Revert the instance IP conflicts

[ganeti-local] / lib / config.py

diff --git a/lib/config.py b/lib/config.py
index 2d849f1..2fea523 100644 (file)
--- a/lib/config.py
+++ b/lib/config.py
@@ -32,9 +32,9 @@ much memory.
 """
 
 import os
 """
 
 import os

-import tempfile

 import random
 import logging
 import random
 import logging

+import time

 
 from ganeti import errors
 from ganeti import locking
 
 from ganeti import errors
 from ganeti import locking

@@ -79,6 +79,7 @@ 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
     # 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

@@ -110,11 +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)

@@ -126,7 +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):

@@ -146,7 +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.
 
     """

@@ -157,27 +159,40 @@ class ConfigWriter:
         lvnames.update(lv_list)
     return lvnames
 
         lvnames.update(lv_list)
     return lvnames
 

-  @locking.ssynchronized(_config_lock, shared=1)
-  def GenerateUniqueID(self, exceptions=None):
-    """Generate an unique disk name.
+  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())
+    existing.update([i.uuid for i in self._AllUUIDObjects() if i.uuid])
+    return existing
+
+  def _GenerateUniqueID(self, exceptions=None):
+    """Generate an unique UUID.

 
     This checks the current node, instances and disk names for
     duplicates.
 
 
     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)
+    @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
 
     """
 
     @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

@@ -191,6 +206,22 @@ class ConfigWriter:
     self._temporary_ids.add(unique_id)
     return unique_id
 
     self._temporary_ids.add(unique_id)
     return unique_id
 

+  @locking.ssynchronized(_config_lock, shared=1)
+  def GenerateUniqueID(self, exceptions=None):
+    """Generate an unique ID.
+
+    This is just a wrapper over the unlocked version.
+
+    """
+    return self._GenerateUniqueID(exceptions=exceptions)
+
+  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.
 

@@ -227,15 +258,64 @@ class ConfigWriter:
 
     return result
 
 
     return result
 

-  @locking.ssynchronized(_config_lock, shared=1)
-  def VerifyConfig(self):
+  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.
 
     """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:

@@ -266,6 +346,12 @@ 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
     for free_port in data.cluster.tcpudp_port_pool:
       if free_port not in ports:
     # cluster-wide pool of free ports
     for free_port in data.cluster.tcpudp_port_pool:
       if free_port not in ports:

@@ -290,13 +376,58 @@ class ConfigWriter:
     if not data.nodes[data.cluster.master_node].master_candidate:
       result.append("Master node is not a master candidate")
 
     if not data.nodes[data.cluster.master_node].master_candidate:
       result.append("Master node is not a master candidate")
 

-    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+    # 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))
 
     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))
+
+    # IP checks
+    ips = { data.cluster.master_ip: ["cluster_ip"] }
+    def _helper(ip, name):
+      if ip in ips:
+        ips[ip].append(name)
+      else:
+        ips[ip] = [name]
+
+    for node in data.nodes.values():
+      _helper(node.primary_ip, "node:%s/primary" % node.name)
+      if node.secondary_ip != node.primary_ip:
+        _helper(node.secondary_ip, "node:%s/secondary" % node.name)
+
+    for ip, owners in ips.items():
+      if len(owners) > 1:
+        result.append("IP address %s is used by multiple owners: %s" %
+                      (ip, ", ".join(owners)))

     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.
 

@@ -392,34 +523,41 @@ class ConfigWriter:
   def _UnlockedComputeDRBDMap(self):
     """Compute the used DRBD minor/nodes.
 
   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
     @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).
+        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):
 
   @locking.ssynchronized(_config_lock)
   def ComputeDRBDMap(self):

@@ -432,7 +570,11 @@ class ConfigWriter:
         an empty list).
 
     """
         an empty list).
 
     """

-    return self._UnlockedComputeDRBDMap()
+    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):

@@ -448,9 +590,12 @@ class ConfigWriter:
 
     """
     assert isinstance(instance, basestring), \
 
     """
     assert isinstance(instance, basestring), \

-           "Invalid argument passed to AllocateDRBDMinor"
+           "Invalid argument '%s' passed to AllocateDRBDMinor" % instance

 
 

-    d_map = self._UnlockedComputeDRBDMap()
+    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]

@@ -469,22 +614,27 @@ 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

@@ -496,6 +646,23 @@ class ConfigWriter:
       if name == instance:
         del self._temporary_drbds[key]
 
       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.

@@ -551,7 +718,7 @@ class ConfigWriter:
     """Get the hypervisor type for this cluster.
 
     """
     """Get the hypervisor type for this cluster.
 
     """

-    return self._config_data.cluster.default_hypervisor
+    return self._config_data.cluster.enabled_hypervisors[0]

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

@@ -580,26 +747,51 @@ 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))
+
+    self._EnsureUUID(instance)
+

     instance.serial_no = 1
     instance.serial_no = 1

+    instance.ctime = instance.mtime = time.time()

     self._config_data.instances[instance.name] = instance
     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()
 
     self._WriteConfig()
 

+  def _EnsureUUID(self, item):
+    """Ensures a given object has a valid UUID.
+
+    @param item: the instance or node to be checked
+
+    """
+    if not item.uuid:
+      item.uuid = self._GenerateUniqueID()
+    elif item.uuid in self._AllIDs(temporary=True):
+      raise errors.ConfigurationError("Cannot add '%s': UUID already in use" %
+                                      (item.name, item.uuid))
+

   def _SetInstanceStatus(self, instance_name, status):
     """Set the instance's status to a given value.
 
     """
   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
       instance.serial_no += 1

+      instance.mtime = time.time()

       self._WriteConfig()
 
   @locking.ssynchronized(_config_lock)
       self._WriteConfig()
 
   @locking.ssynchronized(_config_lock)

@@ -607,7 +799,7 @@ 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):

@@ -617,6 +809,7 @@ 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)

@@ -651,7 +844,7 @@ 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.

@@ -677,10 +870,11 @@ class ConfigWriter:
 
     """
     return utils.MatchNameComponent(short_name,
 
     """
     return utils.MatchNameComponent(short_name,

-                                    self._config_data.instances.keys())
+                                    self._config_data.instances.keys(),
+                                    case_sensitive=False)

 
   def _UnlockedGetInstanceInfo(self, instance_name):
 
   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.
 

@@ -692,9 +886,9 @@ 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.
 
     @param instance_name: name of the instance, e.g.
     an instance are taken from the live systems.
 
     @param instance_name: name of the instance, e.g.

@@ -711,7 +905,7 @@ 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
 
     """

@@ -727,9 +921,12 @@ class ConfigWriter:
     @param node: a Node instance
 
     """
     @param node: a Node instance
 
     """

-    logging.info("Adding node %s to configuration" % node.name)
+    logging.info("Adding node %s to configuration", node.name)
+
+    self._EnsureUUID(node)

 
     node.serial_no = 1
 
     node.serial_no = 1

+    node.ctime = node.mtime = time.time()

     self._config_data.nodes[node.name] = node
     self._config_data.cluster.serial_no += 1
     self._WriteConfig()
     self._config_data.nodes[node.name] = node
     self._config_data.cluster.serial_no += 1
     self._WriteConfig()

@@ -739,7 +936,7 @@ class ConfigWriter:
     """Remove a node from the configuration.
 
     """
     """Remove a node from the configuration.
 
     """

-    logging.info("Removing node %s from configuration" % node_name)
+    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)
 
     if node_name not in self._config_data.nodes:
       raise errors.ConfigurationError("Unknown node '%s'" % node_name)

@@ -754,7 +951,8 @@ class ConfigWriter:
 
     """
     return utils.MatchNameComponent(short_name,
 
     """
     return utils.MatchNameComponent(short_name,

-                                    self._config_data.nodes.keys())
+                                    self._config_data.nodes.keys(),
+                                    case_sensitive=False)

 
   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.

@@ -829,43 +1027,51 @@ class ConfigWriter:
                     for node in self._UnlockedGetNodeList()])
     return my_dict
 
                     for node in self._UnlockedGetNodeList()])
     return my_dict
 

-  def _UnlockedGetMasterCandidateStats(self):
+  def _UnlockedGetMasterCandidateStats(self, exceptions=None):

     """Get the number of current and maximum desired and possible candidates.
 
     """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
     @rtype: tuple

-    @return: tuple of (current, desired and possible)
+    @return: tuple of (current, desired and possible, possible)

 
     """
 
     """

-    mc_now = mc_max = 0
-    for node in self._config_data.nodes.itervalues():
-      if not node.offline:
+    mc_now = mc_should = 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 += 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)
+    mc_should = min(mc_max, self._config_data.cluster.candidate_pool_size)
+    return (mc_now, mc_should, mc_max)

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

-  def GetMasterCandidateStats(self):
+  def GetMasterCandidateStats(self, exceptions=None):

     """Get the number of current and maximum possible candidates.
 
     This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
 
     """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)
 
     """
     @rtype: tuple
     @return: tuple of (current, max)
 
     """

-    return self._UnlockedGetMasterCandidateStats()
+    return self._UnlockedGetMasterCandidateStats(exceptions)

 
   @locking.ssynchronized(_config_lock)
 
   @locking.ssynchronized(_config_lock)

-  def MaintainCandidatePool(self):
+  def MaintainCandidatePool(self, exceptions):

     """Try to grow the candidate pool to the desired size.
 
     """Try to grow the candidate pool to the desired size.
 

+    @type exceptions: list
+    @param exceptions: if passed, list of nodes that should be ignored

     @rtype: list
     @return: list with the adjusted nodes (L{objects.Node} instances)
 
     """
     @rtype: list
     @return: list with the adjusted nodes (L{objects.Node} instances)
 
     """

-    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+    mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats(exceptions)

     mod_list = []
     if mc_now < mc_max:
       node_list = self._config_data.nodes.keys()
     mod_list = []
     if mc_now < mc_max:
       node_list = self._config_data.nodes.keys()

@@ -874,7 +1080,8 @@ class ConfigWriter:
         if mc_now >= mc_max:
           break
         node = self._config_data.nodes[name]
         if mc_now >= mc_max:
           break
         node = self._config_data.nodes[name]

-        if node.master_candidate or node.offline:
+        if (node.master_candidate or node.offline or node.drained or
+            node.name in exceptions):

           continue
         mod_list.append(node)
         node.master_candidate = True
           continue
         mod_list.append(node)
         node.master_candidate = True

@@ -895,19 +1102,26 @@ class ConfigWriter:
 
     """
     self._config_data.serial_no += 1
 
     """
     self._config_data.serial_no += 1

+    self._config_data.mtime = time.time()
+
+  def _AllUUIDObjects(self):
+    """Returns all objects with uuid attributes.
+
+    """
+    return (self._config_data.instances.values() +
+            self._config_data.nodes.values() +
+            [self._config_data.cluster])

 
   def _OpenConfig(self):
     """Read the config data from disk.
 
     """
 
   def _OpenConfig(self):
     """Read the config data from disk.
 
     """

-    f = open(self._cfg_file, 'r')
+    raw_data = utils.ReadFile(self._cfg_file)
+

     try:
     try:

-      try:
-        data = objects.ConfigData.FromDict(serializer.Load(f.read()))
-      except Exception, err:
-        raise errors.ConfigurationError(err)
-    finally:
-      f.close()
+      data = objects.ConfigData.FromDict(serializer.Load(raw_data))
+    except Exception, err:
+      raise errors.ConfigurationError(err)

 
     # Make sure the configuration has the right version
     _ValidateConfig(data)
 
     # Make sure the configuration has the right version
     _ValidateConfig(data)

@@ -916,12 +1130,38 @@ 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
     # reset the last serial as -1 so that the next write will cause
     # ssconf update
     self._last_cluster_serial = -1
 
     self._config_data = data
     # reset the last serial as -1 so that the next write will cause
     # ssconf update
     self._last_cluster_serial = -1
 

-  def _DistributeConfig(self):
+    # And finally run our (custom) config upgrade sequence
+    self._UpgradeConfig()
+
+  def _UpgradeConfig(self):
+    """Run upgrade steps that cannot be done purely in the objects.
+
+    This is because some data elements need uniqueness across the
+    whole configuration, etc.
+
+    @warning: this function will call L{_WriteConfig()}, so it needs
+        to either be called with the lock held or from a safe place
+        (the constructor)
+
+    """
+    modified = False
+    for item in self._AllUUIDObjects():
+      if item.uuid is None:
+        item.uuid = self._GenerateUniqueID()
+        modified = True
+    if modified:
+      self._WriteConfig()
+
+  def _DistributeConfig(self, feedback_fn):

     """Distribute the configuration to the other nodes.
 
     Currently, this only copies the configuration file. In the future,
     """Distribute the configuration to the other nodes.
 
     Currently, this only copies the configuration file. In the future,

@@ -930,6 +1170,7 @@ class ConfigWriter:
     """
     if self._offline:
       return True
     """
     if self._offline:
       return True

+

     bad = False
 
     node_list = []
     bad = False
 
     node_list = []

@@ -950,41 +1191,72 @@ class ConfigWriter:
 
     result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file,
                                             address_list=addr_list)
 
     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)
+    for to_node, to_result in result.items():
+      msg = to_result.fail_msg
+      if msg:
+        msg = ("Copy of file %s to node %s failed: %s" %
+               (self._cfg_file, to_node, msg))
+        logging.error(msg)
+
+        if feedback_fn:
+          feedback_fn(msg)
+

         bad = True
         bad = True

+

     return not bad
 
     return not bad
 

-  def _WriteConfig(self, destination=None):
+  def _WriteConfig(self, destination=None, feedback_fn=None):

     """Write the configuration data to persistent storage.
 
     """
     """Write the configuration data to persistent storage.
 
     """

+    assert feedback_fn is None or callable(feedback_fn)
+
+    # 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()
+
+    # Warn on config errors, but don't abort the save - the
+    # configuration has already been modified, and we can't revert;
+    # the best we can do is to warn the user and save as is, leaving
+    # recovery to the user
+    config_errors = self._UnlockedVerifyConfig()
+    if config_errors:
+      errmsg = ("Configuration data is not consistent: %s" %
+                (", ".join(config_errors)))
+      logging.critical(errmsg)
+      if feedback_fn:
+        feedback_fn(errmsg)
+

     if destination is None:
       destination = self._cfg_file
     self._BumpSerialNo()
     txt = serializer.Dump(self._config_data.ToDict())
     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:
-      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)
+
+    utils.WriteFile(destination, data=txt)
+

     self.write_count += 1
 
     # and redistribute the config file to master candidates
     self.write_count += 1
 
     # and redistribute the config file to master candidates

-    self._DistributeConfig()
+    self._DistributeConfig(feedback_fn)

 
     # Write ssconf files on all nodes (including locally)
     if self._last_cluster_serial < self._config_data.cluster.serial_no:
       if not self._offline:
 
     # Write ssconf files on all nodes (including locally)
     if self._last_cluster_serial < self._config_data.cluster.serial_no:
       if not self._offline:

-        rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(),
-                                              self._UnlockedGetSsconfValues())
+        result = rpc.RpcRunner.call_write_ssconf_files(
+          self._UnlockedGetNodeList(),
+          self._UnlockedGetSsconfValues())
+
+        for nname, nresu in result.items():
+          msg = nresu.fail_msg
+          if msg:
+            errmsg = ("Error while uploading ssconf files to"
+                      " node %s: %s" % (nname, msg))
+            logging.warning(errmsg)
+
+            if feedback_fn:
+              feedback_fn(errmsg)
+

       self._last_cluster_serial = self._config_data.cluster.serial_no
 
   def _UnlockedGetSsconfValues(self):
       self._last_cluster_serial = self._config_data.cluster.serial_no
 
   def _UnlockedGetSsconfValues(self):

@@ -996,52 +1268,44 @@ class ConfigWriter:
 
     """
     fn = "\n".join
 
     """
     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]
     node_names = utils.NiceSort(self._UnlockedGetNodeList())
     node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]

+    node_pri_ips = ["%s %s" % (ninfo.name, ninfo.primary_ip)
+                    for ninfo in node_info]
+    node_snd_ips = ["%s %s" % (ninfo.name, ninfo.secondary_ip)
+                    for ninfo in node_info]

 
 

+    instance_data = fn(instance_names)

     off_data = fn(node.name for node in node_info if node.offline)
     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)
     mc_data = fn(node.name for node in node_info if node.master_candidate)

+    mc_ips_data = fn(node.primary_ip for node in node_info
+                     if node.master_candidate)

     node_data = fn(node_names)
     node_data = fn(node_names)

+    node_pri_ips_data = fn(node_pri_ips)
+    node_snd_ips_data = fn(node_snd_ips)

 
     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_MASTER_CANDIDATES: mc_data,
       constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
       constants.SS_MASTER_CANDIDATES: mc_data,

+      constants.SS_MASTER_CANDIDATES_IPS: mc_ips_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_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_NODE_PRIMARY_IPS: node_pri_ips_data,
+      constants.SS_NODE_SECONDARY_IPS: node_snd_ips_data,

       constants.SS_OFFLINE_NODES: off_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,
       }
 
       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.

@@ -1059,13 +1323,6 @@ class ConfigWriter:
     self._WriteConfig()
 
   @locking.ssynchronized(_config_lock, shared=1)
     self._WriteConfig()
 
   @locking.ssynchronized(_config_lock, shared=1)

-  def GetDefBridge(self):
-    """Return the default bridge.
-
-    """
-    return self._config_data.cluster.default_bridge
-
-  @locking.ssynchronized(_config_lock, shared=1)

   def GetMACPrefix(self):
     """Return the mac prefix.
 
   def GetMACPrefix(self):
     """Return the mac prefix.
 

@@ -1074,7 +1331,7 @@ 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

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

@@ -1083,7 +1340,7 @@ class ConfigWriter:
     return self._config_data.cluster
 
   @locking.ssynchronized(_config_lock)
     return self._config_data.cluster
 
   @locking.ssynchronized(_config_lock)

-  def Update(self, target):
+  def Update(self, target, feedback_fn):

     """Notify function to be called after updates.
 
     This function must be called when an object (as returned by
     """Notify function to be called after updates.
 
     This function must be called when an object (as returned by

@@ -1095,6 +1352,7 @@ class ConfigWriter:
     @param target: an instance of either L{objects.Cluster},
         L{objects.Node} or L{objects.Instance} which is existing in
         the cluster
     @param target: an instance of either L{objects.Cluster},
         L{objects.Node} or L{objects.Instance} which is existing in
         the cluster

+    @param feedback_fn: Callable feedback function

 
     """
     if self._config_data is None:
 
     """
     if self._config_data is None:

@@ -1115,9 +1373,16 @@ class ConfigWriter:
       raise errors.ConfigurationError("Configuration updated since object"
                                       " has been read or unknown object")
     target.serial_no += 1
       raise errors.ConfigurationError("Configuration updated since object"
                                       " has been read or unknown object")
     target.serial_no += 1

+    target.mtime = now = time.time()

 
     if update_serial:
       # for node updates, we need to increase the cluster serial too
       self._config_data.cluster.serial_no += 1
 
     if update_serial:
       # for node updates, we need to increase the cluster serial too
       self._config_data.cluster.serial_no += 1

+      self._config_data.cluster.mtime = now

 
 

-    self._WriteConfig()
+    if isinstance(target, objects.Instance):
+      self._UnlockedReleaseDRBDMinors(target.name)
+      for nic in target.nics:
+        self._temporary_macs.discard(nic.mac)
+
+    self._WriteConfig(feedback_fn=feedback_fn)