self.owned_locks = context.glm.list_owned
self.context = context
self.rpc = rpc_runner
- # Dicts used to declare locking needs to mcpu
+
+ # Dictionaries used to declare locking needs to mcpu
self.needed_locks = None
self.share_locks = dict.fromkeys(locking.LEVELS, 0)
+ self.opportunistic_locks = dict.fromkeys(locking.LEVELS, False)
+
self.add_locks = {}
self.remove_locks = {}
+
# Used to force good behavior when calling helper functions
self.recalculate_locks = {}
+
# logging
self.Log = processor.Log # pylint: disable=C0103
self.LogWarning = processor.LogWarning # pylint: disable=C0103
try:
hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
except Exception, err: # pylint: disable=W0703
- lu.LogWarning("Errors occurred running hooks on %s: %s" % (node_name, err))
+ lu.LogWarning("Errors occurred running hooks on %s: %s",
+ node_name, err)
def _CheckOutputFields(static, dynamic, selected):
"""
if msg is None:
- msg = "can't use instance from outside %s states" % ", ".join(req_states)
+ msg = ("can't use instance from outside %s states" %
+ utils.CommaJoin(req_states))
if instance.admin_state not in req_states:
raise errors.OpPrereqError("Instance '%s' is marked to be %s, %s" %
(instance.name, instance.admin_state, msg),
@param tags: the tags of the network
"""
- env = dict()
+ env = {}
if name:
env["NETWORK_NAME"] = name
if subnet:
locking.LEVEL_INSTANCE: inst_names,
locking.LEVEL_NODEGROUP: [self.group_uuid],
locking.LEVEL_NODE: [],
+
+ # This opcode is run by watcher every five minutes and acquires all nodes
+ # for a group. It doesn't run for a long time, so it's better to acquire
+ # the node allocation lock as well.
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
self.share_locks = _ShareAll()
ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
self.group_info)
err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig)
- _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err))
+ _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err),
+ code=self.ETYPE_WARNING)
for node in node_vol_should:
n_img = node_image[node]
locking.LEVEL_INSTANCE: [],
locking.LEVEL_NODEGROUP: [],
locking.LEVEL_NODE: [],
+
+ # This opcode is acquires all node locks in a group. LUClusterVerifyDisks
+ # starts one instance of this opcode for every group, which means all
+ # nodes will be locked for a short amount of time, so it's better to
+ # acquire the node allocation lock as well.
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
def DeclareLocks(self, level):
def ExpandNames(self):
if self.op.instances:
self.wanted_names = _GetWantedInstances(self, self.op.instances)
+ # Not getting the node allocation lock as only a specific set of
+ # instances (and their nodes) is going to be acquired
self.needed_locks = {
locking.LEVEL_NODE_RES: [],
locking.LEVEL_INSTANCE: self.wanted_names,
self.needed_locks = {
locking.LEVEL_NODE_RES: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
+
+ # This opcode is acquires the node locks for all instances
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
+
self.share_locks = {
locking.LEVEL_NODE_RES: 1,
locking.LEVEL_INSTANCE: 0,
+ locking.LEVEL_NODE_ALLOC: 1,
}
def DeclareLocks(self, level):
def ExpandNames(self):
# FIXME: in the future maybe other cluster params won't require checking on
# all nodes to be modified.
+ # FIXME: This opcode changes cluster-wide settings. Is acquiring all
+ # resource locks the right thing, shouldn't it be the BGL instead?
self.needed_locks = {
locking.LEVEL_NODE: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
locking.LEVEL_NODEGROUP: locking.ALL_SET,
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
- self.share_locks = {
- locking.LEVEL_NODE: 1,
- locking.LEVEL_INSTANCE: 1,
- locking.LEVEL_NODEGROUP: 1,
- }
+ self.share_locks = _ShareAll()
def BuildHooksEnv(self):
"""Build hooks env.
if msg:
msg = ("Copy of file %s to node %s failed: %s" %
(fname, to_node, msg))
- lu.proc.LogWarning(msg)
+ lu.LogWarning(msg)
def _ComputeAncillaryFiles(cluster, redist):
def ExpandNames(self):
self.needed_locks = {
locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
- self.share_locks[locking.LEVEL_NODE] = 1
+ self.share_locks = _ShareAll()
def Exec(self, feedback_fn):
"""Redistribute the configuration.
disks = _ExpandCheckDisks(instance, disks)
if not oneshot:
- lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
+ lu.LogInfo("Waiting for instance %s to sync disks", instance.name)
node = instance.primary_node
max_time = mstat.estimated_time
else:
rem_time = "no time estimate"
- lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
- (disks[i].iv_name, mstat.sync_percent, rem_time))
+ lu.LogInfo("- device %s: %5.2f%% done, %s",
+ disks[i].iv_name, mstat.sync_percent, rem_time)
# if we're done but degraded, let's do a few small retries, to
# make sure we see a stable and not transient situation; therefore
time.sleep(min(60, max_time))
if done:
- lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
+ lu.LogInfo("Instance %s's disks are in sync", instance.name)
+
return not cumul_degraded
locking.LEVEL_NODE: lock_names,
}
+ if not self.op.node_names:
+ # Acquire node allocation lock only if all nodes are affected
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+ self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
+
def CheckPrereq(self):
"""Check prerequisites.
if self.do_locking:
# If any non-static field is requested we need to lock the nodes
lu.needed_locks[locking.LEVEL_NODE] = self.wanted
+ lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
def DeclareLocks(self, lu, level):
pass
def ExpandNames(self):
self.share_locks = _ShareAll()
- self.needed_locks = {}
- if not self.op.nodes:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ if self.op.nodes:
+ self.needed_locks = {
+ locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes),
+ }
else:
- self.needed_locks[locking.LEVEL_NODE] = \
- _GetWantedNodes(self, self.op.nodes)
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
def Exec(self, feedback_fn):
"""Computes the list of nodes and their attributes.
def ExpandNames(self):
self.share_locks = _ShareAll()
- self.needed_locks = {}
if self.op.nodes:
- self.needed_locks[locking.LEVEL_NODE] = \
- _GetWantedNodes(self, self.op.nodes)
+ self.needed_locks = {
+ locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes),
+ }
else:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
def Exec(self, feedback_fn):
"""Computes the list of nodes and their attributes.
def ExpandNames(self):
if self.lock_all:
- self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+
+ # Block allocations when all nodes are locked
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
else:
- self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
+ self.needed_locks = {
+ locking.LEVEL_NODE: self.op.node_name,
+ }
# Since modifying a node can have severe effects on currently running
# operations the resource lock is at least acquired in shared mode
self.needed_locks[locking.LEVEL_NODE_RES] = \
self.needed_locks[locking.LEVEL_NODE]
- # Get node resource and instance locks in shared mode; they are not used
- # for anything but read-only access
- self.share_locks[locking.LEVEL_NODE_RES] = 1
- self.share_locks[locking.LEVEL_INSTANCE] = 1
+ # Get all locks except nodes in shared mode; they are not used for anything
+ # but read-only access
+ self.share_locks = _ShareAll()
+ self.share_locks[locking.LEVEL_NODE] = 0
+ self.share_locks[locking.LEVEL_NODE_RES] = 0
+ self.share_locks[locking.LEVEL_NODE_ALLOC] = 0
if self.lock_instances:
self.needed_locks[locking.LEVEL_INSTANCE] = \
if master_singlehomed and self.op.secondary_ip != node.primary_ip:
if self.op.force and node.name == master.name:
self.LogWarning("Transitioning from single-homed to multi-homed"
- " cluster. All nodes will require a secondary ip.")
+ " cluster; all nodes will require a secondary IP"
+ " address")
else:
raise errors.OpPrereqError("Changing the secondary ip on a"
" single-homed cluster requires the"
elif not master_singlehomed and self.op.secondary_ip == node.primary_ip:
if self.op.force and node.name == master.name:
self.LogWarning("Transitioning from multi-homed to single-homed"
- " cluster. Secondary IPs will have to be removed.")
+ " cluster; secondary IP addresses will have to be"
+ " removed")
else:
raise errors.OpPrereqError("Cannot set the secondary IP to be the"
" same as the primary IP on a multi-homed"
if msg:
is_offline_secondary = (node in instance.secondary_nodes and
result.offline)
- lu.proc.LogWarning("Could not prepare block device %s on node %s"
- " (is_primary=False, pass=1): %s",
- inst_disk.iv_name, node, msg)
+ lu.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=False, pass=1): %s",
+ inst_disk.iv_name, node, msg)
if not (ignore_secondaries or is_offline_secondary):
disks_ok = False
True, idx)
msg = result.fail_msg
if msg:
- lu.proc.LogWarning("Could not prepare block device %s on node %s"
- " (is_primary=True, pass=2): %s",
- inst_disk.iv_name, node, msg)
+ lu.LogWarning("Could not prepare block device %s on node %s"
+ " (is_primary=True, pass=2): %s",
+ inst_disk.iv_name, node, msg)
disks_ok = False
else:
dev_path = result.payload
if not disks_ok:
_ShutdownInstanceDisks(lu, instance)
if force is not None and not force:
- lu.proc.LogWarning("", hint="If the message above refers to a"
- " secondary node,"
- " you can retry the operation using '--force'.")
+ lu.LogWarning("",
+ hint=("If the message above refers to a secondary node,"
+ " you can retry the operation using '--force'"))
raise errors.OpExecError("Disk consistency error")
self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
if self.primary_offline and self.op.ignore_offline_nodes:
- self.proc.LogWarning("Ignoring offline primary node")
+ self.LogWarning("Ignoring offline primary node")
if self.op.hvparams or self.op.beparams:
- self.proc.LogWarning("Overridden parameters are ignored")
+ self.LogWarning("Overridden parameters are ignored")
else:
_CheckNodeOnline(self, instance.primary_node)
if self.primary_offline:
assert self.op.ignore_offline_nodes
- self.proc.LogInfo("Primary node offline, marked instance as started")
+ self.LogInfo("Primary node offline, marked instance as started")
else:
node_current = instance.primary_node
self.cfg.GetNodeInfo(self.instance.primary_node).offline
if self.primary_offline and self.op.ignore_offline_nodes:
- self.proc.LogWarning("Ignoring offline primary node")
+ self.LogWarning("Ignoring offline primary node")
else:
_CheckNodeOnline(self, self.instance.primary_node)
if self.primary_offline:
assert self.op.ignore_offline_nodes
- self.proc.LogInfo("Primary node offline, marked instance as stopped")
+ self.LogInfo("Primary node offline, marked instance as stopped")
else:
result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
msg = result.fail_msg
if msg:
- self.proc.LogWarning("Could not shutdown instance: %s" % msg)
+ self.LogWarning("Could not shutdown instance: %s", msg)
_ShutdownInstanceDisks(self, instance)
def ExpandNames(self):
self._ExpandAndLockInstance()
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
if self.op.nodes:
self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
if self.op.iallocator:
# iallocator will select a new node in the same group
self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
self.needed_locks[locking.LEVEL_NODE_RES] = []
def DeclareLocks(self, level):
for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP):
self.needed_locks[locking.LEVEL_NODE].extend(
self.cfg.GetNodeGroup(group_uuid).members)
+
+ assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
elif not self.op.nodes:
self._LockInstancesNodes(primary_only=False)
elif level == locking.LEVEL_NODE_RES:
# Release unneeded node and node resource locks
_ReleaseLocks(self, locking.LEVEL_NODE, keep=self.op.nodes)
_ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=self.op.nodes)
+ _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
def Exec(self, feedback_fn):
"""Recreate the disks.
# Change the instance lock. This is definitely safe while we hold the BGL.
# Otherwise the new lock would have to be added in acquired mode.
assert self.REQ_BGL
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
self.glm.remove(locking.LEVEL_INSTANCE, old_name)
self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
msg = ("Could not run OS rename script for instance %s on node %s"
" (but the instance has been renamed in Ganeti): %s" %
(inst.name, inst.primary_node, msg))
- self.proc.LogWarning(msg)
+ self.LogWarning(msg)
finally:
_ShutdownInstanceDisks(self, inst)
return self.iq.OldStyleQuery(self)
+def _ExpandNamesForMigration(lu):
+ """Expands names for use with L{TLMigrateInstance}.
+
+ @type lu: L{LogicalUnit}
+
+ """
+ if lu.op.target_node is not None:
+ lu.op.target_node = _ExpandNodeName(lu.cfg, lu.op.target_node)
+
+ lu.needed_locks[locking.LEVEL_NODE] = []
+ lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+ lu.needed_locks[locking.LEVEL_NODE_RES] = []
+ lu.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
+
+ # The node allocation lock is actually only needed for replicated instances
+ # (e.g. DRBD8) and if an iallocator is used.
+ lu.needed_locks[locking.LEVEL_NODE_ALLOC] = []
+
+
+def _DeclareLocksForMigration(lu, level):
+ """Declares locks for L{TLMigrateInstance}.
+
+ @type lu: L{LogicalUnit}
+ @param level: Lock level
+
+ """
+ if level == locking.LEVEL_NODE_ALLOC:
+ assert lu.op.instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
+
+ instance = lu.cfg.GetInstanceInfo(lu.op.instance_name)
+
+ if instance.disk_template in constants.DTS_EXT_MIRROR:
+ if lu.op.target_node is None:
+ lu.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+ else:
+ lu.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
+ lu.op.target_node]
+ del lu.recalculate_locks[locking.LEVEL_NODE]
+ else:
+ lu._LockInstancesNodes() # pylint: disable=W0212
+
+ elif level == locking.LEVEL_NODE:
+ # Node locks are declared together with the node allocation lock
+ assert lu.needed_locks[locking.LEVEL_NODE]
+
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ lu.needed_locks[locking.LEVEL_NODE_RES] = \
+ _CopyLockList(lu.needed_locks[locking.LEVEL_NODE])
+
+
class LUInstanceFailover(LogicalUnit):
"""Failover an instance.
def ExpandNames(self):
self._ExpandAndLockInstance()
+ _ExpandNamesForMigration(self)
- if self.op.target_node is not None:
- self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
-
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- self.needed_locks[locking.LEVEL_NODE_RES] = []
- self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
+ self._migrater = \
+ TLMigrateInstance(self, self.op.instance_name, False, True, False,
+ self.op.ignore_consistency, True,
+ self.op.shutdown_timeout, self.op.ignore_ipolicy)
- ignore_consistency = self.op.ignore_consistency
- shutdown_timeout = self.op.shutdown_timeout
- self._migrater = TLMigrateInstance(self, self.op.instance_name,
- cleanup=False,
- failover=True,
- ignore_consistency=ignore_consistency,
- shutdown_timeout=shutdown_timeout,
- ignore_ipolicy=self.op.ignore_ipolicy)
self.tasklets = [self._migrater]
def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
- if instance.disk_template in constants.DTS_EXT_MIRROR:
- if self.op.target_node is None:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
- else:
- self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
- self.op.target_node]
- del self.recalculate_locks[locking.LEVEL_NODE]
- else:
- self._LockInstancesNodes()
- elif level == locking.LEVEL_NODE_RES:
- # Copy node locks
- self.needed_locks[locking.LEVEL_NODE_RES] = \
- _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+ _DeclareLocksForMigration(self, level)
def BuildHooksEnv(self):
"""Build hooks env.
def ExpandNames(self):
self._ExpandAndLockInstance()
-
- if self.op.target_node is not None:
- self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
-
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
- self.needed_locks[locking.LEVEL_NODE] = []
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ _ExpandNamesForMigration(self)
self._migrater = \
- TLMigrateInstance(self, self.op.instance_name,
- cleanup=self.op.cleanup,
- failover=False,
- fallback=self.op.allow_failover,
- allow_runtime_changes=self.op.allow_runtime_changes,
- ignore_ipolicy=self.op.ignore_ipolicy)
+ TLMigrateInstance(self, self.op.instance_name, self.op.cleanup,
+ False, self.op.allow_failover, False,
+ self.op.allow_runtime_changes,
+ constants.DEFAULT_SHUTDOWN_TIMEOUT,
+ self.op.ignore_ipolicy)
+
self.tasklets = [self._migrater]
def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE:
- instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
- if instance.disk_template in constants.DTS_EXT_MIRROR:
- if self.op.target_node is None:
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
- else:
- self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
- self.op.target_node]
- del self.recalculate_locks[locking.LEVEL_NODE]
- else:
- self._LockInstancesNodes()
- elif level == locking.LEVEL_NODE_RES:
- # Copy node locks
- self.needed_locks[locking.LEVEL_NODE_RES] = \
- _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+ _DeclareLocksForMigration(self, level)
def BuildHooksEnv(self):
"""Build hooks env.
msg = result.fail_msg
if msg:
if self.op.ignore_consistency:
- self.proc.LogWarning("Could not shutdown instance %s on node %s."
- " Proceeding anyway. Please make sure node"
- " %s is down. Error details: %s",
- instance.name, source_node, source_node, msg)
+ self.LogWarning("Could not shutdown instance %s on node %s."
+ " Proceeding anyway. Please make sure node"
+ " %s is down. Error details: %s",
+ instance.name, source_node, source_node, msg)
else:
raise errors.OpExecError("Could not shutdown instance %s on"
" node %s: %s" %
_MIGRATION_POLL_INTERVAL = 1 # seconds
_MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
- def __init__(self, lu, instance_name, cleanup=False,
- failover=False, fallback=False,
- ignore_consistency=False,
- allow_runtime_changes=True,
- shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT,
- ignore_ipolicy=False):
+ def __init__(self, lu, instance_name, cleanup, failover, fallback,
+ ignore_consistency, allow_runtime_changes, shutdown_timeout,
+ ignore_ipolicy):
"""Initializes this class.
"""
errors.ECODE_STATE)
if instance.disk_template in constants.DTS_EXT_MIRROR:
+ assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC)
+
_CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
if self.lu.op.iallocator:
# in the LU
_ReleaseLocks(self.lu, locking.LEVEL_NODE,
keep=[instance.primary_node, self.target_node])
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
else:
+ assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
secondary_nodes = instance.secondary_nodes
if not secondary_nodes:
raise errors.ConfigurationError("No secondary node but using"
"""Run the allocator based on input opcode.
"""
+ assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC)
+
# FIXME: add a self.ignore_ipolicy option
req = iallocator.IAReqRelocate(name=self.instance_name,
relocate_from=[self.instance.primary_node])
osname, node)
-def _CreateInstanceAllocRequest(op, disks, nics, beparams):
+def _CreateInstanceAllocRequest(op, disks, nics, beparams, node_whitelist):
"""Wrapper around IAReqInstanceAlloc.
@param op: The instance opcode
@param disks: The computed disks
@param nics: The computed nics
@param beparams: The full filled beparams
+ @param node_whitelist: List of nodes which should appear as online to the
+ allocator (unless the node is already marked offline)
@returns: A filled L{iallocator.IAReqInstanceAlloc}
spindle_use=spindle_use,
disks=disks,
nics=[n.ToDict() for n in nics],
- hypervisor=op.hypervisor)
+ hypervisor=op.hypervisor,
+ node_whitelist=node_whitelist)
-def _ComputeNics(op, cluster, default_ip, cfg, proc):
+def _ComputeNics(op, cluster, default_ip, cfg, ec_id):
"""Computes the nics.
@param op: The instance opcode
@param cluster: Cluster configuration object
@param default_ip: The default ip to assign
@param cfg: An instance of the configuration object
- @param proc: The executer instance
+ @param ec_id: Execution context ID
@returns: The build up nics
try:
# TODO: We need to factor this out
- cfg.ReserveMAC(mac, proc.GetECId())
+ cfg.ReserveMAC(mac, ec_id)
except errors.ReservationError:
raise errors.OpPrereqError("MAC address %s already in use"
" in cluster" % mac,
# specifying a group on instance creation and then selecting nodes from
# that group
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
- self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
+ if self.op.opportunistic_locking:
+ self.opportunistic_locks[locking.LEVEL_NODE] = True
+ self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
else:
self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
nodelist = [self.op.pnode]
self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
nodelist.append(self.op.snode)
self.needed_locks[locking.LEVEL_NODE] = nodelist
- # Lock resources of instance's primary and secondary nodes (copy to
- # prevent accidential modification)
- self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
# in case of import lock the source node too
if self.op.mode == constants.INSTANCE_IMPORT:
if src_node is None:
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
self.op.src_node = None
if os.path.isabs(src_path):
raise errors.OpPrereqError("Importing an instance from a path"
self.op.src_path = src_path = \
utils.PathJoin(pathutils.EXPORT_DIR, src_path)
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
def _RunAllocator(self):
"""Run the allocator based on input opcode.
"""
+ if self.op.opportunistic_locking:
+ # Only consider nodes for which a lock is held
+ node_whitelist = self.owned_locks(locking.LEVEL_NODE)
+ else:
+ node_whitelist = None
+
#TODO Export network to iallocator so that it chooses a pnode
# in a nodegroup that has the desired network connected to
req = _CreateInstanceAllocRequest(self.op, self.disks,
- self.nics, self.be_full)
+ self.nics, self.be_full,
+ node_whitelist)
ial = iallocator.IAllocator(self.cfg, self.rpc, req)
ial.Run(self.op.iallocator)
if not ial.success:
+ # When opportunistic locks are used only a temporary failure is generated
+ if self.op.opportunistic_locking:
+ ecode = errors.ECODE_TEMP_NORES
+ else:
+ ecode = errors.ECODE_NORES
+
raise errors.OpPrereqError("Can't compute nodes using"
" iallocator '%s': %s" %
(self.op.iallocator, ial.info),
- errors.ECODE_NORES)
+ ecode)
+
self.op.pnode = ial.result[0]
self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
self.op.instance_name, self.op.iallocator,
# NIC buildup
self.nics = _ComputeNics(self.op, cluster, self.hostname1.ip, self.cfg,
- self.proc)
+ self.proc.GetECId())
# disk checks/pre-build
default_vg = self.cfg.GetVGName()
self._RunAllocator()
# Release all unneeded node locks
- _ReleaseLocks(self, locking.LEVEL_NODE,
- keep=filter(None, [self.op.pnode, self.op.snode,
- self.op.src_node]))
- _ReleaseLocks(self, locking.LEVEL_NODE_RES,
- keep=filter(None, [self.op.pnode, self.op.snode,
- self.op.src_node]))
+ keep_locks = filter(None, [self.op.pnode, self.op.snode, self.op.src_node])
+ _ReleaseLocks(self, locking.LEVEL_NODE, keep=keep_locks)
+ _ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=keep_locks)
+ _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES)), \
+ "Node locks differ from node resource locks"
#### node related checks
if baddisks:
raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
" cannot be adopted" %
- (", ".join(baddisks),
+ (utils.CommaJoin(baddisks),
constants.ADOPTABLE_BLOCKDEV_ROOT),
errors.ECODE_INVAL)
assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
self.owned_locks(locking.LEVEL_NODE)), \
"Node locks differ from node resource locks"
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
ht_kind = self.op.hypervisor
if ht_kind in constants.HTS_REQ_PORT:
"""
self.share_locks = _ShareAll()
- self.needed_locks = {}
+ self.needed_locks = {
+ # iallocator will select nodes and even if no iallocator is used,
+ # collisions with LUInstanceCreate should be avoided
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
if self.op.iallocator:
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
+
+ if self.op.opportunistic_locking:
+ self.opportunistic_locks[locking.LEVEL_NODE] = True
+ self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
else:
nodeslist = []
for inst in self.op.instances:
"""
cluster = self.cfg.GetClusterInfo()
default_vg = self.cfg.GetVGName()
+ ec_id = self.proc.GetECId()
+
+ if self.op.opportunistic_locking:
+ # Only consider nodes for which a lock is held
+ node_whitelist = self.owned_locks(locking.LEVEL_NODE)
+ else:
+ node_whitelist = None
+
insts = [_CreateInstanceAllocRequest(op, _ComputeDisks(op, default_vg),
_ComputeNics(op, cluster, None,
- self.cfg, self.proc),
- _ComputeFullBeParams(op, cluster))
+ self.cfg, ec_id),
+ _ComputeFullBeParams(op, cluster),
+ node_whitelist)
for op in self.op.instances]
+
req = iallocator.IAReqMultiInstanceAlloc(instances=insts)
ial = iallocator.IAllocator(self.cfg, self.rpc, req)
if self.op.iallocator is not None:
# iallocator will select a new node in the same group
self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
self.needed_locks[locking.LEVEL_NODE_RES] = []
self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
self.op.iallocator, self.op.remote_node,
- self.op.disks, False, self.op.early_release,
+ self.op.disks, self.op.early_release,
self.op.ignore_ipolicy)
self.tasklets = [self.replacer]
if self.op.iallocator is not None:
assert self.op.remote_node is None
assert not self.needed_locks[locking.LEVEL_NODE]
+ assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
# Lock member nodes of all locked groups
self.needed_locks[locking.LEVEL_NODE] = \
for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
for node_name in self.cfg.GetNodeGroup(group_uuid).members]
else:
+ assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
self._LockInstancesNodes()
+
elif level == locking.LEVEL_NODE_RES:
# Reuse node locks
self.needed_locks[locking.LEVEL_NODE_RES] = \
"""
def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
- disks, delay_iallocator, early_release, ignore_ipolicy):
+ disks, early_release, ignore_ipolicy):
"""Initializes this class.
"""
self.iallocator_name = iallocator_name
self.remote_node = remote_node
self.disks = disks
- self.delay_iallocator = delay_iallocator
self.early_release = early_release
self.ignore_ipolicy = ignore_ipolicy
len(instance.secondary_nodes),
errors.ECODE_FAULT)
- if not self.delay_iallocator:
- self._CheckPrereq2()
-
- def _CheckPrereq2(self):
- """Check prerequisites, second part.
-
- This function should always be part of CheckPrereq. It was separated and is
- now called from Exec because during node evacuation iallocator was only
- called with an unmodified cluster model, not taking planned changes into
- account.
-
- """
instance = self.instance
secondary_node = instance.secondary_nodes[0]
# Release unneeded node and node resource locks
_ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
_ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
# Release any owned node group
- if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
- _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
+ _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
# Check whether disks are valid
for disk_idx in self.disks:
This dispatches the disk replacement to the appropriate handler.
"""
- if self.delay_iallocator:
- self._CheckPrereq2()
-
if __debug__:
# Verify owned locks before starting operation
owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
(owned_nodes, self.node_secondary_ip.keys()))
assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
self.lu.owned_locks(locking.LEVEL_NODE_RES))
+ assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
assert list(owned_instances) == [self.instance_name], \
continue
for node in nodes:
- self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
+ self.lu.LogInfo("Checking disk/%d on %s", idx, node)
self.cfg.SetDiskID(dev, node)
result = _BlockdevFind(self, node, dev, self.instance)
if idx not in self.disks:
continue
- self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
+ self.lu.LogInfo("Adding storage on %s for disk/%d", node_name, idx)
self.cfg.SetDiskID(dev, node_name)
def _RemoveOldStorage(self, node_name, iv_names):
for name, (_, old_lvs, _) in iv_names.iteritems():
- self.lu.LogInfo("Remove logical volumes for %s" % name)
+ self.lu.LogInfo("Remove logical volumes for %s", name)
for lv in old_lvs:
self.cfg.SetDiskID(lv, node_name)
msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
if msg:
- self.lu.LogWarning("Can't remove old LV: %s" % msg,
+ self.lu.LogWarning("Can't remove old LV: %s", msg,
hint="remove unused LVs manually")
def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
# Step: for each lv, detach+rename*2+attach
self.lu.LogStep(4, steps_total, "Changing drbd configuration")
for dev, old_lvs, new_lvs in iv_names.itervalues():
- self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
+ self.lu.LogInfo("Detaching %s drbd from local storage", dev.iv_name)
result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
old_lvs)
self.cfg.SetDiskID(disk, self.target_node)
# Now that the new lvs have the old name, we can add them to the device
- self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
+ self.lu.LogInfo("Adding new mirror component on %s", self.target_node)
result = self.rpc.call_blockdev_addchildren(self.target_node,
(dev, self.instance), new_lvs)
msg = result.fail_msg
# We have new devices, shutdown the drbd on the old secondary
for idx, dev in enumerate(self.instance.disks):
- self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
+ self.lu.LogInfo("Shutting down drbd for disk/%d on old node", idx)
self.cfg.SetDiskID(dev, self.target_node)
msg = self.rpc.call_blockdev_shutdown(self.target_node,
(dev, self.instance)).fail_msg
errors.ECODE_STATE)
except errors.OpPrereqError, err:
if self.op.ignore_consistency:
- self.proc.LogWarning(str(err.args[0]))
+ self.LogWarning(str(err.args[0]))
else:
raise
if self.op.wait_for_sync:
disk_abort = not _WaitForSync(self, instance, disks=[disk])
if disk_abort:
- self.proc.LogWarning("Disk sync-ing has not returned a good"
- " status; please check the instance")
+ self.LogWarning("Disk syncing has not returned a good status; check"
+ " the instance")
if instance.admin_state != constants.ADMINST_UP:
_SafeShutdownInstanceDisks(self, instance, disks=[disk])
elif instance.admin_state != constants.ADMINST_UP:
- self.proc.LogWarning("Not shutting down the disk even if the instance is"
- " not supposed to be running because no wait for"
- " sync mode was requested")
+ self.LogWarning("Not shutting down the disk even if the instance is"
+ " not supposed to be running because no wait for"
+ " sync mode was requested")
assert self.owned_locks(locking.LEVEL_NODE_RES)
assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
def ExpandNames(self):
self._ExpandAndLockInstance()
+ self.needed_locks[locking.LEVEL_NODEGROUP] = []
# Can't even acquire node locks in shared mode as upcoming changes in
# Ganeti 2.6 will start to modify the node object on disk conversion
self.needed_locks[locking.LEVEL_NODE] = []
self.needed_locks[locking.LEVEL_NODE_RES] = []
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ # Look node group to look up the ipolicy
+ self.share_locks[locking.LEVEL_NODEGROUP] = 1
def DeclareLocks(self, level):
- # TODO: Acquire group lock in shared mode (disk parameters)
- if level == locking.LEVEL_NODE:
+ if level == locking.LEVEL_NODEGROUP:
+ assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+ # Acquire locks for the instance's nodegroups optimistically. Needs
+ # to be verified in CheckPrereq
+ self.needed_locks[locking.LEVEL_NODEGROUP] = \
+ self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+ elif level == locking.LEVEL_NODE:
self._LockInstancesNodes()
if self.op.disk_template and self.op.remote_node:
self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
This runs on the master, primary and secondaries.
"""
- args = dict()
+ args = {}
if constants.BE_MINMEM in self.be_new:
args["minmem"] = self.be_new[constants.BE_MINMEM]
if constants.BE_MAXMEM in self.be_new:
netparams = self.cfg.GetGroupNetParams(new_net, pnode)
if netparams is None:
raise errors.OpPrereqError("No netparams found for the network"
- " %s, propably not connected." % new_net,
+ " %s, probably not connected" % new_net,
errors.ECODE_INVAL)
new_params = dict(netparams)
else:
elif (old_net is not None and
(req_link is not None or req_mode is not None)):
raise errors.OpPrereqError("Not allowed to change link or mode of"
- " a NIC that is connected to a network.",
+ " a NIC that is connected to a network",
errors.ECODE_INVAL)
private.params = new_params
This only checks the instance list against the existing names.
"""
- # checking the new params on the primary/secondary nodes
-
+ assert self.op.instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
cluster = self.cluster = self.cfg.GetClusterInfo()
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+
pnode = instance.primary_node
+ assert pnode in self.owned_locks(locking.LEVEL_NODE)
nodelist = list(instance.all_nodes)
pnode_info = self.cfg.GetNodeInfo(pnode)
self.diskparams = self.cfg.GetInstanceDiskParams(instance)
+ #_CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
+ assert pnode_info.group in self.owned_locks(locking.LEVEL_NODEGROUP)
+ group_info = self.cfg.GetNodeGroup(pnode_info.group)
+
+ # dictionary with instance information after the modification
+ ispec = {}
+
# Prepare disk/NIC modifications
self.diskmod = PrepareContainerMods(self.op.disks, None)
self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate)
raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
" more" % constants.MAX_DISKS,
errors.ECODE_STATE)
+ disk_sizes = [disk.size for disk in instance.disks]
+ disk_sizes.extend(params["size"] for (op, idx, params, private) in
+ self.diskmod if op == constants.DDM_ADD)
+ ispec[constants.ISPEC_DISK_COUNT] = len(disk_sizes)
+ ispec[constants.ISPEC_DISK_SIZE] = disk_sizes
if self.op.offline is not None:
if self.op.offline:
ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
self._CreateNewNic, self._ApplyNicMods, None)
self._new_nics = nics
+ ispec[constants.ISPEC_NIC_COUNT] = len(self._new_nics)
else:
self._new_nics = None
+ ispec[constants.ISPEC_NIC_COUNT] = len(instance.nics)
+
+ if not self.op.ignore_ipolicy:
+ ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
+ group_info)
+
+ # Fill ispec with backend parameters
+ ispec[constants.ISPEC_SPINDLE_USE] = \
+ self.be_new.get(constants.BE_SPINDLE_USE, None)
+ ispec[constants.ISPEC_CPU_COUNT] = self.be_new.get(constants.BE_VCPUS,
+ None)
+
+ # Copy ispec to verify parameters with min/max values separately
+ ispec_max = ispec.copy()
+ ispec_max[constants.ISPEC_MEM_SIZE] = \
+ self.be_new.get(constants.BE_MAXMEM, None)
+ res_max = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_max)
+ ispec_min = ispec.copy()
+ ispec_min[constants.ISPEC_MEM_SIZE] = \
+ self.be_new.get(constants.BE_MINMEM, None)
+ res_min = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_min)
+
+ if (res_max or res_min):
+ # FIXME: Improve error message by including information about whether
+ # the upper or lower limit of the parameter fails the ipolicy.
+ msg = ("Instance allocation to group %s (%s) violates policy: %s" %
+ (group_info, group_info.name,
+ utils.CommaJoin(set(res_max + res_min))))
+ raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
def _ConvertPlainToDrbd(self, feedback_fn):
"""Converts an instance from plain to drbd.
def ExpandNames(self):
self.share_locks = _ShareAll()
+
self.needed_locks = {
locking.LEVEL_NODEGROUP: [],
locking.LEVEL_NODE: [],
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
}
self._ExpandAndLockInstance()
locking.LEVEL_NODE: self.wanted,
}
+ if not self.names:
+ lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
def DeclareLocks(self, lu, level):
pass
# - removing the removal operation altogether
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ # Allocations should be stopped while this LU runs with node locks, but
+ # it doesn't have to be exclusive
+ self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
+ self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
def DeclareLocks(self, level):
"""Last minute lock declaration."""
# All nodes are locked anyway, so nothing to do here.
REQ_BGL = False
def ExpandNames(self):
- self.needed_locks = {}
- # We need all nodes to be locked in order for RemoveExport to work, but we
- # don't need to lock the instance itself, as nothing will happen to it (and
- # we can remove exports also for a removed instance)
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ self.needed_locks = {
+ # We need all nodes to be locked in order for RemoveExport to work, but
+ # we don't need to lock the instance itself, as nothing will happen to it
+ # (and we can remove exports also for a removed instance)
+ locking.LEVEL_NODE: locking.ALL_SET,
+
+ # Removing backups is quick, so blocking allocations is justified
+ locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+ }
+
+ # Allocations should be stopped while this LU runs with node locks, but it
+ # doesn't have to be exclusive
+ self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
def Exec(self, feedback_fn):
"""Remove any export.
else:
top_value = self.op.repeat - 1
for i in range(self.op.repeat):
- self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
+ self.LogInfo("Test delay iteration %d/%d", i, top_value)
self._TestDelay()
self.op.mode, errors.ECODE_INVAL)
if self.op.direction == constants.IALLOCATOR_DIR_OUT:
- if self.op.allocator is None:
+ if self.op.iallocator is None:
raise errors.OpPrereqError("Missing allocator name",
errors.ECODE_INVAL)
elif self.op.direction != constants.IALLOCATOR_DIR_IN:
if self.op.direction == constants.IALLOCATOR_DIR_IN:
result = ial.in_text
else:
- ial.Run(self.op.allocator, validate=False)
+ ial.Run(self.op.iallocator, validate=False)
result = ial.out_text
return result
-# Network LUs
class LUNetworkAdd(LogicalUnit):
"""Logical unit for creating networks.
mn = self.cfg.GetMasterNode()
return ([mn], [mn])
+ def CheckArguments(self):
+ if self.op.mac_prefix:
+ self.op.mac_prefix = \
+ utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix)
+
def ExpandNames(self):
self.network_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
- self.needed_locks = {}
- self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid
- def CheckPrereq(self):
- """Check prerequisites.
+ if self.op.conflicts_check:
+ self.share_locks[locking.LEVEL_NODE] = 1
+ self.needed_locks = {
+ locking.LEVEL_NODE: locking.ALL_SET,
+ }
+ else:
+ self.needed_locks = {}
- This checks that the given group name is not an existing node group
- already.
+ self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid
- """
+ def CheckPrereq(self):
if self.op.network is None:
raise errors.OpPrereqError("Network must be given",
errors.ECODE_INVAL)
raise errors.OpPrereqError("Network '%s' already defined" %
self.op.network, errors.ECODE_EXISTS)
- if self.op.mac_prefix:
- utils.NormalizeAndValidateMac(self.op.mac_prefix + ":00:00:00")
-
# Check tag validity
for tag in self.op.tags:
objects.TaggableObject.ValidateTag(tag)
mac_prefix=self.op.mac_prefix,
network_type=self.op.network_type,
uuid=self.network_uuid,
- family=4)
+ family=constants.IP4_VERSION)
# Initialize the associated address pool
try:
pool = network.AddressPool.InitializeNetwork(nobj)
# Check if we need to reserve the nodes and the cluster master IP
# These may not be allocated to any instances in routed mode, as
# they wouldn't function anyway.
- for node in self.cfg.GetAllNodesInfo().values():
- for ip in [node.primary_ip, node.secondary_ip]:
- try:
- pool.Reserve(ip)
- self.LogInfo("Reserved node %s's IP (%s)", node.name, ip)
-
- except errors.AddressPoolError:
- pass
+ if self.op.conflicts_check:
+ for node in self.cfg.GetAllNodesInfo().values():
+ for ip in [node.primary_ip, node.secondary_ip]:
+ try:
+ if pool.Contains(ip):
+ pool.Reserve(ip)
+ self.LogInfo("Reserved IP address of node '%s' (%s)",
+ node.name, ip)
+ except errors.AddressPoolError:
+ self.LogWarning("Cannot reserve IP address of node '%s' (%s)",
+ node.name, ip)
- master_ip = self.cfg.GetClusterInfo().master_ip
- try:
- pool.Reserve(master_ip)
- self.LogInfo("Reserved cluster master IP (%s)", master_ip)
- except errors.AddressPoolError:
- pass
+ master_ip = self.cfg.GetClusterInfo().master_ip
+ try:
+ if pool.Contains(master_ip):
+ pool.Reserve(master_ip)
+ self.LogInfo("Reserved cluster master IP address (%s)", master_ip)
+ except errors.AddressPoolError:
+ self.LogWarning("Cannot reserve cluster master IP address (%s)",
+ master_ip)
if self.op.add_reserved_ips:
for ip in self.op.add_reserved_ips:
self.network_uuid = self.cfg.LookupNetwork(self.op.network_name)
if not self.network_uuid:
- raise errors.OpPrereqError("Network %s not found" % self.op.network_name,
+ raise errors.OpPrereqError(("Network '%s' not found" %
+ self.op.network_name),
errors.ECODE_INVAL)
+
+ self.share_locks[locking.LEVEL_NODEGROUP] = 1
self.needed_locks = {
locking.LEVEL_NETWORK: [self.network_uuid],
+ locking.LEVEL_NODEGROUP: locking.ALL_SET,
}
def CheckPrereq(self):
cluster.
"""
-
# Verify that the network is not conncted.
node_groups = [group.name
for group in self.cfg.GetAllNodeGroupsInfo().values()
- for net in group.networks.keys()
- if net == self.network_uuid]
+ if self.network_uuid in group.networks]
if node_groups:
- self.LogWarning("Nework '%s' is connected to the following"
- " node groups: %s" % (self.op.network_name,
- utils.CommaJoin(utils.NiceSort(node_groups))))
- raise errors.OpPrereqError("Network still connected",
- errors.ECODE_STATE)
+ self.LogWarning("Network '%s' is connected to the following"
+ " node groups: %s" %
+ (self.op.network_name,
+ utils.CommaJoin(utils.NiceSort(node_groups))))
+ raise errors.OpPrereqError("Network still connected", errors.ECODE_STATE)
def BuildHooksEnv(self):
"""Build hooks env.
def ExpandNames(self):
self.network_uuid = self.cfg.LookupNetwork(self.op.network_name)
- self.network = self.cfg.GetNetwork(self.network_uuid)
- if self.network is None:
- raise errors.OpPrereqError("Could not retrieve network '%s' (UUID: %s)" %
- (self.op.network_name, self.network_uuid),
+ if self.network_uuid is None:
+ raise errors.OpPrereqError(("Network '%s' not found" %
+ self.op.network_name),
errors.ECODE_INVAL)
+
self.needed_locks = {
locking.LEVEL_NETWORK: [self.network_uuid],
}
"""Check prerequisites.
"""
+ self.network = self.cfg.GetNetwork(self.network_uuid)
self.gateway = self.network.gateway
self.network_type = self.network.network_type
self.mac_prefix = self.network.mac_prefix
if self.op.mac_prefix == constants.VALUE_NONE:
self.mac_prefix = None
else:
- utils.NormalizeAndValidateMac(self.op.mac_prefix + ":00:00:00")
- self.mac_prefix = self.op.mac_prefix
+ self.mac_prefix = \
+ utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix)
if self.op.gateway6:
if self.op.gateway6 == constants.VALUE_NONE:
# extend cfg.ReserveIp/ReleaseIp with the external flag
if self.op.gateway:
if self.gateway == self.network.gateway:
- self.LogWarning("Gateway is already %s" % self.gateway)
+ self.LogWarning("Gateway is already %s", self.gateway)
else:
if self.gateway:
self.pool.Reserve(self.gateway, external=True)
for ip in self.op.add_reserved_ips:
try:
if self.pool.IsReserved(ip):
- self.LogWarning("IP %s is already reserved" % ip)
+ self.LogWarning("IP address %s is already reserved", ip)
else:
self.pool.Reserve(ip, external=True)
- except errors.AddressPoolError, e:
- self.LogWarning("Cannot reserve ip %s. %s" % (ip, e))
+ except errors.AddressPoolError, err:
+ self.LogWarning("Cannot reserve IP address %s: %s", ip, err)
if self.op.remove_reserved_ips:
for ip in self.op.remove_reserved_ips:
continue
try:
if not self.pool.IsReserved(ip):
- self.LogWarning("IP %s is already unreserved" % ip)
+ self.LogWarning("IP address %s is already unreserved", ip)
else:
self.pool.Release(ip, external=True)
- except errors.AddressPoolError, e:
- self.LogWarning("Cannot release ip %s. %s" % (ip, e))
+ except errors.AddressPoolError, err:
+ self.LogWarning("Cannot release IP address %s: %s", ip, err)
if self.op.mac_prefix:
self.network.mac_prefix = self.mac_prefix
netparams = group.networks[net_uuid]
mode = netparams[constants.NIC_MODE]
link = netparams[constants.NIC_LINK]
- info = group.name + '(' + mode + ', ' + link + ')'
+ info = group.name + "(" + mode + ", " + link + ")"
network_to_groups[net_uuid].append(info)
if do_instances:
"free_count": pool.GetFreeCount(),
"reserved_count": pool.GetReservedCount(),
"map": pool.GetMap(),
- "external_reservations": ", ".join(pool.GetExternalReservations()),
+ "external_reservations":
+ utils.CommaJoin(pool.GetExternalReservations()),
}
return query.NetworkQueryData([self._all_networks[uuid]
self.network_link = self.op.network_link
self.network_uuid = self.cfg.LookupNetwork(self.network_name)
- self.network = self.cfg.GetNetwork(self.network_uuid)
- if self.network is None:
+ if self.network_uuid is None:
raise errors.OpPrereqError("Network %s does not exist" %
self.network_name, errors.ECODE_INVAL)
self.group_uuid = self.cfg.LookupNodeGroup(self.group_name)
- self.group = self.cfg.GetNodeGroup(self.group_uuid)
- if self.group is None:
+ if self.group_uuid is None:
raise errors.OpPrereqError("Group %s does not exist" %
self.group_name, errors.ECODE_INVAL)
}
self.share_locks[locking.LEVEL_INSTANCE] = 1
+ if self.op.conflicts_check:
+ self.needed_locks[locking.LEVEL_NETWORK] = [self.network_uuid]
+ self.share_locks[locking.LEVEL_NETWORK] = 1
+
def DeclareLocks(self, level):
if level == locking.LEVEL_INSTANCE:
assert not self.needed_locks[locking.LEVEL_INSTANCE]
# Lock instances optimistically, needs verification once group lock has
# been acquired
- self.needed_locks[locking.LEVEL_INSTANCE] = \
- self.cfg.GetNodeGroupInstances(self.group_uuid)
+ if self.op.conflicts_check:
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
+ self.cfg.GetNodeGroupInstances(self.group_uuid)
def BuildHooksEnv(self):
- ret = dict()
- ret["GROUP_NAME"] = self.group_name
- ret["GROUP_NETWORK_MODE"] = self.network_mode
- ret["GROUP_NETWORK_LINK"] = self.network_link
- ret.update(_BuildNetworkHookEnvByObject(self.network))
+ ret = {
+ "GROUP_NAME": self.group_name,
+ "GROUP_NETWORK_MODE": self.network_mode,
+ "GROUP_NETWORK_LINK": self.network_link,
+ }
return ret
def BuildHooksNodes(self):
return (nodes, nodes)
def CheckPrereq(self):
- l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2])
- for i in value)
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+
+ assert self.group_uuid in owned_groups
+
+ l = lambda value: utils.CommaJoin("%s: %s/%s" % (i[0], i[1], i[2])
+ for i in value)
- self.netparams = dict()
- self.netparams[constants.NIC_MODE] = self.network_mode
- self.netparams[constants.NIC_LINK] = self.network_link
+ self.netparams = {
+ constants.NIC_MODE: self.network_mode,
+ constants.NIC_LINK: self.network_link,
+ }
objects.NIC.CheckParameterSyntax(self.netparams)
+ self.group = self.cfg.GetNodeGroup(self.group_uuid)
#if self.network_mode == constants.NIC_MODE_BRIDGED:
# _CheckNodeGroupBridgesExist(self, self.network_link, self.group_uuid)
self.connected = False
self.connected = True
return
- pool = network.AddressPool(self.network)
if self.op.conflicts_check:
- groupinstances = []
- for n in self.cfg.GetNodeGroupInstances(self.group_uuid):
- groupinstances.append(self.cfg.GetInstanceInfo(n))
- instances = [(instance.name, idx, nic.ip)
- for instance in groupinstances
- for idx, nic in enumerate(instance.nics)
- if (not nic.network and pool.Contains(nic.ip))]
- if instances:
+ # Check if locked instances are still correct
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+ nobj = self.cfg.GetNetwork(self.network_uuid)
+ pool = network.AddressPool(nobj)
+ conflicting_instances = []
+
+ for (_, instance) in self.cfg.GetMultiInstanceInfo(owned_instances):
+ for idx, nic in enumerate(instance.nics):
+ if pool.Contains(nic.ip):
+ conflicting_instances.append((instance.name, idx, nic.ip))
+
+ if conflicting_instances:
self.LogWarning("Following occurences use IPs from network %s"
" that is about to connect to nodegroup %s: %s" %
(self.network_name, self.group.name,
- l(instances)))
+ l(conflicting_instances)))
raise errors.OpPrereqError("Conflicting IPs found."
" Please remove/modify"
" corresponding NICs",
self.group_name = self.op.group_name
self.network_uuid = self.cfg.LookupNetwork(self.network_name)
- self.network = self.cfg.GetNetwork(self.network_uuid)
- if self.network is None:
+ if self.network_uuid is None:
raise errors.OpPrereqError("Network %s does not exist" %
self.network_name, errors.ECODE_INVAL)
self.group_uuid = self.cfg.LookupNodeGroup(self.group_name)
- self.group = self.cfg.GetNodeGroup(self.group_uuid)
- if self.group is None:
+ if self.group_uuid is None:
raise errors.OpPrereqError("Group %s does not exist" %
self.group_name, errors.ECODE_INVAL)
# Lock instances optimistically, needs verification once group lock has
# been acquired
- self.needed_locks[locking.LEVEL_INSTANCE] = \
+ if self.op.conflicts_check:
+ self.needed_locks[locking.LEVEL_INSTANCE] = \
self.cfg.GetNodeGroupInstances(self.group_uuid)
def BuildHooksEnv(self):
- ret = dict()
- ret["GROUP_NAME"] = self.group_name
- ret.update(_BuildNetworkHookEnvByObject(self.network))
+ ret = {
+ "GROUP_NAME": self.group_name,
+ }
return ret
def BuildHooksNodes(self):
return (nodes, nodes)
def CheckPrereq(self):
- l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2])
- for i in value)
+ owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+ assert self.group_uuid in owned_groups
+
+ l = lambda value: utils.CommaJoin("%s: %s/%s" % (i[0], i[1], i[2])
+ for i in value)
+
+ self.group = self.cfg.GetNodeGroup(self.group_uuid)
self.connected = True
if self.network_uuid not in self.group.networks:
- self.LogWarning("Network '%s' is"
- " not mapped to group '%s'" %
- (self.network_name, self.group.name))
+ self.LogWarning("Network '%s' is not mapped to group '%s'",
+ self.network_name, self.group.name)
self.connected = False
return
if self.op.conflicts_check:
- groupinstances = []
- for n in self.cfg.GetNodeGroupInstances(self.group_uuid):
- groupinstances.append(self.cfg.GetInstanceInfo(n))
- instances = [(instance.name, idx, nic.ip)
- for instance in groupinstances
- for idx, nic in enumerate(instance.nics)
- if nic.network == self.network_name]
- if instances:
+ # Check if locked instances are still correct
+ owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+ _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+ conflicting_instances = []
+
+ for (_, instance) in self.cfg.GetMultiInstanceInfo(owned_instances):
+ for idx, nic in enumerate(instance.nics):
+ if nic.network == self.network_name:
+ conflicting_instances.append((instance.name, idx, nic.ip))
+
+ if conflicting_instances:
self.LogWarning("Following occurences use IPs from network %s"
" that is about to disconnected from the nodegroup"
" %s: %s" %
(self.network_name, self.group.name,
- l(instances)))
+ l(conflicting_instances)))
raise errors.OpPrereqError("Conflicting IPs."
" Please remove/modify"
" corresponding NICS",