import os
import os.path
-import sha
import time
import tempfile
import re
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
- memory, vcpus, nics, disk_template, disks):
+ memory, vcpus, nics, disk_template, disks,
+ bep, hvp, hypervisor):
"""Builds instance related env variables for hooks
This builds the hook environment from individual variables.
@param disk_template: the distk template of the instance
@type disks: list
@param disks: the list of (size, mode) pairs
+ @type bep: dict
+ @param bep: the backend parameters for the instance
+ @type hvp: dict
+ @param hvp: the hypervisor parameters for the instance
+ @type hypervisor: string
+ @param hypervisor: the hypervisor for the instance
@rtype: dict
@return: the hook environment for this instance
"INSTANCE_MEMORY": memory,
"INSTANCE_VCPUS": vcpus,
"INSTANCE_DISK_TEMPLATE": disk_template,
+ "INSTANCE_HYPERVISOR": hypervisor,
}
if nics:
env["INSTANCE_DISK_COUNT"] = disk_count
+ for source, kind in [(bep, "BE"), (hvp, "HV")]:
+ for key, value in source.items():
+ env["INSTANCE_%s_%s" % (kind, key)] = value
+
return env
@return: the hook environment dictionary
"""
- bep = lu.cfg.GetClusterInfo().FillBE(instance)
+ cluster = lu.cfg.GetClusterInfo()
+ bep = cluster.FillBE(instance)
+ hvp = cluster.FillHV(instance)
args = {
'name': instance.name,
'primary_node': instance.primary_node,
'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
'disk_template': instance.disk_template,
'disks': [(disk.size, disk.mode) for disk in instance.disks],
+ 'bep': bep,
+ 'hvp': hvp,
+ 'hypervisor': instance.hypervisor,
}
if override:
args.update(override)
else:
# not candidate and this is not a must-have file
bad = True
- feedback_fn(" - ERROR: non master-candidate has old/wrong file"
- " '%s'" % file_name)
+ feedback_fn(" - ERROR: file '%s' should not exist on non master"
+ " candidates (and the file is outdated)" % file_name)
else:
# all good, except non-master/non-must have combination
if not node_is_mc and not must_have_file:
node_drbd = {}
for minor, instance in all_drbd_map[node].items():
- instance = instanceinfo[instance]
- node_drbd[minor] = (instance.name, instance.admin_up)
+ if instance not in instanceinfo:
+ feedback_fn(" - ERROR: 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
+ # unallocated minor in use)
+ node_drbd[minor] = (instance, False)
+ else:
+ instance = instanceinfo[instance]
+ node_drbd[minor] = (instance.name, instance.admin_up)
result = self._VerifyNode(node_i, file_names, local_checksums,
nresult, feedback_fn, master_files,
node_drbd, vg_name)
}
# FIXME: devise a free space model for file based instances as well
if vg_name is not None:
+ if (constants.NV_VGLIST not in nresult or
+ vg_name not in nresult[constants.NV_VGLIST]):
+ feedback_fn(" - ERROR: node %s didn't return data for the"
+ " volume group '%s' - it is either missing or broken" %
+ (node, vg_name))
+ bad = True
+ continue
node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
- except ValueError:
- feedback_fn(" - ERROR: invalid value returned from node %s" % (node,))
+ except (ValueError, KeyError):
+ feedback_fn(" - ERROR: invalid nodeinfo value returned"
+ " from node %s" % (node,))
bad = True
continue
if isinstance(lvs, basestring):
logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
res_nlvm[node] = lvs
+ continue
elif not isinstance(lvs, dict):
logging.warning("Connection to node %s failed or invalid data"
" returned", node)
_OP_REQP = []
REQ_BGL = False
- def CheckParameters(self):
+ def CheckArguments(self):
"""Check parameters
"""
if self.op.candidate_pool_size is not None:
try:
self.op.candidate_pool_size = int(self.op.candidate_pool_size)
- except ValueError, err:
+ except (ValueError, TypeError), err:
raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
str(err))
if self.op.candidate_pool_size < 1:
"""
if self.op.vg_name is not None:
- if self.op.vg_name != self.cfg.GetVGName():
- self.cfg.SetVGName(self.op.vg_name)
+ new_volume = self.op.vg_name
+ if not new_volume:
+ new_volume = None
+ if new_volume != self.cfg.GetVGName():
+ self.cfg.SetVGName(new_volume)
else:
feedback_fn("Cluster LVM configuration already in desired"
" state, not changing")
self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
if self.op.candidate_pool_size is not None:
self.cluster.candidate_pool_size = self.op.candidate_pool_size
+ # we need to update the pool size here, otherwise the save will fail
+ _AdjustCandidatePool(self)
self.cfg.Update(self.cluster)
- # we want to update nodes after the cluster so that if any errors
- # happen, we have recorded and saved the cluster info
- if self.op.candidate_pool_size is not None:
- _AdjustCandidatePool(self)
-
class LURedistributeConfig(NoHooksLU):
"""Force the redistribution of cluster configuration.
lu.cfg.SetDiskID(dev, node)
retries = 0
+ degr_retries = 10 # in seconds, as we sleep 1 second each time
while True:
max_time = 0
done = True
rem_time = "no time estimate"
lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
(instance.disks[i].iv_name, perc_done, rem_time))
+
+ # if we're done but degraded, let's do a few small retries, to
+ # make sure we see a stable and not transient situation; therefore
+ # we force restart of the loop
+ if (done or oneshot) and cumul_degraded and degr_retries > 0:
+ logging.info("Degraded disks found, %d retries left", degr_retries)
+ degr_retries -= 1
+ time.sleep(1)
+ continue
+
if done or oneshot:
break
selected=self.op.output_fields)
# Lock all nodes, in shared mode
+ # Temporary removal of locks, should be reverted later
+ # TODO: reintroduce locks when they are lighter-weight
self.needed_locks = {}
- self.share_locks[locking.LEVEL_NODE] = 1
- self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+ #self.share_locks[locking.LEVEL_NODE] = 1
+ #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
def CheckPrereq(self):
"""Check prerequisites.
"""
all_os = {}
+ # we build here the list of nodes that didn't fail the RPC (at RPC
+ # level), so that nodes with a non-responding node daemon don't
+ # make all OSes invalid
+ good_nodes = [node_name for node_name in rlist
+ if not rlist[node_name].failed]
for node_name, nr in rlist.iteritems():
if nr.failed or not nr.data:
continue
# build a list of nodes for this os containing empty lists
# for each node in node_list
all_os[os_obj.name] = {}
- for nname in node_list:
+ for nname in good_nodes:
all_os[os_obj.name][nname] = []
all_os[os_obj.name][node_name].append(os_obj)
return all_os
"""Compute the list of OSes.
"""
- node_list = self.acquired_locks[locking.LEVEL_NODE]
- valid_nodes = [node for node in self.cfg.GetOnlineNodeList()
- if node in node_list]
+ valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
node_data = self.rpc.call_os_diagnose(valid_nodes)
if node_data == False:
raise errors.OpExecError("Can't gather the list of OSes")
"master",
"offline",
"drained",
+ "role",
)
def ExpandNames(self):
val = node.drained
elif self._FIELDS_DYNAMIC.Matches(field):
val = live_data[node.name].get(field, None)
+ elif field == "role":
+ if node.name == master_node:
+ val = "M"
+ elif node.master_candidate:
+ val = "C"
+ elif node.drained:
+ val = "D"
+ elif node.offline:
+ val = "O"
+ else:
+ val = "R"
else:
raise errors.ParameterError(field)
node_output.append(val)
" based ping to noded port")
cp_size = self.cfg.GetClusterInfo().candidate_pool_size
- mc_now, _ = self.cfg.GetMasterCandidateStats()
- master_candidate = mc_now < cp_size
+ if self.op.readd:
+ exceptions = [node]
+ else:
+ exceptions = []
+ mc_now, mc_max = self.cfg.GetMasterCandidateStats(exceptions)
+ # the new node will increase mc_max with one, so:
+ mc_max = min(mc_max + 1, cp_size)
+ self.master_candidate = mc_now < mc_max
- self.new_node = objects.Node(name=node,
- primary_ip=primary_ip,
- secondary_ip=secondary_ip,
- master_candidate=master_candidate,
- offline=False, drained=False)
+ if self.op.readd:
+ self.new_node = self.cfg.GetNodeInfo(node)
+ assert self.new_node is not None, "Can't retrieve locked node %s" % node
+ else:
+ self.new_node = objects.Node(name=node,
+ primary_ip=primary_ip,
+ secondary_ip=secondary_ip,
+ master_candidate=self.master_candidate,
+ offline=False, drained=False)
def Exec(self, feedback_fn):
"""Adds the new node to the cluster.
new_node = self.new_node
node = new_node.name
+ # for re-adds, reset the offline/drained/master-candidate flags;
+ # we need to reset here, otherwise offline would prevent RPC calls
+ # later in the procedure; this also means that if the re-add
+ # fails, we are left with a non-offlined, broken node
+ if self.op.readd:
+ new_node.drained = new_node.offline = False
+ self.LogInfo("Readding a node, the offline/drained flags were reset")
+ # if we demote the node, we do cleanup later in the procedure
+ new_node.master_candidate = self.master_candidate
+
+ # notify the user about any possible mc promotion
+ if new_node.master_candidate:
+ self.LogInfo("Node will be a master candidate")
+
# check connectivity
result = self.rpc.call_version([node])[node]
result.Raise()
if self.op.readd:
self.context.ReaddNode(new_node)
+ # make sure we redistribute the config
+ self.cfg.Update(new_node)
+ # and make sure the new node will not have old files around
+ if not new_node.master_candidate:
+ result = self.rpc.call_node_demote_from_mc(new_node.name)
+ msg = result.RemoteFailMsg()
+ if msg:
+ self.LogWarning("Node failed to demote itself from master"
+ " candidate status: %s" % msg)
else:
self.context.AddNode(new_node)
((node.offline and not self.op.offline == False) or
(node.drained and not self.op.drained == False))):
raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
- " to master_candidate")
+ " to master_candidate" % node.name)
return
node.master_candidate = False
changed_mc = True
result.append(("master_candidate", "auto-demotion due to drain"))
+ rrc = self.rpc.call_node_demote_from_mc(node.name)
+ msg = rrc.RemoteFailMsg()
+ if msg:
+ self.LogWarning("Node failed to demote itself: %s" % msg)
if node.offline:
node.offline = False
result.append(("offline", "clear offline status due to drain"))
for hypervisor in cluster.enabled_hypervisors]),
"beparams": cluster.beparams,
"candidate_pool_size": cluster.candidate_pool_size,
+ "default_bridge": cluster.default_bridge,
+ "master_netdev": cluster.master_netdev,
+ "volume_group_name": cluster.volume_group_name,
+ "file_storage_dir": cluster.file_storage_dir,
}
return result
assert self.instance is not None, \
"Cannot retrieve locked instance %s" % self.op.instance_name
+ # extra beparams
+ self.beparams = getattr(self.op, "beparams", {})
+ if self.beparams:
+ if not isinstance(self.beparams, dict):
+ raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
+ " dict" % (type(self.beparams), ))
+ # fill the beparams dict
+ utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
+ self.op.beparams = self.beparams
+
+ # extra hvparams
+ self.hvparams = getattr(self.op, "hvparams", {})
+ if self.hvparams:
+ if not isinstance(self.hvparams, dict):
+ raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
+ " dict" % (type(self.hvparams), ))
+
+ # check hypervisor parameter syntax (locally)
+ cluster = self.cfg.GetClusterInfo()
+ utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
+ filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
+ instance.hvparams)
+ filled_hvp.update(self.hvparams)
+ hv_type = hypervisor.GetHypervisor(instance.hypervisor)
+ hv_type.CheckParameterSyntax(filled_hvp)
+ _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
+ self.op.hvparams = self.hvparams
+
_CheckNodeOnline(self, instance.primary_node)
bep = self.cfg.GetClusterInfo().FillBE(instance)
# check bridges existance
_CheckInstanceBridgesExist(self, instance)
- _CheckNodeFreeMemory(self, instance.primary_node,
- "starting instance %s" % instance.name,
- bep[constants.BE_MEMORY], instance.hypervisor)
+ remote_info = self.rpc.call_instance_info(instance.primary_node,
+ instance.name,
+ instance.hypervisor)
+ remote_info.Raise()
+ if not remote_info.data:
+ _CheckNodeFreeMemory(self, instance.primary_node,
+ "starting instance %s" % instance.name,
+ bep[constants.BE_MEMORY], instance.hypervisor)
def Exec(self, feedback_fn):
"""Start the instance.
_StartInstanceDisks(self, instance, force)
- result = self.rpc.call_instance_start(node_current, instance)
+ result = self.rpc.call_instance_start(node_current, instance,
+ self.hvparams, self.beparams)
msg = result.RemoteFailMsg()
if msg:
_ShutdownInstanceDisks(self, instance)
" full reboot: %s" % msg)
_ShutdownInstanceDisks(self, instance)
_StartInstanceDisks(self, instance, ignore_secondaries)
- result = self.rpc.call_instance_start(node_current, instance)
+ result = self.rpc.call_instance_start(node_current, instance, None, None)
msg = result.RemoteFailMsg()
if msg:
_ShutdownInstanceDisks(self, instance)
remote_info = self.rpc.call_instance_info(instance.primary_node,
instance.name,
instance.hypervisor)
- if remote_info.failed or remote_info.data:
+ remote_info.Raise()
+ if remote_info.data:
raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
(self.op.instance_name,
instance.primary_node))
val = live_data[instance.name].get("memory", "?")
else:
val = "-"
+ elif field == "vcpus":
+ val = i_be[constants.BE_VCPUS]
elif field == "disk_template":
val = instance.disk_template
elif field == "ip":
- val = instance.nics[0].ip
+ if instance.nics:
+ val = instance.nics[0].ip
+ else:
+ val = None
elif field == "bridge":
- val = instance.nics[0].bridge
+ if instance.nics:
+ val = instance.nics[0].bridge
+ else:
+ val = None
elif field == "mac":
- val = instance.nics[0].mac
+ if instance.nics:
+ val = instance.nics[0].mac
+ else:
+ val = None
elif field == "sda_size" or field == "sdb_size":
idx = ord(field[2]) - ord('a')
try:
else:
assert False, "Unhandled NIC parameter"
else:
- assert False, "Unhandled variable parameter"
+ assert False, ("Declared but unhandled variable parameter '%s'" %
+ field)
else:
- raise errors.ParameterError(field)
+ assert False, "Declared but unhandled parameter '%s'" % field
iout.append(val)
output.append(iout)
target_node = secondary_nodes[0]
_CheckNodeOnline(self, target_node)
_CheckNodeNotDrained(self, target_node)
- # check memory requirements on the secondary node
- _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
- instance.name, bep[constants.BE_MEMORY],
- instance.hypervisor)
+
+ if instance.admin_up:
+ # check memory requirements on the secondary node
+ _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
+ instance.name, bep[constants.BE_MEMORY],
+ instance.hypervisor)
+ else:
+ self.LogInfo("Not checking memory on the secondary node as"
+ " instance will not be started")
# check bridge existance
brlist = [nic.bridge for nic in instance.nics]
raise errors.OpExecError("Can't activate the instance's disks")
feedback_fn("* starting the instance on the target node")
- result = self.rpc.call_instance_start(target_node, instance)
+ result = self.rpc.call_instance_start(target_node, instance, None, None)
msg = result.RemoteFailMsg()
if msg:
_ShutdownInstanceDisks(self, instance)
continue
msg = info.RemoteFailMsg()
if msg:
- raise errors.OpPrereqError("Hypervisor parameter validation failed:"
- " %s" % msg)
+ raise errors.OpPrereqError("Hypervisor parameter validation"
+ " failed on node %s: %s" % (node, msg))
class LUCreateInstance(LogicalUnit):
self.op.hvparams)
hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
hv_type.CheckParameterSyntax(filled_hvp)
+ self.hv_full = filled_hvp
# fill and remember the beparams dict
utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
disk_template=self.op.disk_template,
disks=[(d["size"], d["mode"]) for d in self.disks],
+ bep=self.be_full,
+ hvp=self.hv_full,
+ hypervisor=self.op.hypervisor,
))
nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
# os verification
result = self.rpc.call_os_get(pnode.name, self.op.os_type)
result.Raise()
- if not isinstance(result.data, objects.OS):
+ if not isinstance(result.data, objects.OS) or not result.data:
raise errors.OpPrereqError("OS '%s' not in supported os list for"
" primary node" % self.op.os_type)
self.cfg.Update(iobj)
logging.info("Starting instance %s on node %s", instance, pnode_name)
feedback_fn("* starting instance...")
- result = self.rpc.call_instance_start(pnode_name, iobj)
+ result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
msg = result.RemoteFailMsg()
if msg:
raise errors.OpExecError("Could not start instance: %s" % msg)
new_net_id)
new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
logical_id=new_alone_id,
- children=dev.children)
+ children=dev.children,
+ size=dev.size)
try:
_CreateSingleBlockDev(self, new_node, instance, new_drbd,
_GetInstanceInfoText(instance), False)
- except errors.BlockDeviceError:
+ except errors.GenericError:
self.cfg.ReleaseDRBDMinors(instance.name)
raise
"sstatus": dev_sstatus,
"children": dev_children,
"mode": dev.mode,
+ "size": dev.size,
}
return data
self.warn.append("Can't get info from primary node %s" % pnode)
else:
if not instance_info.failed and instance_info.data:
- current_mem = instance_info.data['memory']
+ current_mem = int(instance_info.data['memory'])
else:
# Assume instance not running
# (there is a slight race condition here, but it's not very probable,
self.cfg.SetDiskID(disk, src_node)
try:
- for disk in instance.disks:
+ for idx, disk in enumerate(instance.disks):
# new_dev_name will be a snapshot of an lvm leaf of the one we passed
new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
if new_dev_name.failed or not new_dev_name.data:
- self.LogWarning("Could not snapshot block device %s on node %s",
- disk.logical_id[1], src_node)
+ self.LogWarning("Could not snapshot disk/%d on node %s",
+ idx, src_node)
snap_disks.append(False)
else:
new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
finally:
if self.op.shutdown and instance.admin_up:
- result = self.rpc.call_instance_start(src_node, instance)
+ result = self.rpc.call_instance_start(src_node, instance, None, None)
msg = result.RemoteFailMsg()
if msg:
_ShutdownInstanceDisks(self, instance)
result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
instance, cluster_name, idx)
if result.failed or not result.data:
- self.LogWarning("Could not export block device %s from node %s to"
- " node %s", dev.logical_id[1], src_node,
- dst_node.name)
+ self.LogWarning("Could not export disk/%d from node %s to"
+ " node %s", idx, src_node, dst_node.name)
msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
if msg:
- self.LogWarning("Could not remove snapshot block device %s from node"
- " %s: %s", dev.logical_id[1], src_node, msg)
+ self.LogWarning("Could not remove snapshot for disk/%d from node"
+ " %s: %s", idx, src_node, msg)
result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
if result.failed or not result.data:
"disk_template": iinfo.disk_template,
"hypervisor": iinfo.hypervisor,
}
+ pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
+ pir["disks"])
instance_data[iinfo.name] = pir
data["instances"] = instance_data
projects / ganeti-local / blobdiff