Fix gnt-os for offline nodes

[ganeti-local] / lib / config.py

diff --git a/lib/config.py b/lib/config.py
index 739f761..986b03a 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" %

@@ -76,6 +84,7 @@ class ConfigWriter:
     # better to raise an error before starting to modify the config
     # file than after it was modified
     self._my_hostname = utils.HostInfo().name
     # 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

@@ -155,13 +164,13 @@ 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 = set()

@@ -185,6 +194,9 @@ class ConfigWriter:
   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 +208,9 @@ 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."""

@@ -214,7 +229,8 @@ class ConfigWriter:
 
   @locking.ssynchronized(_config_lock, shared=1)
   def VerifyConfig(self):
 
   @locking.ssynchronized(_config_lock, shared=1)
   def VerifyConfig(self):

-    """Stub verify function.
+    """Verify function.
+

     """
     result = []
     seen_macs = []
     """
     result = []
     seen_macs = []

@@ -251,7 +267,7 @@ class ConfigWriter:
         ports[net_port].append((instance.name, "network port"))
 
     # cluster-wide pool of free ports
         ports[net_port].append((instance.name, "network port"))
 
     # cluster-wide pool of free ports

-    for free_port in self._config_data.cluster.tcpudp_port_pool:
+    for free_port in data.cluster.tcpudp_port_pool:

       if free_port not in ports:
         ports[free_port] = []
       ports[free_port].append(("cluster", "port marked as free"))
       if free_port not in ports:
         ports[free_port] = []
       ports[free_port].append(("cluster", "port marked as free"))

@@ -267,10 +283,17 @@ 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")
+
+    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+    if mc_now < mc_max:
+      result.append("Not enough master candidates: actual %d, target %d" %
+                    (mc_now, mc_max))

 
     return result
 
 
     return result
 

@@ -369,9 +392,9 @@ class ConfigWriter:
   def _ComputeDRBDMap(self, instance):
     """Compute the used DRBD minor/nodes.
 
   def _ComputeDRBDMap(self, instance):
     """Compute the used DRBD minor/nodes.
 

-    Return: dictionary of node_name: dict of minor: instance_name. The
-    returned dict will have all the nodes in it (even if with an empty
-    list).
+    @return: dictionary of node_name: dict of minor: instance_name;
+        the returned dict will have all the nodes in it (even if with
+        an empty list).

 
     """
     def _AppendUsedPorts(instance_name, disk, used):
 
     """
     def _AppendUsedPorts(instance_name, disk, used):

@@ -513,9 +536,9 @@ 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 +548,9 @@ 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")

@@ -620,9 +644,8 @@ 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()

@@ -653,11 +676,11 @@ class ConfigWriter:
     It takes the information from the configuration file. Other informations of
     an instance are taken from the live systems.
 
     It takes the information from the configuration file. Other informations of
     an instance are taken from the live systems.
 

-    Args:
-      instance: name of the instance, ex instance1.example.com
+    @param instance_name: name of the instance, e.g.
+        I{instance1.example.com}

 
 

-    Returns:
-      the instance object
+    @rtype: L{objects.Instance}
+    @return: the instance object

 
     """
     return self._UnlockedGetInstanceInfo(instance_name)
 
     """
     return self._UnlockedGetInstanceInfo(instance_name)

@@ -679,8 +702,8 @@ 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 +738,13 @@ 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 +757,12 @@ 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 +770,10 @@ 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 +787,20 @@ 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 +808,67 @@ class ConfigWriter:
                     for node in self._UnlockedGetNodeList()])
     return my_dict
 
                     for node in self._UnlockedGetNodeList()])
     return my_dict
 

+  def _UnlockedGetMasterCandidateStats(self):
+    """Get the number of current and maximum desired and possible candidates.
+
+    @rtype: tuple
+    @return: tuple of (current, desired and possible)
+
+    """
+    mc_now = mc_max = 0
+    for node in self._config_data.nodes.itervalues():
+      if not node.offline:
+        mc_max += 1
+      if node.master_candidate:
+        mc_now += 1
+    mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size)
+    return (mc_now, mc_max)
+
+  @locking.ssynchronized(_config_lock, shared=1)
+  def GetMasterCandidateStats(self):
+    """Get the number of current and maximum possible candidates.
+
+    This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
+
+    @rtype: tuple
+    @return: tuple of (current, max)
+
+    """
+    return self._UnlockedGetMasterCandidateStats()
+
+  @locking.ssynchronized(_config_lock)
+  def MaintainCandidatePool(self):
+    """Try to grow the candidate pool to the desired size.
+
+    @rtype: list
+    @return: list with the adjusted nodes (L{objects.Node} instances)
+
+    """
+    mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
+    mod_list = []
+    if mc_now < mc_max:
+      node_list = self._config_data.nodes.keys()
+      random.shuffle(node_list)
+      for name in node_list:
+        if mc_now >= mc_max:
+          break
+        node = self._config_data.nodes[name]
+        if node.master_candidate or node.offline:
+          continue
+        mod_list.append(node)
+        node.master_candidate = True
+        node.serial_no += 1
+        mc_now += 1
+      if mc_now != mc_max:
+        # this should not happen
+        logging.warning("Warning: MaintainCandidatePool didn't manage to"
+                        " fill the candidate pool (%d/%d)", mc_now, mc_max)
+      if mod_list:
+        self._config_data.cluster.serial_no += 1
+        self._WriteConfig()
+
+    return mod_list
+

   def _BumpSerialNo(self):
     """Bump up the serial number of the config.
 
   def _BumpSerialNo(self):
     """Bump up the serial number of the config.
 

@@ -777,10 +878,6 @@ 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:

@@ -799,7 +896,7 @@ class ConfigWriter:
       raise errors.ConfigurationError("Incomplete configuration"
                                       " (missing cluster.rsahostkeypub)")
     self._config_data = data
       raise errors.ConfigurationError("Incomplete configuration"
                                       " (missing cluster.rsahostkeypub)")
     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 +910,26 @@ 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)

@@ -856,7 +956,7 @@ 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)

@@ -867,11 +967,31 @@ class ConfigWriter:
       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):

+    """Return the values needed by ssconf.
+
+    @rtype: dict
+    @return: a dictionary with keys the ssconf names and values their
+        associated value
+
+    """
+    fn = "\n".join
+    node_names = utils.NiceSort(self._UnlockedGetNodeList())
+    node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]
+
+    off_data = fn(node.name for node in node_info if node.offline)
+    mc_data = fn(node.name for node in node_info if node.master_candidate)
+    node_data = fn(node_names)
+
+    cluster = self._config_data.cluster

     return {
     return {

-      "cluster_name": self._config_data.cluster.cluster_name,
-      "master_ip": self._config_data.cluster.master_ip,
-      "master_netdev": self._config_data.cluster.master_netdev,
-      "master_node": self._config_data.cluster.master_node,
+      constants.SS_CLUSTER_NAME: cluster.cluster_name,
+      constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
+      constants.SS_MASTER_CANDIDATES: mc_data,
+      constants.SS_MASTER_IP: cluster.master_ip,
+      constants.SS_MASTER_NETDEV: cluster.master_netdev,
+      constants.SS_MASTER_NODE: cluster.master_node,
+      constants.SS_NODE_LIST: node_data,
+      constants.SS_OFFLINE_NODES: off_data,

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

@@ -934,8 +1054,8 @@ class ConfigWriter:
   def GetClusterInfo(self):
     """Returns informations about the cluster
 
   def GetClusterInfo(self):
     """Returns informations 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

@@ -950,25 +1070,32 @@ 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,"
                                    " cannot save.")
     """
     if self._config_data is None:
       raise errors.ProgrammerError("Configuration file not read,"
                                    " cannot save.")

-    update_serial = True
+    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()
     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()
     elif isinstance(target, objects.Instance):
       test = target in self._config_data.instances.values()

-      update_serial = False

     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")
     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:
     if update_serial:

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

 
     self._WriteConfig()
 
     self._WriteConfig()