_ComputeAncillaryFiles(cluster, True)
# Never re-distribute configuration file from here
filemap = [
(online_nodes, files_all),
- (online_nodes, files_all_opt),
(vm_nodes, files_vm),
]
_RedistributeAncillaryFiles(self)
+class LUClusterActivateMasterIp(NoHooksLU):
+ """Activate the master IP on the master node.
+
+ """
+ def Exec(self, feedback_fn):
+ """Activate the master IP.
+
+ """
+ master = self.cfg.GetMasterNode()
+ self.rpc.call_node_activate_master_ip(master)
+
+
+class LUClusterDeactivateMasterIp(NoHooksLU):
+ """Deactivate the master IP on the master node.
+
+ """
+ def Exec(self, feedback_fn):
+ """Deactivate the master IP.
+
+ """
+ master = self.cfg.GetMasterNode()
+ self.rpc.call_node_deactivate_master_ip(master)
+
+
def _WaitForSync(lu, instance, disks=None, oneshot=False):
"""Sleep and poll for an instance's disk to sync.
raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
utils.CommaJoin(errs))
+
class _OsQuery(_QueryBase):
FIELDS = query.OS_FIELDS
node = self.cfg.GetNodeInfo(self.op.node_name)
assert node is not None
- instance_list = self.cfg.GetInstanceList()
-
masternode = self.cfg.GetMasterNode()
if node.name == masternode:
raise errors.OpPrereqError("Node is the master node, failover to another"
" node is required", errors.ECODE_INVAL)
- for instance_name in instance_list:
- instance = self.cfg.GetInstanceInfo(instance_name)
+ for instance_name, instance in self.cfg.GetAllInstancesInfo():
if node.name in instance.all_nodes:
raise errors.OpPrereqError("Instance %s is still running on the node,"
" please remove first" % instance_name,
def ExpandNames(self, lu):
lu.needed_locks = {}
- lu.share_locks[locking.LEVEL_NODE] = 1
+ lu.share_locks = _ShareAll()
if self.names:
self.wanted = _GetWantedNodes(lu, self.names)
query.NQ_LIVE in self.requested_data)
if self.do_locking:
- # if we don't request only static fields, we need to lock the nodes
+ # If any non-static field is requested we need to lock the nodes
lu.needed_locks[locking.LEVEL_NODE] = self.wanted
def DeclareLocks(self, lu, level):
"""Logical unit for querying nodes.
"""
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
REQ_BGL = False
def CheckArguments(self):
"""Computes the list of nodes and their attributes.
"""
- nodenames = self.glm.list_owned(locking.LEVEL_NODE)
+ nodenames = self.owned_locks(locking.LEVEL_NODE)
volumes = self.rpc.call_node_volumes(nodenames)
ilist = self.cfg.GetAllInstancesInfo()
"""Computes the list of nodes and their attributes.
"""
- self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
+ self.nodes = self.owned_locks(locking.LEVEL_NODE)
# Always get name to sort by
if constants.SF_NAME in self.op.output_fields:
# via the node before it's locked, requiring verification later on
lu.needed_locks[locking.LEVEL_NODEGROUP] = \
set(group_uuid
- for instance_name in
- lu.glm.list_owned(locking.LEVEL_INSTANCE)
- for group_uuid in
- lu.cfg.GetInstanceNodeGroups(instance_name))
+ for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
+ for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
elif level == locking.LEVEL_NODE:
- lu._LockInstancesNodes() # pylint: disable-msg=W0212
+ lu._LockInstancesNodes() # pylint: disable=W0212
@staticmethod
def _CheckGroupLocks(lu):
- owned_instances = frozenset(lu.glm.list_owned(locking.LEVEL_INSTANCE))
- owned_groups = frozenset(lu.glm.list_owned(locking.LEVEL_NODEGROUP))
+ owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
+ owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
# Check if node groups for locked instances are still correct
for instance_name in owned_instances:
- inst_groups = lu.cfg.GetInstanceNodeGroups(instance_name)
- if not owned_groups.issuperset(inst_groups):
- raise errors.OpPrereqError("Instance %s's node groups changed since"
- " locks were acquired, current groups are"
- " are '%s', owning groups '%s'; retry the"
- " operation" %
- (instance_name,
- utils.CommaJoin(inst_groups),
- utils.CommaJoin(owned_groups)),
- errors.ECODE_STATE)
+ _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
def _GetQueryData(self, lu):
"""Computes the list of instances and their attributes.
"""Query for resources/items of a certain kind.
"""
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
REQ_BGL = False
def CheckArguments(self):
qcls = _GetQueryImplementation(self.op.what)
- self.impl = qcls(self.op.filter, self.op.fields, False)
+ self.impl = qcls(self.op.filter, self.op.fields, self.op.use_locking)
def ExpandNames(self):
self.impl.ExpandNames(self)
"""Query for resources/items of a certain kind.
"""
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
REQ_BGL = False
def CheckArguments(self):
# later in the procedure; this also means that if the re-add
# fails, we are left with a non-offlined, broken node
if self.op.readd:
- new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
+ new_node.drained = new_node.offline = False # pylint: disable=W0201
self.LogInfo("Readding a node, the offline/drained flags were reset")
# if we demote the node, we do cleanup later in the procedure
new_node.master_candidate = self.master_candidate
node_verify_list = [self.cfg.GetMasterNode()]
node_verify_param = {
- constants.NV_NODELIST: [node],
+ constants.NV_NODELIST: ([node], {}),
# TODO: do a node-net-test as well?
}
instances_keep = []
# Build list of instances to release
- for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
- instance = self.context.cfg.GetInstanceInfo(instance_name)
+ 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)
_ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
- assert (set(self.glm.list_owned(locking.LEVEL_INSTANCE)) ==
+ assert (set(self.owned_locks(locking.LEVEL_INSTANCE)) ==
set(instances_keep))
def BuildHooksEnv(self):
"""Logical unit for querying instances.
"""
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
REQ_BGL = False
def CheckArguments(self):
# 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)) ==
+ assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
frozenset([self.op.node_name]))
return ResultWithJobs(jobs)
target_node = self.target_node
source_node = self.source_node
+ # Check for hypervisor version mismatch and warn the user.
+ nodeinfo = self.rpc.call_node_info([source_node, target_node],
+ None, self.instance.hypervisor)
+ src_info = nodeinfo[source_node]
+ dst_info = nodeinfo[target_node]
+
+ if ((constants.HV_NODEINFO_KEY_VERSION in src_info.payload) and
+ (constants.HV_NODEINFO_KEY_VERSION in dst_info.payload)):
+ src_version = src_info.payload[constants.HV_NODEINFO_KEY_VERSION]
+ dst_version = dst_info.payload[constants.HV_NODEINFO_KEY_VERSION]
+ if src_version != dst_version:
+ self.feedback_fn("* warning: hypervisor version mismatch between"
+ " source (%s) and target (%s) node" %
+ (src_version, dst_version))
+
self.feedback_fn("* checking disk consistency between source and target")
for dev in instance.disks:
if not _CheckDiskConsistency(self.lu, dev, target_node, False):
if disk_template not in req_size_dict:
raise errors.ProgrammerError("Disk template '%s' size requirement"
- " is unknown" % disk_template)
+ " is unknown" % disk_template)
return req_size_dict[disk_template]
if disk_template not in req_size_dict:
raise errors.ProgrammerError("Disk template '%s' size requirement"
- " is unknown" % disk_template)
+ " is unknown" % disk_template)
return req_size_dict[disk_template]
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
self.op.src_node = None
if os.path.isabs(src_path):
- raise errors.OpPrereqError("Importing an instance from an absolute"
- " path requires a source node option",
+ raise errors.OpPrereqError("Importing an instance from a path"
+ " requires a source node option",
errors.ECODE_INVAL)
else:
self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
src_path = self.op.src_path
if src_node is None:
- locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
+ locked_nodes = self.owned_locks(locking.LEVEL_NODE)
exp_list = self.rpc.call_export_list(locked_nodes)
found = False
for node in exp_list:
joinargs.append(self.op.instance_name)
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
self.instance_file_storage_dir = utils.PathJoin(*joinargs)
def CheckPrereq(self):
# 'fake' LV disks with the old data, plus the new unique_id
tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
rename_to = []
- for t_dsk, a_dsk in zip (tmp_disks, self.disks):
+ for t_dsk, a_dsk in zip(tmp_disks, self.disks):
rename_to.append(t_dsk.logical_id)
t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
self.cfg.SetDiskID(t_dsk, pnode_name)
disk_abort = not _WaitForSync(self, iobj)
elif iobj.disk_template in constants.DTS_INT_MIRROR:
# make sure the disks are not degraded (still sync-ing is ok)
- time.sleep(15)
feedback_fn("* checking mirrors status")
disk_abort = not _WaitForSync(self, iobj, oneshot=True)
else:
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:
+ pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
+ not self.op.wait_for_sync)
+ if pause_sync:
+ feedback_fn("* pausing disk sync to install instance OS")
+ result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+ iobj.disks, True)
+ for idx, success in enumerate(result.payload):
+ if not success:
+ logging.warn("pause-sync of instance %s for disk %d failed",
+ instance, idx)
+
feedback_fn("* running the instance OS create scripts...")
# FIXME: pass debug option from opcode to backend
- result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
- self.op.debug_level)
- result.Raise("Could not add os for instance %s"
- " on node %s" % (instance, pnode_name))
+ os_add_result = \
+ self.rpc.call_instance_os_add(pnode_name, iobj, False,
+ self.op.debug_level)
+ if pause_sync:
+ feedback_fn("* resuming disk sync")
+ result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+ iobj.disks, False)
+ for idx, success in enumerate(result.payload):
+ if not success:
+ logging.warn("resume-sync of instance %s for disk %d failed",
+ instance, idx)
+
+ os_add_result.Raise("Could not add os for instance %s"
+ " on node %s" % (instance, pnode_name))
elif self.op.mode == constants.INSTANCE_IMPORT:
feedback_fn("* running the instance OS import scripts...")
# Lock member nodes of all locked groups
self.needed_locks[locking.LEVEL_NODE] = [node_name
- for group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
for node_name in self.cfg.GetNodeGroup(group_uuid).members]
else:
self._LockInstancesNodes()
assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
self.op.iallocator is None)
- owned_groups = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
if owned_groups:
- groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
- 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(owned_groups)))
+ _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
return LogicalUnit.CheckPrereq(self)
return remote_node_name
def _FindFaultyDisks(self, node_name):
+ """Wrapper for L{_FindFaultyInstanceDisks}.
+
+ """
return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
node_name, True)
if remote_node is None:
self.remote_node_info = None
else:
- assert remote_node in self.lu.glm.list_owned(locking.LEVEL_NODE), \
+ assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
"Remote node '%s' is not locked" % remote_node
self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
if __debug__:
# Verify owned locks before starting operation
- owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
- assert set(owned_locks) == set(self.node_secondary_ip), \
+ owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
+ assert set(owned_nodes) == set(self.node_secondary_ip), \
("Incorrect node locks, owning %s, expected %s" %
- (owned_locks, self.node_secondary_ip.keys()))
+ (owned_nodes, self.node_secondary_ip.keys()))
- owned_locks = self.lu.glm.list_owned(locking.LEVEL_INSTANCE)
- assert list(owned_locks) == [self.instance_name], \
+ owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
+ assert list(owned_instances) == [self.instance_name], \
"Instance '%s' not locked" % self.instance_name
assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
if __debug__:
# Verify owned locks
- owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
+ owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
nodes = frozenset(self.node_secondary_ip)
- assert ((self.early_release and not owned_locks) or
- (not self.early_release and not (set(owned_locks) - nodes))), \
+ assert ((self.early_release and not owned_nodes) or
+ (not self.early_release and not (set(owned_nodes) - nodes))), \
("Not owning the correct locks, early_release=%s, owned=%r,"
- " nodes=%r" % (self.early_release, owned_locks, nodes))
+ " nodes=%r" % (self.early_release, owned_nodes, nodes))
return result
self.lu.LogWarning("Can't remove old LV: %s" % msg,
hint="remove unused LVs manually")
- def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable-msg=W0613
+ def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
"""Replace a disk on the primary or secondary for DRBD 8.
The algorithm for replace is quite complicated:
"""
steps_total = 6
+ pnode = self.instance.primary_node
+
# Step: check device activation
self.lu.LogStep(1, steps_total, "Check device existence")
self._CheckDisksExistence([self.instance.primary_node])
" soon as possible"))
self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
- result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
- self.node_secondary_ip,
- self.instance.disks)\
- [self.instance.primary_node]
+ result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
+ self.instance.disks)[pnode]
msg = result.fail_msg
if msg:
def CheckPrereq(self):
# Verify locks
- owned_instances = self.glm.list_owned(locking.LEVEL_INSTANCE)
- owned_nodes = self.glm.list_owned(locking.LEVEL_NODE)
- owned_groups = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
+ owned_nodes = self.owned_locks(locking.LEVEL_NODE)
+ owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
assert owned_nodes == self.lock_nodes
"""
if self.wanted_names is None:
assert self.op.use_locking, "Locking was not used"
- self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
+ self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
- self.wanted_instances = [self.cfg.GetInstanceInfo(name)
- for name in self.wanted_names]
+ self.wanted_instances = \
+ map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
def _ComputeBlockdevStatus(self, node, instance_name, dev):
"""Returns the status of a block device
if msg:
# Assume the primary node is unreachable and go ahead
self.warn.append("Can't get info from primary node %s: %s" %
- (pnode, msg))
+ (pnode, msg))
elif not isinstance(pninfo.payload.get("memory_free", None), int):
self.warn.append("Node data from primary node %s doesn't contain"
" free memory information" % pnode)
}
+class LUInstanceChangeGroup(LogicalUnit):
+ HPATH = "instance-change-group"
+ HTYPE = constants.HTYPE_INSTANCE
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ self.share_locks = _ShareAll()
+ self.needed_locks = {
+ locking.LEVEL_NODEGROUP: [],
+ locking.LEVEL_NODE: [],
+ }
+
+ self._ExpandAndLockInstance()
+
+ if self.op.target_groups:
+ self.req_target_uuids = map(self.cfg.LookupNodeGroup,
+ self.op.target_groups)
+ else:
+ self.req_target_uuids = None
+
+ self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+ if self.req_target_uuids:
+ lock_groups = set(self.req_target_uuids)
+
+ # Lock all groups used by instance optimistically; this requires going
+ # via the node before it's locked, requiring verification later on
+ instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+ lock_groups.update(instance_groups)
+ else:
+ # No target groups, need to lock all of them
+ lock_groups = locking.ALL_SET
+
+ self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
+
+ elif level == locking.LEVEL_NODE:
+ if self.req_target_uuids:
+ # Lock all nodes used by instances
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ self._LockInstancesNodes()
+
+ # Lock all nodes in all potential target groups
+ lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name))
+ member_nodes = [node_name
+ for group in lock_groups
+ for node_name in self.cfg.GetNodeGroup(group).members]
+ self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
+ else:
+ # Lock all nodes as all groups are potential targets
+ self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+ def CheckPrereq(self):
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
+
+ assert (self.req_target_uuids is None or
+ owned_groups.issuperset(self.req_target_uuids))
+ assert owned_instances == set([self.op.instance_name])
+
+ # Get instance information
+ self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
+ # Check if node groups for locked instance are still correct
+ assert owned_nodes.issuperset(self.instance.all_nodes), \
+ ("Instance %s's nodes changed while we kept the lock" %
+ self.op.instance_name)
+
+ inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
+ owned_groups)
+
+ if self.req_target_uuids:
+ # User requested specific target groups
+ self.target_uuids = self.req_target_uuids
+ else:
+ # All groups except those used by the instance are potential targets
+ self.target_uuids = owned_groups - inst_groups
+
+ conflicting_groups = self.target_uuids & inst_groups
+ if conflicting_groups:
+ raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
+ " used by the instance '%s'" %
+ (utils.CommaJoin(conflicting_groups),
+ self.op.instance_name),
+ errors.ECODE_INVAL)
+
+ if not self.target_uuids:
+ raise errors.OpPrereqError("There are no possible target groups",
+ errors.ECODE_INVAL)
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ """
+ assert self.target_uuids
+
+ env = {
+ "TARGET_GROUPS": " ".join(self.target_uuids),
+ }
+
+ env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+
+ return env
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ mn = self.cfg.GetMasterNode()
+ return ([mn], [mn])
+
+ def Exec(self, feedback_fn):
+ instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
+
+ assert instances == [self.op.instance_name], "Instance not locked"
+
+ ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
+ instances=instances, target_groups=list(self.target_uuids))
+
+ ial.Run(self.op.iallocator)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute solution for changing group of"
+ " instance '%s' using iallocator '%s': %s" %
+ (self.op.instance_name, self.op.iallocator,
+ ial.info),
+ errors.ECODE_NORES)
+
+ jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
+
+ self.LogInfo("Iallocator returned %s job(s) for changing group of"
+ " instance '%s'", len(jobs), self.op.instance_name)
+
+ return ResultWithJobs(jobs)
+
+
class LUBackupQuery(NoHooksLU):
"""Query the exports list
that node.
"""
- self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
+ self.nodes = self.owned_locks(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.glm.list_owned(locking.LEVEL_NODE)
+ locked_nodes = self.owned_locks(locking.LEVEL_NODE)
exportlist = self.rpc.call_export_list(locked_nodes)
found = False
for node in exportlist:
"""
assert self.needed_locks[locking.LEVEL_NODEGROUP]
- assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+ assert (frozenset(self.owned_locks(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)
+ actual_locks = self.owned_locks(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'" %
def ExpandNames(self):
self.gq.ExpandNames(self)
+ def DeclareLocks(self, level):
+ self.gq.DeclareLocks(self, level)
+
def Exec(self, feedback_fn):
return self.gq.OldStyleQuery(self)
return result
-
class LUGroupRemove(LogicalUnit):
HPATH = "group-remove"
HTYPE = constants.HTYPE_GROUP
utils.CommaJoin(self.req_target_uuids)),
errors.ECODE_INVAL)
- if not self.op.iallocator:
- # Use default iallocator
- self.op.iallocator = self.cfg.GetDefaultIAllocator()
-
- if not self.op.iallocator:
- raise errors.OpPrereqError("No iallocator was specified, neither in the"
- " opcode nor as a cluster-wide default",
- errors.ECODE_INVAL)
+ self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
self.share_locks = _ShareAll()
self.needed_locks = {
# via the node before it's locked, requiring verification later on
lock_groups.update(group_uuid
for instance_name in
- self.glm.list_owned(locking.LEVEL_INSTANCE)
+ self.owned_locks(locking.LEVEL_INSTANCE)
for group_uuid in
self.cfg.GetInstanceNodeGroups(instance_name))
else:
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
self._LockInstancesNodes()
- # Lock all nodes in group to be evacuated
- assert self.group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
- member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
+ # Lock all nodes in group to be evacuated and target groups
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ assert self.group_uuid in owned_groups
+ member_nodes = [node_name
+ for group in owned_groups
+ for node_name in self.cfg.GetNodeGroup(group).members]
self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
def CheckPrereq(self):
- owned_instances = frozenset(self.glm.list_owned(locking.LEVEL_INSTANCE))
- owned_groups = frozenset(self.glm.list_owned(locking.LEVEL_NODEGROUP))
- owned_nodes = frozenset(self.glm.list_owned(locking.LEVEL_NODE))
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
assert owned_groups.issuperset(self.req_target_uuids)
assert self.group_uuid in owned_groups
# Check if locked instances are still correct
- wanted_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
- if owned_instances != wanted_instances:
- raise errors.OpPrereqError("Instances in node group to be evacuated (%s)"
- " changed since locks were acquired, wanted"
- " %s, have %s; retry the operation" %
- (self.group_uuid,
- utils.CommaJoin(wanted_instances),
- utils.CommaJoin(owned_instances)),
- errors.ECODE_STATE)
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
# Get instance information
- self.instances = dict((name, self.cfg.GetInstanceInfo(name))
- for name in owned_instances)
+ self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
# Check if node groups for locked instances are still correct
for instance_name in owned_instances:
inst = self.instances[instance_name]
- assert self.group_uuid in self.cfg.GetInstanceNodeGroups(instance_name), \
- "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
assert owned_nodes.issuperset(inst.all_nodes), \
"Instance %s's nodes changed while we kept the lock" % instance_name
- inst_groups = self.cfg.GetInstanceNodeGroups(instance_name)
- if not owned_groups.issuperset(inst_groups):
- raise errors.OpPrereqError("Instance %s's node groups changed since"
- " locks were acquired, current groups"
- " are '%s', owning groups '%s'; retry the"
- " operation" %
- (instance_name,
- utils.CommaJoin(inst_groups),
- utils.CommaJoin(owned_groups)),
- errors.ECODE_STATE)
+ inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
+ owned_groups)
+
+ assert self.group_uuid in inst_groups, \
+ "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
if self.req_target_uuids:
# User requested specific target groups
if group_uuid != self.group_uuid]
if not self.target_uuids:
- raise errors.OpExecError("There are no possible target groups")
+ raise errors.OpPrereqError("There are no possible target groups",
+ errors.ECODE_INVAL)
def BuildHooksEnv(self):
"""Build hooks env.
"""
mn = self.cfg.GetMasterNode()
- assert self.group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
+ assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
return (run_nodes, run_nodes)
def Exec(self, feedback_fn):
- instances = list(self.glm.list_owned(locking.LEVEL_INSTANCE))
+ instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
assert self.group_uuid not in self.target_uuids
return ResultWithJobs(jobs)
-class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
+class TagsLU(NoHooksLU): # pylint: disable=W0223
"""Generic tags LU.
This is an abstract class which is the parent of all the other tags LUs.
# Wait for client to close
try:
try:
- # pylint: disable-msg=E1101
+ # pylint: disable=E1101
# Instance of '_socketobject' has no ... member
conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
conn.recv(1)
easy usage
"""
- # pylint: disable-msg=R0902
+ # pylint: disable=R0902
# lots of instance attributes
def __init__(self, cfg, rpc, mode, **kwargs):
self.hypervisor = None
self.relocate_from = None
self.name = None
- self.evac_nodes = None
self.instances = None
self.evac_mode = None
self.target_groups = []
}
return request
- def _AddEvacuateNodes(self):
- """Add evacuate nodes data to allocator structure.
-
- """
- request = {
- "evac_nodes": self.evac_nodes
- }
- return request
-
def _AddNodeEvacuate(self):
"""Get data for node-evacuate requests.
_STRING_LIST = ht.TListOf(ht.TString)
_JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
- # pylint: disable-msg=E1101
+ # pylint: disable=E1101
# Class '...' has no 'OP_ID' member
"OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
opcodes.OpInstanceMigrate.OP_ID,
(_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),
(self._result_check, self.result),
errors.ECODE_INVAL)
- if self.mode in (constants.IALLOCATOR_MODE_RELOC,
- constants.IALLOCATOR_MODE_MEVAC):
+ if self.mode == constants.IALLOCATOR_MODE_RELOC:
+ assert self.relocate_from is not None
+ assert self.required_nodes == 1
+
node2group = dict((name, ndata["group"])
for (name, ndata) in self.in_data["nodes"].items())
fn = compat.partial(self._NodesToGroups, node2group,
self.in_data["nodegroups"])
- 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)
+ instance = self.cfg.GetInstanceInfo(self.name)
+ request_groups = fn(self.relocate_from + [instance.primary_node])
+ result_groups = fn(rdict["result"] + [instance.primary_node])
+
+ if self.success and not set(result_groups).issubset(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_NODE_EVAC:
assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
self.op.name = fname
self.relocate_from = \
list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
- elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
- 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:
name=self.op.name,
relocate_from=list(self.relocate_from),
)
- elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
- 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,
projects / ganeti-local / blobdiff