# W0201 since most LU attributes are defined in CheckPrereq or similar
# functions
-# C0302: since we have waaaay to many lines in this module
+# C0302: since we have waaaay too many lines in this module
import os
import os.path
from ganeti import qlang
from ganeti import opcodes
from ganeti import ht
+from ganeti import rpc
import ganeti.masterd.instance # pylint: disable=W0611
+#: 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.
HTYPE = None
REQ_BGL = True
- def __init__(self, processor, op, context, rpc):
+ def __init__(self, processor, op, context, rpc_runner):
"""Constructor for LogicalUnit.
This needs to be overridden in derived classes in order to check op
# readability alias
self.owned_locks = context.glm.list_owned
self.context = context
- self.rpc = rpc
+ self.rpc = rpc_runner
# Dicts used to declare locking needs to mcpu
self.needed_locks = None
self.share_locks = dict.fromkeys(locking.LEVELS, 0)
self.op.instance_name)
self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
- def _LockInstancesNodes(self, primary_only=False):
+ def _LockInstancesNodes(self, primary_only=False,
+ level=locking.LEVEL_NODE):
"""Helper function to declare instances' nodes for locking.
This function should be called after locking one or more instances to lock
@type primary_only: boolean
@param primary_only: only lock primary nodes of locked instances
+ @param level: Which lock level to use for locking nodes
"""
- assert locking.LEVEL_NODE in self.recalculate_locks, \
+ assert level in self.recalculate_locks, \
"_LockInstancesNodes helper function called with no nodes to recalculate"
# TODO: check if we're really been called with the instance locks held
if not primary_only:
wanted_nodes.extend(instance.secondary_nodes)
- if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
- self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
- elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
- self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
+ if self.recalculate_locks[level] == constants.LOCKS_REPLACE:
+ self.needed_locks[level] = wanted_nodes
+ elif self.recalculate_locks[level] == constants.LOCKS_APPEND:
+ self.needed_locks[level].extend(wanted_nodes)
+ else:
+ raise errors.ProgrammerError("Unknown recalculation mode")
- del self.recalculate_locks[locking.LEVEL_NODE]
+ del self.recalculate_locks[level]
class NoHooksLU(LogicalUnit): # pylint: disable=W0223
#: Attribute holding field definitions
FIELDS = None
- def __init__(self, filter_, fields, use_locking):
+ def __init__(self, qfilter, fields, use_locking):
"""Initializes this class.
"""
self.use_locking = use_locking
- self.query = query.Query(self.FIELDS, fields, filter_=filter_,
+ self.query = query.Query(self.FIELDS, fields, qfilter=qfilter,
namefield="name")
self.requested_data = self.query.RequestedData()
self.names = self.query.RequestedNames()
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:
"""Runs the post-hook for an opcode on a single node.
"""
- hm = lu.proc.hmclass(lu.rpc.call_hooks_runner, lu)
+ hm = lu.proc.BuildHooksManager(lu)
try:
hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
except:
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.
+
+ @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)
- 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 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,
return []
-def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
+def _FindFaultyInstanceDisks(cfg, rpc_runner, instance, node_name, prereq):
faulty = []
for dev in instance.disks:
cfg.SetDiskID(dev, node_name)
- result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
+ result = rpc_runner.call_blockdev_getmirrorstatus(node_name, instance.disks)
result.Raise("Failed to get disk status from node %s" % node_name,
prereq=prereq, ecode=errors.ECODE_ENVIRON)
"""Destroys the cluster.
"""
- master = self.cfg.GetMasterNode()
+ master_params = self.cfg.GetMasterNetworkParameters()
# Run post hooks on master node before it's removed
- _RunPostHook(self, master)
+ _RunPostHook(self, master_params.name)
- result = self.rpc.call_node_stop_master(master, False)
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master role")
- return master
+ return master_params.name
def _VerifyCertificate(filename):
self.op and self._feedback_fn to be available.)
"""
- TCLUSTER = "cluster"
- TNODE = "node"
- TINSTANCE = "instance"
-
- ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
- ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
- ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
- ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
- ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
- EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
- EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
- EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
- EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
- EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
- EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
- EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
- ENODEDRBD = (TNODE, "ENODEDRBD")
- ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
- ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
- ENODEHOOKS = (TNODE, "ENODEHOOKS")
- ENODEHV = (TNODE, "ENODEHV")
- ENODELVM = (TNODE, "ENODELVM")
- ENODEN1 = (TNODE, "ENODEN1")
- ENODENET = (TNODE, "ENODENET")
- ENODEOS = (TNODE, "ENODEOS")
- ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
- ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
- ENODERPC = (TNODE, "ENODERPC")
- ENODESSH = (TNODE, "ENODESSH")
- ENODEVERSION = (TNODE, "ENODEVERSION")
- ENODESETUP = (TNODE, "ENODESETUP")
- ENODETIME = (TNODE, "ENODETIME")
- ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
ETYPE_FIELD = "code"
ETYPE_ERROR = "ERROR"
"""
ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
- itype, etxt = ecode
+ itype, etxt, _ = ecode
# first complete the msg
if args:
msg = msg % args
# and finally report it via the feedback_fn
self._feedback_fn(" - %s" % msg) # Mix-in. pylint: disable=E1101
- def _ErrorIf(self, cond, *args, **kwargs):
+ def _ErrorIf(self, cond, ecode, *args, **kwargs):
"""Log an error message if the passed condition is True.
"""
cond = (bool(cond)
or self.op.debug_simulate_errors) # pylint: disable=E1101
+
+ # If the error code is in the list of ignored errors, demote the error to a
+ # warning
+ (_, etxt, _) = ecode
+ if etxt in self.op.ignore_errors: # pylint: disable=E1101
+ kwargs[self.ETYPE_FIELD] = self.ETYPE_WARNING
+
if cond:
- self._Error(*args, **kwargs)
+ self._Error(ecode, *args, **kwargs)
+
# do not mark the operation as failed for WARN cases only
if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
self.bad = self.bad or cond
groups = self.cfg.GetNodeGroupList()
# Verify global configuration
- jobs.append([opcodes.OpClusterVerifyConfig()])
+ jobs.append([
+ opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
+ ])
# Always depend on global verification
depends_fn = lambda: [(-len(jobs), [])]
jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
- depends=depends_fn())]
+ ignore_errors=self.op.ignore_errors,
+ depends=depends_fn())]
for group in groups)
# Fix up all parameters
utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
hv_class.CheckParameterSyntax(hv_params)
except errors.GenericError, err:
- self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
+ self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg % str(err))
def ExpandNames(self):
# Information can be safely retrieved as the BGL is acquired in exclusive
feedback_fn("* Verifying cluster config")
for msg in self.cfg.VerifyConfig():
- self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
+ self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg)
feedback_fn("* Verifying cluster certificate files")
for cert_filename in constants.ALL_CERT_FILES:
(errcode, msg) = _VerifyCertificate(cert_filename)
- self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
+ self._ErrorIf(errcode, constants.CV_ECLUSTERCERT, None, msg, code=errcode)
feedback_fn("* Verifying hypervisor parameters")
["no instances"])))
for node in dangling_nodes]
- self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
+ self._ErrorIf(bool(dangling_nodes), constants.CV_ECLUSTERDANGLINGNODES,
+ None,
"the following nodes (and their instances) belong to a non"
" existing group: %s", utils.CommaJoin(pretty_dangling))
- self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
+ self._ErrorIf(bool(no_node_instances), constants.CV_ECLUSTERDANGLINGINST,
+ None,
"the following instances have a non-existing primary-node:"
" %s", utils.CommaJoin(no_node_instances))
# main result, nresult should be a non-empty dict
test = not nresult or not isinstance(nresult, dict)
- _ErrorIf(test, self.ENODERPC, node,
+ _ErrorIf(test, constants.CV_ENODERPC, node,
"unable to verify node: no data returned")
if test:
return False
test = not (remote_version and
isinstance(remote_version, (list, tuple)) and
len(remote_version) == 2)
- _ErrorIf(test, self.ENODERPC, node,
+ _ErrorIf(test, constants.CV_ENODERPC, node,
"connection to node returned invalid data")
if test:
return False
test = local_version != remote_version[0]
- _ErrorIf(test, self.ENODEVERSION, node,
+ _ErrorIf(test, constants.CV_ENODEVERSION, node,
"incompatible protocol versions: master %s,"
" node %s", local_version, remote_version[0])
if test:
# full package version
self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
- self.ENODEVERSION, node,
+ constants.CV_ENODEVERSION, node,
"software version mismatch: master %s, node %s",
constants.RELEASE_VERSION, remote_version[1],
code=self.ETYPE_WARNING)
if ninfo.vm_capable and isinstance(hyp_result, dict):
for hv_name, hv_result in hyp_result.iteritems():
test = hv_result is not None
- _ErrorIf(test, self.ENODEHV, node,
+ _ErrorIf(test, constants.CV_ENODEHV, node,
"hypervisor %s verify failure: '%s'", hv_name, hv_result)
hvp_result = nresult.get(constants.NV_HVPARAMS, None)
if ninfo.vm_capable and isinstance(hvp_result, list):
for item, hv_name, hv_result in hvp_result:
- _ErrorIf(True, self.ENODEHV, node,
+ _ErrorIf(True, constants.CV_ENODEHV, node,
"hypervisor %s parameter verify failure (source %s): %s",
hv_name, item, hv_result)
test = nresult.get(constants.NV_NODESETUP,
["Missing NODESETUP results"])
- _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
+ _ErrorIf(test, constants.CV_ENODESETUP, node, "node setup error: %s",
"; ".join(test))
return True
try:
ntime_merged = utils.MergeTime(ntime)
except (ValueError, TypeError):
- _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
+ _ErrorIf(True, constants.CV_ENODETIME, node, "Node returned invalid time")
return
if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
else:
ntime_diff = None
- _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
+ _ErrorIf(ntime_diff is not None, constants.CV_ENODETIME, node,
"Node time diverges by at least %s from master node time",
ntime_diff)
# checks vg existence and size > 20G
vglist = nresult.get(constants.NV_VGLIST, None)
test = not vglist
- _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
+ _ErrorIf(test, constants.CV_ENODELVM, node, "unable to check volume groups")
if not test:
vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
constants.MIN_VG_SIZE)
- _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
+ _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus)
# check pv names
pvlist = nresult.get(constants.NV_PVLIST, None)
test = pvlist is None
- _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
+ _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node")
if not test:
# check that ':' is not present in PV names, since it's a
# special character for lvcreate (denotes the range of PEs to
# use on the PV)
for _, pvname, owner_vg in pvlist:
test = ":" in pvname
- _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
- " '%s' of VG '%s'", pvname, owner_vg)
+ _ErrorIf(test, constants.CV_ENODELVM, node,
+ "Invalid character ':' in PV '%s' of VG '%s'",
+ pvname, owner_vg)
def _VerifyNodeBridges(self, ninfo, nresult, bridges):
"""Check the node bridges.
missing = nresult.get(constants.NV_BRIDGES, None)
test = not isinstance(missing, list)
- _ErrorIf(test, self.ENODENET, node,
+ _ErrorIf(test, constants.CV_ENODENET, node,
"did not return valid bridge information")
if not test:
- _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
- utils.CommaJoin(sorted(missing)))
+ _ErrorIf(bool(missing), constants.CV_ENODENET, node,
+ "missing bridges: %s" % utils.CommaJoin(sorted(missing)))
+
+ def _VerifyNodeUserScripts(self, ninfo, nresult):
+ """Check the results of user scripts presence and executability on the node
+
+ @type ninfo: L{objects.Node}
+ @param ninfo: the node to check
+ @param nresult: the remote results for the node
+
+ """
+ node = ninfo.name
+
+ test = not constants.NV_USERSCRIPTS in nresult
+ self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
+ "did not return user scripts information")
+
+ broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
+ if not test:
+ self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
+ "user scripts not present or not executable: %s" %
+ utils.CommaJoin(sorted(broken_scripts)))
def _VerifyNodeNetwork(self, ninfo, nresult):
"""Check the node network connectivity results.
_ErrorIf = self._ErrorIf # pylint: disable=C0103
test = constants.NV_NODELIST not in nresult
- _ErrorIf(test, self.ENODESSH, node,
+ _ErrorIf(test, constants.CV_ENODESSH, node,
"node hasn't returned node ssh connectivity data")
if not test:
if nresult[constants.NV_NODELIST]:
for a_node, a_msg in nresult[constants.NV_NODELIST].items():
- _ErrorIf(True, self.ENODESSH, node,
+ _ErrorIf(True, constants.CV_ENODESSH, node,
"ssh communication with node '%s': %s", a_node, a_msg)
test = constants.NV_NODENETTEST not in nresult
- _ErrorIf(test, self.ENODENET, node,
+ _ErrorIf(test, constants.CV_ENODENET, node,
"node hasn't returned node tcp connectivity data")
if not test:
if nresult[constants.NV_NODENETTEST]:
nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
for anode in nlist:
- _ErrorIf(True, self.ENODENET, node,
+ _ErrorIf(True, constants.CV_ENODENET, node,
"tcp communication with node '%s': %s",
anode, nresult[constants.NV_NODENETTEST][anode])
test = constants.NV_MASTERIP not in nresult
- _ErrorIf(test, self.ENODENET, node,
+ _ErrorIf(test, constants.CV_ENODENET, node,
"node hasn't returned node master IP reachability data")
if not test:
if not nresult[constants.NV_MASTERIP]:
msg = "the master node cannot reach the master IP (not configured?)"
else:
msg = "cannot reach the master IP"
- _ErrorIf(True, self.ENODENET, node, msg)
+ _ErrorIf(True, constants.CV_ENODENET, node, msg)
def _VerifyInstance(self, instance, instanceconfig, node_image,
diskstatus):
continue
for volume in node_vol_should[node]:
test = volume not in n_img.volumes
- _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
+ _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, self.EINSTANCEDOWN, instance,
+ _ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
"instance not running on its primary node %s",
node_current)
# node here
snode = node_image[nname]
bad_snode = snode.ghost or snode.offline
- _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
- self.EINSTANCEFAULTYDISK, instance,
+ _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),
- self.EINSTANCEFAULTYDISK, instance,
+ _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)
def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
test = ((node not in node_vol_should or
volume not in node_vol_should[node]) and
not reserved.Matches(volume))
- self._ErrorIf(test, self.ENODEORPHANLV, node,
+ self._ErrorIf(test, constants.CV_ENODEORPHANLV, node,
"volume %s is unknown", volume)
def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
# 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, self.ENODEN1, node,
+ self._ErrorIf(test, constants.CV_ENODEN1, node,
"not enough memory to accomodate instance failovers"
" should node %s fail (%dMiB needed, %dMiB available)",
prinode, needed_mem, n_img.mfree)
@classmethod
def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
- (files_all, files_all_opt, files_mc, files_vm)):
+ (files_all, files_opt, files_mc, files_vm)):
"""Verifies file checksums collected from all nodes.
@param errorif: Callback for reporting errors
@param all_nvinfo: RPC results
"""
- assert (len(files_all | files_all_opt | files_mc | files_vm) ==
- sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
- "Found file listed in more than one file list"
-
# Define functions determining which nodes to consider for a file
files2nodefn = [
(files_all, None),
- (files_all_opt, None),
(files_mc, lambda node: (node.master_candidate or
node.name == master_node)),
(files_vm, lambda node: node.vm_capable),
frozenset(map(operator.attrgetter("name"), filenodes)))
for filename in files)
- assert set(nodefiles) == (files_all | files_all_opt | files_mc | files_vm)
+ assert set(nodefiles) == (files_all | files_mc | files_vm)
fileinfo = dict((filename, {}) for filename in nodefiles)
ignore_nodes = set()
node_files = nresult.payload.get(constants.NV_FILELIST, None)
test = not (node_files and isinstance(node_files, dict))
- errorif(test, cls.ENODEFILECHECK, node.name,
+ errorif(test, constants.CV_ENODEFILECHECK, node.name,
"Node did not return file checksum data")
if test:
ignore_nodes.add(node.name)
# Nodes missing file
missing_file = expected_nodes - with_file
- if filename in files_all_opt:
+ if filename in files_opt:
# All or no nodes
errorif(missing_file and missing_file != expected_nodes,
- cls.ECLUSTERFILECHECK, None,
+ constants.CV_ECLUSTERFILECHECK, None,
"File %s is optional, but it must exist on all or no"
" nodes (not found on %s)",
filename, utils.CommaJoin(utils.NiceSort(missing_file)))
else:
- # Non-optional files
- errorif(missing_file, cls.ECLUSTERFILECHECK, None,
+ errorif(missing_file, constants.CV_ECLUSTERFILECHECK, None,
"File %s is missing from node(s) %s", filename,
utils.CommaJoin(utils.NiceSort(missing_file)))
# Warn if a node has a file it shouldn't
unexpected = with_file - expected_nodes
errorif(unexpected,
- cls.ECLUSTERFILECHECK, None,
+ constants.CV_ECLUSTERFILECHECK, None,
"File %s should not exist on node(s) %s",
filename, utils.CommaJoin(utils.NiceSort(unexpected)))
else:
variants = []
- errorif(test, cls.ECLUSTERFILECHECK, None,
+ errorif(test, constants.CV_ECLUSTERFILECHECK, None,
"File %s found with %s different checksums (%s)",
filename, len(checksums), "; ".join(variants))
if drbd_helper:
helper_result = nresult.get(constants.NV_DRBDHELPER, None)
test = (helper_result == None)
- _ErrorIf(test, self.ENODEDRBDHELPER, node,
+ _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
"no drbd usermode helper returned")
if helper_result:
status, payload = helper_result
test = not status
- _ErrorIf(test, self.ENODEDRBDHELPER, node,
+ _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
"drbd usermode helper check unsuccessful: %s", payload)
test = status and (payload != drbd_helper)
- _ErrorIf(test, self.ENODEDRBDHELPER, node,
+ _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
"wrong drbd usermode helper: %s", payload)
# compute the DRBD minors
node_drbd = {}
for minor, instance in drbd_map[node].items():
test = instance not in instanceinfo
- _ErrorIf(test, self.ECLUSTERCFG, None,
+ _ErrorIf(test, constants.CV_ECLUSTERCFG, None,
"ghost instance '%s' in temporary DRBD map", instance)
# ghost instance should not be running, but otherwise we
# don't give double warnings (both ghost instance and
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, [])
test = not isinstance(used_minors, (tuple, list))
- _ErrorIf(test, self.ENODEDRBD, node,
+ _ErrorIf(test, constants.CV_ENODEDRBD, node,
"cannot parse drbd status file: %s", str(used_minors))
if test:
# we cannot check drbd status
for minor, (iname, must_exist) in node_drbd.items():
test = minor not in used_minors and must_exist
- _ErrorIf(test, self.ENODEDRBD, node,
+ _ErrorIf(test, constants.CV_ENODEDRBD, node,
"drbd minor %d of instance %s is not active", minor, iname)
for minor in used_minors:
test = minor not in node_drbd
- _ErrorIf(test, self.ENODEDRBD, node,
+ _ErrorIf(test, constants.CV_ENODEDRBD, node,
"unallocated drbd minor %d is in use", minor)
def _UpdateNodeOS(self, ninfo, nresult, nimg):
not compat.all(isinstance(v, list) and len(v) == 7
for v in remote_os))
- _ErrorIf(test, self.ENODEOS, node,
+ _ErrorIf(test, constants.CV_ENODEOS, node,
"node hasn't returned valid OS data")
nimg.os_fail = test
for os_name, os_data in nimg.oslist.items():
assert os_data, "Empty OS status for OS %s?!" % os_name
f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
- _ErrorIf(not f_status, self.ENODEOS, node,
+ _ErrorIf(not f_status, constants.CV_ENODEOS, node,
"Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
- _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
+ _ErrorIf(len(os_data) > 1, constants.CV_ENODEOS, node,
"OS '%s' has multiple entries (first one shadows the rest): %s",
os_name, utils.CommaJoin([v[0] for v in os_data]))
# comparisons with the 'base' image
test = os_name not in base.oslist
- _ErrorIf(test, self.ENODEOS, node,
+ _ErrorIf(test, constants.CV_ENODEOS, node,
"Extra OS %s not present on reference node (%s)",
os_name, base.name)
if test:
("variants list", f_var, b_var),
("parameters", beautify_params(f_param),
beautify_params(b_param))]:
- _ErrorIf(a != b, self.ENODEOS, node,
+ _ErrorIf(a != b, constants.CV_ENODEOS, node,
"OS %s for %s differs from reference node %s: [%s] vs. [%s]",
kind, os_name, base.name,
utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
# check any missing OSes
missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
- _ErrorIf(missing, self.ENODEOS, node,
+ _ErrorIf(missing, constants.CV_ENODEOS, node,
"OSes present on reference node %s but missing on this node: %s",
base.name, utils.CommaJoin(missing))
if ((ninfo.master_candidate or ninfo.master_capable) and
constants.NV_OOB_PATHS in nresult):
for path_result in nresult[constants.NV_OOB_PATHS]:
- self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
+ self._ErrorIf(path_result, constants.CV_ENODEOOBPATH, node, path_result)
def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
"""Verifies and updates the node volume data.
if vg_name is None:
pass
elif isinstance(lvdata, basestring):
- _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
+ _ErrorIf(True, constants.CV_ENODELVM, node, "LVM problem on node: %s",
utils.SafeEncode(lvdata))
elif not isinstance(lvdata, dict):
- _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
+ _ErrorIf(True, constants.CV_ENODELVM, node,
+ "rpc call to node failed (lvlist)")
else:
nimg.volumes = lvdata
nimg.lvm_fail = False
"""
idata = nresult.get(constants.NV_INSTANCELIST, None)
test = not isinstance(idata, list)
- self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
- " (instancelist): %s", utils.SafeEncode(str(idata)))
+ self._ErrorIf(test, constants.CV_ENODEHV, ninfo.name,
+ "rpc call to node failed (instancelist): %s",
+ utils.SafeEncode(str(idata)))
if test:
nimg.hyp_fail = True
else:
# try to read free memory (from the hypervisor)
hv_info = nresult.get(constants.NV_HVINFO, None)
test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
- _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
+ _ErrorIf(test, constants.CV_ENODEHV, node,
+ "rpc call to node failed (hvinfo)")
if not test:
try:
nimg.mfree = int(hv_info["memory_free"])
except (ValueError, TypeError):
- _ErrorIf(True, self.ENODERPC, node,
+ _ErrorIf(True, constants.CV_ENODERPC, node,
"node returned invalid nodeinfo, check hypervisor")
# FIXME: devise a free space model for file based instances as well
if vg_name is not None:
test = (constants.NV_VGLIST not in nresult or
vg_name not in nresult[constants.NV_VGLIST])
- _ErrorIf(test, self.ENODELVM, node,
+ _ErrorIf(test, constants.CV_ENODELVM, node,
"node didn't return data for the volume group '%s'"
" - it is either missing or broken", vg_name)
if not test:
try:
nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
except (ValueError, TypeError):
- _ErrorIf(True, self.ENODERPC, node,
+ _ErrorIf(True, constants.CV_ENODERPC, node,
"node returned invalid LVM info, check LVM status")
def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
data = len(disks) * [(False, "node offline")]
else:
msg = nres.fail_msg
- _ErrorIf(msg, self.ENODERPC, nname,
+ _ErrorIf(msg, constants.CV_ENODERPC, nname,
"while getting disk information: %s", msg)
if msg:
# No data from this node
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 = {}
feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
+ user_scripts = []
+ if self.cfg.GetUseExternalMipScript():
+ user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
+
node_verify_param = {
constants.NV_FILELIST:
utils.UniqueSequence(filename
constants.NV_MASTERIP: (master_node, master_ip),
constants.NV_OSLIST: None,
constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
+ constants.NV_USERSCRIPTS: user_scripts,
}
if vg_name is not None:
feedback_fn("* Verifying node %s (%s)" % (node, ntype))
msg = all_nvinfo[node].fail_msg
- _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
+ _ErrorIf(msg, constants.CV_ENODERPC, node, "while contacting node: %s",
+ msg)
if msg:
nimg.rpc_fail = True
continue
nimg.call_ok = self._VerifyNode(node_i, nresult)
self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
self._VerifyNodeNetwork(node_i, nresult)
+ self._VerifyNodeUserScripts(node_i, nresult)
self._VerifyOob(node_i, nresult)
if nimg.vm_capable:
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, self.EINSTANCEWRONGNODE, inst,
+ _ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
"instance should not run on node %s", node_i.name)
- _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
+ _ErrorIf(not test, constants.CV_ENODEORPHANINSTANCE, node_i.name,
"node is running unknown instance %s", inst)
for node, result in extra_lv_nvinfo.items():
pnode = inst_config.primary_node
pnode_img = node_image[pnode]
_ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
- self.ENODERPC, pnode, "instance %s, connection to"
+ constants.CV_ENODERPC, pnode, "instance %s, connection to"
" primary node failed", instance)
- _ErrorIf(inst_config.admin_up and pnode_img.offline,
- self.EINSTANCEBADNODE, instance,
+ _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)
if not inst_config.secondary_nodes:
i_non_redundant.append(instance)
- _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
+ _ErrorIf(len(inst_config.secondary_nodes) > 1,
+ constants.CV_EINSTANCELAYOUT,
instance, "instance has multiple secondary nodes: %s",
utils.CommaJoin(inst_config.secondary_nodes),
code=self.ETYPE_WARNING)
key=lambda (_, nodes): pnode in nodes,
reverse=True)]
- self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
+ self._ErrorIf(len(instance_groups) > 1,
+ constants.CV_EINSTANCESPLITGROUPS,
instance, "instance has primary and secondary nodes in"
" different groups: %s", utils.CommaJoin(pretty_list),
code=self.ETYPE_WARNING)
for snode in inst_config.secondary_nodes:
s_img = node_image[snode]
- _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
- "instance %s, connection to secondary node failed", instance)
+ _ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC,
+ snode, "instance %s, connection to secondary node failed",
+ instance)
if s_img.offline:
inst_nodes_offline.append(snode)
# warn that the instance lives on offline nodes
- _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
+ _ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance,
"instance has offline secondary node(s) %s",
utils.CommaJoin(inst_nodes_offline))
# ... or ghost/non-vm_capable nodes
for node in inst_config.all_nodes:
- _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
- "instance lives on ghost node %s", node)
- _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
+ _ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE,
+ instance, "instance lives on ghost node %s", node)
+ _ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE,
instance, "instance lives on non-vm_capable node %s", node)
feedback_fn("* Verifying orphan volumes")
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 = hooks_results[node_name]
msg = res.fail_msg
test = msg and not res.offline
- self._ErrorIf(test, self.ENODEHOOKS, node_name,
+ self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
"Communication failure in hooks execution: %s", msg)
if res.offline or msg:
# No need to investigate payload if node is offline or gave
continue
for script, hkr, output in res.payload:
test = hkr == constants.HKR_FAIL
- self._ErrorIf(test, self.ENODEHOOKS, node_name,
+ self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
"Script %s failed, output:", script)
if test:
output = self._HOOKS_INDENT_RE.sub(" ", output)
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 = {
- locking.LEVEL_NODE: 1,
+ 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]
"""
clustername = self.op.name
- ip = self.ip
+ new_ip = self.ip
# shutdown the master IP
- master = self.cfg.GetMasterNode()
- result = self.rpc.call_node_stop_master(master, False)
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master role")
try:
cluster = self.cfg.GetClusterInfo()
cluster.cluster_name = clustername
- cluster.master_ip = ip
+ cluster.master_ip = new_ip
self.cfg.Update(cluster, feedback_fn)
# update the known hosts file
ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
node_list = self.cfg.GetOnlineNodeList()
try:
- node_list.remove(master)
+ node_list.remove(master_params.name)
except ValueError:
pass
_UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
finally:
- result = self.rpc.call_node_start_master(master, False, False)
+ master_params.ip = new_ip
+ result = self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
msg = result.fail_msg
if msg:
self.LogWarning("Could not re-enable the master role on"
return clustername
+def _ValidateNetmask(cfg, netmask):
+ """Checks if a netmask is valid.
+
+ @type cfg: L{config.ConfigWriter}
+ @param cfg: The cluster configuration
+ @type netmask: int
+ @param netmask: the netmask to be verified
+ @raise errors.OpPrereqError: if the validation fails
+
+ """
+ ip_family = cfg.GetPrimaryIPFamily()
+ try:
+ ipcls = netutils.IPAddress.GetClassFromIpFamily(ip_family)
+ except errors.ProgrammerError:
+ raise errors.OpPrereqError("Invalid primary ip family: %s." %
+ ip_family)
+ if not ipcls.ValidateNetmask(netmask):
+ raise errors.OpPrereqError("CIDR netmask (%s) not valid" %
+ (netmask))
+
+
class LUClusterSetParams(LogicalUnit):
"""Change the parameters of the cluster.
if self.op.remove_uids:
uidpool.CheckUidPool(self.op.remove_uids)
+ 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:
if self.op.reserved_lvs is not None:
self.cluster.reserved_lvs = self.op.reserved_lvs
+ if self.op.use_external_mip_script is not None:
+ self.cluster.use_external_mip_script = self.op.use_external_mip_script
+
def helper_os(aname, mods, desc):
desc += " OS list"
lst = getattr(self.cluster, aname)
helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
if self.op.master_netdev:
- master = self.cfg.GetMasterNode()
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
feedback_fn("Shutting down master ip on the current netdev (%s)" %
self.cluster.master_netdev)
- result = self.rpc.call_node_stop_master(master, False)
+ result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+ master_params, ems)
result.Raise("Could not disable the master ip")
feedback_fn("Changing master_netdev from %s to %s" %
- (self.cluster.master_netdev, self.op.master_netdev))
+ (master_params.netdev, self.op.master_netdev))
self.cluster.master_netdev = self.op.master_netdev
+ if self.op.master_netmask:
+ master_params = self.cfg.GetMasterNetworkParameters()
+ feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
+ result = self.rpc.call_node_change_master_netmask(master_params.name,
+ master_params.netmask,
+ self.op.master_netmask,
+ master_params.ip,
+ master_params.netdev)
+ if result.fail_msg:
+ msg = "Could not change the master IP netmask: %s" % result.fail_msg
+ feedback_fn(msg)
+
+ self.cluster.master_netmask = self.op.master_netmask
+
self.cfg.Update(self.cluster, feedback_fn)
if self.op.master_netdev:
+ master_params = self.cfg.GetMasterNetworkParameters()
feedback_fn("Starting the master ip on the new master netdev (%s)" %
self.op.master_netdev)
- result = self.rpc.call_node_start_master(master, False, False)
+ ems = self.cfg.GetUseExternalMipScript()
+ result = self.rpc.call_node_activate_master_ip(master_params.name,
+ master_params, ems)
if result.fail_msg:
self.LogWarning("Could not re-enable the master ip on"
" the master, please restart manually: %s",
constants.SSH_KNOWN_HOSTS_FILE,
constants.CONFD_HMAC_KEY,
constants.CLUSTER_DOMAIN_SECRET_FILE,
+ constants.SPICE_CERT_FILE,
+ constants.SPICE_CACERT_FILE,
+ constants.RAPI_USERS_FILE,
])
if not redist:
if cluster.modify_etc_hosts:
files_all.add(constants.ETC_HOSTS)
- # Files which must either exist on all nodes or on none
- files_all_opt = set([
+ # Files which are optional, these must:
+ # - be present in one other category as well
+ # - either exist or not exist on all nodes of that category (mc, vm all)
+ files_opt = set([
constants.RAPI_USERS_FILE,
])
# Files which should only be on master candidates
files_mc = set()
+
if not redist:
files_mc.add(constants.CLUSTER_CONF_FILE)
+ # FIXME: this should also be replicated but Ganeti doesn't support files_mc
+ # replication
+ files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
+
# Files which should only be on VM-capable nodes
files_vm = set(filename
for hv_name in cluster.enabled_hypervisors
- for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
+
+ files_opt |= set(filename
+ for hv_name in cluster.enabled_hypervisors
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
- # Filenames must be unique
- assert (len(files_all | files_all_opt | files_mc | files_vm) ==
- sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
+ # Filenames in each category must be unique
+ all_files_set = files_all | files_mc | files_vm
+ assert (len(all_files_set) ==
+ sum(map(len, [files_all, files_mc, files_vm]))), \
"Found file listed in more than one file list"
- return (files_all, files_all_opt, files_mc, files_vm)
+ # Optional files must be present in one other category
+ assert all_files_set.issuperset(files_opt), \
+ "Optional file not in a different required list"
+
+ return (files_all, files_opt, files_mc, files_vm)
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
nodelist.remove(master_info.name)
# Gather file lists
- (files_all, files_all_opt, files_mc, files_vm) = \
+ (files_all, _, files_mc, files_vm) = \
_ComputeAncillaryFiles(cluster, True)
# Never re-distribute configuration file from here
filemap = [
(online_nodes, files_all),
- (online_nodes, files_all_opt),
(vm_nodes, files_vm),
]
_RedistributeAncillaryFiles(self)
+class LUClusterActivateMasterIp(NoHooksLU):
+ """Activate the master IP on the master node.
+
+ """
+ def Exec(self, feedback_fn):
+ """Activate the master IP.
+
+ """
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ 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):
+ """Deactivate the master IP on the master node.
+
+ """
+ def Exec(self, feedback_fn):
+ """Deactivate the master IP.
+
+ """
+ master_params = self.cfg.GetMasterNetworkParameters()
+ ems = self.cfg.GetUseExternalMipScript()
+ 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):
"""Sleep and poll for an instance's disk to sync.
modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+ "Not owning BGL"
+
# Promote nodes to master candidate as needed
_AdjustCandidatePool(self, exceptions=[node.name])
self.context.RemoveNode(node.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:
def ExpandNames(self):
self.nq.ExpandNames(self)
+ def DeclareLocks(self, level):
+ self.nq.DeclareLocks(self, level)
+
def Exec(self, feedback_fn):
return self.nq.OldStyleQuery(self)
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] = \
def CheckArguments(self):
qcls = _GetQueryImplementation(self.op.what)
- self.impl = qcls(self.op.filter, self.op.fields, self.op.use_locking)
+ self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
def ExpandNames(self):
self.impl.ExpandNames(self)
new_node = self.new_node
node = new_node.name
+ assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+ "Not owning BGL"
+
# We adding a new node so we assume it's powered
new_node.powered = True
self.lock_all = self.op.auto_promote and self.might_demote
self.lock_instances = self.op.secondary_ip is not None
+ def _InstanceFilter(self, instance):
+ """Filter for getting affected instances.
+
+ """
+ return (instance.disk_template in constants.DTS_INT_MIRROR and
+ self.op.node_name in instance.all_nodes)
+
def ExpandNames(self):
if self.lock_all:
self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
else:
self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
- if self.lock_instances:
- self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+ # 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]
- def DeclareLocks(self, level):
- # If we have locked all instances, before waiting to lock nodes, release
- # all the ones living on nodes unrelated to the current operation.
- if level == locking.LEVEL_NODE and self.lock_instances:
- self.affected_instances = []
- if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
- instances_keep = []
-
- # Build list of instances to release
- locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
- for instance_name, instance in self.cfg.GetMultiInstanceInfo(locked_i):
- if (instance.disk_template in constants.DTS_INT_MIRROR and
- self.op.node_name in instance.all_nodes):
- instances_keep.append(instance_name)
- self.affected_instances.append(instance)
-
- _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
-
- assert (set(self.owned_locks(locking.LEVEL_INSTANCE)) ==
- set(instances_keep))
+ # 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))
def BuildHooksEnv(self):
"""Build hooks env.
"""
node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
+ if self.lock_instances:
+ affected_instances = \
+ self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
+
+ # Verify instance locks
+ owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
+ wanted_instances = frozenset(affected_instances.keys())
+ if wanted_instances - owned_instances:
+ raise errors.OpPrereqError("Instances affected by changing node %s's"
+ " secondary IP address have changed since"
+ " locks were acquired, wanted '%s', have"
+ " '%s'; retry the operation" %
+ (self.op.node_name,
+ utils.CommaJoin(wanted_instances),
+ utils.CommaJoin(owned_instances)),
+ errors.ECODE_STATE)
+ else:
+ affected_instances = None
+
if (self.op.master_candidate is not None or
self.op.drained is not None or
self.op.offline is not None):
if old_role == self._ROLE_OFFLINE and new_role != old_role:
# Trying to transition out of offline status
- result = self.rpc.call_version([node.name])[node.name]
+ # TODO: Use standard RPC runner, but make sure it works when the node is
+ # still marked offline
+ result = rpc.BootstrapRunner().call_version([node.name])[node.name]
if result.fail_msg:
raise errors.OpPrereqError("Node %s is being de-offlined but fails"
" to report its version: %s" %
raise errors.OpPrereqError("Cannot change the secondary ip on a single"
" homed cluster", errors.ECODE_INVAL)
+ assert not (frozenset(affected_instances) -
+ self.owned_locks(locking.LEVEL_INSTANCE))
+
if node.offline:
- if self.affected_instances:
- raise errors.OpPrereqError("Cannot change secondary ip: offline"
- " node has instances (%s) configured"
- " to use it" % self.affected_instances)
+ if affected_instances:
+ raise errors.OpPrereqError("Cannot change secondary IP address:"
+ " offline node has instances (%s)"
+ " configured to use it" %
+ utils.CommaJoin(affected_instances.keys()))
else:
# 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 self.affected_instances:
- _CheckInstanceDown(self, instance, "cannot change secondary ip")
+ for instance in affected_instances.values():
+ _CheckInstanceState(self, instance, INSTANCE_DOWN,
+ msg="cannot change secondary ip")
_CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
if master.name != node.name:
"ndparams": cluster.ndparams,
"candidate_pool_size": cluster.candidate_pool_size,
"master_netdev": cluster.master_netdev,
+ "master_netmask": cluster.master_netmask,
+ "use_external_mip_script": cluster.use_external_mip_script,
"volume_group_name": cluster.volume_group_name,
"drbd_usermode_helper": cluster.drbd_usermode_helper,
"file_storage_dir": cluster.file_storage_dir,
_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'" %
errors.ECODE_NORES)
+def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
+ """Checks if nodes have enough physical CPUs
+
+ This function checks if all given nodes have the needed number of
+ physical CPUs. In case any node has less CPUs or we cannot get the
+ information from the node, this function raises an OpPrereqError
+ exception.
+
+ @type lu: C{LogicalUnit}
+ @param lu: a logical unit from which we get configuration data
+ @type nodenames: C{list}
+ @param nodenames: the list of node names to check
+ @type requested: C{int}
+ @param requested: the minimum acceptable number of physical CPUs
+ @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
+ or we cannot check the node
+
+ """
+ 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)
+ (_, _, (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'" %
+ (node, num_cpus), errors.ECODE_ENVIRON)
+ if requested > num_cpus:
+ raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
+ "required" % (node, num_cpus, requested),
+ errors.ECODE_NORES)
+
+
class LUInstanceStartup(LogicalUnit):
"""Starts an instance.
# 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.
_StartInstanceDisks(self, instance, force)
- result = self.rpc.call_instance_start(node_current, instance,
- self.op.hvparams, self.op.beparams,
- self.op.startup_paused)
+ result = \
+ self.rpc.call_instance_start(node_current,
+ (instance, self.op.hvparams,
+ self.op.beparams),
+ self.op.startup_paused)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, 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
self.LogInfo("Instance %s was already stopped, starting now",
instance.name)
_StartInstanceDisks(self, instance, ignore_secondaries)
- result = self.rpc.call_instance_start(node_current, instance,
- None, None, False)
+ result = self.rpc.call_instance_start(node_current,
+ (instance, None, None), False)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, instance)
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
try:
feedback_fn("Running the instance OS create scripts...")
# FIXME: pass debug option from opcode to backend
- result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
- self.op.debug_level,
- osparams=self.os_inst)
+ result = self.rpc.call_instance_os_add(inst.primary_node,
+ (inst, self.os_inst), True,
+ self.op.debug_level)
result.Raise("Could not install OS for instance %s on node %s" %
(inst.name, inst.primary_node))
finally:
# 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
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)
_ShutdownInstanceDisks(self, instance)
raise errors.OpExecError("Can't activate the instance's disks")
- result = self.rpc.call_instance_start(target_node, instance,
- None, None, False)
+ result = self.rpc.call_instance_start(target_node,
+ (instance, None, None), False)
msg = result.fail_msg
if msg:
_ShutdownInstanceDisks(self, instance)
@ivar shutdown_timeout: In case of failover timeout of the shutdown
"""
+
+ # Constants
+ _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,
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"
"""
instance = self.instance
target_node = self.target_node
+ source_node = self.source_node
migration_info = self.migration_info
- abort_result = self.rpc.call_finalize_migration(target_node,
- instance,
- migration_info,
- False)
+ abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
+ instance,
+ migration_info,
+ False)
abort_msg = abort_result.fail_msg
if abort_msg:
logging.error("Aborting migration failed on target node %s: %s",
# Don't raise an exception here, as we stil have to try to revert the
# disk status, even if this step failed.
+ abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
+ instance, False, self.live)
+ abort_msg = abort_result.fail_msg
+ if abort_msg:
+ logging.error("Aborting migration failed on source node %s: %s",
+ source_node, abort_msg)
+
def _ExecMigration(self):
"""Migrate an instance.
target_node = self.target_node
source_node = self.source_node
+ # Check for hypervisor version mismatch and warn the user.
+ nodeinfo = self.rpc.call_node_info([source_node, target_node],
+ None, [self.instance.hypervisor])
+ 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" %
+ (src_version, dst_version))
+
self.feedback_fn("* checking disk consistency between source and target")
for dev in instance.disks:
if not _CheckDiskConsistency(self.lu, dev, target_node, False):
raise errors.OpExecError("Could not migrate instance %s: %s" %
(instance.name, msg))
+ self.feedback_fn("* starting memory transfer")
+ last_feedback = time.time()
+ while True:
+ result = self.rpc.call_instance_get_migration_status(source_node,
+ instance)
+ msg = result.fail_msg
+ ms = result.payload # MigrationStatus instance
+ if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
+ logging.error("Instance migration failed, trying to revert"
+ " disk status: %s", msg)
+ self.feedback_fn("Migration failed, aborting")
+ self._AbortMigration()
+ self._RevertDiskStatus()
+ raise errors.OpExecError("Could not migrate instance %s: %s" %
+ (instance.name, msg))
+
+ if result.payload.status != constants.HV_MIGRATION_ACTIVE:
+ self.feedback_fn("* memory transfer complete")
+ break
+
+ if (utils.TimeoutExpired(last_feedback,
+ self._MIGRATION_FEEDBACK_INTERVAL) and
+ ms.transferred_ram is not None):
+ mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
+ self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
+ last_feedback = time.time()
+
+ time.sleep(self._MIGRATION_POLL_INTERVAL)
+
+ result = self.rpc.call_instance_finalize_migration_src(source_node,
+ instance,
+ True,
+ self.live)
+ msg = result.fail_msg
+ if msg:
+ logging.error("Instance migration succeeded, but finalization failed"
+ " on the source node: %s", msg)
+ raise errors.OpExecError("Could not finalize instance migration: %s" %
+ msg)
+
instance.primary_node = target_node
+
# distribute new instance config to the other nodes
self.cfg.Update(instance, self.feedback_fn)
- result = self.rpc.call_finalize_migration(target_node,
- instance,
- migration_info,
- True)
+ result = self.rpc.call_instance_finalize_migration_dst(target_node,
+ instance,
+ migration_info,
+ True)
msg = result.fail_msg
if msg:
- logging.error("Instance migration succeeded, but finalization failed:"
- " %s", msg)
+ logging.error("Instance migration succeeded, but finalization failed"
+ " on the target node: %s", msg)
raise errors.OpExecError("Could not finalize instance migration: %s" %
msg)
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",
self.feedback_fn("* starting the instance on the target node %s" %
target_node)
- result = self.rpc.call_instance_start(target_node, instance, None, None,
+ result = self.rpc.call_instance_start(target_node, (instance, None, None),
False)
msg = result.fail_msg
if msg:
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]))
- dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
- logical_id=(vgnames[1], names[1]))
+ 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]),
+ 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:
constants.DT_DISKLESS: {},
constants.DT_PLAIN: _compute(disks, 0),
# 128 MB are added for drbd metadata for each disk
- constants.DT_DRBD8: _compute(disks, 128),
+ constants.DT_DRBD8: _compute(disks, DRBD_META_SIZE),
constants.DT_FILE: {},
constants.DT_SHARED_FILE: {},
}
constants.DT_DISKLESS: None,
constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
# 128 MB are added for drbd metadata for each disk
- constants.DT_DRBD8: sum(d[constants.IDISK_SIZE] + 128 for d in disks),
+ constants.DT_DRBD8:
+ sum(d[constants.IDISK_SIZE] + DRBD_META_SIZE for d in disks),
constants.DT_FILE: None,
constants.DT_SHARED_FILE: 0,
constants.DT_BLOCK: 0,
"""
nodenames = _FilterVmNodes(lu, nodenames)
- hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
- hvname,
- hvparams)
+
+ cluster = lu.cfg.GetClusterInfo()
+ hvfull = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams)
+
+ hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames, hvname, hvfull)
for node in nodenames:
info = hvinfo[node]
if info.offline:
"""
nodenames = _FilterVmNodes(lu, nodenames)
- result = lu.rpc.call_os_validate(required, nodenames, osname,
+ result = lu.rpc.call_os_validate(nodenames, required, osname,
[constants.OS_VALIDATE_PARAMETERS],
osparams)
for node, nres in result.items():
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,
if einfo.has_option(constants.INISECT_INS, "disk_template"):
self.op.disk_template = einfo.get(constants.INISECT_INS,
"disk_template")
+ if self.op.disk_template not in constants.DISK_TEMPLATES:
+ raise errors.OpPrereqError("Disk template specified in configuration"
+ " file is not one of the allowed values:"
+ " %s" % " ".join(constants.DISK_TEMPLATES))
else:
raise errors.OpPrereqError("No disk template specified and the export"
" is missing the disk_template information",
errors.ECODE_INVAL)
if not self.op.disks:
- if einfo.has_option(constants.INISECT_INS, "disk_count"):
- disks = []
- # TODO: import the disk iv_name too
- for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
+ disks = []
+ # TODO: import the disk iv_name too
+ for idx in range(constants.MAX_DISKS):
+ if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
disks.append({constants.IDISK_SIZE: disk_sz})
- self.op.disks = disks
- else:
+ self.op.disks = disks
+ if not disks and self.op.disk_template != constants.DT_DISKLESS:
raise errors.OpPrereqError("No disk info specified and the export"
" is missing the disk information",
errors.ECODE_INVAL)
- if (not self.op.nics and
- einfo.has_option(constants.INISECT_INS, "nic_count")):
+ if not self.op.nics:
nics = []
- for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
- ndict = {}
- for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
- v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
- ndict[name] = v
- nics.append(ndict)
+ for idx in range(constants.MAX_NICS):
+ if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
+ ndict = {}
+ for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
+ v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
+ ndict[name] = v
+ nics.append(ndict)
+ else:
+ break
self.op.nics = nics
if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
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:
raise errors.OpPrereqError("Cluster does not support lvm-based"
" instances", errors.ECODE_STATE)
- if self.op.hypervisor is None:
+ if (self.op.hypervisor is None or
+ self.op.hypervisor == constants.VALUE_AUTO):
self.op.hypervisor = self.cfg.GetHypervisorType()
cluster = self.cfg.GetClusterInfo()
_CheckGlobalHvParams(self.op.hvparams)
# fill and remember the beparams dict
+ default_beparams = cluster.beparams[constants.PP_DEFAULT]
+ 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)
for idx, nic in enumerate(self.op.nics):
nic_mode_req = nic.get(constants.INIC_MODE, None)
nic_mode = nic_mode_req
- if nic_mode is None:
+ if nic_mode is None or nic_mode == constants.VALUE_AUTO:
nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
# in routed mode, for the first nic, the default ip is 'auto'
# Build nic parameters
link = nic.get(constants.INIC_LINK, None)
+ if link == constants.VALUE_AUTO:
+ link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
nicparams = {}
if nic_mode_req:
- nicparams[constants.NIC_MODE] = nic_mode_req
+ nicparams[constants.NIC_MODE] = nic_mode
if link:
nicparams[constants.NIC_LINK] = link
self.disks.append(new_disk)
if self.op.mode == constants.INSTANCE_IMPORT:
-
- # Check that the new instance doesn't have less disks than the export
- instance_disks = len(self.disks)
- export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
- if instance_disks < export_disks:
- raise errors.OpPrereqError("Not enough disks to import."
- " (instance: %d, export: %d)" %
- (instance_disks, export_disks),
- errors.ECODE_INVAL)
-
disk_images = []
- for idx in range(export_disks):
+ for idx in range(len(self.disks)):
option = "disk%d_dump" % idx
if export_info.has_option(constants.INISECT_INS, option):
# FIXME: are the old os-es, disk sizes, etc. useful?
self.src_images = disk_images
old_name = export_info.get(constants.INISECT_INS, "name")
- try:
- exp_nic_count = export_info.getint(constants.INISECT_INS, "nic_count")
- except (TypeError, ValueError), err:
- raise errors.OpPrereqError("Invalid export file, nic_count is not"
- " an integer: %s" % str(err),
- errors.ECODE_STATE)
if self.op.instance_name == old_name:
for idx, nic in enumerate(self.nics):
- if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
+ if nic.mac == constants.VALUE_AUTO:
nic_mac_ini = "nic%d_mac" % idx
nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
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:
feedback_fn("* running the instance OS create scripts...")
# FIXME: pass debug option from opcode to backend
- result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
- self.op.debug_level)
+ os_add_result = \
+ self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
+ self.op.debug_level)
if pause_sync:
feedback_fn("* resuming disk sync")
result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
logging.warn("resume-sync of instance %s for disk %d failed",
instance, idx)
- result.Raise("Could not add os for instance %s"
- " on node %s" % (instance, pnode_name))
+ os_add_result.Raise("Could not add os for instance %s"
+ " on node %s" % (instance, pnode_name))
elif self.op.mode == constants.INSTANCE_IMPORT:
feedback_fn("* running the instance OS import scripts...")
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...")
- result = self.rpc.call_instance_start(pnode_name, iobj,
- None, None, False)
+ result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
+ False)
result.Raise("Could not start instance")
return list(iobj.all_nodes)
REQ_BGL = False
def ExpandNames(self):
+ self.share_locks = _ShareAll()
self._ExpandAndLockInstance()
def CheckPrereq(self):
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=128,
- logical_id=(vg_meta, names[1]))
+ lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
+ 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,
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
# local check
hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
_CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
- self.hv_new = hv_new # the new actual values
+ self.hv_proposed = self.hv_new = hv_new # the new actual values
self.hv_inst = i_hvdict # the new dict (without defaults)
else:
+ self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
+ instance.hvparams)
self.hv_new = self.hv_inst = {}
# beparams processing
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_new = be_new # the new actual values
+ self.be_proposed = self.be_new = be_new # the new actual values
self.be_inst = i_bedict # the new dict (without defaults)
else:
self.be_new = self.be_inst = {}
+ self.be_proposed = cluster.SimpleFillBE(instance.beparams)
be_old = cluster.FillBE(instance)
+ # CPU param validation -- checking every time a paramtere is
+ # changed to cover all cases where either CPU mask or vcpus have
+ # changed
+ if (constants.BE_VCPUS in self.be_proposed and
+ constants.HV_CPU_MASK in self.hv_proposed):
+ cpu_list = \
+ utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
+ # Verify mask is consistent with number of vCPUs. Can skip this
+ # test if only 1 entry in the CPU mask, which means same mask
+ # is applied to all vCPUs.
+ if (len(cpu_list) > 1 and
+ len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
+ raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
+ " CPU mask [%s]" %
+ (self.be_proposed[constants.BE_VCPUS],
+ self.hv_proposed[constants.HV_CPU_MASK]),
+ errors.ECODE_INVAL)
+
+ # Only perform this test if a new CPU mask is given
+ if constants.HV_CPU_MASK in self.hv_new:
+ # Calculate the largest CPU number requested
+ max_requested_cpu = max(map(max, cpu_list))
+ # Check that all of the instance's nodes have enough physical CPUs to
+ # satisfy the requested CPU mask
+ _CheckNodesPhysicalCPUs(self, instance.all_nodes,
+ max_requested_cpu + 1, instance.hypervisor)
+
# osparams processing
if self.op.osparams:
i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
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)
- result = self.rpc.call_instance_start(src_node, instance,
- None, None, False)
+ result = self.rpc.call_instance_start(src_node,
+ (instance, None, None), False)
msg = result.fail_msg
if msg:
feedback_fn("Failed to start instance: %s" % msg)
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
+ mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
- # FIXME: Depends on side-effects of modifying the result of
- # C{cfg.GetAllNodesInfo}
-
- 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
# pylint: disable=R0902
# lots of instance attributes
- def __init__(self, cfg, rpc, mode, **kwargs):
+ def __init__(self, cfg, rpc_runner, mode, **kwargs):
self.cfg = cfg
- self.rpc = rpc
+ self.rpc = rpc_runner
# init buffer variables
self.in_text = self.out_text = self.in_data = self.out_data = None
# init all input fields so that pylint is happy
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,