Implement BuildHooksEnv for NoHooksLU

[ganeti-local] / lib / config.py

diff --git a/lib/config.py b/lib/config.py
index 0f4749d..712f599 100644 (file)
--- a/lib/config.py
+++ b/lib/config.py
@@ -49,6 +49,14 @@ _config_lock = locking.SharedLock()
 
 
 def _ValidateConfig(data):
 
 
 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" %
   if data.version != constants.CONFIG_VERSION:
     raise errors.ConfigurationError("Cluster configuration version"
                                     " mismatch, got %s instead of %s" %

@@ -71,11 +79,13 @@ class ConfigWriter:
       self._cfg_file = cfg_file
     self._temporary_ids = set()
     self._temporary_drbds = {}
       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
     # 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
     self._OpenConfig()
 
   # this method needs to be static, so that we can call it on the class

@@ -101,11 +111,12 @@ class ConfigWriter:
       byte2 = random.randrange(0, 256)
       byte3 = random.randrange(0, 256)
       mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
       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")
         break
       retries -= 1
     else:
       raise errors.ConfigurationError("Can't generate unique MAC")

+    self._temporary_macs.add(mac)

     return mac
 
   @locking.ssynchronized(_config_lock, shared=1)
     return mac
 
   @locking.ssynchronized(_config_lock, shared=1)

@@ -117,7 +128,7 @@ class ConfigWriter:
 
     """
     all_macs = self._AllMACs()
 
     """
     all_macs = self._AllMACs()

-    return mac in all_macs
+    return mac in all_macs or mac in self._temporary_macs

 
   @locking.ssynchronized(_config_lock, shared=1)
   def GenerateDRBDSecret(self):
 
   @locking.ssynchronized(_config_lock, shared=1)
   def GenerateDRBDSecret(self):

@@ -137,7 +148,7 @@ class ConfigWriter:
       raise errors.ConfigurationError("Can't generate unique DRBD secret")
     return secret
 
       raise errors.ConfigurationError("Can't generate unique DRBD secret")
     return secret
 

-  def _ComputeAllLVs(self):
+  def _AllLVs(self):

     """Compute the list of all LVs.
 
     """
     """Compute the list of all LVs.
 
     """

@@ -148,6 +159,23 @@ class ConfigWriter:
         lvnames.update(lv_list)
     return lvnames
 
         lvnames.update(lv_list)
     return lvnames
 

+  def _AllIDs(self, include_temporary):
+    """Compute the list of all UUIDs and names we have.
+
+    @type include_temporary: boolean
+    @param include_temporary: whether to include the _temporary_ids set
+    @rtype: set
+    @return: a set of IDs
+
+    """
+    existing = set()
+    if include_temporary:
+      existing.update(self._temporary_ids)
+    existing.update(self._AllLVs())
+    existing.update(self._config_data.instances.keys())
+    existing.update(self._config_data.nodes.keys())
+    return existing
+

   @locking.ssynchronized(_config_lock, shared=1)
   def GenerateUniqueID(self, exceptions=None):
     """Generate an unique disk name.
   @locking.ssynchronized(_config_lock, shared=1)
   def GenerateUniqueID(self, exceptions=None):
     """Generate an unique disk name.

@@ -155,20 +183,16 @@ class ConfigWriter:
     This checks the current node, instances and disk names for
     duplicates.
 
     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()
-    existing.update(self._temporary_ids)
-    existing.update(self._ComputeAllLVs())
-    existing.update(self._config_data.instances.keys())
-    existing.update(self._config_data.nodes.keys())
+    existing = self._AllIDs(include_temporary=True)

     if exceptions is not None:
       existing.update(exceptions)
     retries = 64
     if exceptions is not None:
       existing.update(exceptions)
     retries = 64

@@ -182,9 +206,19 @@ class ConfigWriter:
     self._temporary_ids.add(unique_id)
     return unique_id
 
     self._temporary_ids.add(unique_id)
     return unique_id
 

+  def _CleanupTemporaryIDs(self):
+    """Cleanups the _temporary_ids structure.
+
+    """
+    existing = self._AllIDs(include_temporary=False)
+    self._temporary_ids = self._temporary_ids - existing
+

   def _AllMACs(self):
     """Return all MACs present in the config.
 
   def _AllMACs(self):
     """Return all MACs present in the config.
 

+    @rtype: list
+    @return: the list of all MACs
+

     """
     result = []
     for instance in self._config_data.instances.values():
     """
     result = []
     for instance in self._config_data.instances.values():

@@ -196,6 +230,9 @@ class ConfigWriter:
   def _AllDRBDSecrets(self):
     """Return all DRBD secrets present in the config.
 
   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."""
     """
     def helper(disk, result):
       """Recursively gather secrets from this disk."""

@@ -212,14 +249,64 @@ class ConfigWriter:
 
     return result
 
 
     return result
 

-  @locking.ssynchronized(_config_lock, shared=1)
-  def VerifyConfig(self):
-    """Stub verify function.
+  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
     """
     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:
     for instance_name in data.instances:
       instance = data.instances[instance_name]
       if instance.primary_node not in data.nodes:

@@ -250,8 +337,14 @@ class ConfigWriter:
           ports[net_port] = []
         ports[net_port].append((instance.name, "network port"))
 
           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
     # 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"))
       if free_port not in ports:
         ports[free_port] = []
       ports[free_port].append(("cluster", "port marked as free"))

@@ -267,13 +360,48 @@ class ConfigWriter:
 
     # highest used tcp port check
     if keys:
 
     # 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" %
         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")
+
+    # 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
 
 
     return result
 

+  @locking.ssynchronized(_config_lock, shared=1)
+  def VerifyConfig(self):
+    """Verify function.
+
+    This is just a wrapper over L{_UnlockedVerifyConfig}.
+
+    @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.
 
   def _UnlockedSetDiskID(self, disk, node_name):
     """Convert the unique ID to the ID needed on the target nodes.
 

@@ -366,37 +494,61 @@ class ConfigWriter:
     self._WriteConfig()
     return port
 
     self._WriteConfig()
     return port
 

-  def _ComputeDRBDMap(self, instance):
+  def _UnlockedComputeDRBDMap(self):

     """Compute the used DRBD minor/nodes.
 
     """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).
+    @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):
 
     """
     def _AppendUsedPorts(instance_name, disk, used):

+      duplicates = []

       if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
       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"
+        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]:
           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
+            duplicates.append((node, port, instance_name, used[node][port]))
+          else:
+            used[node][port] = instance_name

       if disk.children:
         for child in disk.children:
       if disk.children:
         for child in disk.children:

-          _AppendUsedPorts(instance_name, child, used)
+          duplicates.extend(_AppendUsedPorts(instance_name, child, used))
+      return duplicates

 
 

+    duplicates = []

     my_dict = dict((node, {}) for node in self._config_data.nodes)
     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:
     for instance in self._config_data.instances.itervalues():
       for disk in instance.disks:

-        _AppendUsedPorts(instance.name, disk, my_dict)
-    return my_dict
+        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):
 
   @locking.ssynchronized(_config_lock)
   def AllocateDRBDMinor(self, nodes, instance):

@@ -407,8 +559,17 @@ class ConfigWriter:
     multiple minors. The result is the list of minors, in the same
     order as the passed nodes.
 
     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
+

     """
     """

-    d_map = self._ComputeDRBDMap(instance)
+    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]
     result = []
     for nname in nodes:
       ndata = d_map[nname]

@@ -427,31 +588,55 @@ class ConfigWriter:
         minor = keys[-1] + 1
       else:
         minor = ffree
         minor = keys[-1] + 1
       else:
         minor = ffree

-      result.append(minor)
+      # 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
       ndata[minor] = instance

-      assert (nname, minor) not in self._temporary_drbds, \
-             "Attempt to reuse reserved DRBD minor"
-      self._temporary_drbds[(nname, 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
 
     logging.debug("Request to allocate drbd minors, input: %s, returning %s",
                   nodes, result)
     return result
 

-  @locking.ssynchronized(_config_lock)
-  def ReleaseDRBDMinors(self, instance):
+  def _UnlockedReleaseDRBDMinors(self, instance):

     """Release temporary drbd minors allocated for a given 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
 
     """
     @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]
 
     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.
   @locking.ssynchronized(_config_lock, shared=1)
   def GetConfigVersion(self):
     """Get the configuration version.

@@ -513,9 +698,9 @@ class ConfigWriter:
   def GetHostKey(self):
     """Return the rsa hostkey from the config.
 
   def GetHostKey(self):
     """Return the rsa hostkey from the config.
 

-    Args: None
+    @rtype: string
+    @return: the rsa hostkey

 
 

-    Returns: rsa hostkey

     """
     return self._config_data.cluster.rsahostkeypub
 
     """
     return self._config_data.cluster.rsahostkeypub
 

@@ -525,8 +710,9 @@ class ConfigWriter:
 
     This should be used after creating a new instance.
 
 
     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")
     """
     if not isinstance(instance, objects.Instance):
       raise errors.ProgrammerError("Invalid type passed to AddInstance")

@@ -535,25 +721,33 @@ class ConfigWriter:
       all_lvs = instance.MapLVsByNode()
       logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
 
       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
     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 _SetInstanceStatus(self, instance_name, status):
     """Set the instance's status to a given value.
 
     """
     self._WriteConfig()
 
   def _SetInstanceStatus(self, instance_name, status):
     """Set the instance's status to a given value.
 
     """

-    if status not in ("up", "down"):
-      raise errors.ProgrammerError("Invalid status '%s' passed to"
-                                   " ConfigWriter._SetInstanceStatus()" %
-                                   status)
+    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)
     instance = self._config_data.instances[instance_name]
 
     if instance_name not in self._config_data.instances:
       raise errors.ConfigurationError("Unknown instance '%s'" %
                                       instance_name)
     instance = self._config_data.instances[instance_name]

-    if instance.status != status:
-      instance.status = status
+    if instance.admin_up != status:
+      instance.admin_up = status

       instance.serial_no += 1
       self._WriteConfig()
 
       instance.serial_no += 1
       self._WriteConfig()
 

@@ -562,7 +756,7 @@ class ConfigWriter:
     """Mark the instance status to up in the config.
 
     """
     """Mark the instance status to up in the config.
 
     """

-    self._SetInstanceStatus(instance_name, "up")
+    self._SetInstanceStatus(instance_name, True)

 
   @locking.ssynchronized(_config_lock)
   def RemoveInstance(self, instance_name):
 
   @locking.ssynchronized(_config_lock)
   def RemoveInstance(self, instance_name):

@@ -572,6 +766,7 @@ class ConfigWriter:
     if instance_name not in self._config_data.instances:
       raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
     del self._config_data.instances[instance_name]
     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)
     self._WriteConfig()
 
   @locking.ssynchronized(_config_lock)

@@ -606,7 +801,7 @@ class ConfigWriter:
     """Mark the status of an instance to down in the configuration.
 
     """
     """Mark the status of an instance to down in the configuration.
 
     """

-    self._SetInstanceStatus(instance_name, "down")
+    self._SetInstanceStatus(instance_name, False)

 
   def _UnlockedGetInstanceList(self):
     """Get the list of instances.
 
   def _UnlockedGetInstanceList(self):
     """Get the list of instances.

@@ -620,9 +815,8 @@ class ConfigWriter:
   def GetInstanceList(self):
     """Get the list of instances.
 
   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()
 
     """
     return self._UnlockedGetInstanceList()

@@ -636,7 +830,7 @@ class ConfigWriter:
                                     self._config_data.instances.keys())
 
   def _UnlockedGetInstanceInfo(self, instance_name):
                                     self._config_data.instances.keys())
 
   def _UnlockedGetInstanceInfo(self, instance_name):

-    """Returns informations about an instance.
+    """Returns information about an instance.

 
     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.
 

@@ -648,16 +842,16 @@ class ConfigWriter:
 
   @locking.ssynchronized(_config_lock, shared=1)
   def GetInstanceInfo(self, 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.
 
     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)
 
     """
     return self._UnlockedGetInstanceInfo(instance_name)

@@ -667,7 +861,7 @@ class ConfigWriter:
     """Get the configuration of all instances.
 
     @rtype: dict
     """Get the configuration of all instances.
 
     @rtype: dict

-    @returns: dict of (instance, instance_info), where instance_info is what
+    @return: dict of (instance, instance_info), where instance_info is what

               would GetInstanceInfo return for the node
 
     """
               would GetInstanceInfo return for the node
 
     """

@@ -679,8 +873,8 @@ class ConfigWriter:
   def AddNode(self, node):
     """Add a node to the configuration.
 
   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)
 
     """
     logging.info("Adding node %s to configuration" % node.name)

@@ -715,11 +909,13 @@ class ConfigWriter:
   def _UnlockedGetNodeInfo(self, node_name):
     """Get the configuration of a node, as stored in the config.
 
   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

 
     """
     if node_name not in self._config_data.nodes:
 
     """
     if node_name not in self._config_data.nodes:

@@ -732,9 +928,12 @@ class ConfigWriter:
   def GetNodeInfo(self, node_name):
     """Get the configuration of a node, as stored in the config.
 
   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)
 
     """
     return self._UnlockedGetNodeInfo(node_name)

@@ -742,7 +941,10 @@ class ConfigWriter:
   def _UnlockedGetNodeList(self):
     """Return the list of nodes which are in the configuration.
 
   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

 
     """
     return self._config_data.nodes.keys()
 
     """
     return self._config_data.nodes.keys()

@@ -756,11 +958,20 @@ class ConfigWriter:
     return self._UnlockedGetNodeList()
 
   @locking.ssynchronized(_config_lock, shared=1)
     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
   def GetAllNodesInfo(self):
     """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
 
     """
               would GetNodeInfo return for the node
 
     """

@@ -768,6 +979,73 @@ class ConfigWriter:
                     for node in self._UnlockedGetNodeList()])
     return my_dict
 
                     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.
 
   def _BumpSerialNo(self):
     """Bump up the serial number of the config.
 

@@ -777,10 +1055,6 @@ class ConfigWriter:
   def _OpenConfig(self):
     """Read the config data from disk.
 
   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.
-

     """
     f = open(self._cfg_file, 'r')
     try:
     """
     f = open(self._cfg_file, 'r')
     try:

@@ -798,8 +1072,12 @@ class ConfigWriter:
         not hasattr(data.cluster, 'rsahostkeypub')):
       raise errors.ConfigurationError("Incomplete configuration"
                                       " (missing cluster.rsahostkeypub)")
         not hasattr(data.cluster, 'rsahostkeypub')):
       raise errors.ConfigurationError("Incomplete configuration"
                                       " (missing cluster.rsahostkeypub)")

+
+    # Upgrade configuration if needed
+    data.UpgradeConfig()
+

     self._config_data = data
     self._config_data = data

-    # init the last serial as -1 so that the next write will cause
+    # reset the last serial as -1 so that the next write will cause

     # ssconf update
     self._last_cluster_serial = -1
 
     # ssconf update
     self._last_cluster_serial = -1
 

@@ -813,23 +1091,26 @@ class ConfigWriter:
     if self._offline:
       return True
     bad = False
     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
     # 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

-    address_list = [self._UnlockedGetNodeInfo(name).primary_ip
-                    for name in nodelist]
+    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.RpcRunner.call_upload_file(nodelist, self._cfg_file,
-                                            address_list=address_list)
-    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)
       if not result[node]:
         logging.error("copy of file %s to node %s failed",
                       self._cfg_file, node)

@@ -840,6 +1121,15 @@ class ConfigWriter:
     """Write the configuration data to persistent storage.
 
     """
     """Write the configuration data to persistent storage.
 
     """

+    # first, cleanup the _temporary_ids set, if an ID is now in the
+    # other objects it should be discarded to prevent unbounded growth
+    # of that structure
+    self._CleanupTemporaryIDs()
+    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()
     if destination is None:
       destination = self._cfg_file
     self._BumpSerialNo()

@@ -856,7 +1146,7 @@ class ConfigWriter:
     os.rename(name, destination)
     self.write_count += 1
 
     os.rename(name, destination)
     self.write_count += 1
 

-    # and redistribute the config file
+    # and redistribute the config file to master candidates

     self._DistributeConfig()
 
     # Write ssconf files on all nodes (including locally)
     self._DistributeConfig()
 
     # Write ssconf files on all nodes (including locally)

@@ -874,43 +1164,34 @@ class ConfigWriter:
         associated value
 
     """
         associated value
 
     """

-    node_list = "\n".join(utils.NiceSort(self._UnlockedGetNodeList()))
+    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 = self._config_data.cluster

+    cluster_tags = fn(cluster.GetTags())

     return {
       constants.SS_CLUSTER_NAME: cluster.cluster_name,
     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_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_MASTER_IP: cluster.master_ip,
       constants.SS_MASTER_NETDEV: cluster.master_netdev,
       constants.SS_MASTER_NODE: cluster.master_node,

-      constants.SS_NODE_LIST: node_list,
+      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)
-  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.
-
-    @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={},
-                                           serial_no=1)
-    self._WriteConfig()
-

   @locking.ssynchronized(_config_lock, shared=1)
   def GetVGName(self):
     """Return the volume group name.
   @locking.ssynchronized(_config_lock, shared=1)
   def GetVGName(self):
     """Return the volume group name.

@@ -943,10 +1224,10 @@ class ConfigWriter:
 
   @locking.ssynchronized(_config_lock, shared=1)
   def GetClusterInfo(self):
 
   @locking.ssynchronized(_config_lock, shared=1)
   def GetClusterInfo(self):

-    """Returns informations about the cluster
+    """Returns information about the cluster

 
 

-    Returns:
-      the cluster object
+    @rtype: L{objects.Cluster}
+    @return: the cluster object

 
     """
     return self._config_data.cluster
 
     """
     return self._config_data.cluster

@@ -961,6 +1242,10 @@ class ConfigWriter:
     that all modified objects will be saved, but the target argument
     is the one the caller wants to ensure that it's saved.
 
     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,"
     """
     if self._config_data is None:
       raise errors.ProgrammerError("Configuration file not read,"

@@ -985,4 +1270,9 @@ class ConfigWriter:
       # for node updates, we need to increase the cluster serial too
       self._config_data.cluster.serial_no += 1
 
       # 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()
     self._WriteConfig()