from ganeti import rpc
from ganeti import objects
from ganeti import serializer
-from ganeti import ssconf
_config_lock = locking.SharedLock()
-def ValidateConfig():
- sstore = ssconf.SimpleStore()
+def _ValidateConfig(data):
+ """Verifies that a configuration objects looks valid.
- if sstore.GetConfigVersion() != constants.CONFIG_VERSION:
+ This only verifies the version of the configuration.
+
+ @raise errors.ConfigurationError: if the version differs from what
+ we expect
+
+ """
+ if data.version != constants.CONFIG_VERSION:
raise errors.ConfigurationError("Cluster configuration version"
" mismatch, got %s instead of %s" %
- (sstore.GetConfigVersion(),
+ (data.version,
constants.CONFIG_VERSION))
self.write_count = 0
self._lock = _config_lock
self._config_data = None
- self._config_time = None
- self._config_size = None
- self._config_inode = None
self._offline = offline
if cfg_file is None:
self._cfg_file = constants.CLUSTER_CONF_FILE
# better to raise an error before starting to modify the config
# file than after it was modified
self._my_hostname = utils.HostInfo().name
+ self._last_cluster_serial = -1
+ self._OpenConfig()
# this method needs to be static, so that we can call it on the class
@staticmethod
This should check the current instances for duplicates.
"""
- self._OpenConfig()
prefix = self._config_data.cluster.mac_prefix
all_macs = self._AllMACs()
retries = 64
check for potential collisions elsewhere.
"""
- self._OpenConfig()
all_macs = self._AllMACs()
return mac in all_macs
This checks the current disks for duplicates.
"""
- self._OpenConfig()
all_secrets = self._AllDRBDSecrets()
retries = 64
while retries > 0:
"""Compute the list of all LVs.
"""
- self._OpenConfig()
lvnames = set()
for instance in self._config_data.instances.values():
node_data = instance.MapLVsByNode()
This checks the current node, instances and disk names for
duplicates.
- Args:
- - exceptions: a list with some other names which should be checked
- for uniqueness (used for example when you want to get
- more than one id at one time without adding each one in
- turn to the config file
+ @param exceptions: a list with some other names which should be checked
+ for uniqueness (used for example when you want to get
+ more than one id at one time without adding each one in
+ turn to the config file)
- Returns: the unique id as a string
+ @rtype: string
+ @return: the unique id
"""
existing = set()
def _AllMACs(self):
"""Return all MACs present in the config.
- """
- self._OpenConfig()
+ @rtype: list
+ @return: the list of all MACs
+ """
result = []
for instance in self._config_data.instances.values():
for nic in instance.nics:
def _AllDRBDSecrets(self):
"""Return all DRBD secrets present in the config.
+ @rtype: list
+ @return: the list of all DRBD secrets
+
"""
def helper(disk, result):
"""Recursively gather secrets from this disk."""
@locking.ssynchronized(_config_lock, shared=1)
def VerifyConfig(self):
- """Stub verify function.
- """
- self._OpenConfig()
+ """Verify function.
+ """
result = []
seen_macs = []
ports = {}
ports[net_port].append((instance.name, "network port"))
# cluster-wide pool of free ports
- for free_port in self._config_data.cluster.tcpudp_port_pool:
+ for free_port in data.cluster.tcpudp_port_pool:
if free_port not in ports:
ports[free_port] = []
ports[free_port].append(("cluster", "port marked as free"))
# highest used tcp port check
if keys:
- if keys[-1] > self._config_data.cluster.highest_used_port:
+ if keys[-1] > data.cluster.highest_used_port:
result.append("Highest used port mismatch, saved %s, computed %s" %
- (self._config_data.cluster.highest_used_port,
- keys[-1]))
+ (data.cluster.highest_used_port, keys[-1]))
+
+ if not data.nodes[data.cluster.master_node].master_candidate:
+ result.append("Master node is not a master candidate")
+
+ mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+ if mc_now < mc_max:
+ result.append("Not enough master candidates: actual %d, target %d" %
+ (mc_now, mc_max))
return result
if not isinstance(port, int):
raise errors.ProgrammerError("Invalid type passed for port")
- self._OpenConfig()
self._config_data.cluster.tcpudp_port_pool.add(port)
self._WriteConfig()
"""Returns a copy of the current port list.
"""
- self._OpenConfig()
return self._config_data.cluster.tcpudp_port_pool.copy()
@locking.ssynchronized(_config_lock)
highest_used_port).
"""
- self._OpenConfig()
-
# If there are TCP/IP ports configured, we use them first.
if self._config_data.cluster.tcpudp_port_pool:
port = self._config_data.cluster.tcpudp_port_pool.pop()
def _ComputeDRBDMap(self, instance):
"""Compute the used DRBD minor/nodes.
- Return: dictionary of node_name: dict of minor: instance_name. The
- returned dict will have all the nodes in it (even if with an empty
- list).
+ @return: dictionary of node_name: dict of minor: instance_name;
+ the returned dict will have all the nodes in it (even if with
+ an empty list).
"""
def _AppendUsedPorts(instance_name, disk, used):
order as the passed nodes.
"""
- self._OpenConfig()
-
d_map = self._ComputeDRBDMap(instance)
result = []
for nname in nodes:
@return: Cluster name
"""
- self._OpenConfig()
return self._config_data.cluster.cluster_name
@locking.ssynchronized(_config_lock, shared=1)
@return: Master hostname
"""
- self._OpenConfig()
return self._config_data.cluster.master_node
@locking.ssynchronized(_config_lock, shared=1)
@return: Master IP
"""
- self._OpenConfig()
return self._config_data.cluster.master_ip
@locking.ssynchronized(_config_lock, shared=1)
"""Get the master network device for this cluster.
"""
- self._OpenConfig()
return self._config_data.cluster.master_netdev
@locking.ssynchronized(_config_lock, shared=1)
"""Get the file storage dir for this cluster.
"""
- self._OpenConfig()
return self._config_data.cluster.file_storage_dir
@locking.ssynchronized(_config_lock, shared=1)
"""Get the hypervisor type for this cluster.
"""
- self._OpenConfig()
- return self._config_data.cluster.hypervisor
+ return self._config_data.cluster.default_hypervisor
@locking.ssynchronized(_config_lock, shared=1)
def GetHostKey(self):
"""Return the rsa hostkey from the config.
- Args: None
+ @rtype: string
+ @return: the rsa hostkey
- Returns: rsa hostkey
"""
- self._OpenConfig()
return self._config_data.cluster.rsahostkeypub
@locking.ssynchronized(_config_lock)
This should be used after creating a new instance.
- Args:
- instance: the instance object
+ @type instance: L{objects.Instance}
+ @param instance: the instance object
+
"""
if not isinstance(instance, objects.Instance):
raise errors.ProgrammerError("Invalid type passed to AddInstance")
all_lvs = instance.MapLVsByNode()
logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
- self._OpenConfig()
instance.serial_no = 1
self._config_data.instances[instance.name] = instance
- self._config_data.cluster.serial_no += 1
self._WriteConfig()
def _SetInstanceStatus(self, instance_name, status):
raise errors.ProgrammerError("Invalid status '%s' passed to"
" ConfigWriter._SetInstanceStatus()" %
status)
- self._OpenConfig()
if instance_name not in self._config_data.instances:
raise errors.ConfigurationError("Unknown instance '%s'" %
"""Remove the instance from the configuration.
"""
- self._OpenConfig()
-
if instance_name not in self._config_data.instances:
raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
del self._config_data.instances[instance_name]
- self._config_data.cluster.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock)
rename.
"""
- self._OpenConfig()
if old_name not in self._config_data.instances:
raise errors.ConfigurationError("Unknown instance '%s'" % old_name)
inst = self._config_data.instances[old_name]
disk.iv_name))
self._config_data.instances[inst.name] = inst
- self._config_data.cluster.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock)
This function is for internal use, when the config lock is already held.
"""
- self._OpenConfig()
return self._config_data.instances.keys()
@locking.ssynchronized(_config_lock, shared=1)
def GetInstanceList(self):
"""Get the list of instances.
- Returns:
- array of instances, ex. ['instance2.example.com','instance1.example.com']
- these contains all the instances, also the ones in Admin_down state
+ @return: array of instances, ex. ['instance2.example.com',
+ 'instance1.example.com']
"""
return self._UnlockedGetInstanceList()
"""Attempt to expand an incomplete instance name.
"""
- self._OpenConfig()
-
return utils.MatchNameComponent(short_name,
self._config_data.instances.keys())
This function is for internal use, when the config lock is already held.
"""
- self._OpenConfig()
-
if instance_name not in self._config_data.instances:
return None
It takes the information from the configuration file. Other informations of
an instance are taken from the live systems.
- Args:
- instance: name of the instance, ex instance1.example.com
+ @param instance_name: name of the instance, e.g.
+ I{instance1.example.com}
- Returns:
- the instance object
+ @rtype: L{objects.Instance}
+ @return: the instance object
"""
return self._UnlockedGetInstanceInfo(instance_name)
def AddNode(self, node):
"""Add a node to the configuration.
- Args:
- node: an object.Node instance
+ @type node: L{objects.Node}
+ @param node: a Node instance
"""
logging.info("Adding node %s to configuration" % node.name)
- self._OpenConfig()
node.serial_no = 1
self._config_data.nodes[node.name] = node
self._config_data.cluster.serial_no += 1
"""
logging.info("Removing node %s from configuration" % node_name)
- self._OpenConfig()
if node_name not in self._config_data.nodes:
raise errors.ConfigurationError("Unknown node '%s'" % node_name)
"""Attempt to expand an incomplete instance name.
"""
- self._OpenConfig()
-
return utils.MatchNameComponent(short_name,
self._config_data.nodes.keys())
def _UnlockedGetNodeInfo(self, node_name):
"""Get the configuration of a node, as stored in the config.
- This function is for internal use, when the config lock is already held.
+ This function is for internal use, when the config lock is already
+ held.
- Args: node: nodename (tuple) of the node
+ @param node_name: the node name, e.g. I{node1.example.com}
- Returns: the node object
+ @rtype: L{objects.Node}
+ @return: the node object
"""
- self._OpenConfig()
-
if node_name not in self._config_data.nodes:
return None
def GetNodeInfo(self, node_name):
"""Get the configuration of a node, as stored in the config.
- Args: node: nodename (tuple) of the node
+ This is just a locked wrapper over L{_UnlockedGetNodeInfo}.
- Returns: the node object
+ @param node_name: the node name, e.g. I{node1.example.com}
+
+ @rtype: L{objects.Node}
+ @return: the node object
"""
return self._UnlockedGetNodeInfo(node_name)
def _UnlockedGetNodeList(self):
"""Return the list of nodes which are in the configuration.
- This function is for internal use, when the config lock is already held.
+ This function is for internal use, when the config lock is already
+ held.
+
+ @rtype: list
"""
- self._OpenConfig()
return self._config_data.nodes.keys()
"""Get the configuration of all nodes.
@rtype: dict
- @returns: dict of (node, node_info), where node_info is what
+ @return: dict of (node, node_info), where node_info is what
would GetNodeInfo return for the node
"""
for node in self._UnlockedGetNodeList()])
return my_dict
+ def _UnlockedGetMasterCandidateStats(self):
+ """Get the number of current and maximum desired and possible candidates.
+
+ @rtype: tuple
+ @return: tuple of (current, desired and possible)
+
+ """
+ mc_now = mc_max = 0
+ for node in self._config_data.nodes.itervalues():
+ if not node.offline:
+ mc_max += 1
+ if node.master_candidate:
+ mc_now += 1
+ mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size)
+ return (mc_now, mc_max)
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterCandidateStats(self):
+ """Get the number of current and maximum possible candidates.
+
+ This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
+
+ @rtype: tuple
+ @return: tuple of (current, max)
+
+ """
+ return self._UnlockedGetMasterCandidateStats()
+
+ @locking.ssynchronized(_config_lock)
+ def MaintainCandidatePool(self):
+ """Try to grow the candidate pool to the desired size.
+
+ @rtype: list
+ @return: list with the adjusted nodes (L{objects.Node} instances)
+
+ """
+ mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+ mod_list = []
+ if mc_now < mc_max:
+ node_list = self._config_data.nodes.keys()
+ random.shuffle(node_list)
+ for name in node_list:
+ if mc_now >= mc_max:
+ break
+ node = self._config_data.nodes[name]
+ if node.master_candidate or node.offline:
+ continue
+ mod_list.append(node)
+ node.master_candidate = True
+ node.serial_no += 1
+ mc_now += 1
+ if mc_now != mc_max:
+ # this should not happen
+ logging.warning("Warning: MaintainCandidatePool didn't manage to"
+ " fill the candidate pool (%d/%d)", mc_now, mc_max)
+ if mod_list:
+ self._config_data.cluster.serial_no += 1
+ self._WriteConfig()
+
+ return mod_list
+
def _BumpSerialNo(self):
"""Bump up the serial number of the config.
def _OpenConfig(self):
"""Read the config data from disk.
- In case we already have configuration data and the config file has
- the same mtime as when we read it, we skip the parsing of the
- file, since de-serialisation could be slow.
-
"""
- try:
- st = os.stat(self._cfg_file)
- except OSError, err:
- raise errors.ConfigurationError("Can't stat config file: %s" % err)
- if (self._config_data is not None and
- self._config_time is not None and
- self._config_time == st.st_mtime and
- self._config_size == st.st_size and
- self._config_inode == st.st_ino):
- # data is current, so skip loading of config file
- return
-
- # Make sure the configuration has the right version
- ValidateConfig()
-
f = open(self._cfg_file, 'r')
try:
try:
raise errors.ConfigurationError(err)
finally:
f.close()
+
+ # Make sure the configuration has the right version
+ _ValidateConfig(data)
+
if (not hasattr(data, 'cluster') or
not hasattr(data.cluster, 'rsahostkeypub')):
raise errors.ConfigurationError("Incomplete configuration"
" (missing cluster.rsahostkeypub)")
self._config_data = data
- self._config_time = st.st_mtime
- self._config_size = st.st_size
- self._config_inode = st.st_ino
+ # reset the last serial as -1 so that the next write will cause
+ # ssconf update
+ self._last_cluster_serial = -1
def _DistributeConfig(self):
"""Distribute the configuration to the other nodes.
if self._offline:
return True
bad = False
- nodelist = self._UnlockedGetNodeList()
- myhostname = self._my_hostname
- try:
- nodelist.remove(myhostname)
- except ValueError:
- pass
+ node_list = []
+ addr_list = []
+ myhostname = self._my_hostname
+ # we can skip checking whether _UnlockedGetNodeInfo returns None
+ # since the node list comes from _UnlocketGetNodeList, and we are
+ # called with the lock held, so no modifications should take place
+ # in between
+ for node_name in self._UnlockedGetNodeList():
+ if node_name == myhostname:
+ continue
+ node_info = self._UnlockedGetNodeInfo(node_name)
+ if not node_info.master_candidate:
+ continue
+ node_list.append(node_info.name)
+ addr_list.append(node_info.primary_ip)
- result = rpc.call_upload_file(nodelist, self._cfg_file)
- for node in nodelist:
+ result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file,
+ address_list=addr_list)
+ for node in node_list:
if not result[node]:
logging.error("copy of file %s to node %s failed",
self._cfg_file, node)
# we don't need to do os.close(fd) as f.close() did it
os.rename(name, destination)
self.write_count += 1
- # re-set our cache as not to re-read the config file
- try:
- st = os.stat(destination)
- except OSError, err:
- raise errors.ConfigurationError("Can't stat config file: %s" % err)
- self._config_time = st.st_mtime
- self._config_size = st.st_size
- self._config_inode = st.st_ino
- # and redistribute the config file
+
+ # and redistribute the config file to master candidates
self._DistributeConfig()
+ # Write ssconf files on all nodes (including locally)
+ if self._last_cluster_serial < self._config_data.cluster.serial_no:
+ if not self._offline:
+ rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(),
+ self._UnlockedGetSsconfValues())
+ self._last_cluster_serial = self._config_data.cluster.serial_no
+
+ def _UnlockedGetSsconfValues(self):
+ """Return the values needed by ssconf.
+
+ @rtype: dict
+ @return: a dictionary with keys the ssconf names and values their
+ associated value
+
+ """
+ fn = "\n".join
+ node_names = utils.NiceSort(self._UnlockedGetNodeList())
+ node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]
+
+ off_data = fn(node.name for node in node_info if node.offline)
+ mc_data = fn(node.name for node in node_info if node.master_candidate)
+ node_data = fn(node_names)
+
+ cluster = self._config_data.cluster
+ return {
+ constants.SS_CLUSTER_NAME: cluster.cluster_name,
+ constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
+ constants.SS_MASTER_CANDIDATES: mc_data,
+ constants.SS_MASTER_IP: cluster.master_ip,
+ constants.SS_MASTER_NETDEV: cluster.master_netdev,
+ constants.SS_MASTER_NODE: cluster.master_node,
+ constants.SS_NODE_LIST: node_data,
+ constants.SS_OFFLINE_NODES: off_data,
+ }
+
@locking.ssynchronized(_config_lock)
def InitConfig(self, version, cluster_config, master_node_config):
"""Create the initial cluster configuration.
"""Return the volume group name.
"""
- self._OpenConfig()
return self._config_data.cluster.volume_group_name
@locking.ssynchronized(_config_lock)
"""Set the volume group name.
"""
- self._OpenConfig()
self._config_data.cluster.volume_group_name = vg_name
self._config_data.cluster.serial_no += 1
self._WriteConfig()
"""Return the default bridge.
"""
- self._OpenConfig()
return self._config_data.cluster.default_bridge
@locking.ssynchronized(_config_lock, shared=1)
"""Return the mac prefix.
"""
- self._OpenConfig()
return self._config_data.cluster.mac_prefix
@locking.ssynchronized(_config_lock, shared=1)
def GetClusterInfo(self):
"""Returns informations about the cluster
- Returns:
- the cluster object
+ @rtype: L{objects.Cluster}
+ @return: the cluster object
"""
- self._OpenConfig()
-
return self._config_data.cluster
@locking.ssynchronized(_config_lock)
that all modified objects will be saved, but the target argument
is the one the caller wants to ensure that it's saved.
+ @param target: an instance of either L{objects.Cluster},
+ L{objects.Node} or L{objects.Instance} which is existing in
+ the cluster
+
"""
if self._config_data is None:
raise errors.ProgrammerError("Configuration file not read,"
" cannot save.")
+ update_serial = False
if isinstance(target, objects.Cluster):
test = target == self._config_data.cluster
elif isinstance(target, objects.Node):
test = target in self._config_data.nodes.values()
+ update_serial = True
elif isinstance(target, objects.Instance):
test = target in self._config_data.instances.values()
else:
" has been read or unknown object")
target.serial_no += 1
+ if update_serial:
+ # for node updates, we need to increase the cluster serial too
+ self._config_data.cluster.serial_no += 1
+
self._WriteConfig()