-#!/usr/bin/python
+#
#
# Copyright (C) 2006, 2007 Google Inc.
"""Configuration management for Ganeti
-This module provides the interface to the ganeti cluster configuration.
-
+This module provides the interface to the Ganeti cluster configuration.
-The configuration data is stored on every node but is updated on the
-master only. After each update, the master distributes the data to the
-other nodes.
+The configuration data is stored on every node but is updated on the master
+only. After each update, the master distributes the data to the other nodes.
-Currently the data storage format is pickle as yaml was initially not
-available, then we used it but it was a memory-eating slow beast, so
-we reverted to pickle using custom Unpicklers.
+Currently, the data storage format is JSON. YAML was slow and consuming too
+much memory.
"""
import os
-import socket
import tempfile
import random
+import logging
from ganeti import errors
-from ganeti import logger
+from ganeti import locking
from ganeti import utils
from ganeti import constants
from ganeti import rpc
from ganeti import objects
+from ganeti import serializer
+
+
+_config_lock = locking.SharedLock()
+
+
+def _ValidateConfig(data):
+ """Verifies that a configuration objects looks valid.
+
+ 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" %
+ (data.version,
+ constants.CONFIG_VERSION))
class ConfigWriter:
"""
def __init__(self, cfg_file=None, offline=False):
+ 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
else:
self._cfg_file = cfg_file
+ self._temporary_ids = set()
+ self._temporary_drbds = {}
+ self._temporary_macs = set()
+ # 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
+ # 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
"""
return os.path.exists(constants.CLUSTER_CONF_FILE)
+ @locking.ssynchronized(_config_lock, shared=1)
def GenerateMAC(self):
"""Generate a MAC for an instance.
This should check the current instances for duplicates.
"""
- self._OpenConfig()
- self._ReleaseLock()
prefix = self._config_data.cluster.mac_prefix
all_macs = self._AllMACs()
retries = 64
byte2 = random.randrange(0, 256)
byte3 = random.randrange(0, 256)
mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
- if mac not in all_macs:
+ if mac not in all_macs and mac not in self._temporary_macs:
break
retries -= 1
else:
- raise errors.ConfigurationError, ("Can't generate unique MAC")
+ raise errors.ConfigurationError("Can't generate unique MAC")
+ self._temporary_macs.add(mac)
return mac
+ @locking.ssynchronized(_config_lock, shared=1)
+ def IsMacInUse(self, mac):
+ """Predicate: check if the specified MAC is in use in the Ganeti cluster.
+
+ This only checks instances managed by this cluster, it does not
+ check for potential collisions elsewhere.
+
+ """
+ all_macs = self._AllMACs()
+ return mac in all_macs or mac in self._temporary_macs
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GenerateDRBDSecret(self):
+ """Generate a DRBD secret.
+
+ This checks the current disks for duplicates.
+
+ """
+ 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.
+
+ """
+ lvnames = set()
+ for instance in self._config_data.instances.values():
+ node_data = instance.MapLVsByNode()
+ for lv_list in node_data.values():
+ lvnames.update(lv_list)
+ return lvnames
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GenerateUniqueID(self, exceptions=None):
+ """Generate an unique disk name.
+
+ This checks the current node, instances and disk names for
+ duplicates.
+
+ @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)
+
+ @rtype: string
+ @return: the unique id
+
+ """
+ existing = set()
+ existing.update(self._temporary_ids)
+ existing.update(self._ComputeAllLVs())
+ existing.update(self._config_data.instances.keys())
+ existing.update(self._config_data.nodes.keys())
+ if exceptions is not None:
+ existing.update(exceptions)
+ retries = 64
+ while retries > 0:
+ unique_id = utils.NewUUID()
+ if unique_id not in existing and unique_id is not None:
+ break
+ else:
+ raise errors.ConfigurationError("Not able generate an unique ID"
+ " (last tried ID: %s" % unique_id)
+ self._temporary_ids.add(unique_id)
+ return unique_id
+
def _AllMACs(self):
"""Return all MACs present in the config.
- """
- self._OpenConfig()
- self._ReleaseLock()
+ @rtype: list
+ @return: the list of all MACs
+ """
result = []
for instance in self._config_data.instances.values():
for nic in instance.nics:
return result
- def VerifyConfig(self):
- """Stub verify function.
+ def _AllDRBDSecrets(self):
+ """Return all DRBD secrets present in the config.
+
+ @rtype: list
+ @return: the list of all DRBD secrets
+
"""
- self._OpenConfig()
- self._ReleaseLock()
+ 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
+ def _CheckDiskIDs(self, disk, l_ids, p_ids):
+ """Compute duplicate disk IDs
+
+ @type disk: L{objects.Disk}
+ @param disk: the disk at which to start searching
+ @type l_ids: list
+ @param l_ids: list of current logical ids
+ @type p_ids: list
+ @param p_ids: list of current physical ids
+ @rtype: list
+ @return: a list of error messages
+
+ """
+ result = []
+ if disk.logical_id is not None:
+ if disk.logical_id in l_ids:
+ result.append("duplicate logical id %s" % str(disk.logical_id))
+ else:
+ l_ids.append(disk.logical_id)
+ if disk.physical_id is not None:
+ if disk.physical_id in p_ids:
+ result.append("duplicate physical id %s" % str(disk.physical_id))
+ else:
+ p_ids.append(disk.physical_id)
+
+ if disk.children:
+ for child in disk.children:
+ result.extend(self._CheckDiskIDs(child, l_ids, p_ids))
+ return result
+
+ def _UnlockedVerifyConfig(self):
+ """Verify function.
+
+ @rtype: list
+ @return: a list of error messages; a non-empty list signifies
+ configuration errors
+
+ """
result = []
seen_macs = []
+ ports = {}
data = self._config_data
+ seen_lids = []
+ seen_pids = []
+
+ # global cluster checks
+ if not data.cluster.enabled_hypervisors:
+ result.append("enabled hypervisors list doesn't have any entries")
+ invalid_hvs = set(data.cluster.enabled_hypervisors) - constants.HYPER_TYPES
+ if invalid_hvs:
+ result.append("enabled hypervisors contains invalid entries: %s" %
+ invalid_hvs)
+
+ if data.cluster.master_node not in data.nodes:
+ result.append("cluster has invalid primary node '%s'" %
+ data.cluster.master_node)
+
+ # per-instance checks
for instance_name in data.instances:
instance = data.instances[instance_name]
if instance.primary_node not in data.nodes:
- result.append("Instance '%s' has invalid primary node '%s'" %
+ result.append("instance '%s' has invalid primary node '%s'" %
(instance_name, instance.primary_node))
for snode in instance.secondary_nodes:
if snode not in data.nodes:
- result.append("Instance '%s' has invalid secondary node '%s'" %
+ result.append("instance '%s' has invalid secondary node '%s'" %
(instance_name, snode))
for idx, nic in enumerate(instance.nics):
if nic.mac in seen_macs:
- result.append("Instance '%s' has NIC %d mac %s duplicate" %
+ result.append("instance '%s' has NIC %d mac %s duplicate" %
(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"))
+
+ # instance disk verify
+ for idx, disk in enumerate(instance.disks):
+ result.extend(["instance '%s' disk %d error: %s" %
+ (instance.name, idx, msg) for msg in disk.Verify()])
+ result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids))
+
+ # cluster-wide pool of free ports
+ 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"))
+
+ # 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] > data.cluster.highest_used_port:
+ result.append("Highest used port mismatch, saved %s, computed %s" %
+ (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")
+
+ # master candidate checks
+ 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))
+
+ # node checks
+ for node in data.nodes.values():
+ if [node.master_candidate, node.drained, node.offline].count(True) > 1:
+ result.append("Node %s state is invalid: master_candidate=%s,"
+ " drain=%s, offline=%s" %
+ (node.name, node.master_candidate, node.drain,
+ node.offline))
+
+ # drbd minors check
+ d_map, duplicates = self._UnlockedComputeDRBDMap()
+ for node, minor, instance_a, instance_b in duplicates:
+ result.append("DRBD minor %d on node %s is assigned twice to instances"
+ " %s and %s" % (minor, node, instance_a, instance_b))
+
return result
+ @locking.ssynchronized(_config_lock, shared=1)
+ def VerifyConfig(self):
+ """Verify function.
+
+ This is just a wrapper over L{_UnlockedVerifyConfig}.
- def SetDiskID(self, disk, node_name):
+ @rtype: list
+ @return: a list of error messages; a non-empty list signifies
+ configuration errors
+
+ """
+ return self._UnlockedVerifyConfig()
+
+ def _UnlockedSetDiskID(self, disk, node_name):
"""Convert the unique ID to the ID needed on the target nodes.
This is used only for drbd, which needs ip/port configuration.
this helps when the only the top device is passed to the remote
node.
+ This function is for internal use, when the config lock is already held.
+
"""
if disk.children:
for child in disk.children:
- self.SetDiskID(child, node_name)
+ self._UnlockedSetDiskID(child, node_name)
if disk.logical_id is None and disk.physical_id is not None:
return
- if disk.dev_type == "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)
- pnode_info = self.GetNodeInfo(pnode)
- snode_info = self.GetNodeInfo(snode)
+ raise errors.ConfigurationError("DRBD device not knowing node %s" %
+ node_name)
+ pnode_info = self._UnlockedGetNodeInfo(pnode)
+ snode_info = self._UnlockedGetNodeInfo(snode)
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
+ @locking.ssynchronized(_config_lock)
+ def SetDiskID(self, disk, node_name):
+ """Convert the unique ID to the ID needed on the target nodes.
+
+ This is used only for drbd, which needs ip/port configuration.
+
+ The routine descends down and updates its children also, because
+ this helps when the only the top device is passed to the remote
+ node.
+
+ """
+ return self._UnlockedSetDiskID(disk, node_name)
+
+ @locking.ssynchronized(_config_lock)
def AddTcpUdpPort(self, port):
"""Adds a new port to the available port pool.
"""
if not isinstance(port, int):
- raise errors.ProgrammerError, ("Invalid type passed for port")
+ raise errors.ProgrammerError("Invalid type passed for port")
- self._OpenConfig()
self._config_data.cluster.tcpudp_port_pool.add(port)
self._WriteConfig()
+ @locking.ssynchronized(_config_lock, shared=1)
def GetPortList(self):
"""Returns a copy of the current port list.
"""
- self._OpenConfig()
- self._ReleaseLock()
return self._config_data.cluster.tcpudp_port_pool.copy()
+ @locking.ssynchronized(_config_lock)
def AllocatePort(self):
"""Allocate a port.
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()
else:
port = self._config_data.cluster.highest_used_port + 1
if port >= constants.LAST_DRBD_PORT:
- raise errors.ConfigurationError, ("The highest used port is greater"
- " than %s. Aborting." %
- constants.LAST_DRBD_PORT)
+ raise errors.ConfigurationError("The highest used port is greater"
+ " than %s. Aborting." %
+ constants.LAST_DRBD_PORT)
self._config_data.cluster.highest_used_port = port
self._WriteConfig()
return port
+ def _UnlockedComputeDRBDMap(self):
+ """Compute the used DRBD minor/nodes.
+
+ @rtype: (dict, 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), and a list of duplicates; if the duplicates
+ list is not empty, the configuration is corrupted and its caller
+ should raise an exception
+
+ """
+ def _AppendUsedPorts(instance_name, disk, used):
+ duplicates = []
+ if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
+ node_a, node_b, _, minor_a, minor_b = disk.logical_id[:5]
+ for node, port in ((node_a, minor_a), (node_b, minor_b)):
+ assert node in used, ("Node '%s' of instance '%s' not found"
+ " in node list" % (node, instance_name))
+ if port in used[node]:
+ duplicates.append((node, port, instance_name, used[node][port]))
+ else:
+ used[node][port] = instance_name
+ if disk.children:
+ for child in disk.children:
+ duplicates.extend(_AppendUsedPorts(instance_name, child, used))
+ return duplicates
+
+ duplicates = []
+ my_dict = dict((node, {}) for node in self._config_data.nodes)
+ for instance in self._config_data.instances.itervalues():
+ for disk in instance.disks:
+ duplicates.extend(_AppendUsedPorts(instance.name, disk, my_dict))
+ for (node, minor), instance in self._temporary_drbds.iteritems():
+ if minor in my_dict[node] and my_dict[node][minor] != instance:
+ duplicates.append((node, minor, instance, my_dict[node][minor]))
+ else:
+ my_dict[node][minor] = instance
+ return my_dict, duplicates
+
+ @locking.ssynchronized(_config_lock)
+ def ComputeDRBDMap(self):
+ """Compute the used DRBD minor/nodes.
+
+ This is just a wrapper over L{_UnlockedComputeDRBDMap}.
+
+ @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).
+
+ """
+ d_map, duplicates = self._UnlockedComputeDRBDMap()
+ if duplicates:
+ raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
+ str(duplicates))
+ return d_map
+
+ @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.
+
+ @type instance: string
+ @param instance: the instance for which we allocate minors
+
+ """
+ assert isinstance(instance, basestring), \
+ "Invalid argument '%s' passed to AllocateDRBDMinor" % instance
+
+ d_map, duplicates = self._UnlockedComputeDRBDMap()
+ if duplicates:
+ raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
+ str(duplicates))
+ 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
+ # double-check minor against current instances
+ assert minor not in d_map[nname], \
+ ("Attempt to reuse allocated DRBD minor %d on node %s,"
+ " already allocated to instance %s" %
+ (minor, nname, d_map[nname][minor]))
+ ndata[minor] = instance
+ # double-check minor against reservation
+ r_key = (nname, minor)
+ assert r_key not in self._temporary_drbds, \
+ ("Attempt to reuse reserved DRBD minor %d on node %s,"
+ " reserved for instance %s" %
+ (minor, nname, self._temporary_drbds[r_key]))
+ self._temporary_drbds[r_key] = instance
+ result.append(minor)
+ logging.debug("Request to allocate drbd minors, input: %s, returning %s",
+ nodes, result)
+ return result
+
+ def _UnlockedReleaseDRBDMinors(self, instance):
+ """Release temporary drbd minors allocated for a given instance.
+
+ @type instance: string
+ @param instance: the instance for which temporary minors should be
+ released
+
+ """
+ assert isinstance(instance, basestring), \
+ "Invalid argument passed to ReleaseDRBDMinors"
+ for key, name in self._temporary_drbds.items():
+ if name == instance:
+ del self._temporary_drbds[key]
+
+ @locking.ssynchronized(_config_lock)
+ def ReleaseDRBDMinors(self, instance):
+ """Release temporary drbd minors allocated for a given instance.
+
+ This should be called on the error paths, on the success paths
+ it's automatically called by the ConfigWriter add and update
+ functions.
+
+ This function is just a wrapper over L{_UnlockedReleaseDRBDMinors}.
+
+ @type instance: string
+ @param instance: the instance for which temporary minors should be
+ released
+
+ """
+ self._UnlockedReleaseDRBDMinors(instance)
+
+ @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
+
+ """
+ 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
+
+ """
+ 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
+
+ """
+ return self._config_data.cluster.master_ip
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetMasterNetdev(self):
+ """Get the master network device for this cluster.
+
+ """
+ return self._config_data.cluster.master_netdev
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetFileStorageDir(self):
+ """Get the file storage dir for this cluster.
+
+ """
+ return self._config_data.cluster.file_storage_dir
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetHypervisorType(self):
+ """Get the hypervisor type for this cluster.
+
+ """
+ return self._config_data.cluster.enabled_hypervisors[0]
+
+ @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()
- self._ReleaseLock()
return self._config_data.cluster.rsahostkeypub
+ @locking.ssynchronized(_config_lock)
def AddInstance(self, instance):
"""Add an instance to the config.
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")
- self._OpenConfig()
+ if instance.disk_template != constants.DT_DISKLESS:
+ all_lvs = instance.MapLVsByNode()
+ logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
+
+ all_macs = self._AllMACs()
+ for nic in instance.nics:
+ if nic.mac in all_macs:
+ raise errors.ConfigurationError("Cannot add instance %s:"
+ " MAC address '%s' already in use." % (instance.name, nic.mac))
+
+ instance.serial_no = 1
self._config_data.instances[instance.name] = instance
+ self._config_data.cluster.serial_no += 1
+ self._UnlockedReleaseDRBDMinors(instance.name)
+ for nic in instance.nics:
+ self._temporary_macs.discard(nic.mac)
self._WriteConfig()
- def MarkInstanceUp(self, instance_name):
- """Mark the instance status to up in the config.
+ def _SetInstanceStatus(self, instance_name, status):
+ """Set the instance's status to a given value.
"""
- self._OpenConfig()
+ assert isinstance(status, bool), \
+ "Invalid status '%s' passed to SetInstanceStatus" % (status,)
if instance_name not in self._config_data.instances:
- raise errors.ConfigurationError, ("Unknown instance '%s'" %
- instance_name)
+ raise errors.ConfigurationError("Unknown instance '%s'" %
+ instance_name)
instance = self._config_data.instances[instance_name]
- instance.status = "up"
- self._WriteConfig()
+ if instance.admin_up != status:
+ instance.admin_up = status
+ instance.serial_no += 1
+ self._WriteConfig()
+ @locking.ssynchronized(_config_lock)
+ def MarkInstanceUp(self, instance_name):
+ """Mark the instance status to up in the config.
+
+ """
+ self._SetInstanceStatus(instance_name, True)
+
+ @locking.ssynchronized(_config_lock)
def RemoveInstance(self, instance_name):
"""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)
+ 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)
+ def RenameInstance(self, old_name, new_name):
+ """Rename an instance.
+
+ This needs to be done in ConfigWriter and not by RemoveInstance
+ combined with AddInstance as only we can guarantee an atomic
+ rename.
+
+ """
+ if old_name not in self._config_data.instances:
+ raise errors.ConfigurationError("Unknown instance '%s'" % old_name)
+ inst = self._config_data.instances[old_name]
+ del self._config_data.instances[old_name]
+ inst.name = new_name
+
+ for disk in inst.disks:
+ if disk.dev_type == constants.LD_FILE:
+ # rename the file paths in logical and physical id
+ file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1]))
+ disk.physical_id = disk.logical_id = (disk.logical_id[0],
+ os.path.join(file_storage_dir,
+ inst.name,
+ disk.iv_name))
+
+ self._config_data.instances[inst.name] = inst
+ self._WriteConfig()
+
+ @locking.ssynchronized(_config_lock)
def MarkInstanceDown(self, instance_name):
"""Mark the status of an instance to down in the configuration.
"""
- self._OpenConfig()
+ self._SetInstanceStatus(instance_name, False)
- if instance_name not in self._config_data.instances:
- raise errors.ConfigurationError, ("Unknown instance '%s'" %
- instance_name)
- instance = self._config_data.instances[instance_name]
- instance.status = "down"
- self._WriteConfig()
+ def _UnlockedGetInstanceList(self):
+ """Get the list of instances.
+
+ This function is for internal use, when the config lock is already held.
+ """
+ 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']
"""
- self._OpenConfig()
- self._ReleaseLock()
-
- return self._config_data.instances.keys()
+ return self._UnlockedGetInstanceList()
+ @locking.ssynchronized(_config_lock, shared=1)
def ExpandInstanceName(self, short_name):
"""Attempt to expand an incomplete instance name.
"""
- self._OpenConfig()
- self._ReleaseLock()
-
return utils.MatchNameComponent(short_name,
self._config_data.instances.keys())
+ def _UnlockedGetInstanceInfo(self, instance_name):
+ """Returns information about an instance.
+
+ This function is for internal use, when the config lock is already held.
+
+ """
+ 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.
+ """Returns information about an instance.
- It takes the information from the configuration file. Other informations of
+ It takes the information from the configuration file. Other information 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
"""
- self._OpenConfig()
- self._ReleaseLock()
+ 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
+ @return: 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):
"""Add a node to the configuration.
- Args:
- node: an object.Node instance
+ @type node: L{objects.Node}
+ @param node: a Node instance
"""
- self._OpenConfig()
+ logging.info("Adding node %s to configuration" % node.name)
+
+ node.serial_no = 1
self._config_data.nodes[node.name] = node
+ self._config_data.cluster.serial_no += 1
self._WriteConfig()
+ @locking.ssynchronized(_config_lock)
def RemoveNode(self, node_name):
"""Remove a node from the configuration.
"""
- self._OpenConfig()
+ logging.info("Removing node %s from configuration" % node_name)
+
if node_name not in self._config_data.nodes:
- raise errors.ConfigurationError, ("Unknown node '%s'" % node_name)
+ 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)
def ExpandNodeName(self, short_name):
"""Attempt to expand an incomplete instance name.
"""
- self._OpenConfig()
- self._ReleaseLock()
-
return utils.MatchNameComponent(short_name,
self._config_data.nodes.keys())
- def GetNodeInfo(self, node_name):
+ def _UnlockedGetNodeInfo(self, node_name):
"""Get the configuration of a node, as stored in the config.
- Args: node: nodename (tuple) of the node
+ This function is for internal use, when the config lock is already
+ held.
- Returns: the node object
+ @param node_name: the node name, e.g. I{node1.example.com}
- """
- self._OpenConfig()
- self._ReleaseLock()
+ @rtype: L{objects.Node}
+ @return: the node object
+ """
if node_name not in self._config_data.nodes:
return None
return self._config_data.nodes[node_name]
- def GetNodeList(self):
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeInfo(self, node_name):
+ """Get the configuration of a node, as stored in the config.
+
+ This is just a locked wrapper over L{_UnlockedGetNodeInfo}.
+
+ @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.
+
+ @rtype: list
+
"""
- self._OpenConfig()
- self._ReleaseLock()
return self._config_data.nodes.keys()
- def DumpConfig(self):
- """Return the entire configuration of the cluster.
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetNodeList(self):
+ """Return the list of nodes which are in the configuration.
+
"""
- self._OpenConfig()
- self._ReleaseLock()
- return self._config_data
+ return self._UnlockedGetNodeList()
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetOnlineNodeList(self):
+ """Return the list of nodes which are online.
+
+ """
+ all_nodes = [self._UnlockedGetNodeInfo(node)
+ for node in self._UnlockedGetNodeList()]
+ return [node.name for node in all_nodes if not node.offline]
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetAllNodesInfo(self):
+ """Get the configuration of all nodes.
+
+ @rtype: dict
+ @return: dict of (node, node_info), where node_info is what
+ would GetNodeInfo return for the node
+
+ """
+ my_dict = dict([(node, self._UnlockedGetNodeInfo(node))
+ for node in self._UnlockedGetNodeList()])
+ return my_dict
+
+ def _UnlockedGetMasterCandidateStats(self, exceptions=None):
+ """Get the number of current and maximum desired and possible candidates.
+
+ @type exceptions: list
+ @param exceptions: if passed, list of nodes that should be ignored
+ @rtype: tuple
+ @return: tuple of (current, desired and possible)
+
+ """
+ mc_now = mc_max = 0
+ for node in self._config_data.nodes.values():
+ if exceptions and node.name in exceptions:
+ continue
+ if not (node.offline or node.drained):
+ 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, exceptions=None):
+ """Get the number of current and maximum possible candidates.
+
+ This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
+
+ @type exceptions: list
+ @param exceptions: if passed, list of nodes that should be ignored
+ @rtype: tuple
+ @return: tuple of (current, max)
+
+ """
+ return self._UnlockedGetMasterCandidateStats(exceptions)
+
+ @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 or node.drained:
+ 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.
"""
- self._config_data.cluster.serial_no += 1
+ self._config_data.serial_no += 1
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
f = open(self._cfg_file, 'r')
try:
try:
- data = objects.ConfigObject.Load(f)
+ data = objects.ConfigData.FromDict(serializer.Load(f.read()))
except Exception, err:
- raise errors.ConfigurationError, err
+ 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, 'config_version')):
- raise errors.ConfigurationError, ("Incomplete configuration"
- " (missing cluster.config_version)")
- if data.cluster.config_version != constants.CONFIG_VERSION:
- raise errors.ConfigurationError, ("Cluster configuration version"
- " mismatch, got %s instead of %s" %
- (data.cluster.config_version,
- constants.CONFIG_VERSION))
+ 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
-
- def _ReleaseLock(self):
- """xxxx
- """
+ # 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.GetNodeList()
- myhostname = socket.gethostname()
- tgt_list = []
- for node in nodelist:
- nodeinfo = self.GetNodeInfo(node)
- if nodeinfo.name == myhostname:
+ 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
- tgt_list.append(node)
-
- result = rpc.call_upload_file(tgt_list, self._cfg_file)
- for node in tgt_list:
- if not result[node]:
- logger.Error("copy of file %s to node %s failed" %
- (self._cfg_file, node))
+ 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.RpcRunner.call_upload_file(node_list, self._cfg_file,
+ address_list=addr_list)
+ for to_node, to_result in result.items():
+ msg = to_result.RemoteFailMsg()
+ if msg:
+ msg = ("Copy of file %s to node %s failed: %s" %
+ (self._cfg_file, to_node, msg))
+ logging.error(msg)
bad = True
return not bad
"""Write the configuration data to persistent storage.
"""
+ config_errors = self._UnlockedVerifyConfig()
+ if config_errors:
+ raise errors.ConfigurationError("Configuration data is not"
+ " consistent: %s" %
+ (", ".join(config_errors)))
if destination is None:
destination = self._cfg_file
self._BumpSerialNo()
+ txt = serializer.Dump(self._config_data.ToDict())
dir_name, file_name = os.path.split(destination)
fd, name = tempfile.mkstemp('.newconfig', file_name, dir_name)
f = os.fdopen(fd, 'w')
try:
- self._config_data.Dump(f)
+ f.write(txt)
os.fsync(f.fileno())
finally:
f.close()
# we don't need to do os.close(fd) as f.close() did it
os.rename(name, destination)
- self._DistributeConfig()
+ self.write_count += 1
- def InitConfig(self, node, primary_ip, secondary_ip,
- clustername, hostkeypub, mac_prefix, vg_name, def_bridge):
- """Create the initial cluster configuration.
-
- It will contain the current node, which will also be the master
- node, and no instances or operating systmes.
-
- 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
- clustername: the name of the cluster
- hostkeypub: the public hostkey of this host
-
- """
- hu_port = constants.FIRST_DRBD_PORT - 1
- globalconfig = objects.Cluster(config_version=constants.CONFIG_VERSION,
- serial_no=1,
- name=clustername,
- 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},
- instances={},
- cluster=globalconfig)
- self._WriteConfig()
+ # and redistribute the config file to master candidates
+ self._DistributeConfig()
- def GetClusterName(self):
- """Return the cluster name.
+ # Write ssconf files on all nodes (including locally)
+ if self._last_cluster_serial < self._config_data.cluster.serial_no:
+ if not self._offline:
+ result = rpc.RpcRunner.call_write_ssconf_files(\
+ self._UnlockedGetNodeList(),
+ self._UnlockedGetSsconfValues())
+ for nname, nresu in result.items():
+ msg = nresu.RemoteFailMsg()
+ if msg:
+ logging.warning("Error while uploading ssconf files to"
+ " node %s: %s", nname, msg)
+ 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
"""
- self._OpenConfig()
- self._ReleaseLock()
- return self._config_data.cluster.name
-
+ fn = "\n".join
+ instance_names = utils.NiceSort(self._UnlockedGetInstanceList())
+ node_names = utils.NiceSort(self._UnlockedGetNodeList())
+ node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]
+
+ instance_data = fn(instance_names)
+ off_data = fn(node.name for node in node_info if node.offline)
+ on_data = fn(node.name for node in node_info if not 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
+ cluster_tags = fn(cluster.GetTags())
+ return {
+ constants.SS_CLUSTER_NAME: cluster.cluster_name,
+ constants.SS_CLUSTER_TAGS: cluster_tags,
+ 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,
+ constants.SS_ONLINE_NODES: on_data,
+ constants.SS_INSTANCE_LIST: instance_data,
+ constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION,
+ }
+
+ @locking.ssynchronized(_config_lock, shared=1)
def GetVGName(self):
"""Return the volume group name.
"""
- self._OpenConfig()
- self._ReleaseLock()
return self._config_data.cluster.volume_group_name
- def GetDefBridge(self):
- """Return the default bridge.
+ @locking.ssynchronized(_config_lock)
+ def SetVGName(self, vg_name):
+ """Set the volume group name.
"""
- self._OpenConfig()
- self._ReleaseLock()
- return self._config_data.cluster.default_bridge
+ self._config_data.cluster.volume_group_name = vg_name
+ self._config_data.cluster.serial_no += 1
+ self._WriteConfig()
+ @locking.ssynchronized(_config_lock, shared=1)
def GetMACPrefix(self):
"""Return the mac prefix.
"""
- self._OpenConfig()
- self._ReleaseLock()
return self._config_data.cluster.mac_prefix
+
+ @locking.ssynchronized(_config_lock, shared=1)
+ def GetClusterInfo(self):
+ """Returns information about the cluster
+
+ @rtype: L{objects.Cluster}
+ @return: the cluster object
+
+ """
+ return self._config_data.cluster
+
+ @locking.ssynchronized(_config_lock)
+ def Update(self, target):
+ """Notify function to be called after updates.
+
+ This function must be called when an object (as returned by
+ GetInstanceInfo, GetNodeInfo, GetCluster) has been updated and the
+ caller wants the modifications saved to the backing store. Note
+ 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:
+ raise errors.ProgrammerError("Invalid object type (%s) passed to"
+ " ConfigWriter.Update" % type(target))
+ if not test:
+ raise errors.ConfigurationError("Configuration updated since object"
+ " 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
+
+ if isinstance(target, objects.Instance):
+ self._UnlockedReleaseDRBDMinors(target.name)
+ for nic in target.nics:
+ self._temporary_macs.discard(nic.mac)
+
+ self._WriteConfig()