@classmethod
def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
- (files_all, files_all_opt, files_mc, files_vm)):
+ (files_all, files_opt, files_mc, files_vm)):
"""Verifies file checksums collected from all nodes.
@param errorif: Callback for reporting errors
@param all_nvinfo: RPC results
"""
- node_names = frozenset(node.name for node in nodeinfo if not node.offline)
+ # Define functions determining which nodes to consider for a file
+ files2nodefn = [
+ (files_all, None),
+ (files_mc, lambda node: (node.master_candidate or
+ node.name == master_node)),
+ (files_vm, lambda node: node.vm_capable),
+ ]
- assert master_node in node_names
- assert (len(files_all | files_all_opt | files_mc | files_vm) ==
- sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
- "Found file listed in more than one file list"
+ # Build mapping from filename to list of nodes which should have the file
+ nodefiles = {}
+ for (files, fn) in files2nodefn:
+ if fn is None:
+ filenodes = nodeinfo
+ else:
+ filenodes = filter(fn, nodeinfo)
+ nodefiles.update((filename,
+ frozenset(map(operator.attrgetter("name"), filenodes)))
+ for filename in files)
- # Define functions determining which nodes to consider for a file
- file2nodefn = dict([(filename, fn)
- for (files, fn) in [(files_all, None),
- (files_all_opt, None),
- (files_mc, lambda node: (node.master_candidate or
- node.name == master_node)),
- (files_vm, lambda node: node.vm_capable)]
- for filename in files])
+ assert set(nodefiles) == (files_all | files_mc | files_vm)
- fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
+ fileinfo = dict((filename, {}) for filename in nodefiles)
+ ignore_nodes = set()
for node in nodeinfo:
if node.offline:
+ ignore_nodes.add(node.name)
continue
nresult = all_nvinfo[node.name]
errorif(test, cls.ENODEFILECHECK, node.name,
"Node did not return file checksum data")
if test:
+ ignore_nodes.add(node.name)
continue
+ # Build per-checksum mapping from filename to nodes having it
for (filename, checksum) in node_files.items():
- # Check if the file should be considered for a node
- fn = file2nodefn[filename]
- if fn is None or fn(node):
- fileinfo[filename].setdefault(checksum, set()).add(node.name)
+ assert filename in nodefiles
+ fileinfo[filename].setdefault(checksum, set()).add(node.name)
for (filename, checksums) in fileinfo.items():
assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
# Nodes having the file
with_file = frozenset(node_name
for nodes in fileinfo[filename].values()
- for node_name in nodes)
+ for node_name in nodes) - ignore_nodes
+
+ expected_nodes = nodefiles[filename] - ignore_nodes
# Nodes missing file
- missing_file = node_names - with_file
+ missing_file = expected_nodes - with_file
- if filename in files_all_opt:
+ if filename in files_opt:
# All or no nodes
- errorif(missing_file and missing_file != node_names,
+ errorif(missing_file and missing_file != expected_nodes,
cls.ECLUSTERFILECHECK, None,
"File %s is optional, but it must exist on all or no"
" nodes (not found on %s)",
filename, utils.CommaJoin(utils.NiceSort(missing_file)))
else:
+ # Non-optional files
errorif(missing_file, cls.ECLUSTERFILECHECK, None,
"File %s is missing from node(s) %s", filename,
utils.CommaJoin(utils.NiceSort(missing_file)))
+ # Warn if a node has a file it shouldn't
+ unexpected = with_file - expected_nodes
+ errorif(unexpected,
+ cls.ECLUSTERFILECHECK, None,
+ "File %s should not exist on node(s) %s",
+ filename, utils.CommaJoin(utils.NiceSort(unexpected)))
+
# See if there are multiple versions of the file
test = len(checksums) > 1
if test:
return instdisk
+ @staticmethod
+ def _SshNodeSelector(group_uuid, all_nodes):
+ """Create endless iterators for all potential SSH check hosts.
+
+ """
+ nodes = [node for node in all_nodes
+ if (node.group != group_uuid and
+ not node.offline)]
+ keyfunc = operator.attrgetter("group")
+
+ return map(itertools.cycle,
+ [sorted(map(operator.attrgetter("name"), names))
+ for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
+ keyfunc)])
+
+ @classmethod
+ def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
+ """Choose which nodes should talk to which other nodes.
+
+ We will make nodes contact all nodes in their group, and one node from
+ every other group.
+
+ @warning: This algorithm has a known issue if one node group is much
+ smaller than others (e.g. just one node). In such a case all other
+ nodes will talk to the single node.
+
+ """
+ online_nodes = sorted(node.name for node in group_nodes if not node.offline)
+ sel = cls._SshNodeSelector(group_uuid, all_nodes)
+
+ return (online_nodes,
+ dict((name, sorted([i.next() for i in sel]))
+ for name in online_nodes))
+
def BuildHooksEnv(self):
"""Build hooks env.
feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
- # We will make nodes contact all nodes in their group, and one node from
- # every other group.
- # TODO: should it be a *random* node, different every time?
- online_nodes = [node.name for node in node_data_list if not node.offline]
- other_group_nodes = {}
-
- for name in sorted(self.all_node_info):
- node = self.all_node_info[name]
- if (node.group not in other_group_nodes
- and node.group != self.group_uuid
- and not node.offline):
- other_group_nodes[node.group] = node.name
-
node_verify_param = {
constants.NV_FILELIST:
utils.UniqueSequence(filename
for files in filemap
for filename in files),
- constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
+ constants.NV_NODELIST:
+ self._SelectSshCheckNodes(node_data_list, self.group_uuid,
+ self.all_node_info.values()),
constants.NV_HYPERVISOR: hypervisors,
constants.NV_HVPARAMS:
_GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
self._ErrorIf(test, self.ENODEHOOKS, node_name,
"Communication failure in hooks execution: %s", msg)
if res.offline or msg:
- # No need to investigate payload if node is offline or gave an error.
- # override manually lu_result here as _ErrorIf only
- # overrides self.bad
- lu_result = 1
+ # No need to investigate payload if node is offline or gave
+ # an error.
continue
for script, hkr, output in res.payload:
test = hkr == constants.HKR_FAIL
if test:
output = self._HOOKS_INDENT_RE.sub(" ", output)
feedback_fn("%s" % output)
- lu_result = 0
+ lu_result = False
return lu_result
constants.SSH_KNOWN_HOSTS_FILE,
constants.CONFD_HMAC_KEY,
constants.CLUSTER_DOMAIN_SECRET_FILE,
+ constants.RAPI_USERS_FILE,
])
if not redist:
files_all.update(constants.ALL_CERT_FILES)
files_all.update(ssconf.SimpleStore().GetFileList())
+ else:
+ # we need to ship at least the RAPI certificate
+ files_all.add(constants.RAPI_CERT_FILE)
if cluster.modify_etc_hosts:
files_all.add(constants.ETC_HOSTS)
- # Files which must either exist on all nodes or on none
- files_all_opt = set([
+ # Files which are optional, these must:
+ # - be present in one other category as well
+ # - either exist or not exist on all nodes of that category (mc, vm all)
+ files_opt = set([
constants.RAPI_USERS_FILE,
])
# Files which should only be on VM-capable nodes
files_vm = set(filename
for hv_name in cluster.enabled_hypervisors
- for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
+ for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
+
+ 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 = self.cfg.GetMasterNode()
+ self.rpc.call_node_activate_master_ip(master)
+
+
+class LUClusterDeactivateMasterIp(NoHooksLU):
+ """Deactivate the master IP on the master node.
+
+ """
+ def Exec(self, feedback_fn):
+ """Deactivate the master IP.
+
+ """
+ master = self.cfg.GetMasterNode()
+ self.rpc.call_node_deactivate_master_ip(master)
+
+
def _WaitForSync(lu, instance, disks=None, oneshot=False):
"""Sleep and poll for an instance's disk to sync.
node_verify_list = [self.cfg.GetMasterNode()]
node_verify_param = {
- constants.NV_NODELIST: [node],
+ constants.NV_NODELIST: ([node], {}),
# TODO: do a node-net-test as well?
}
projects / ganeti-local / blobdiff