import os
import tempfile
import random
+import logging
from ganeti import errors
from ganeti import locking
-from ganeti import logger
from ganeti import utils
from ganeti import constants
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()
-
- if sstore.GetConfigVersion() != constants.CONFIG_VERSION:
+def _ValidateConfig(data):
+ 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))
else:
self._cfg_file = cfg_file
self._temporary_ids = set()
+ self._temporary_drbds = {}
# Note: in order to prevent errors when resolving our name in
# _DistributeConfig, we compute it here once and reuse it; it's
# better to raise an error before starting to modify the config
all_macs = self._AllMACs()
return mac in all_macs
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GenerateDRBDSecret(self):
+ """Generate a DRBD secret.
+
+ This checks the current disks for duplicates.
+
+ """
+ self._OpenConfig()
+ all_secrets = self._AllDRBDSecrets()
+ retries = 64
+ while retries > 0:
+ secret = utils.GenerateSecret()
+ if secret not in all_secrets:
+ break
+ retries -= 1
+ else:
+ raise errors.ConfigurationError("Can't generate unique DRBD secret")
+ return secret
+
def _ComputeAllLVs(self):
"""Compute the list of all LVs.
return result
+ def _AllDRBDSecrets(self):
+ """Return all DRBD secrets present in the config.
+
+ """
+ def helper(disk, result):
+ """Recursively gather secrets from this disk."""
+ if disk.dev_type == constants.DT_DRBD8:
+ result.append(disk.logical_id[5])
+ if disk.children:
+ for child in disk.children:
+ helper(child, result)
+
+ result = []
+ for instance in self._config_data.instances.values():
+ for disk in instance.disks:
+ helper(disk, result)
+
+ return result
+
@locking.ssynchronized(_config_lock, shared=1)
def VerifyConfig(self):
"""Stub verify function.
result = []
seen_macs = []
+ ports = {}
data = self._config_data
for instance_name in data.instances:
instance = data.instances[instance_name]
(instance_name, idx, nic.mac))
else:
seen_macs.append(nic.mac)
+
+ # gather the drbd ports for duplicate checks
+ for dsk in instance.disks:
+ if dsk.dev_type in constants.LDS_DRBD:
+ tcp_port = dsk.logical_id[2]
+ if tcp_port not in ports:
+ ports[tcp_port] = []
+ ports[tcp_port].append((instance.name, "drbd disk %s" % dsk.iv_name))
+ # gather network port reservation
+ net_port = getattr(instance, "network_port", None)
+ if net_port is not None:
+ if net_port not in ports:
+ ports[net_port] = []
+ 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:
+ if free_port not in ports:
+ ports[free_port] = []
+ ports[free_port].append(("cluster", "port marked as free"))
+
+ # compute tcp/udp duplicate ports
+ keys = ports.keys()
+ keys.sort()
+ for pnum in keys:
+ pdata = ports[pnum]
+ if len(pdata) > 1:
+ txt = ", ".join(["%s/%s" % val for val in pdata])
+ result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt))
+
+ # highest used tcp port check
+ if keys:
+ if keys[-1] > self._config_data.cluster.highest_used_port:
+ result.append("Highest used port mismatch, saved %s, computed %s" %
+ (self._config_data.cluster.highest_used_port,
+ keys[-1]))
+
return result
def _UnlockedSetDiskID(self, disk, node_name):
if disk.logical_id is None and disk.physical_id is not None:
return
- if disk.dev_type in constants.LDS_DRBD:
- pnode, snode, port = disk.logical_id
+ if disk.dev_type == constants.LD_DRBD8:
+ pnode, snode, port, pminor, sminor, secret = disk.logical_id
if node_name not in (pnode, snode):
raise errors.ConfigurationError("DRBD device not knowing node %s" %
node_name)
if pnode_info is None or snode_info is None:
raise errors.ConfigurationError("Can't find primary or secondary node"
" for %s" % str(disk))
+ p_data = (pnode_info.secondary_ip, port)
+ s_data = (snode_info.secondary_ip, port)
if pnode == node_name:
- disk.physical_id = (pnode_info.secondary_ip, port,
- snode_info.secondary_ip, port)
+ disk.physical_id = p_data + s_data + (pminor, secret)
else: # it must be secondary, we tested above
- disk.physical_id = (snode_info.secondary_ip, port,
- pnode_info.secondary_ip, port)
+ disk.physical_id = s_data + p_data + (sminor, secret)
else:
disk.physical_id = disk.logical_id
return
self._WriteConfig()
return port
+ 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).
+
+ """
+ def _AppendUsedPorts(instance_name, disk, used):
+ if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
+ nodeA, nodeB, dummy, minorA, minorB = disk.logical_id[:5]
+ for node, port in ((nodeA, minorA), (nodeB, minorB)):
+ assert node in used, "Instance node not found in node list"
+ if port in used[node]:
+ raise errors.ProgrammerError("DRBD minor already used:"
+ " %s/%s, %s/%s" %
+ (node, port, instance_name,
+ used[node][port]))
+
+ used[node][port] = instance_name
+ if disk.children:
+ for child in disk.children:
+ _AppendUsedPorts(instance_name, child, used)
+
+ my_dict = dict((node, {}) for node in self._config_data.nodes)
+ for (node, minor), instance in self._temporary_drbds.iteritems():
+ my_dict[node][minor] = instance
+ for instance in self._config_data.instances.itervalues():
+ for disk in instance.disks:
+ _AppendUsedPorts(instance.name, disk, my_dict)
+ return my_dict
+
+ @locking.ssynchronized(_config_lock)
+ def AllocateDRBDMinor(self, nodes, instance):
+ """Allocate a drbd minor.
+
+ The free minor will be automatically computed from the existing
+ devices. A node can be given multiple times in order to allocate
+ multiple minors. The result is the list of minors, in the same
+ order as the passed nodes.
+
+ """
+ self._OpenConfig()
+
+ d_map = self._ComputeDRBDMap(instance)
+ result = []
+ for nname in nodes:
+ ndata = d_map[nname]
+ if not ndata:
+ # no minors used, we can start at 0
+ result.append(0)
+ ndata[0] = instance
+ self._temporary_drbds[(nname, 0)] = instance
+ continue
+ keys = ndata.keys()
+ keys.sort()
+ ffree = utils.FirstFree(keys)
+ if ffree is None:
+ # return the next minor
+ # TODO: implement high-limit check
+ minor = keys[-1] + 1
+ else:
+ minor = ffree
+ result.append(minor)
+ ndata[minor] = instance
+ assert (nname, minor) not in self._temporary_drbds, \
+ "Attempt to reuse reserved DRBD minor"
+ self._temporary_drbds[(nname, minor)] = instance
+ logging.debug("Request to allocate drbd minors, input: %s, returning %s",
+ nodes, result)
+ return result
+
+ @locking.ssynchronized(_config_lock)
+ def ReleaseDRBDMinors(self, instance):
+ """Release temporary drbd minors allocated for a given instance.
+
+ This should be called on both the error paths and on the success
+ paths (after the instance has been added or updated).
+
+ @type instance: string
+ @param instance: the instance for which temporary minors should be
+ released
+
+ """
+ for key, name in self._temporary_drbds.items():
+ if name == instance:
+ del self._temporary_drbds[key]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetConfigVersion(self):
+ """Get the configuration version.
+
+ @return: Config version
+
+ """
+ return self._config_data.version
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetClusterName(self):
+ """Get cluster name.
+
+ @return: Cluster name
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.cluster_name
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterNode(self):
+ """Get the hostname of the master node for this cluster.
+
+ @return: Master hostname
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.master_node
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterIP(self):
+ """Get the IP of the master node for this cluster.
+
+ @return: Master IP
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.master_ip
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterNetdev(self):
+ """Get the master network device for this cluster.
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.master_netdev
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetFileStorageDir(self):
+ """Get the file storage dir for this cluster.
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.file_storage_dir
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetHypervisorType(self):
+ """Get the hypervisor type for this cluster.
+
+ """
+ self._OpenConfig()
+ return self._config_data.cluster.hypervisor
+
@locking.ssynchronized(_config_lock, shared=1)
def GetHostKey(self):
"""Return the rsa hostkey from the config.
if instance.disk_template != constants.DT_DISKLESS:
all_lvs = instance.MapLVsByNode()
- logger.Info("Instance '%s' DISK_LAYOUT: %s" % (instance.name, all_lvs))
+ 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):
instance = self._config_data.instances[instance_name]
if instance.status != status:
instance.status = status
+ instance.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock)
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)
disk.iv_name))
self._config_data.instances[inst.name] = inst
+ self._config_data.cluster.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock)
"""
self._SetInstanceStatus(instance_name, "down")
+ def _UnlockedGetInstanceList(self):
+ """Get the list of instances.
+
+ 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.
these contains all the instances, also the ones in Admin_down state
"""
- self._OpenConfig()
-
- return self._config_data.instances.keys()
+ return self._UnlockedGetInstanceList()
@locking.ssynchronized(_config_lock, shared=1)
def ExpandInstanceName(self, short_name):
return utils.MatchNameComponent(short_name,
self._config_data.instances.keys())
+ def _UnlockedGetInstanceInfo(self, instance_name):
+ """Returns informations about an instance.
+
+ 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
+
+ return self._config_data.instances[instance_name]
+
@locking.ssynchronized(_config_lock, shared=1)
def GetInstanceInfo(self, instance_name):
"""Returns informations about an instance.
the instance object
"""
- self._OpenConfig()
+ return self._UnlockedGetInstanceInfo(instance_name)
- if instance_name not in self._config_data.instances:
- return None
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetAllInstancesInfo(self):
+ """Get the configuration of all instances.
- return self._config_data.instances[instance_name]
+ @rtype: dict
+ @returns: dict of (instance, instance_info), where instance_info is what
+ would GetInstanceInfo return for the node
+
+ """
+ my_dict = dict([(instance, self._UnlockedGetInstanceInfo(instance))
+ for instance in self._UnlockedGetInstanceList()])
+ return my_dict
@locking.ssynchronized(_config_lock)
def AddNode(self, node):
node: an object.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
self._WriteConfig()
@locking.ssynchronized(_config_lock)
"""Remove a node from the configuration.
"""
+ 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)
del self._config_data.nodes[node_name]
+ self._config_data.cluster.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock, shared=1)
return self._UnlockedGetNodeList()
@locking.ssynchronized(_config_lock, shared=1)
- def DumpConfig(self):
- """Return the entire configuration of the cluster.
+ def GetAllNodesInfo(self):
+ """Get the configuration of all nodes.
+
+ @rtype: dict
+ @returns: dict of (node, node_info), where node_info is what
+ would GetNodeInfo return for the node
+
"""
- self._OpenConfig()
- return self._config_data
+ my_dict = dict([(node, self._UnlockedGetNodeInfo(node))
+ for node in self._UnlockedGetNodeList()])
+ return my_dict
def _BumpSerialNo(self):
"""Bump up the serial number of the config.
"""
- self._config_data.cluster.serial_no += 1
+ self._config_data.serial_no += 1
def _OpenConfig(self):
"""Read the config data from disk.
# 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"
result = rpc.call_upload_file(nodelist, self._cfg_file)
for node in nodelist:
if not result[node]:
- logger.Error("copy of file %s to node %s failed" %
- (self._cfg_file, node))
+ logging.error("copy of file %s to node %s failed",
+ self._cfg_file, node)
bad = True
return not bad
self._DistributeConfig()
@locking.ssynchronized(_config_lock)
- def InitConfig(self, node, primary_ip, secondary_ip,
- hostkeypub, mac_prefix, vg_name, def_bridge):
+ def InitConfig(self, version, cluster_config, master_node_config):
"""Create the initial cluster configuration.
It will contain the current node, which will also be the master
- node, and no instances or operating systmes.
+ node, and no instances.
- Args:
- node: the nodename of the initial node
- primary_ip: the IP address of the current host
- secondary_ip: the secondary IP of the current host or None
- hostkeypub: the public hostkey of this host
-
- """
- hu_port = constants.FIRST_DRBD_PORT - 1
- globalconfig = objects.Cluster(serial_no=1,
- rsahostkeypub=hostkeypub,
- highest_used_port=hu_port,
- mac_prefix=mac_prefix,
- volume_group_name=vg_name,
- default_bridge=def_bridge,
- tcpudp_port_pool=set())
- if secondary_ip is None:
- secondary_ip = primary_ip
- nodeconfig = objects.Node(name=node, primary_ip=primary_ip,
- secondary_ip=secondary_ip)
-
- self._config_data = objects.ConfigData(nodes={node: nodeconfig},
+ @type version: int
+ @param version: Configuration version
+ @type cluster_config: objects.Cluster
+ @param cluster_config: Cluster configuration
+ @type master_node_config: objects.Node
+ @param master_node_config: Master node configuration
+
+ """
+ nodes = {
+ master_node_config.name: master_node_config,
+ }
+
+ self._config_data = objects.ConfigData(version=version,
+ cluster=cluster_config,
+ nodes=nodes,
instances={},
- cluster=globalconfig)
+ serial_no=1)
self._WriteConfig()
@locking.ssynchronized(_config_lock, shared=1)
"""
self._OpenConfig()
self._config_data.cluster.volume_group_name = vg_name
+ self._config_data.cluster.serial_no += 1
self._WriteConfig()
@locking.ssynchronized(_config_lock, shared=1)
if not test:
raise errors.ConfigurationError("Configuration updated since object"
" has been read or unknown object")
+ target.serial_no += 1
+
self._WriteConfig()