#: Size of DRBD meta block device
DRBD_META_SIZE = 128
+# States of instance
+INSTANCE_UP = [constants.ADMINST_UP]
+INSTANCE_DOWN = [constants.ADMINST_DOWN]
+INSTANCE_OFFLINE = [constants.ADMINST_OFFLINE]
+INSTANCE_ONLINE = [constants.ADMINST_DOWN, constants.ADMINST_UP]
+INSTANCE_NOT_RUNNING = [constants.ADMINST_DOWN, constants.ADMINST_OFFLINE]
+
class ResultWithJobs:
"""Data container for LU results with jobs.
return dict.fromkeys(locking.LEVELS, 1)
+def _MakeLegacyNodeInfo(data):
+ """Formats the data returned by L{rpc.RpcRunner.call_node_info}.
+
+ Converts the data into a single dictionary. This is fine for most use cases,
+ but some require information from more than one volume group or hypervisor.
+
+ """
+ (bootid, (vg_info, ), (hv_info, )) = data
+
+ return utils.JoinDisjointDicts(utils.JoinDisjointDicts(vg_info, hv_info), {
+ "bootid": bootid,
+ })
+
+
def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
"""Checks if the owned node groups are still correct for an instance.
else:
should_release = None
- if should_release:
+ owned = lu.owned_locks(level)
+ if not owned:
+ # Not owning any lock at this level, do nothing
+ pass
+
+ elif should_release:
retain = []
release = []
# Determine which locks to release
- for name in lu.owned_locks(level):
+ for name in owned:
if should_release(name):
release.append(name)
else:
strict=True)
-def _CheckInstanceDown(lu, instance, reason):
- """Ensure that an instance is not running."""
- if instance.admin_up:
- raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
- (instance.name, reason), errors.ECODE_STATE)
+def _CheckInstanceState(lu, instance, req_states, msg=None):
+ """Ensure that an instance is in one of the required states.
- pnode = instance.primary_node
- ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
- ins_l.Raise("Can't contact node %s for instance information" % pnode,
- prereq=True, ecode=errors.ECODE_ENVIRON)
+ @param lu: the LU on behalf of which we make the check
+ @param instance: the instance to check
+ @param msg: if passed, should be a message to replace the default one
+ @raise errors.OpPrereqError: if the instance is not in the required state
+
+ """
+ if msg is None:
+ msg = "can't use instance from outside %s states" % ", ".join(req_states)
+ if instance.admin_state not in req_states:
+ raise errors.OpPrereqError("Instance %s is marked to be %s, %s" %
+ (instance, instance.admin_state, msg),
+ errors.ECODE_STATE)
+
+ if constants.ADMINST_UP not in req_states:
+ pnode = instance.primary_node
+ ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
+ ins_l.Raise("Can't contact node %s for instance information" % pnode,
+ prereq=True, ecode=errors.ECODE_ENVIRON)
- if instance.name in ins_l.payload:
- raise errors.OpPrereqError("Instance %s is running, %s" %
- (instance.name, reason), errors.ECODE_STATE)
+ if instance.name in ins_l.payload:
+ raise errors.OpPrereqError("Instance %s is running, %s" %
+ (instance.name, msg), errors.ECODE_STATE)
def _ExpandItemName(fn, name, kind):
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
- memory, vcpus, nics, disk_template, disks,
+ minmem, maxmem, vcpus, nics, disk_template, disks,
bep, hvp, hypervisor_name, tags):
"""Builds instance related env variables for hooks
@param secondary_nodes: list of secondary nodes as strings
@type os_type: string
@param os_type: the name of the instance's OS
- @type status: boolean
- @param status: the should_run status of the instance
- @type memory: string
- @param memory: the memory size of the instance
+ @type status: string
+ @param status: the desired status of the instance
+ @type minmem: string
+ @param minmem: the minimum memory size of the instance
+ @type maxmem: string
+ @param maxmem: the maximum memory size of the instance
@type vcpus: string
@param vcpus: the count of VCPUs the instance has
@type nics: list
@return: the hook environment for this instance
"""
- if status:
- str_status = "up"
- else:
- str_status = "down"
env = {
"OP_TARGET": name,
"INSTANCE_NAME": name,
"INSTANCE_PRIMARY": primary_node,
"INSTANCE_SECONDARIES": " ".join(secondary_nodes),
"INSTANCE_OS_TYPE": os_type,
- "INSTANCE_STATUS": str_status,
- "INSTANCE_MEMORY": memory,
+ "INSTANCE_STATUS": status,
+ "INSTANCE_MINMEM": minmem,
+ "INSTANCE_MAXMEM": maxmem,
+ # TODO(2.7) remove deprecated "memory" value
+ "INSTANCE_MEMORY": maxmem,
"INSTANCE_VCPUS": vcpus,
"INSTANCE_DISK_TEMPLATE": disk_template,
"INSTANCE_HYPERVISOR": hypervisor_name,
}
-
if nics:
nic_count = len(nics)
for idx, (ip, mac, mode, link) in enumerate(nics):
"primary_node": instance.primary_node,
"secondary_nodes": instance.secondary_nodes,
"os_type": instance.os,
- "status": instance.admin_up,
- "memory": bep[constants.BE_MEMORY],
+ "status": instance.admin_state,
+ "maxmem": bep[constants.BE_MAXMEM],
+ "minmem": bep[constants.BE_MINMEM],
"vcpus": bep[constants.BE_VCPUS],
"nics": _NICListToTuple(lu, instance.nics),
"disk_template": instance.disk_template,
_ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance,
"volume %s missing on node %s", volume, node)
- if instanceconfig.admin_up:
+ if instanceconfig.admin_state == constants.ADMINST_UP:
pri_img = node_image[node_current]
test = instance not in pri_img.instances and not pri_img.offline
_ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
# node here
snode = node_image[nname]
bad_snode = snode.ghost or snode.offline
- _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
+ _ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
+ not success and not bad_snode,
constants.CV_EINSTANCEFAULTYDISK, instance,
"couldn't retrieve status for disk/%s on %s: %s",
idx, nname, bdev_status)
- _ErrorIf((instanceconfig.admin_up and success and
- bdev_status.ldisk_status == constants.LDS_FAULTY),
+ _ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
+ success and bdev_status.ldisk_status == constants.LDS_FAULTY),
constants.CV_EINSTANCEFAULTYDISK, instance,
"disk/%s on %s is faulty", idx, nname)
# we already list instances living on such nodes, and that's
# enough warning
continue
+ #TODO(dynmem): use MINMEM for checking
+ #TODO(dynmem): also consider ballooning out other instances
for prinode, instances in n_img.sbp.items():
needed_mem = 0
for instance in instances:
bep = cluster_info.FillBE(instance_cfg[instance])
if bep[constants.BE_AUTO_BALANCE]:
- needed_mem += bep[constants.BE_MEMORY]
+ needed_mem += bep[constants.BE_MAXMEM]
test = n_img.mfree < needed_mem
self._ErrorIf(test, constants.CV_ENODEN1, node,
"not enough memory to accomodate instance failovers"
node_drbd[minor] = (instance, False)
else:
instance = instanceinfo[instance]
- node_drbd[minor] = (instance.name, instance.admin_up)
+ node_drbd[minor] = (instance.name,
+ instance.admin_state == constants.ADMINST_UP)
# and now check them
used_minors = nresult.get(constants.NV_DRBDLIST, [])
i_non_redundant = [] # Non redundant instances
i_non_a_balanced = [] # Non auto-balanced instances
+ i_offline = 0 # Count of offline instances
n_offline = 0 # Count of offline nodes
n_drained = 0 # Count of nodes being drained
node_vol_should = {}
non_primary_inst = set(nimg.instances).difference(nimg.pinst)
for inst in non_primary_inst:
+ # FIXME: investigate best way to handle offline insts
+ if inst.admin_state == constants.ADMINST_OFFLINE:
+ if verbose:
+ feedback_fn("* Skipping offline instance %s" % inst.name)
+ i_offline += 1
+ continue
test = inst in self.all_inst_info
_ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
"instance should not run on node %s", node_i.name)
constants.CV_ENODERPC, pnode, "instance %s, connection to"
" primary node failed", instance)
- _ErrorIf(inst_config.admin_up and pnode_img.offline,
+ _ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
+ pnode_img.offline,
constants.CV_EINSTANCEBADNODE, instance,
"instance is marked as running and lives on offline node %s",
inst_config.primary_node)
feedback_fn(" - NOTICE: %d non-auto-balanced instance(s) found."
% len(i_non_a_balanced))
+ if i_offline:
+ feedback_fn(" - NOTICE: %d offline instance(s) found." % i_offline)
+
if n_offline:
feedback_fn(" - NOTICE: %d offline node(s) found." % n_offline)
res_missing = {}
nv_dict = _MapInstanceDisksToNodes([inst
- for inst in self.instances.values()
- if inst.admin_up])
+ for inst in self.instances.values()
+ if inst.admin_state == constants.ADMINST_UP])
if nv_dict:
nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
if self.op.instances:
self.wanted_names = _GetWantedInstances(self, self.op.instances)
self.needed_locks = {
- locking.LEVEL_NODE: [],
+ locking.LEVEL_NODE_RES: [],
locking.LEVEL_INSTANCE: self.wanted_names,
}
- self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+ self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
else:
self.wanted_names = None
self.needed_locks = {
- locking.LEVEL_NODE: locking.ALL_SET,
+ locking.LEVEL_NODE_RES: locking.ALL_SET,
locking.LEVEL_INSTANCE: locking.ALL_SET,
}
- self.share_locks = _ShareAll()
+ self.share_locks = {
+ locking.LEVEL_NODE_RES: 1,
+ locking.LEVEL_INSTANCE: 0,
+ }
def DeclareLocks(self, level):
- if level == locking.LEVEL_NODE and self.wanted_names is not None:
- self._LockInstancesNodes(primary_only=True)
+ if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
+ self._LockInstancesNodes(primary_only=True, level=level)
def CheckPrereq(self):
"""Check prerequisites.
for idx, disk in enumerate(instance.disks):
per_node_disks[pnode].append((instance, idx, disk))
+ assert not (frozenset(per_node_disks.keys()) -
+ self.owned_locks(locking.LEVEL_NODE_RES)), \
+ "Not owning correct locks"
+ assert not self.owned_locks(locking.LEVEL_NODE)
+
changed = []
for node, dskl in per_node_disks.items():
newl = [v[2].Copy() for v in dskl]
if self.op.master_netmask is not None:
_ValidateNetmask(self.cfg, self.op.master_netmask)
+ if self.op.diskparams:
+ for dt_params in self.op.diskparams.values():
+ utils.ForceDictType(dt_params, constants.DISK_DT_TYPES)
+
def ExpandNames(self):
# FIXME: in the future maybe other cluster params won't require checking on
# all nodes to be modified.
self.cluster = cluster = self.cfg.GetClusterInfo()
# validate params changes
if self.op.beparams:
+ objects.UpgradeBeParams(self.op.beparams)
utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
else:
self.new_hvparams[hv_name].update(hv_dict)
+ # disk template parameters
+ self.new_diskparams = objects.FillDict(cluster.diskparams, {})
+ if self.op.diskparams:
+ for dt_name, dt_params in self.op.diskparams.items():
+ if dt_name not in self.op.diskparams:
+ self.new_diskparams[dt_name] = dt_params
+ else:
+ self.new_diskparams[dt_name].update(dt_params)
+
# os hypervisor parameters
self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
if self.op.os_hvp:
self.cluster.osparams = self.new_osp
if self.op.ndparams:
self.cluster.ndparams = self.new_ndparams
+ if self.op.diskparams:
+ self.cluster.diskparams = self.new_diskparams
if self.op.candidate_pool_size is not None:
self.cluster.candidate_pool_size = self.op.candidate_pool_size
_AdjustCandidatePool(self, [])
if self.op.maintain_node_health is not None:
+ if self.op.maintain_node_health and not constants.ENABLE_CONFD:
+ feedback_fn("Note: CONFD was disabled at build time, node health"
+ " maintenance is not useful (still enabling it)")
self.cluster.maintain_node_health = self.op.maintain_node_health
if self.op.prealloc_wipe_disks is not None:
"""
master_params = self.cfg.GetMasterNetworkParameters()
ems = self.cfg.GetUseExternalMipScript()
- self.rpc.call_node_activate_master_ip(master_params.name,
- master_params, ems)
+ result = self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
+ result.Raise("Could not activate the master IP")
class LUClusterDeactivateMasterIp(NoHooksLU):
"""
master_params = self.cfg.GetMasterNetworkParameters()
ems = self.cfg.GetUseExternalMipScript()
- self.rpc.call_node_deactivate_master_ip(master_params.name, master_params,
- ems)
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
+ result.Raise("Could not deactivate the master IP")
def _WaitForSync(lu, instance, disks=None, oneshot=False):
raise errors.OpPrereqError("Node is the master node, failover to another"
" node is required", errors.ECODE_INVAL)
- for instance_name, instance in self.cfg.GetAllInstancesInfo():
+ for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
if node.name in instance.all_nodes:
raise errors.OpPrereqError("Instance %s is still running on the node,"
" please remove first" % instance_name,
# filter out non-vm_capable nodes
toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
- node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
- lu.cfg.GetHypervisorType())
- live_data = dict((name, nresult.payload)
+ node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
+ [lu.cfg.GetHypervisorType()])
+ live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
for (name, nresult) in node_data.items()
if not nresult.fail_msg and nresult.payload)
else:
selected=self.op.output_fields)
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
+
if not self.op.nodes:
self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
else:
selected=self.op.output_fields)
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
if self.op.nodes:
self.needed_locks[locking.LEVEL_NODE] = \
else:
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
+
if self.lock_instances:
self.needed_locks[locking.LEVEL_INSTANCE] = \
frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
# On online nodes, check that no instances are running, and that
# the node has the new ip and we can reach it.
for instance in affected_instances.values():
- _CheckInstanceDown(self, instance, "cannot change secondary ip")
+ _CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot change secondary ip")
_CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
if master.name != node.name:
_ShutdownInstanceDisks.
"""
- _CheckInstanceDown(lu, instance, "cannot shutdown disks")
+ _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
_ShutdownInstanceDisks(lu, instance, disks=disks)
we cannot check the node
"""
- nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
+ 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)
+ (_, _, (hv_info, )) = nodeinfo[node].payload
+
+ free_mem = hv_info.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),
or we cannot check the node
"""
- nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
+ nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
for node in nodenames:
info = nodeinfo[node]
info.Raise("Cannot get current information from node %s" % node,
prereq=True, ecode=errors.ECODE_ENVIRON)
- vg_free = info.payload.get("vg_free", None)
+ (_, (vg_info, ), _) = info.payload
+ vg_free = vg_info.get("vg_free", None)
if not isinstance(vg_free, int):
raise errors.OpPrereqError("Can't compute free disk space on node"
" %s for vg %s, result was '%s'" %
or we cannot check the node
"""
- nodeinfo = lu.rpc.call_node_info(nodenames, None, hypervisor_name)
+ nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
for node in nodenames:
info = nodeinfo[node]
info.Raise("Cannot get current information from node %s" % node,
prereq=True, ecode=errors.ECODE_ENVIRON)
- num_cpus = info.payload.get("cpu_total", None)
+ (_, _, (hv_info, )) = info.payload
+ num_cpus = hv_info.get("cpu_total", None)
if not isinstance(num_cpus, int):
raise errors.OpPrereqError("Can't compute the number of physical CPUs"
" on node %s, result was '%s'" %
# extra beparams
if self.op.beparams:
# fill the beparams dict
+ objects.UpgradeBeParams(self.op.beparams)
utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
def ExpandNames(self):
hv_type.CheckParameterSyntax(filled_hvp)
_CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
+ _CheckInstanceState(self, instance, INSTANCE_ONLINE)
+
self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
if self.primary_offline and self.op.ignore_offline_nodes:
if not remote_info.payload: # not running already
_CheckNodeFreeMemory(self, instance.primary_node,
"starting instance %s" % instance.name,
- bep[constants.BE_MEMORY], instance.hypervisor)
+ bep[constants.BE_MAXMEM], instance.hypervisor)
def Exec(self, feedback_fn):
"""Start the instance.
self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
-
+ _CheckInstanceState(self, instance, INSTANCE_ONLINE)
_CheckNodeOnline(self, instance.primary_node)
# check bridges existence
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
+
self.primary_offline = \
self.cfg.GetNodeInfo(self.instance.primary_node).offline
raise errors.OpPrereqError("Instance '%s' has no disks" %
self.op.instance_name,
errors.ECODE_INVAL)
- _CheckInstanceDown(self, instance, "cannot reinstall")
+ _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
if self.op.os_type is not None:
# OS verification
# otherwise we need to lock all nodes for disk re-creation
primary_only = bool(self.op.nodes)
self._LockInstancesNodes(primary_only=primary_only)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
self.op.instance_name, errors.ECODE_INVAL)
# if we replace nodes *and* the old primary is offline, we don't
# check
- assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
+ assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
+ assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
if not (self.op.nodes and old_pnode.offline):
- _CheckInstanceDown(self, instance, "cannot recreate disks")
+ _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+ msg="cannot recreate disks")
if not self.op.disks:
self.op.disks = range(len(instance.disks))
"""
instance = self.instance
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
to_skip = []
mods = [] # keeps track of needed logical_id changes
instance = self.cfg.GetInstanceInfo(self.op.instance_name)
assert instance is not None
_CheckNodeOnline(self, instance.primary_node)
- _CheckInstanceDown(self, instance, "cannot rename")
+ _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+ msg="cannot rename")
self.instance = instance
new_name = self.op.new_name
if self.op.name_check:
hostname = netutils.GetHostname(name=new_name)
- if hostname != new_name:
+ if hostname.name != new_name:
self.LogInfo("Resolved given name '%s' to '%s'", new_name,
hostname.name)
if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
def ExpandNames(self):
self._ExpandAndLockInstance()
self.needed_locks[locking.LEVEL_NODE] = []
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
" node %s: %s" %
(instance.name, instance.primary_node, msg))
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+ assert not (set(instance.all_nodes) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Not owning correct locks"
+
_RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
target_node = _ExpandNodeName(self.cfg, self.op.target_node)
self.op.target_node = target_node
self.needed_locks[locking.LEVEL_NODE] = [target_node]
+ self.needed_locks[locking.LEVEL_NODE_RES] = []
self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes(primary_only=True)
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
_CheckNodeNotDrained(self, target_node)
_CheckNodeVmCapable(self, target_node)
- if instance.admin_up:
+ if instance.admin_state == constants.ADMINST_UP:
# check memory requirements on the secondary node
_CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
- instance.name, bep[constants.BE_MEMORY],
+ instance.name, bep[constants.BE_MAXMEM],
instance.hypervisor)
else:
self.LogInfo("Not checking memory on the secondary node as"
self.LogInfo("Shutting down instance %s on source node %s",
instance.name, source_node)
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
result = self.rpc.call_instance_shutdown(source_node, instance,
self.op.shutdown_timeout)
msg = result.fail_msg
_RemoveDisks(self, instance, target_node=source_node)
# Only start the instance if it's marked as up
- if instance.admin_up:
+ if instance.admin_state == constants.ADMINST_UP:
self.LogInfo("Starting instance %s on node %s",
instance.name, target_node)
assert instance is not None
self.instance = instance
- if (not self.cleanup and not instance.admin_up and not self.failover and
- self.fallback):
- self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
- " to failover")
+ if (not self.cleanup and
+ not instance.admin_state == constants.ADMINST_UP and
+ not self.failover and self.fallback):
+ self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
+ " switching to failover")
self.failover = True
if instance.disk_template not in constants.DTS_MIRRORED:
i_be = self.cfg.GetClusterInfo().FillBE(instance)
# check memory requirements on the secondary node
- if not self.failover or instance.admin_up:
+ if not self.failover or instance.admin_state == constants.ADMINST_UP:
_CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
- instance.name, i_be[constants.BE_MEMORY],
+ instance.name, i_be[constants.BE_MAXMEM],
instance.hypervisor)
else:
self.lu.LogInfo("Not checking memory on the secondary node as"
# 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]
+ None, [self.instance.hypervisor])
+ for ninfo in nodeinfo.values():
+ ninfo.Raise("Unable to retrieve node information from node '%s'" %
+ ninfo.node)
+ (_, _, (src_info, )) = nodeinfo[source_node].payload
+ (_, _, (dst_info, )) = nodeinfo[target_node].payload
+
+ if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
+ (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
+ src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
+ dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
if src_version != dst_version:
self.feedback_fn("* warning: hypervisor version mismatch between"
" source (%s) and target (%s) node" %
source_node = instance.primary_node
target_node = self.target_node
- if instance.admin_up:
+ if instance.admin_state == constants.ADMINST_UP:
self.feedback_fn("* checking disk consistency between source and target")
for dev in instance.disks:
# for drbd, these are drbd over lvm
self.cfg.Update(instance, self.feedback_fn)
# Only start the instance if it's marked as up
- if instance.admin_up:
+ if instance.admin_state == constants.ADMINST_UP:
self.feedback_fn("* activating the instance's disks on target node %s" %
target_node)
logging.info("Starting instance %s on node %s",
return results
+def _ComputeLDParams(disk_template, disk_params):
+ """Computes Logical Disk parameters from Disk Template parameters.
+
+ @type disk_template: string
+ @param disk_template: disk template, one of L{constants.DISK_TEMPLATES}
+ @type disk_params: dict
+ @param disk_params: disk template parameters; dict(template_name -> parameters
+ @rtype: list(dict)
+ @return: a list of dicts, one for each node of the disk hierarchy. Each dict
+ contains the LD parameters of the node. The tree is flattened in-order.
+
+ """
+ if disk_template not in constants.DISK_TEMPLATES:
+ raise errors.ProgrammerError("Unknown disk template %s" % disk_template)
+
+ result = list()
+ dt_params = disk_params[disk_template]
+ if disk_template == constants.DT_DRBD8:
+ drbd_params = {
+ constants.RESYNC_RATE: dt_params[constants.DRBD_RESYNC_RATE]
+ }
+
+ drbd_params = \
+ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_DRBD8],
+ drbd_params)
+
+ result.append(drbd_params)
+
+ # data LV
+ data_params = {
+ constants.STRIPES: dt_params[constants.DRBD_DATA_STRIPES],
+ }
+ data_params = \
+ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+ data_params)
+ result.append(data_params)
+
+ # metadata LV
+ meta_params = {
+ constants.STRIPES: dt_params[constants.DRBD_META_STRIPES],
+ }
+ meta_params = \
+ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+ meta_params)
+ result.append(meta_params)
+
+ elif (disk_template == constants.DT_FILE or
+ disk_template == constants.DT_SHARED_FILE):
+ result.append(constants.DISK_LD_DEFAULTS[constants.LD_FILE])
+
+ elif disk_template == constants.DT_PLAIN:
+ params = {
+ constants.STRIPES: dt_params[constants.LV_STRIPES],
+ }
+ params = \
+ objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+ params)
+ result.append(params)
+
+ elif disk_template == constants.DT_BLOCK:
+ result.append(constants.DISK_LD_DEFAULTS[constants.LD_BLOCKDEV])
+
+ return result
+
+
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
- iv_name, p_minor, s_minor):
+ iv_name, p_minor, s_minor, drbd_params, data_params,
+ meta_params):
"""Generate a drbd8 device complete with its children.
"""
assert len(vgnames) == len(names) == 2
port = lu.cfg.AllocatePort()
shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
+
dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
- logical_id=(vgnames[0], names[0]))
+ logical_id=(vgnames[0], names[0]),
+ params=data_params)
dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
- logical_id=(vgnames[1], names[1]))
+ logical_id=(vgnames[1], names[1]),
+ params=meta_params)
drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
logical_id=(primary, secondary, port,
p_minor, s_minor,
shared_secret),
children=[dev_data, dev_meta],
- iv_name=iv_name)
+ iv_name=iv_name, params=drbd_params)
return drbd_dev
instance_name, primary_node,
secondary_nodes, disk_info,
file_storage_dir, file_driver,
- base_index, feedback_fn):
+ base_index, feedback_fn, disk_params):
"""Generate the entire disk layout for a given template type.
"""
vgname = lu.cfg.GetVGName()
disk_count = len(disk_info)
disks = []
+ ld_params = _ComputeLDParams(template_name, disk_params)
if template_name == constants.DT_DISKLESS:
pass
elif template_name == constants.DT_PLAIN:
size=disk[constants.IDISK_SIZE],
logical_id=(vg, names[idx]),
iv_name="disk/%d" % disk_index,
- mode=disk[constants.IDISK_MODE])
+ mode=disk[constants.IDISK_MODE],
+ params=ld_params[0])
disks.append(disk_dev)
elif template_name == constants.DT_DRBD8:
+ drbd_params, data_params, meta_params = ld_params
if len(secondary_nodes) != 1:
raise errors.ProgrammerError("Wrong template configuration")
remote_node = secondary_nodes[0]
[data_vg, meta_vg],
names[idx * 2:idx * 2 + 2],
"disk/%d" % disk_index,
- minors[idx * 2], minors[idx * 2 + 1])
+ minors[idx * 2], minors[idx * 2 + 1],
+ drbd_params, data_params, meta_params)
disk_dev.mode = disk[constants.IDISK_MODE]
disks.append(disk_dev)
elif template_name == constants.DT_FILE:
logical_id=(file_driver,
"%s/disk%d" % (file_storage_dir,
disk_index)),
- mode=disk[constants.IDISK_MODE])
+ mode=disk[constants.IDISK_MODE],
+ params=ld_params[0])
disks.append(disk_dev)
elif template_name == constants.DT_SHARED_FILE:
if len(secondary_nodes) != 0:
logical_id=(file_driver,
"%s/disk%d" % (file_storage_dir,
disk_index)),
- mode=disk[constants.IDISK_MODE])
+ mode=disk[constants.IDISK_MODE],
+ params=ld_params[0])
disks.append(disk_dev)
elif template_name == constants.DT_BLOCK:
if len(secondary_nodes) != 0:
logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
disk[constants.IDISK_ADOPT]),
iv_name="disk/%d" % disk_index,
- mode=disk[constants.IDISK_MODE])
+ mode=disk[constants.IDISK_MODE],
+ params=ld_params[0])
disks.append(disk_dev)
else:
" continuing anyway: %s", device.iv_name, node, msg)
all_result = False
+ # if this is a DRBD disk, return its port to the pool
+ if device.dev_type in constants.LDS_DRBD:
+ tcp_port = device.logical_id[2]
+ lu.cfg.AddTcpUdpPort(tcp_port)
+
if instance.disk_template == constants.DT_FILE:
file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
if target_node:
self.add_locks[locking.LEVEL_INSTANCE] = instance_name
if self.op.iallocator:
+ # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
+ # 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
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:
tags=self.op.tags,
os=self.op.os_type,
vcpus=self.be_full[constants.BE_VCPUS],
- memory=self.be_full[constants.BE_MEMORY],
+ memory=self.be_full[constants.BE_MAXMEM],
disks=self.disks,
nics=nics,
hypervisor=self.op.hypervisor,
secondary_nodes=self.secondaries,
status=self.op.start,
os_type=self.op.os_type,
- memory=self.be_full[constants.BE_MEMORY],
+ minmem=self.be_full[constants.BE_MINMEM],
+ maxmem=self.be_full[constants.BE_MAXMEM],
vcpus=self.be_full[constants.BE_VCPUS],
nics=_NICListToTuple(self, self.nics),
disk_template=self.op.disk_template,
for name, value in einfo.items(constants.INISECT_BEP):
if name not in self.op.beparams:
self.op.beparams[name] = value
+ # Compatibility for the old "memory" be param
+ if name == constants.BE_MEMORY:
+ if constants.BE_MAXMEM not in self.op.beparams:
+ self.op.beparams[constants.BE_MAXMEM] = value
+ if constants.BE_MINMEM not in self.op.beparams:
+ self.op.beparams[constants.BE_MINMEM] = value
else:
# try to read the parameters old style, from the main section
for name in constants.BES_PARAMETERS:
for param, value in self.op.beparams.iteritems():
if value == constants.VALUE_AUTO:
self.op.beparams[param] = default_beparams[param]
+ objects.UpgradeBeParams(self.op.beparams)
utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
self.be_full = cluster.SimpleFillBE(self.op.beparams)
if self.op.iallocator is not None:
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]))
+
#### node related checks
# check primary node
_CheckNodeVmCapable(self, self.op.snode)
self.secondaries.append(self.op.snode)
+ snode = self.cfg.GetNodeInfo(self.op.snode)
+ if pnode.group != snode.group:
+ self.LogWarning("The primary and secondary nodes are in two"
+ " different node groups; the disk parameters"
+ " from the first disk's node group will be"
+ " used")
+
nodenames = [pnode.name] + self.secondaries
+ # disk parameters (not customizable at instance or node level)
+ # just use the primary node parameters, ignoring the secondary.
+ self.diskparams = self.cfg.GetNodeGroup(pnode.group).diskparams
+
if not self.adopt_disks:
# Check lv size requirements, if not adopting
req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
_CheckNicsBridgesExist(self, self.nics, self.pnode.name)
# memory check on primary node
+ #TODO(dynmem): use MINMEM for checking
if self.op.start:
_CheckNodeFreeMemory(self, self.pnode.name,
"creating instance %s" % self.op.instance_name,
- self.be_full[constants.BE_MEMORY],
+ self.be_full[constants.BE_MAXMEM],
self.op.hypervisor)
self.dry_run_result = list(nodenames)
instance = self.op.instance_name
pnode_name = self.pnode.name
+ assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "Node locks differ from node resource locks"
+
ht_kind = self.op.hypervisor
if ht_kind in constants.HTS_REQ_PORT:
network_port = self.cfg.AllocatePort()
self.instance_file_storage_dir,
self.op.file_driver,
0,
- feedback_fn)
+ feedback_fn,
+ self.diskparams)
iobj = objects.Instance(name=instance, os=self.op.os_type,
primary_node=pnode_name,
nics=self.nics, disks=disks,
disk_template=self.op.disk_template,
- admin_up=False,
+ admin_state=constants.ADMINST_DOWN,
network_port=network_port,
beparams=self.op.beparams,
hvparams=self.op.hvparams,
raise errors.OpExecError("There are some degraded disks for"
" this instance")
+ # Release all node resource locks
+ _ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
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:
raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
% self.op.mode)
+ assert not self.owned_locks(locking.LEVEL_NODE_RES)
+
if self.op.start:
- iobj.admin_up = True
+ iobj.admin_state = constants.ADMINST_UP
self.cfg.Update(iobj, feedback_fn)
logging.info("Starting instance %s on node %s", instance, pnode_name)
feedback_fn("* starting instance...")
node_insts.Raise("Can't get node information from %s" % node)
if instance.name not in node_insts.payload:
- if instance.admin_up:
+ if instance.admin_state == constants.ADMINST_UP:
state = constants.INSTST_ERRORDOWN
- else:
+ elif instance.admin_state == constants.ADMINST_DOWN:
state = constants.INSTST_ADMINDOWN
+ else:
+ state = constants.INSTST_ADMINOFFLINE
raise errors.OpExecError("Instance %s is not running (state %s)" %
(instance.name, state))
self._ExpandAndLockInstance()
assert locking.LEVEL_NODE not in self.needed_locks
+ assert locking.LEVEL_NODE_RES not in self.needed_locks
assert locking.LEVEL_NODEGROUP not in self.needed_locks
assert self.op.iallocator is None or self.op.remote_node is None, \
# iallocator will select a new node in the same group
self.needed_locks[locking.LEVEL_NODEGROUP] = []
+ 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)
assert not self.needed_locks[locking.LEVEL_NODEGROUP]
self.share_locks[locking.LEVEL_NODEGROUP] = 1
+ # Lock all groups used by instance optimistically; this requires going
+ # via the node before it's locked, requiring verification later on
self.needed_locks[locking.LEVEL_NODEGROUP] = \
self.cfg.GetInstanceNodeGroups(self.op.instance_name)
for node_name in self.cfg.GetNodeGroup(group_uuid).members]
else:
self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Reuse node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE]
def BuildHooksEnv(self):
"""Build hooks env.
assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
self.op.iallocator is None)
+ # Verify if node group locks are still correct
owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
if owned_groups:
_CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
if not self.disks:
self.disks = range(len(self.instance.disks))
+ # TODO: compute disk parameters
+ primary_node_info = self.cfg.GetNodeInfo(instance.primary_node)
+ secondary_node_info = self.cfg.GetNodeInfo(secondary_node)
+ if primary_node_info.group != secondary_node_info.group:
+ self.lu.LogInfo("The instance primary and secondary nodes are in two"
+ " different node groups; the disk parameters of the"
+ " primary node's group will be applied.")
+
+ self.diskparams = self.cfg.GetNodeGroup(primary_node_info.group).diskparams
+
for node in check_nodes:
_CheckNodeOnline(self.lu, node)
self.target_node]
if node_name is not None)
- # Release unneeded node locks
+ # 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)
# Release any owned node group
if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
assert set(owned_nodes) == set(self.node_secondary_ip), \
("Incorrect node locks, owning %s, expected %s" %
(owned_nodes, self.node_secondary_ip.keys()))
+ assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
+ self.lu.owned_locks(locking.LEVEL_NODE_RES))
owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
assert list(owned_instances) == [self.instance_name], \
feedback_fn("Replacing disk(s) %s for %s" %
(utils.CommaJoin(self.disks), self.instance.name))
- activate_disks = (not self.instance.admin_up)
+ activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
# Activate the instance disks if we're replacing them on a down instance
if activate_disks:
if activate_disks:
_SafeShutdownInstanceDisks(self.lu, self.instance)
+ assert not self.lu.owned_locks(locking.LEVEL_NODE)
+
if __debug__:
# Verify owned locks
- owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
+ owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
nodes = frozenset(self.node_secondary_ip)
assert ((self.early_release and not owned_nodes) or
(not self.early_release and not (set(owned_nodes) - nodes))), \
lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
names = _GenerateUniqueNames(self.lu, lv_names)
+ _, data_p, meta_p = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
+
vg_data = dev.children[0].logical_id[0]
lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
- logical_id=(vg_data, names[0]))
+ logical_id=(vg_data, names[0]), params=data_p)
vg_meta = dev.children[1].logical_id[0]
lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
- logical_id=(vg_meta, names[1]))
+ logical_id=(vg_meta, names[1]), params=meta_p)
new_lvs = [lv_data, lv_meta]
old_lvs = [child.Copy() for child in dev.children]
"volumes"))
raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
- cstep = 5
+ cstep = itertools.count(5)
+
if self.early_release:
- self.lu.LogStep(cstep, steps_total, "Removing old storage")
- cstep += 1
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
self._RemoveOldStorage(self.target_node, iv_names)
- # WARNING: we release both node locks here, do not do other RPCs
- # than WaitForSync to the primary node
- _ReleaseLocks(self.lu, locking.LEVEL_NODE,
- names=[self.target_node, self.other_node])
+ # TODO: Check if releasing locks early still makes sense
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+ else:
+ # Release all resource locks except those used by the instance
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+ keep=self.node_secondary_ip.keys())
+
+ # Release all node locks while waiting for sync
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
+ # TODO: Can the instance lock be downgraded here? Take the optional disk
+ # shutdown in the caller into consideration.
# Wait for sync
# This can fail as the old devices are degraded and _WaitForSync
# does a combined result over all disks, so we don't check its return value
- self.lu.LogStep(cstep, steps_total, "Sync devices")
- cstep += 1
+ self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
_WaitForSync(self.lu, self.instance)
# Check all devices manually
# Step: remove old storage
if not self.early_release:
- self.lu.LogStep(cstep, steps_total, "Removing old storage")
- cstep += 1
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
self._RemoveOldStorage(self.target_node, iv_names)
def _ExecDrbd8Secondary(self, feedback_fn):
iv_names[idx] = (dev, dev.children, new_net_id)
logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
new_net_id)
+ drbd_params, _, _ = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
logical_id=new_alone_id,
children=dev.children,
- size=dev.size)
+ size=dev.size,
+ params=drbd_params)
try:
_CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
_GetInstanceInfoText(self.instance), False)
self.cfg.Update(self.instance, feedback_fn)
+ # Release all node locks (the configuration has been updated)
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
# and now perform the drbd attach
self.lu.LogInfo("Attaching primary drbds to new secondary"
" (standalone => connected)")
to_node, msg,
hint=("please do a gnt-instance info to see the"
" status of disks"))
- cstep = 5
+
+ cstep = itertools.count(5)
+
if self.early_release:
- self.lu.LogStep(cstep, steps_total, "Removing old storage")
- cstep += 1
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
self._RemoveOldStorage(self.target_node, iv_names)
- # WARNING: we release all node locks here, do not do other RPCs
- # than WaitForSync to the primary node
- _ReleaseLocks(self.lu, locking.LEVEL_NODE,
- names=[self.instance.primary_node,
- self.target_node,
- self.new_node])
+ # TODO: Check if releasing locks early still makes sense
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+ else:
+ # Release all resource locks except those used by the instance
+ _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+ keep=self.node_secondary_ip.keys())
+
+ # TODO: Can the instance lock be downgraded here? Take the optional disk
+ # shutdown in the caller into consideration.
# Wait for sync
# This can fail as the old devices are degraded and _WaitForSync
# does a combined result over all disks, so we don't check its return value
- self.lu.LogStep(cstep, steps_total, "Sync devices")
- cstep += 1
+ self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
_WaitForSync(self.lu, self.instance)
# Check all devices manually
# Step: remove old storage
if not self.early_release:
- self.lu.LogStep(cstep, steps_total, "Removing old storage")
+ self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
self._RemoveOldStorage(self.target_node, iv_names)
"""
# Check whether any instance on this node has faulty disks
for inst in _GetNodeInstances(self.cfg, self.op.node_name):
- if not inst.admin_up:
+ if inst.admin_state != constants.ADMINST_UP:
continue
check_nodes = set(inst.all_nodes)
check_nodes.discard(self.op.node_name)
"""
REQ_BGL = False
+ _MODE2IALLOCATOR = {
+ constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
+ constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
+ constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
+ }
+ assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
+ assert (frozenset(_MODE2IALLOCATOR.values()) ==
+ constants.IALLOCATOR_NEVAC_MODES)
+
def CheckArguments(self):
_CheckIAllocatorOrNode(self, "iallocator", "remote_node")
raise errors.OpPrereqError("Can not use evacuated node as a new"
" secondary node", errors.ECODE_INVAL)
- if self.op.mode != constants.IALLOCATOR_NEVAC_SEC:
+ if self.op.mode != constants.NODE_EVAC_SEC:
raise errors.OpPrereqError("Without the use of an iallocator only"
" secondary instances can be evacuated",
errors.ECODE_INVAL)
locking.LEVEL_NODE: [],
}
+ # Determine nodes (via group) optimistically, needs verification once locks
+ # have been acquired
+ self.lock_nodes = self._DetermineNodes()
+
+ def _DetermineNodes(self):
+ """Gets the list of nodes to operate on.
+
+ """
if self.op.remote_node is None:
# Iallocator will choose any node(s) in the same group
group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
group_nodes = frozenset([self.op.remote_node])
# Determine nodes to be locked
- self.lock_nodes = set([self.op.node_name]) | group_nodes
+ return 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
+ assert self.op.mode in constants.NODE_EVAC_MODES
- if self.op.mode == constants.IALLOCATOR_NEVAC_PRI:
+ if self.op.mode == constants.NODE_EVAC_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:
+ elif self.op.mode == constants.NODE_EVAC_SEC:
# Secondary instances only
inst_fn = _GetNodeSecondaryInstances
else:
# All instances
- assert self.op.mode == constants.IALLOCATOR_NEVAC_ALL
+ assert self.op.mode == constants.NODE_EVAC_ALL
inst_fn = _GetNodeInstances
+ # TODO: In 2.6, change the iallocator interface to take an evacuation mode
+ # per instance
+ raise errors.OpPrereqError("Due to an issue with the iallocator"
+ " interface it is not possible to evacuate"
+ " all instances at once; specify explicitly"
+ " whether to evacuate primary or secondary"
+ " instances",
+ errors.ECODE_INVAL)
return inst_fn(self.cfg, self.op.node_name)
set(i.name for i in self._DetermineInstances())
elif level == locking.LEVEL_NODEGROUP:
- # Lock node groups optimistically, needs verification once nodes have
- # been acquired
+ # Lock node groups for all potential target nodes optimistically, needs
+ # verification once nodes have been acquired
self.needed_locks[locking.LEVEL_NODEGROUP] = \
self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
owned_nodes = self.owned_locks(locking.LEVEL_NODE)
owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
- assert owned_nodes == self.lock_nodes
+ need_nodes = self._DetermineNodes()
+
+ if not owned_nodes.issuperset(need_nodes):
+ raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
+ " locks were acquired, current nodes are"
+ " are '%s', used to be '%s'; retry the"
+ " operation" %
+ (self.op.node_name,
+ utils.CommaJoin(need_nodes),
+ utils.CommaJoin(owned_nodes)),
+ errors.ECODE_STATE)
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'" %
+ " current groups are '%s', used to be '%s';"
+ " retry the operation" %
(utils.CommaJoin(wanted_groups),
utils.CommaJoin(owned_groups)))
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'" %
+ " used to be '%s'; retry the operation" %
(self.op.node_name,
utils.CommaJoin(self.instance_names),
utils.CommaJoin(owned_instances)))
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,
+ evac_mode=self._MODE2IALLOCATOR[self.op.mode],
instances=list(self.instance_names))
ial.Run(self.op.iallocator)
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
+ assert self.op.mode == constants.NODE_EVAC_SEC
jobs = [
[opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
remote_node=self.op.remote_node,
(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))
+ failreason = utils.CommaJoin("%s (%s)" % (name, reason)
+ for (name, reason) in failed)
+ lu.LogWarning("Unable to evacuate instances %s", failreason)
+ raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
if moved:
lu.LogInfo("Instances to be moved: %s",
def ExpandNames(self):
self._ExpandAndLockInstance()
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
def DeclareLocks(self, level):
if level == locking.LEVEL_NODE:
self._LockInstancesNodes()
+ elif level == locking.LEVEL_NODE_RES:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
instance = self.instance
disk = self.disk
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+ assert (self.owned_locks(locking.LEVEL_NODE) ==
+ self.owned_locks(locking.LEVEL_NODE_RES))
+
disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
if not disks_ok:
raise errors.OpExecError("Cannot activate block device to grow")
+ feedback_fn("Growing disk %s of instance '%s' by %s" %
+ (self.op.disk, instance.name,
+ utils.FormatUnit(self.op.amount, "h")))
+
# First run all grow ops in dry-run mode
for node in instance.all_nodes:
self.cfg.SetDiskID(disk, node)
disk.RecordGrow(self.op.amount)
self.cfg.Update(instance, feedback_fn)
+
+ # Changes have been recorded, release node lock
+ _ReleaseLocks(self, locking.LEVEL_NODE)
+
+ # Downgrade lock while waiting for sync
+ self.glm.downgrade(locking.LEVEL_INSTANCE)
+
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")
- if not instance.admin_up:
+ if instance.admin_state != constants.ADMINST_UP:
_SafeShutdownInstanceDisks(self, instance, disks=[disk])
- elif not instance.admin_up:
+ 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")
+ assert self.owned_locks(locking.LEVEL_NODE_RES)
+ assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+
class LUInstanceQueryData(NoHooksLU):
"""Query runtime instance data.
if remote_info and "state" in remote_info:
remote_state = "up"
else:
- remote_state = "down"
-
- if instance.admin_up:
- config_state = "up"
- else:
- config_state = "down"
+ if instance.admin_state == constants.ADMINST_UP:
+ remote_state = "down"
+ else:
+ remote_state = instance.admin_state
disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
instance.disks)
result[instance.name] = {
"name": instance.name,
- "config_state": config_state,
+ "config_state": instance.admin_state,
"run_state": remote_state,
"pnode": instance.primary_node,
"snodes": instance.secondary_nodes,
def CheckArguments(self):
if not (self.op.nics or self.op.disks or self.op.disk_template or
- self.op.hvparams or self.op.beparams or self.op.os_name):
+ self.op.hvparams or self.op.beparams or self.op.os_name or
+ self.op.online_inst or self.op.offline_inst):
raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
if self.op.hvparams:
def ExpandNames(self):
self._ExpandAndLockInstance()
+ # 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
def DeclareLocks(self, level):
if self.op.disk_template and self.op.remote_node:
self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
+ elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
+ # Copy node locks
+ self.needed_locks[locking.LEVEL_NODE_RES] = \
+ self.needed_locks[locking.LEVEL_NODE][:]
def BuildHooksEnv(self):
"""Build hooks env.
"""
args = dict()
- if constants.BE_MEMORY in self.be_new:
- args["memory"] = self.be_new[constants.BE_MEMORY]
+ if constants.BE_MINMEM in self.be_new:
+ args["minmem"] = self.be_new[constants.BE_MINMEM]
+ if constants.BE_MAXMEM in self.be_new:
+ args["maxmem"] = self.be_new[constants.BE_MAXMEM]
if constants.BE_VCPUS in self.be_new:
args["vcpus"] = self.be_new[constants.BE_VCPUS]
# TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
"Cannot retrieve locked instance %s" % self.op.instance_name
pnode = instance.primary_node
nodelist = list(instance.all_nodes)
+ pnode_info = self.cfg.GetNodeInfo(pnode)
+ self.diskparams = self.cfg.GetNodeGroup(pnode_info.group).diskparams
# OS change
if self.op.os_name and not self.op.force:
" %s to %s" % (instance.disk_template,
self.op.disk_template),
errors.ECODE_INVAL)
- _CheckInstanceDown(self, instance, "cannot change disk template")
+ _CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot change disk template")
if self.op.disk_template in constants.DTS_INT_MIRROR:
if self.op.remote_node == pnode:
raise errors.OpPrereqError("Given new secondary node %s is the same"
required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
_CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
+ snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
+ if pnode_info.group != snode_info.group:
+ self.LogWarning("The primary and secondary nodes are in two"
+ " different node groups; the disk parameters"
+ " from the first disk's node group will be"
+ " used")
+
# hvparams processing
if self.op.hvparams:
hv_type = instance.hypervisor
if self.op.beparams:
i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
use_none=True)
+ objects.UpgradeBeParams(i_bedict)
utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
be_new = cluster.SimpleFillBE(i_bedict)
self.be_proposed = self.be_new = be_new # the new actual values
self.warn = []
- if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
- be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
+ be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
mem_check_list = [pnode]
if be_new[constants.BE_AUTO_BALANCE]:
# either we changed auto_balance to yes or it was from before
instance_info = self.rpc.call_instance_info(pnode, instance.name,
instance.hypervisor)
nodeinfo = self.rpc.call_node_info(mem_check_list, None,
- instance.hypervisor)
+ [instance.hypervisor])
pninfo = nodeinfo[pnode]
msg = pninfo.fail_msg
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):
- self.warn.append("Node data from primary node %s doesn't contain"
- " free memory information" % pnode)
- elif instance_info.fail_msg:
- self.warn.append("Can't get instance runtime information: %s" %
- instance_info.fail_msg)
else:
- if instance_info.payload:
- current_mem = int(instance_info.payload["memory"])
+ (_, _, (pnhvinfo, )) = pninfo.payload
+ if not isinstance(pnhvinfo.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:
+ self.warn.append("Can't get instance runtime information: %s" %
+ instance_info.fail_msg)
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"])
- if miss_mem > 0:
- raise errors.OpPrereqError("This change will prevent the instance"
- " from starting, due to %d MB of memory"
- " missing on its primary node" % miss_mem,
- errors.ECODE_NORES)
+ if instance_info.payload:
+ 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)
+ # TODO: Describe race condition
+ current_mem = 0
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
+ pnhvinfo["memory_free"])
+ if miss_mem > 0:
+ raise errors.OpPrereqError("This change will prevent the instance"
+ " from starting, due to %d MB of memory"
+ " missing on its primary node" %
+ miss_mem,
+ errors.ECODE_NORES)
if be_new[constants.BE_AUTO_BALANCE]:
for node, nres in nodeinfo.items():
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):
+ (_, _, (nhvinfo, )) = nres.payload
+ if not isinstance(nhvinfo.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"]:
+ #TODO(dynmem): do the appropriate check involving MINMEM
+ elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
raise errors.OpPrereqError("This change will prevent the instance"
" from failover to its secondary node"
" %s, due to not enough memory" % node,
if len(instance.disks) == 1:
raise errors.OpPrereqError("Cannot remove the last disk of"
" an instance", errors.ECODE_INVAL)
- _CheckInstanceDown(self, instance, "cannot remove disks")
+ _CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot remove disks")
if (disk_op == constants.DDM_ADD and
len(instance.disks) >= constants.MAX_DISKS):
(disk_op, len(instance.disks)),
errors.ECODE_INVAL)
- return
+ # disabling the instance
+ if self.op.offline_inst:
+ _CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot change instance state to offline")
+
+ # enabling the instance
+ if self.op.online_inst:
+ _CheckInstanceState(self, instance, INSTANCE_OFFLINE,
+ msg="cannot make instance go online")
def _ConvertPlainToDrbd(self, feedback_fn):
"""Converts an instance from plain to drbd.
pnode = instance.primary_node
snode = self.op.remote_node
+ assert instance.disk_template == constants.DT_PLAIN
+
# create a fake disk info for _GenerateDiskTemplate
disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
constants.IDISK_VG: d.logical_id[0]}
for d in instance.disks]
new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
instance.name, pnode, [snode],
- disk_info, None, None, 0, feedback_fn)
+ disk_info, None, None, 0, feedback_fn,
+ self.diskparams)
info = _GetInstanceInfoText(instance)
feedback_fn("Creating aditional volumes...")
# first, create the missing data and meta devices
instance.disks = new_disks
self.cfg.Update(instance, feedback_fn)
+ # Release node locks while waiting for sync
+ _ReleaseLocks(self, locking.LEVEL_NODE)
+
# disks are created, waiting for sync
disk_abort = not _WaitForSync(self, instance,
oneshot=not self.op.wait_for_sync)
raise errors.OpExecError("There are some degraded disks for"
" this instance, please cleanup manually")
+ # Node resource locks will be released by caller
+
def _ConvertDrbdToPlain(self, feedback_fn):
"""Converts an instance from drbd to plain.
"""
instance = self.instance
+
assert len(instance.secondary_nodes) == 1
+ assert instance.disk_template == constants.DT_DRBD8
+
pnode = instance.primary_node
snode = instance.secondary_nodes[0]
feedback_fn("Converting template to plain")
instance.disk_template = constants.DT_PLAIN
self.cfg.Update(instance, feedback_fn)
+ # Release locks in case removing disks takes a while
+ _ReleaseLocks(self, locking.LEVEL_NODE)
+
feedback_fn("Removing volumes on the secondary node...")
for disk in old_disks:
self.cfg.SetDiskID(disk, snode)
self.LogWarning("Could not remove metadata for disk %d on node %s,"
" continuing anyway: %s", idx, pnode, msg)
+ # this is a DRBD disk, return its port to the pool
+ for disk in old_disks:
+ tcp_port = disk.logical_id[2]
+ self.cfg.AddTcpUdpPort(tcp_port)
+
+ # Node resource locks will be released by caller
+
def Exec(self, feedback_fn):
"""Modifies an instance.
for warn in self.warn:
feedback_fn("WARNING: %s" % warn)
+ assert ((self.op.disk_template is None) ^
+ bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
+ "Not owning any node resource locks"
+
result = []
instance = self.instance
# disk changes
self.LogWarning("Could not remove disk/%d on node %s: %s,"
" continuing anyway", device_idx, node, msg)
result.append(("disk/%d" % device_idx, "remove"))
+
+ # if this is a DRBD disk, return its port to the pool
+ if device.dev_type in constants.LDS_DRBD:
+ tcp_port = device.logical_id[2]
+ self.cfg.AddTcpUdpPort(tcp_port)
elif disk_op == constants.DDM_ADD:
# add a new disk
if instance.disk_template in (constants.DT_FILE,
[disk_dict],
file_path,
file_driver,
- disk_idx_base, feedback_fn)[0]
+ disk_idx_base,
+ feedback_fn,
+ self.diskparams)[0]
instance.disks.append(new_disk)
info = _GetInstanceInfoText(instance)
disk_dict[constants.IDISK_MODE]))
if self.op.disk_template:
+ if __debug__:
+ check_nodes = set(instance.all_nodes)
+ if self.op.remote_node:
+ check_nodes.add(self.op.remote_node)
+ for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
+ owned = self.owned_locks(level)
+ assert not (check_nodes - owned), \
+ ("Not owning the correct locks, owning %r, expected at least %r" %
+ (owned, check_nodes))
+
r_shut = _ShutdownInstanceDisks(self, instance)
if not r_shut:
raise errors.OpExecError("Cannot shutdown instance disks, unable to"
raise
result.append(("disk_template", self.op.disk_template))
+ assert instance.disk_template == self.op.disk_template, \
+ ("Expected disk template '%s', found '%s'" %
+ (self.op.disk_template, instance.disk_template))
+
+ # Release node and resource locks if there are any (they might already have
+ # been released during disk conversion)
+ _ReleaseLocks(self, locking.LEVEL_NODE)
+ _ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
# NIC changes
for nic_op, nic_dict in self.op.nics:
if nic_op == constants.DDM_REMOVE:
for key, val in self.op.osparams.iteritems():
result.append(("os/%s" % key, val))
+ # online/offline instance
+ if self.op.online_inst:
+ self.cfg.MarkInstanceDown(instance.name)
+ result.append(("admin_state", constants.ADMINST_DOWN))
+ if self.op.offline_inst:
+ self.cfg.MarkInstanceOffline(instance.name)
+ result.append(("admin_state", constants.ADMINST_OFFLINE))
+
self.cfg.Update(instance, feedback_fn)
+ assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
+ self.owned_locks(locking.LEVEL_NODE)), \
+ "All node locks should have been released by now"
+
return result
_DISK_CONVERSIONS = {
"Cannot retrieve locked instance %s" % self.op.instance_name
_CheckNodeOnline(self, self.instance.primary_node)
- if (self.op.remove_instance and self.instance.admin_up and
+ if (self.op.remove_instance and
+ self.instance.admin_state == constants.ADMINST_UP and
not self.op.shutdown):
raise errors.OpPrereqError("Can not remove instance without shutting it"
" down before")
for disk in instance.disks:
self.cfg.SetDiskID(disk, src_node)
- activate_disks = (not instance.admin_up)
+ activate_disks = (instance.admin_state != constants.ADMINST_UP)
if activate_disks:
# Activate the instance disks if we'exporting a stopped instance
helper.CreateSnapshots()
try:
- if (self.op.shutdown and instance.admin_up and
+ if (self.op.shutdown and
+ instance.admin_state == constants.ADMINST_UP and
not self.op.remove_instance):
assert not activate_disks
feedback_fn("Starting instance %s" % instance.name)
if self.op.ndparams:
utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
+ if self.op.diskparams:
+ for templ in constants.DISK_TEMPLATES:
+ if templ not in self.op.diskparams:
+ self.op.diskparams[templ] = {}
+ utils.ForceDictType(self.op.diskparams[templ], constants.DISK_DT_TYPES)
+ else:
+ self.op.diskparams = self.cfg.GetClusterInfo().diskparams
+
def BuildHooksEnv(self):
"""Build hooks env.
group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
uuid=self.group_uuid,
alloc_policy=self.op.alloc_policy,
- ndparams=self.op.ndparams)
+ ndparams=self.op.ndparams,
+ diskparams=self.op.diskparams)
self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
del self.remove_locks[locking.LEVEL_NODEGROUP]
"""Assign nodes to a new group.
"""
- for node in self.op.nodes:
- self.node_data[node].group = self.group_uuid
-
- # FIXME: Depends on side-effects of modifying the result of
- # C{cfg.GetAllNodesInfo}
+ mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
- self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
+ self.cfg.AssignGroupNodes(mods)
@staticmethod
def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
def CheckArguments(self):
all_changes = [
self.op.ndparams,
+ self.op.diskparams,
self.op.alloc_policy,
]
utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
self.new_ndparams = new_ndparams
+ if self.op.diskparams:
+ self.new_diskparams = dict()
+ for templ in constants.DISK_TEMPLATES:
+ if templ not in self.op.diskparams:
+ self.op.diskparams[templ] = {}
+ new_templ_params = _GetUpdatedParams(self.group.diskparams[templ],
+ self.op.diskparams[templ])
+ utils.ForceDictType(new_templ_params, constants.DISK_DT_TYPES)
+ self.new_diskparams[templ] = new_templ_params
+
def BuildHooksEnv(self):
"""Build hooks env.
self.group.ndparams = self.new_ndparams
result.append(("ndparams", str(self.group.ndparams)))
+ if self.op.diskparams:
+ self.group.diskparams = self.new_diskparams
+ result.append(("diskparams", str(self.group.diskparams)))
+
if self.op.alloc_policy:
self.group.alloc_policy = self.op.alloc_policy
else:
hypervisor_name = cluster_info.enabled_hypervisors[0]
- node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
- hypervisor_name)
+ node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
+ [hypervisor_name])
node_iinfo = \
self.rpc.call_all_instances_info(node_list,
cluster_info.enabled_hypervisors)
@param node_results: the basic node structures as filled from the config
"""
+ #TODO(dynmem): compute the right data on MAX and MIN memory
# make a copy of the current dict
node_results = dict(node_results)
for nname, nresult in node_data.items():
nresult.Raise("Can't get data for node %s" % nname)
node_iinfo[nname].Raise("Can't get node instance info from node %s" %
nname)
- remote_info = nresult.payload
+ remote_info = _MakeLegacyNodeInfo(nresult.payload)
for attr in ["memory_total", "memory_free", "memory_dom0",
"vg_size", "vg_free", "cpu_total"]:
i_p_mem = i_p_up_mem = 0
for iinfo, beinfo in i_list:
if iinfo.primary_node == nname:
- i_p_mem += beinfo[constants.BE_MEMORY]
+ i_p_mem += beinfo[constants.BE_MAXMEM]
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_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
+ i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
remote_info["memory_free"] -= max(0, i_mem_diff)
- if iinfo.admin_up:
- i_p_up_mem += beinfo[constants.BE_MEMORY]
+ if iinfo.admin_state == constants.ADMINST_UP:
+ i_p_up_mem += beinfo[constants.BE_MAXMEM]
# compute memory used by instances
pnr_dyn = {
nic_data.append(nic_dict)
pir = {
"tags": list(iinfo.GetTags()),
- "admin_up": iinfo.admin_up,
+ "admin_state": iinfo.admin_state,
"vcpus": beinfo[constants.BE_VCPUS],
- "memory": beinfo[constants.BE_MEMORY],
+ "memory": beinfo[constants.BE_MAXMEM],
"os": iinfo.os,
"nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
"nics": nic_data,