"""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
import tempfile
import shutil
import itertools
+import operator
from ganeti import ssh
from ganeti import utils
from ganeti import opcodes
from ganeti import ht
-import ganeti.masterd.instance # pylint: disable-msg=W0611
-
-
-def _SupportsOob(cfg, node):
- """Tells if node supports OOB.
-
- @type cfg: L{config.ConfigWriter}
- @param cfg: The cluster configuration
- @type node: L{objects.Node}
- @param node: The node
- @return: The OOB script if supported or an empty string otherwise
-
- """
- return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
+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
sort_by_name=self.sort_by_name)
+def _ShareAll():
+ """Returns a dict declaring all lock levels shared.
+
+ """
+ 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.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: The cluster configuration
+ @type node: L{objects.Node}
+ @param node: The node
+ @return: The OOB script if supported or an empty string otherwise
+
+ """
+ return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
+
+
def _GetWantedNodes(lu, nodes):
"""Returns list of checked and expanded node names.
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)
assert not lu.glm.is_owned(level), "No locks should be owned"
+def _MapInstanceDisksToNodes(instances):
+ """Creates a map from (node, volume) to instance name.
+
+ @type instances: list of L{objects.Instance}
+ @rtype: dict; tuple of (node name, volume name) as key, instance name as value
+
+ """
+ return dict(((node, vol), inst.name)
+ for inst in instances
+ for (node, vols) in inst.MapLVsByNode().items()
+ for vol in vols)
+
+
def _RunPostHook(lu, node_name):
"""Runs the post-hook for an opcode on a single node.
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)
bep = cluster.FillBE(instance)
hvp = cluster.FillHV(instance)
args = {
- 'name': instance.name,
- 'primary_node': instance.primary_node,
- 'secondary_nodes': instance.secondary_nodes,
- 'os_type': instance.os,
- 'status': instance.admin_up,
- 'memory': bep[constants.BE_MEMORY],
- 'vcpus': bep[constants.BE_VCPUS],
- 'nics': _NICListToTuple(lu, instance.nics),
- 'disk_template': instance.disk_template,
- 'disks': [(disk.size, disk.mode) for disk in instance.disks],
- 'bep': bep,
- 'hvp': hvp,
- 'hypervisor_name': instance.hypervisor,
- 'tags': instance.tags,
+ "name": instance.name,
+ "primary_node": instance.primary_node,
+ "secondary_nodes": instance.secondary_nodes,
+ "os_type": instance.os,
+ "status": instance.admin_up,
+ "memory": bep[constants.BE_MEMORY],
+ "vcpus": bep[constants.BE_VCPUS],
+ "nics": _NICListToTuple(lu, instance.nics),
+ "disk_template": instance.disk_template,
+ "disks": [(disk.size, disk.mode) for disk in instance.disks],
+ "bep": bep,
+ "hvp": hvp,
+ "hypervisor_name": instance.hypervisor,
+ "tags": instance.tags,
}
if override:
args.update(override)
- return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
+ return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
def _AdjustCandidatePool(lu, exceptions):
@param name: OS name passed by the user, to check for validity
"""
+ variant = objects.OS.GetVariant(name)
if not os_obj.supported_variants:
+ if variant:
+ raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
+ " passed)" % (os_obj.name, variant),
+ errors.ECODE_INVAL)
return
- variant = objects.OS.GetVariant(name)
if not variant:
raise errors.OpPrereqError("OS name must include a variant",
errors.ECODE_INVAL)
" 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
def _VerifyCertificate(filename):
- """Verifies a certificate for LUClusterVerifyConfig.
+ """Verifies a certificate for L{LUClusterVerifyConfig}.
@type filename: string
@param filename: Path to PEM file
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.
"""
- REQ_BGL = False
+ REQ_BGL = True
def _VerifyHVP(self, hvp_data):
"""Verifies locally the syntax of the hypervisor parameters.
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
+ 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()
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 catched by VerifyGroup, which only acts on
+ # 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()
"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):
# This raises errors.OpPrereqError on its own:
self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
- all_node_info = self.cfg.GetAllNodesInfo()
- all_inst_info = self.cfg.GetAllInstancesInfo()
+ # Get instances in node group; this is unsafe and needs verification later
+ inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
- node_names = set(node.name
- for node in all_node_info.values()
- if node.group == self.group_uuid)
+ self.needed_locks = {
+ locking.LEVEL_INSTANCE: inst_names,
+ locking.LEVEL_NODEGROUP: [self.group_uuid],
+ locking.LEVEL_NODE: [],
+ }
- inst_names = [inst.name
- for inst in all_inst_info.values()
- if inst.primary_node in node_names]
+ self.share_locks = _ShareAll()
- # 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 inst_names:
- if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
- node_names.update(all_inst_info[inst].secondary_nodes)
+ 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)
- self.needed_locks = {
- locking.LEVEL_NODEGROUP: [self.group_uuid],
- locking.LEVEL_NODE: list(node_names),
- locking.LEVEL_INSTANCE: inst_names,
- }
+ all_inst_info = self.cfg.GetAllInstancesInfo()
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ # 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.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)
- def CheckPrereq(self):
- self.all_node_info = self.cfg.GetAllNodesInfo()
- self.all_inst_info = self.cfg.GetAllInstancesInfo()
+ self.needed_locks[locking.LEVEL_NODE] = nodes
- group_nodes = set(node.name
- for node in self.all_node_info.values()
- if node.group == self.group_uuid)
+ def CheckPrereq(self):
+ assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+ self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
- group_instances = set(inst.name
- for inst in self.all_inst_info.values()
- if inst.primary_node in group_nodes)
+ 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" %
+ raise errors.OpPrereqError("Missing lock for nodes: %s" %
utils.CommaJoin(unlocked_nodes))
if unlocked_instances:
- raise errors.OpPrereqError("missing lock for instances: %s" %
+ 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)
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)
+ # 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]
if nresult.fail_msg or not nresult.payload:
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)",
+ "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?"
_ErrorIf(len(os_data) > 1, self.ENODEOS, node,
"OS '%s' has multiple entries (first one shadows the rest): %s",
os_name, utils.CommaJoin([v[0] for v in os_data]))
- # this will catched in backend too
- _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
- and not f_var, self.ENODEOS, node,
- "OS %s with API at least %d does not declare any variant",
- os_name, constants.OS_API_V15)
# comparisons with the 'base' image
test = os_name not in base.oslist
_ErrorIf(test, self.ENODEOS, node,
"""
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.
"""Build hooks nodes.
"""
- 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 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
+ feedback_fn("* Empty node group, skipping verification")
+ 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()),
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,
self.cfg.GetClusterName())
else:
extra_lv_nvinfo = {}
- nvinfo_endtime = time.time()
all_drbd_map = self.cfg.ComputeDRBDMap()
and hook results
"""
- # We only really run POST phase hooks, and are only interested in
- # their results
- if phase == constants.HOOKS_PHASE_POST:
+ # We only really run POST phase hooks, only for non-empty groups,
+ # and are only interested in their results
+ if not self.my_node_names:
+ # empty node group
+ pass
+ elif phase == constants.HOOKS_PHASE_POST:
# Used to change hooks' output to proper indentation
feedback_fn("* Hooks Results")
assert hooks_results, "invalid result from hooks"
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
self._ErrorIf(test, self.ENODEHOOKS, node_name,
"Script %s failed, output:", script)
if test:
- output = self._HOOKS_INDENT_RE.sub(' ', output)
+ output = self._HOOKS_INDENT_RE.sub(" ", output)
feedback_fn("%s" % output)
- lu_result = 0
+ lu_result = False
- return lu_result
+ return lu_result
class LUClusterVerifyDisks(NoHooksLU):
REQ_BGL = False
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self.needed_locks = {
- locking.LEVEL_NODE: locking.ALL_SET,
- locking.LEVEL_INSTANCE: locking.ALL_SET,
- }
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ locking.LEVEL_NODEGROUP: locking.ALL_SET,
+ }
+
+ def Exec(self, feedback_fn):
+ 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)]
+ for group in group_names])
+
+
+class LUGroupVerifyDisks(NoHooksLU):
+ """Verifies the status of all disks in a node group.
+
+ """
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ # Raises errors.OpPrereqError on its own if group can't be found
+ self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+ self.share_locks = _ShareAll()
+ self.needed_locks = {
+ locking.LEVEL_INSTANCE: [],
+ locking.LEVEL_NODEGROUP: [],
+ locking.LEVEL_NODE: [],
+ }
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_INSTANCE:
+ assert not self.needed_locks[locking.LEVEL_INSTANCE]
+
+ # Lock instances optimistically, needs verification once node and group
+ # locks have been acquired
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
+ self.cfg.GetNodeGroupInstances(self.group_uuid)
+
+ elif level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ set([self.group_uuid] +
+ # Lock all groups used by instances optimistically; this requires
+ # going via the node before it's locked, requiring verification
+ # later on
+ [group_uuid
+ 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
+ # actual instances
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ self._LockInstancesNodes()
+
+ # Lock all nodes in group to be verified
+ 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.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
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+ # Get instance information
+ 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 owned_nodes.issuperset(inst.all_nodes), \
+ "Instance %s's nodes changed while we kept the lock" % instance_name
+
+ 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.
missing volumes
"""
- result = res_nodes, res_instances, res_missing = {}, [], {}
+ res_nodes = {}
+ res_instances = set()
+ res_missing = {}
- nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
- instances = self.cfg.GetAllInstancesInfo().values()
+ nv_dict = _MapInstanceDisksToNodes([inst
+ for inst in self.instances.values()
+ if inst.admin_up])
- nv_dict = {}
- for inst in instances:
- inst_lvs = {}
- if not inst.admin_up:
- continue
- inst.MapLVsByNode(inst_lvs)
- # transform { iname: {node: [vol,],},} to {(node, vol): iname}
- for node, vol_list in inst_lvs.iteritems():
- for vol in vol_list:
- nv_dict[(node, vol)] = inst
-
- if not nv_dict:
- return result
-
- node_lvs = self.rpc.call_lv_list(nodes, [])
- for node, node_res in node_lvs.items():
- if node_res.offline:
- continue
- msg = node_res.fail_msg
- if msg:
- logging.warning("Error enumerating LVs on node %s: %s", node, msg)
- res_nodes[node] = msg
- continue
+ if nv_dict:
+ nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
+ set(self.cfg.GetVmCapableNodeList()))
- lvs = node_res.payload
- for lv_name, (_, _, lv_online) in lvs.items():
- inst = nv_dict.pop((node, lv_name), None)
- if (not lv_online and inst is not None
- and inst.name not in res_instances):
- res_instances.append(inst.name)
+ node_lvs = self.rpc.call_lv_list(nodes, [])
- # any leftover items in nv_dict are missing LVs, let's arrange the
- # data better
- for key, inst in nv_dict.iteritems():
- if inst.name not in res_missing:
- res_missing[inst.name] = []
- res_missing[inst.name].append(key)
+ for (node, node_res) in node_lvs.items():
+ if node_res.offline:
+ continue
- return result
+ msg = node_res.fail_msg
+ if msg:
+ logging.warning("Error enumerating LVs on node %s: %s", node, msg)
+ res_nodes[node] = msg
+ continue
+
+ for lv_name, (_, _, lv_online) in node_res.payload.items():
+ inst = nv_dict.pop((node, lv_name), None)
+ if not (lv_online or inst is None):
+ res_instances.add(inst)
+
+ # any leftover items in nv_dict are missing LVs, let's arrange the data
+ # better
+ for key, inst in nv_dict.iteritems():
+ res_missing.setdefault(inst, []).append(key)
+
+ return (res_nodes, list(res_instances), res_missing)
class LUClusterRepairDiskSizes(NoHooksLU):
locking.LEVEL_NODE: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
}
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ self.share_locks = _ShareAll()
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.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:
if self.op.drbd_helper:
# checks given drbd helper on all nodes
helpers = self.rpc.call_drbd_helper(node_list)
- for node in node_list:
- ninfo = self.cfg.GetNodeInfo(node)
+ for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
if ninfo.offline:
self.LogInfo("Not checking drbd helper on offline node %s", node)
continue
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])
- # 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]))), \
+ files_opt |= set(filename
+ for hv_name in cluster.enabled_hypervisors
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
+
+ # 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.
if self.op.command in self._SKIP_MASTER:
assert self.master_node not in self.op.node_names
- for node_name in self.op.node_names:
- node = self.cfg.GetNodeInfo(node_name)
-
+ for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
if node is None:
raise errors.OpPrereqError("Node %s not found" % node_name,
errors.ECODE_NOENT)
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.GetInstanceInfo(iname) for iname
- in self.cfg.GetInstanceList()]
-
- lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
+ ilist = self.cfg.GetAllInstancesInfo()
+ vol2inst = _MapInstanceDisksToNodes(ilist.values())
output = []
for node in nodenames:
self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
continue
- node_vols = nresult.payload[:]
- node_vols.sort(key=lambda vol: vol['dev'])
+ node_vols = sorted(nresult.payload,
+ key=operator.itemgetter("dev"))
for vol in node_vols:
node_output = []
if field == "node":
val = node
elif field == "phys":
- val = vol['dev']
+ val = vol["dev"]
elif field == "vg":
- val = vol['vg']
+ val = vol["vg"]
elif field == "name":
- val = vol['name']
+ val = vol["name"]
elif field == "size":
- val = int(float(vol['size']))
+ val = int(float(vol["size"]))
elif field == "instance":
- for inst in ilist:
- if node not in lv_by_node[inst]:
- continue
- if vol['name'] in lv_by_node[inst][node]:
- val = inst.name
- break
- else:
- val = '-'
+ val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
else:
raise errors.ParameterError(field)
node_output.append(str(val))
"""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:
def ExpandNames(self, lu):
lu.needed_locks = {}
- lu.share_locks[locking.LEVEL_INSTANCE] = 1
- lu.share_locks[locking.LEVEL_NODE] = 1
+ lu.share_locks = _ShareAll()
if self.names:
self.wanted = _GetWantedInstances(lu, self.names)
query.IQ_LIVE in self.requested_data)
if self.do_locking:
lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
+ lu.needed_locks[locking.LEVEL_NODEGROUP] = []
lu.needed_locks[locking.LEVEL_NODE] = []
lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.do_grouplocks = (self.do_locking and
+ query.IQ_NODES in self.requested_data)
+
def DeclareLocks(self, lu, level):
- if level == locking.LEVEL_NODE and self.do_locking:
- lu._LockInstancesNodes() # pylint: disable-msg=W0212
+ if self.do_locking:
+ if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
+ assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
+
+ # Lock all groups used by instances optimistically; this requires going
+ # 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.owned_locks(locking.LEVEL_INSTANCE)
+ for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
+ elif level == locking.LEVEL_NODE:
+ lu._LockInstancesNodes() # pylint: disable=W0212
+
+ @staticmethod
+ def _CheckGroupLocks(lu):
+ 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:
+ _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
def _GetQueryData(self, lu):
"""Computes the list of instances and their attributes.
"""
+ if self.do_grouplocks:
+ self._CheckGroupLocks(lu)
+
cluster = lu.cfg.GetClusterInfo()
all_info = lu.cfg.GetAllInstancesInfo()
else:
consinfo = None
+ if query.IQ_NODES in self.requested_data:
+ node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
+ instance_list)))
+ nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
+ groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
+ for uuid in set(map(operator.attrgetter("group"),
+ nodes.values())))
+ else:
+ nodes = None
+ groups = None
+
return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
disk_usage, offline_nodes, bad_nodes,
- live_data, wrongnode_inst, consinfo)
+ live_data, wrongnode_inst, consinfo,
+ nodes, groups)
class LUQuery(NoHooksLU):
"""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):
self.changed_primary_ip = False
- for existing_node_name in node_list:
- existing_node = cfg.GetNodeInfo(existing_node_name)
-
+ for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
if self.op.readd and node == existing_node_name:
if existing_node.secondary_ip != secondary_ip:
raise errors.OpPrereqError("Readded node doesn't have the same IP"
# 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):
nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
nodeinfo[node].Raise("Can't get data from node %s" % node,
prereq=True, ecode=errors.ECODE_ENVIRON)
- free_mem = nodeinfo[node].payload.get('memory_free', None)
+ free_mem = nodeinfo[node].payload.get("memory_free", None)
if not isinstance(free_mem, int):
raise errors.OpPrereqError("Can't compute free memory on node %s, result"
" was '%s'" % (node, free_mem),
_StartInstanceDisks(self, instance, force)
result = self.rpc.call_instance_start(node_current, instance,
- self.op.hvparams, self.op.beparams)
+ self.op.hvparams, self.op.beparams,
+ self.op.startup_paused)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, instance)
self.LogInfo("Instance %s was already stopped, starting now",
instance.name)
_StartInstanceDisks(self, instance, ignore_secondaries)
- result = self.rpc.call_instance_start(node_current, instance, None, None)
+ result = self.rpc.call_instance_start(node_current, instance,
+ None, None, False)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, instance)
"""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")
+ instance = self.instance
to_skip = []
- for idx, disk in enumerate(self.instance.disks):
+ mods = [] # keeps track of needed logical_id changes
+
+ for idx, disk in enumerate(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
+ # need to update the nodes and minors
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)
+ assert len(disk.logical_id) == 6 # otherwise disk internals
+ # have changed
+ (_, _, old_port, _, _, old_secret) = disk.logical_id
+ new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
+ new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
+ new_minors[0], new_minors[1], old_secret)
+ assert len(disk.logical_id) == len(new_id)
+ mods.append((idx, new_id))
+
+ # now that we have passed all asserts above, we can apply the mods
+ # in a single run (to avoid partial changes)
+ for idx, new_id in mods:
+ instance.disks[idx].logical_id = new_id
+
+ # change primary node, if needed
+ if self.op.nodes:
+ 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")
if self.op.nodes:
- self.cfg.Update(self.instance, feedback_fn)
+ self.cfg.Update(instance, feedback_fn)
- _CreateDisks(self, self.instance, to_skip=to_skip)
+ _CreateDisks(self, instance, to_skip=to_skip)
class LUInstanceRename(LogicalUnit):
old_name = inst.name
rename_file_storage = False
- if (inst.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE) and
+ if (inst.disk_template in constants.DTS_FILEBASED and
self.op.new_name != inst.name):
old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
rename_file_storage = True
"""Logical unit for querying instances.
"""
- # pylint: disable-msg=W0142
+ # pylint: disable=W0142
REQ_BGL = False
def CheckArguments(self):
_ShutdownInstanceDisks(self, instance)
raise errors.OpExecError("Can't activate the instance's disks")
- result = self.rpc.call_instance_start(target_node, instance, None, None)
+ result = self.rpc.call_instance_start(target_node, instance,
+ None, None, False)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, instance)
def ExpandNames(self):
self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ self.share_locks = _ShareAll()
self.needed_locks = {
locking.LEVEL_NODE: [self.op.node_name],
}
# 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):
self.feedback_fn("* checking disk consistency between source and target")
for dev in instance.disks:
# for drbd, these are drbd over lvm
- if not _CheckDiskConsistency(self, dev, target_node, False):
- if not self.ignore_consistency:
+ if not _CheckDiskConsistency(self.lu, dev, target_node, False):
+ if primary_node.offline:
+ self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
+ " target node %s" %
+ (primary_node.name, dev.iv_name, target_node))
+ elif not self.ignore_consistency:
raise errors.OpExecError("Disk %s is degraded on target node,"
" aborting failover" % dev.iv_name)
else:
(instance.name, source_node, msg))
self.feedback_fn("* deactivating the instance's disks on source node")
- if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
- raise errors.OpExecError("Can't shut down the instance's disks.")
+ if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
+ raise errors.OpExecError("Can't shut down the instance's disks")
instance.primary_node = target_node
# distribute new instance config to the other nodes
# Only start the instance if it's marked as up
if instance.admin_up:
- self.feedback_fn("* activating the instance's disks on target node")
+ self.feedback_fn("* activating the instance's disks on target node %s" %
+ target_node)
logging.info("Starting instance %s on node %s",
instance.name, target_node)
- disks_ok, _ = _AssembleInstanceDisks(self, instance,
+ disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
ignore_secondaries=True)
if not disks_ok:
- _ShutdownInstanceDisks(self, instance)
+ _ShutdownInstanceDisks(self.lu, instance)
raise errors.OpExecError("Can't activate the instance's disks")
- self.feedback_fn("* starting the instance on the target node")
- result = self.rpc.call_instance_start(target_node, instance, None, None)
+ self.feedback_fn("* starting the instance on the target node %s" %
+ target_node)
+ result = self.rpc.call_instance_start(target_node, instance, None, None,
+ False)
msg = result.fail_msg
if msg:
- _ShutdownInstanceDisks(self, instance)
+ _ShutdownInstanceDisks(self.lu, instance)
raise errors.OpExecError("Could not start instance %s on node %s: %s" %
(instance.name, target_node, msg))
# directly, or through an iallocator.
self.all_nodes = [self.source_node, self.target_node]
- self.nodes_ip = {
- self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
- self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
- }
+ self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
+ in self.cfg.GetMultiNodeInfo(self.all_nodes))
if self.failover:
feedback_fn("Failover instance %s" % self.instance.name)
pnode = target_node
all_nodes = [pnode]
- if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
+ if instance.disk_template in constants.DTS_FILEBASED:
file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
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]
raise errors.OpPrereqError("Invalid file driver name '%s'" %
self.op.file_driver, errors.ECODE_INVAL)
- if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
- raise errors.OpPrereqError("File storage directory path not absolute",
- errors.ECODE_INVAL)
+ if self.op.disk_template == constants.DT_FILE:
+ opcodes.RequireFileStorage()
+ elif self.op.disk_template == constants.DT_SHARED_FILE:
+ opcodes.RequireSharedFileStorage()
### Node/iallocator related checks
_CheckIAllocatorOrNode(self, "iallocator", "pnode")
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:
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
if name in os_defs and os_defs[name] == self.op.osparams[name]:
del self.op.osparams[name]
+ def _CalculateFileStorageDir(self):
+ """Calculate final instance file storage dir.
+
+ """
+ # file storage dir calculation/check
+ self.instance_file_storage_dir = None
+ if self.op.disk_template in constants.DTS_FILEBASED:
+ # build the full file storage dir path
+ joinargs = []
+
+ if self.op.disk_template == constants.DT_SHARED_FILE:
+ get_fsd_fn = self.cfg.GetSharedFileStorageDir
+ else:
+ get_fsd_fn = self.cfg.GetFileStorageDir
+
+ cfg_storagedir = get_fsd_fn()
+ if not cfg_storagedir:
+ raise errors.OpPrereqError("Cluster file storage dir not defined")
+ joinargs.append(cfg_storagedir)
+
+ if self.op.file_storage_dir is not None:
+ joinargs.append(self.op.file_storage_dir)
+
+ joinargs.append(self.op.instance_name)
+
+ # pylint: disable=W0142
+ self.instance_file_storage_dir = utils.PathJoin(*joinargs)
+
def CheckPrereq(self):
"""Check prerequisites.
"""
+ self._CalculateFileStorageDir()
+
if self.op.mode == constants.INSTANCE_IMPORT:
export_info = self._ReadExportInfo()
self._ReadExportParams(export_info)
disk_images = []
for idx in range(export_disks):
- option = 'disk%d_dump' % idx
+ option = "disk%d_dump" % idx
if export_info.has_option(constants.INISECT_INS, option):
# FIXME: are the old os-es, disk sizes, etc. useful?
export_name = export_info.get(constants.INISECT_INS, option)
self.src_images = disk_images
- old_name = export_info.get(constants.INISECT_INS, 'name')
+ old_name = export_info.get(constants.INISECT_INS, "name")
try:
- exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
+ exp_nic_count = export_info.getint(constants.INISECT_INS, "nic_count")
except (TypeError, ValueError), err:
raise errors.OpPrereqError("Invalid export file, nic_count is not"
" an integer: %s" % str(err),
if self.op.instance_name == old_name:
for idx, nic in enumerate(self.nics):
if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
- nic_mac_ini = 'nic%d_mac' % idx
+ nic_mac_ini = "nic%d_mac" % idx
nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
# ENDIF: self.op.mode == constants.INSTANCE_IMPORT
else:
network_port = None
- if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
- # this is needed because os.path.join does not accept None arguments
- if self.op.file_storage_dir is None:
- string_file_storage_dir = ""
- else:
- string_file_storage_dir = self.op.file_storage_dir
-
- # build the full file storage dir path
- if self.op.disk_template == constants.DT_SHARED_FILE:
- get_fsd_fn = self.cfg.GetSharedFileStorageDir
- else:
- get_fsd_fn = self.cfg.GetFileStorageDir
-
- file_storage_dir = utils.PathJoin(get_fsd_fn(),
- string_file_storage_dir, instance)
- else:
- file_storage_dir = ""
-
disks = _GenerateDiskTemplate(self,
self.op.disk_template,
instance, pnode_name,
self.secondaries,
self.disks,
- file_storage_dir,
+ self.instance_file_storage_dir,
self.op.file_driver,
0,
feedback_fn)
hvparams=self.op.hvparams,
hypervisor=self.op.hypervisor,
osparams=self.op.osparams,
- tags=self.op.tags,
)
+ 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
# '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...")
self.cfg.Update(iobj, feedback_fn)
logging.info("Starting instance %s on node %s", instance, pnode_name)
feedback_fn("* starting instance...")
- result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
+ result = self.rpc.call_instance_start(pnode_name, iobj,
+ None, None, False)
result.Raise("Could not start instance")
return list(iobj.all_nodes)
# 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)
ial = IAllocator(lu.cfg, lu.rpc,
mode=constants.IALLOCATOR_MODE_RELOC,
name=instance_name,
- relocate_from=relocate_from)
+ relocate_from=list(relocate_from))
ial.Run(iallocator_name)
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)
instance.FindDisk(disk_idx)
# Get secondary node IP addresses
- self.node_secondary_ip = \
- dict((node_name, self.cfg.GetNodeInfo(node_name).secondary_ip)
- for node_name in touched_nodes)
+ self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
+ in self.cfg.GetMultiNodeInfo(touched_nodes))
def Exec(self, feedback_fn):
"""Execute disk replacement.
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
(node_name, self.instance.name))
def _CreateNewStorage(self, node_name):
+ """Create new storage on the primary or secondary node.
+
+ This is only used for same-node replaces, not for changing the
+ secondary node, hence we don't want to modify the existing disk.
+
+ """
iv_names = {}
for idx, dev in enumerate(self.instance.disks):
logical_id=(vg_meta, names[1]))
new_lvs = [lv_data, lv_meta]
- old_lvs = dev.children
+ old_lvs = [child.Copy() for child in dev.children]
iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
# we pass force_create=True to force the LVM creation
self.lu.LogWarning("Can't remove old LV: %s" % msg,
hint="remove unused LVs manually")
- def _ExecDrbd8DiskOnly(self, feedback_fn):
+ 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:
rename_new_to_old)
result.Raise("Can't rename new LVs on node %s" % self.target_node)
+ # Intermediate steps of in memory modifications
for old, new in zip(old_lvs, new_lvs):
new.logical_id = old.logical_id
self.cfg.SetDiskID(new, self.target_node)
+ # We need to modify old_lvs so that removal later removes the
+ # right LVs, not the newly added ones; note that old_lvs is a
+ # copy here
for disk in old_lvs:
disk.logical_id = ren_fn(disk, temp_suffix)
self.cfg.SetDiskID(disk, self.target_node)
"volumes"))
raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
- dev.children = new_lvs
-
- self.cfg.Update(self.instance, feedback_fn)
-
cstep = 5
if self.early_release:
self.lu.LogStep(cstep, steps_total, "Removing old storage")
"""
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:
(self.op.name, self.op.node_name))
-class LUNodeEvacStrategy(NoHooksLU):
- """Computes the node evacuation strategy.
+class LUNodeEvacuate(NoHooksLU):
+ """Evacuates instances off a list of nodes.
"""
REQ_BGL = False
_CheckIAllocatorOrNode(self, "iallocator", "remote_node")
def ExpandNames(self):
- self.op.nodes = _GetWantedNodes(self, self.op.nodes)
- self.needed_locks = locks = {}
+ self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
+
+ if self.op.remote_node is not None:
+ self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
+ assert self.op.remote_node
+
+ if self.op.remote_node == self.op.node_name:
+ raise errors.OpPrereqError("Can not use evacuated node as a new"
+ " secondary node", errors.ECODE_INVAL)
+
+ if self.op.mode != constants.IALLOCATOR_NEVAC_SEC:
+ raise errors.OpPrereqError("Without the use of an iallocator only"
+ " secondary instances can be evacuated",
+ errors.ECODE_INVAL)
+
+ # Declare locks
+ self.share_locks = _ShareAll()
+ self.needed_locks = {
+ locking.LEVEL_INSTANCE: [],
+ locking.LEVEL_NODEGROUP: [],
+ locking.LEVEL_NODE: [],
+ }
+
if self.op.remote_node is None:
- locks[locking.LEVEL_NODE] = locking.ALL_SET
+ # Iallocator will choose any node(s) in the same group
+ group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
else:
- self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
- locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
+ group_nodes = frozenset([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 []
+ # Determine nodes to be locked
+ self.lock_nodes = set([self.op.node_name]) | group_nodes
+
+ def _DetermineInstances(self):
+ """Builds list of instances to operate on.
+
+ """
+ assert self.op.mode in constants.IALLOCATOR_NEVAC_MODES
+
+ if self.op.mode == constants.IALLOCATOR_NEVAC_PRI:
+ # Primary instances only
+ inst_fn = _GetNodePrimaryInstances
+ assert self.op.remote_node is None, \
+ "Evacuating primary instances requires iallocator"
+ elif self.op.mode == constants.IALLOCATOR_NEVAC_SEC:
+ # Secondary instances only
+ inst_fn = _GetNodeSecondaryInstances
+ else:
+ # All instances
+ assert self.op.mode == constants.IALLOCATOR_NEVAC_ALL
+ inst_fn = _GetNodeInstances
+
+ return inst_fn(self.cfg, self.op.node_name)
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_INSTANCE:
+ # Lock instances optimistically, needs verification once node and group
+ # locks have been acquired
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
+ set(i.name for i in self._DetermineInstances())
+
+ elif level == locking.LEVEL_NODEGROUP:
+ # Lock node groups optimistically, needs verification once nodes have
+ # been acquired
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
+
+ elif level == locking.LEVEL_NODE:
+ self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
+
+ def CheckPrereq(self):
+ # Verify locks
+ 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
+
+ wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
+ if owned_groups != wanted_groups:
+ raise errors.OpExecError("Node groups changed since locks were acquired,"
+ " current groups are '%s', used to be '%s'" %
+ (utils.CommaJoin(wanted_groups),
+ utils.CommaJoin(owned_groups)))
+
+ # Determine affected instances
+ self.instances = self._DetermineInstances()
+ self.instance_names = [i.name for i in self.instances]
+
+ if set(self.instance_names) != owned_instances:
+ raise errors.OpExecError("Instances on node '%s' changed since locks"
+ " were acquired, current instances are '%s',"
+ " used to be '%s'" %
+ (self.op.node_name,
+ utils.CommaJoin(self.instance_names),
+ utils.CommaJoin(owned_instances)))
+
+ if self.instance_names:
+ self.LogInfo("Evacuating instances from node '%s': %s",
+ self.op.node_name,
+ utils.CommaJoin(utils.NiceSort(self.instance_names)))
+ else:
+ self.LogInfo("No instances to evacuate from node '%s'",
+ self.op.node_name)
if self.op.remote_node is not None:
- result = []
- for i in instances:
+ for i in self.instances:
if i.primary_node == self.op.remote_node:
raise errors.OpPrereqError("Node %s is the primary node of"
" instance %s, cannot use it as"
" secondary" %
(self.op.remote_node, i.name),
errors.ECODE_INVAL)
- result.append([i.name, self.op.remote_node])
- else:
- ial = IAllocator(self.cfg, self.rpc,
- mode=constants.IALLOCATOR_MODE_MEVAC,
- evac_nodes=self.op.nodes)
- ial.Run(self.op.iallocator, validate=True)
+
+ def Exec(self, feedback_fn):
+ assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
+
+ if not self.instance_names:
+ # No instances to evacuate
+ jobs = []
+
+ elif self.op.iallocator is not None:
+ # TODO: Implement relocation to other group
+ ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
+ evac_mode=self.op.mode,
+ instances=list(self.instance_names))
+
+ ial.Run(self.op.iallocator)
+
if not ial.success:
- raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
- errors.ECODE_NORES)
- result = ial.result
- return result
+ raise errors.OpPrereqError("Can't compute node evacuation using"
+ " iallocator '%s': %s" %
+ (self.op.iallocator, ial.info),
+ errors.ECODE_NORES)
+
+ jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
+
+ elif self.op.remote_node is not None:
+ assert self.op.mode == constants.IALLOCATOR_NEVAC_SEC
+ jobs = [
+ [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
+ remote_node=self.op.remote_node,
+ disks=[],
+ mode=constants.REPLACE_DISK_CHG,
+ early_release=self.op.early_release)]
+ for instance_name in self.instance_names
+ ]
+
+ else:
+ raise errors.ProgrammerError("No iallocator or remote node")
+
+ return ResultWithJobs(jobs)
+
+
+def _SetOpEarlyRelease(early_release, op):
+ """Sets C{early_release} flag on opcodes if available.
+
+ """
+ try:
+ op.early_release = early_release
+ except AttributeError:
+ assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
+
+ return op
+
+
+def _NodeEvacDest(use_nodes, group, nodes):
+ """Returns group or nodes depending on caller's choice.
+
+ """
+ if use_nodes:
+ return utils.CommaJoin(nodes)
+ else:
+ return group
+
+
+def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
+ """Unpacks the result of change-group and node-evacuate iallocator requests.
+
+ Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
+ L{constants.IALLOCATOR_MODE_CHG_GROUP}.
+
+ @type lu: L{LogicalUnit}
+ @param lu: Logical unit instance
+ @type alloc_result: tuple/list
+ @param alloc_result: Result from iallocator
+ @type early_release: bool
+ @param early_release: Whether to release locks early if possible
+ @type use_nodes: bool
+ @param use_nodes: Whether to display node names instead of groups
+
+ """
+ (moved, failed, jobs) = alloc_result
+
+ if failed:
+ lu.LogWarning("Unable to evacuate instances %s",
+ utils.CommaJoin("%s (%s)" % (name, reason)
+ for (name, reason) in failed))
+
+ if moved:
+ lu.LogInfo("Instances to be moved: %s",
+ utils.CommaJoin("%s (to %s)" %
+ (name, _NodeEvacDest(use_nodes, group, nodes))
+ for (name, group, nodes) in moved))
+
+ return [map(compat.partial(_SetOpEarlyRelease, early_release),
+ map(opcodes.OpCode.LoadOpCode, ops))
+ for ops in jobs]
class LUInstanceGrowDisk(LogicalUnit):
self.wanted_names = None
if self.op.use_locking:
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ self.share_locks = _ShareAll()
if self.wanted_names is None:
self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
self.needed_locks[locking.LEVEL_NODE] = []
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
def DeclareLocks(self, level):
"""
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
dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
if dev.children:
- dev_children = [self._ComputeDiskStatus(instance, snode, child)
- for child in dev.children]
+ dev_children = map(compat.partial(self._ComputeDiskStatus,
+ instance, snode),
+ dev.children)
else:
dev_children = []
cluster = self.cfg.GetClusterInfo()
- for instance in self.wanted_instances:
- if not self.op.static:
+ pri_nodes = self.cfg.GetMultiNodeInfo(i.primary_node
+ for i in self.wanted_instances)
+ for instance, (_, pnode) in zip(self.wanted_instances, pri_nodes):
+ if self.op.static or pnode.offline:
+ remote_state = None
+ if pnode.offline:
+ self.LogWarning("Primary node %s is marked offline, returning static"
+ " information only for instance %s" %
+ (pnode.name, instance.name))
+ else:
remote_info = self.rpc.call_instance_info(instance.primary_node,
instance.name,
instance.hypervisor)
remote_state = "up"
else:
remote_state = "down"
- else:
- remote_state = None
+
if instance.admin_up:
config_state = "up"
else:
config_state = "down"
- disks = [self._ComputeDiskStatus(instance, None, device)
- for device in instance.disks]
+ disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
+ instance.disks)
result[instance.name] = {
"name": instance.name,
raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
errors.ECODE_INVAL)
- nic_bridge = nic_dict.get('bridge', None)
+ nic_bridge = nic_dict.get("bridge", None)
nic_link = nic_dict.get(constants.INIC_LINK, None)
if nic_bridge and nic_link:
raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
" at the same time", errors.ECODE_INVAL)
elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
- nic_dict['bridge'] = None
+ nic_dict["bridge"] = None
elif nic_link and nic_link.lower() == constants.VALUE_NONE:
nic_dict[constants.INIC_LINK] = None
"""
args = dict()
if constants.BE_MEMORY in self.be_new:
- args['memory'] = self.be_new[constants.BE_MEMORY]
+ args["memory"] = self.be_new[constants.BE_MEMORY]
if constants.BE_VCPUS in self.be_new:
- args['vcpus'] = self.be_new[constants.BE_VCPUS]
+ args["vcpus"] = self.be_new[constants.BE_VCPUS]
# TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
# information at all.
if self.op.nics:
- args['nics'] = []
+ args["nics"] = []
nic_override = dict(self.op.nics)
for idx, nic in enumerate(self.instance.nics):
if idx in nic_override:
nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
mode = nicparams[constants.NIC_MODE]
link = nicparams[constants.NIC_LINK]
- args['nics'].append((ip, mac, mode, link))
+ args["nics"].append((ip, mac, mode, link))
if constants.DDM_ADD in nic_override:
ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
nicparams = self.nic_pnew[constants.DDM_ADD]
mode = nicparams[constants.NIC_MODE]
link = nicparams[constants.NIC_LINK]
- args['nics'].append((ip, mac, mode, link))
+ args["nics"].append((ip, mac, mode, link))
elif constants.DDM_REMOVE in nic_override:
- del args['nics'][-1]
+ del args["nics"][-1]
env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
if self.op.disk_template:
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))
- elif not isinstance(pninfo.payload.get('memory_free', None), int):
+ (pnode, msg))
+ elif not isinstance(pninfo.payload.get("memory_free", None), int):
self.warn.append("Node data from primary node %s doesn't contain"
" free memory information" % pnode)
elif instance_info.fail_msg:
instance_info.fail_msg)
else:
if instance_info.payload:
- current_mem = int(instance_info.payload['memory'])
+ current_mem = int(instance_info.payload["memory"])
else:
# Assume instance not running
# (there is a slight race condition here, but it's not very probable,
# and we have no other way to check)
current_mem = 0
miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
- pninfo.payload['memory_free'])
+ pninfo.payload["memory_free"])
if miss_mem > 0:
raise errors.OpPrereqError("This change will prevent the instance"
" from starting, due to %d MB of memory"
continue
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):
+ 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']:
+ elif be_new[constants.BE_MEMORY] > nres.payload["memory_free"]:
raise errors.OpPrereqError("This change will prevent the instance"
" from failover to its secondary node"
" %s, due to not enough memory" % node,
for key in constants.NICS_PARAMETERS
if key in nic_dict])
- if 'bridge' in nic_dict:
- update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
+ if "bridge" in nic_dict:
+ update_params_dict[constants.NIC_LINK] = nic_dict["bridge"]
new_nic_params = _GetUpdatedParams(old_nic_params,
update_params_dict)
else:
nic_ip = old_nic_ip
if nic_ip is None:
- raise errors.OpPrereqError('Cannot set the nic ip to None'
- ' on a routed nic', errors.ECODE_INVAL)
+ raise errors.OpPrereqError("Cannot set the nic ip to None"
+ " on a routed nic", errors.ECODE_INVAL)
if constants.INIC_MAC in nic_dict:
nic_mac = nic_dict[constants.INIC_MAC]
if nic_mac is None:
- raise errors.OpPrereqError('Cannot set the nic mac to None',
+ raise errors.OpPrereqError("Cannot set the nic mac to None",
errors.ECODE_INVAL)
elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
# otherwise generate the mac
}
+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:
not self.op.remove_instance):
assert not activate_disks
feedback_fn("Starting instance %s" % instance.name)
- result = self.rpc.call_instance_start(src_node, instance, None, None)
+ result = self.rpc.call_instance_start(src_node, instance,
+ None, None, False)
msg = result.fail_msg
if msg:
feedback_fn("Failed to start instance: %s" % msg)
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
return self.op.new_name
-class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
+class LUGroupEvacuate(LogicalUnit):
+ HPATH = "group-evacuate"
+ HTYPE = constants.HTYPE_GROUP
+ REQ_BGL = False
+
+ def ExpandNames(self):
+ # This raises errors.OpPrereqError on its own:
+ self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+ if self.op.target_groups:
+ self.req_target_uuids = map(self.cfg.LookupNodeGroup,
+ self.op.target_groups)
+ else:
+ self.req_target_uuids = []
+
+ if self.group_uuid in self.req_target_uuids:
+ raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
+ " as a target group (targets are %s)" %
+ (self.group_uuid,
+ utils.CommaJoin(self.req_target_uuids)),
+ errors.ECODE_INVAL)
+
+ self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
+
+ self.share_locks = _ShareAll()
+ self.needed_locks = {
+ locking.LEVEL_INSTANCE: [],
+ locking.LEVEL_NODEGROUP: [],
+ locking.LEVEL_NODE: [],
+ }
+
+ def DeclareLocks(self, level):
+ if level == locking.LEVEL_INSTANCE:
+ assert not self.needed_locks[locking.LEVEL_INSTANCE]
+
+ # Lock instances optimistically, needs verification once node and group
+ # locks have been acquired
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
+ self.cfg.GetNodeGroupInstances(self.group_uuid)
+
+ elif level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+ if self.req_target_uuids:
+ lock_groups = set([self.group_uuid] + self.req_target_uuids)
+
+ # Lock all groups used by instances optimistically; this requires going
+ # via the node before it's locked, requiring verification later on
+ lock_groups.update(group_uuid
+ for instance_name in
+ self.owned_locks(locking.LEVEL_INSTANCE)
+ for group_uuid in
+ self.cfg.GetInstanceNodeGroups(instance_name))
+ 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:
+ # This will only lock the nodes in the group to be evacuated which
+ # contain actual instances
+ self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+ self._LockInstancesNodes()
+
+ # 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.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
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+ # Get instance information
+ 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 owned_nodes.issuperset(inst.all_nodes), \
+ "Instance %s's nodes changed while we kept the lock" % instance_name
+
+ 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
+ self.target_uuids = self.req_target_uuids
+ else:
+ # All groups except the one to be evacuated are potential targets
+ self.target_uuids = [group_uuid for group_uuid in owned_groups
+ if group_uuid != self.group_uuid]
+
+ if not self.target_uuids:
+ raise errors.OpPrereqError("There are no possible target groups",
+ errors.ECODE_INVAL)
+
+ def BuildHooksEnv(self):
+ """Build hooks env.
+
+ """
+ return {
+ "GROUP_NAME": self.op.group_name,
+ "TARGET_GROUPS": " ".join(self.target_uuids),
+ }
+
+ def BuildHooksNodes(self):
+ """Build hooks nodes.
+
+ """
+ mn = self.cfg.GetMasterNode()
+
+ 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.owned_locks(locking.LEVEL_INSTANCE))
+
+ assert self.group_uuid not in self.target_uuids
+
+ ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
+ instances=instances, target_groups=self.target_uuids)
+
+ ial.Run(self.op.iallocator)
+
+ if not ial.success:
+ raise errors.OpPrereqError("Can't compute group evacuation using"
+ " iallocator '%s': %s" %
+ (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 evacuating node group %s",
+ len(jobs), self.op.group_name)
+
+ return ResultWithJobs(jobs)
+
+
+class TagsLU(NoHooksLU): # pylint: disable=W0223
"""Generic tags LU.
This is an abstract class which is the parent of all the other tags LUs.
TagsLU.ExpandNames(self)
# Share locks as this is only a read operation
- self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+ self.share_locks = _ShareAll()
def Exec(self, feedback_fn):
"""Returns the tag list.
# 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.reloc_mode = None
+ self.evac_mode = None
self.target_groups = []
# computed fields
self.required_nodes = None
hypervisor_name = self.hypervisor
elif self.mode == constants.IALLOCATOR_MODE_RELOC:
hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
- elif self.mode in (constants.IALLOCATOR_MODE_MEVAC,
- constants.IALLOCATOR_MODE_MRELOC):
+ else:
hypervisor_name = cluster_info.enabled_hypervisors[0]
node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
nname)
remote_info = nresult.payload
- for attr in ['memory_total', 'memory_free', 'memory_dom0',
- 'vg_size', 'vg_free', 'cpu_total']:
+ for attr in ["memory_total", "memory_free", "memory_dom0",
+ "vg_size", "vg_free", "cpu_total"]:
if attr not in remote_info:
raise errors.OpExecError("Node '%s' didn't return attribute"
" '%s'" % (nname, attr))
if iinfo.name not in node_iinfo[nname].payload:
i_used_mem = 0
else:
- i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
+ i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
- remote_info['memory_free'] -= max(0, i_mem_diff)
+ remote_info["memory_free"] -= max(0, i_mem_diff)
if iinfo.admin_up:
i_p_up_mem += beinfo[constants.BE_MEMORY]
# compute memory used by instances
pnr_dyn = {
- "total_memory": remote_info['memory_total'],
- "reserved_memory": remote_info['memory_dom0'],
- "free_memory": remote_info['memory_free'],
- "total_disk": remote_info['vg_size'],
- "free_disk": remote_info['vg_free'],
- "total_cpus": remote_info['cpu_total'],
+ "total_memory": remote_info["memory_total"],
+ "reserved_memory": remote_info["memory_dom0"],
+ "free_memory": remote_info["memory_free"],
+ "total_disk": remote_info["vg_size"],
+ "free_disk": remote_info["vg_free"],
+ "total_cpus": remote_info["cpu_total"],
"i_pri_memory": i_p_mem,
"i_pri_up_memory": i_p_up_mem,
}
}
return request
- def _AddEvacuateNodes(self):
- """Add evacuate nodes data to allocator structure.
+ def _AddNodeEvacuate(self):
+ """Get data for node-evacuate requests.
"""
- request = {
- "evac_nodes": self.evac_nodes
+ return {
+ "instances": self.instances,
+ "evac_mode": self.evac_mode,
}
- return request
- def _AddMultiRelocate(self):
- """Get data for multi-relocate requests.
+ def _AddChangeGroup(self):
+ """Get data for node-evacuate requests.
"""
return {
"instances": self.instances,
- "reloc_mode": self.reloc_mode,
"target_groups": self.target_groups,
}
self.in_text = serializer.Dump(self.in_data)
_STRING_LIST = ht.TListOf(ht.TString)
+ _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
+ # pylint: disable=E1101
+ # Class '...' has no 'OP_ID' member
+ "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
+ opcodes.OpInstanceMigrate.OP_ID,
+ opcodes.OpInstanceReplaceDisks.OP_ID])
+ })))
+
+ _NEVAC_MOVED = \
+ ht.TListOf(ht.TAnd(ht.TIsLength(3),
+ ht.TItems([ht.TNonEmptyString,
+ ht.TNonEmptyString,
+ ht.TListOf(ht.TNonEmptyString),
+ ])))
+ _NEVAC_FAILED = \
+ ht.TListOf(ht.TAnd(ht.TIsLength(2),
+ ht.TItems([ht.TNonEmptyString,
+ ht.TMaybeString,
+ ])))
+ _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
+ ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
+
_MODE_DATA = {
constants.IALLOCATOR_MODE_ALLOC:
(_AddNewInstance,
(_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_MRELOC:
- (_AddMultiRelocate, [
+ constants.IALLOCATOR_MODE_NODE_EVAC:
+ (_AddNodeEvacuate, [
+ ("instances", _STRING_LIST),
+ ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
+ ], _NEVAC_RESULT),
+ constants.IALLOCATOR_MODE_CHG_GROUP:
+ (_AddChangeGroup, [
("instances", _STRING_LIST),
- ("reloc_mode", ht.TElemOf(constants.IALLOCATOR_MRELOC_MODES)),
("target_groups", _STRING_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])
- })))),
+ ], _NEVAC_RESULT),
}
def Run(self, name, validate=True, call_fn=None):
(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.out_data = rdict
elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
fname = _ExpandInstanceName(self.cfg, self.op.name)
self.op.name = fname
- self.relocate_from = 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 == constants.IALLOCATOR_MODE_MRELOC:
- if self.op.instances:
- self.op.instances = _GetWantedInstances(self, self.op.instances)
- else:
- raise errors.OpPrereqError("Missing instances to relocate",
- errors.ECODE_INVAL)
+ self.relocate_from = \
+ list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
+ 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)
name=self.op.name,
relocate_from=list(self.relocate_from),
)
- elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
+ elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
ial = IAllocator(self.cfg, self.rpc,
mode=self.op.mode,
- evac_nodes=self.op.evac_nodes)
- elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
+ 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,
- reloc_mode=self.op.reloc_mode,
- target_groups=self.op.target_groups)
+ evac_mode=self.op.evac_mode)
else:
raise errors.ProgrammerError("Uncatched mode %s in"
" LUTestAllocator.Exec", self.op.mode)
projects / ganeti-local / blobdiff