Move all iallocator functions into a class

[ganeti-local] / lib / cmdlib.py

diff --git a/lib/cmdlib.py b/lib/cmdlib.py
index 85f1651..d20f0d1 100644 (file)
--- a/lib/cmdlib.py
+++ b/lib/cmdlib.py
@@ -30,7 +30,6 @@ import time
 import tempfile
 import re
 import platform
 import tempfile
 import re
 import platform

-import simplejson

 
 from ganeti import rpc
 from ganeti import ssh
 
 from ganeti import rpc
 from ganeti import ssh

@@ -43,14 +42,7 @@ from ganeti import constants
 from ganeti import objects
 from ganeti import opcodes
 from ganeti import ssconf
 from ganeti import objects
 from ganeti import opcodes
 from ganeti import ssconf

-
-
-# Check whether the simplejson module supports indentation
-_JSON_INDENT = 2
-try:
-  simplejson.dumps(1, indent=_JSON_INDENT)
-except TypeError:
-  _JSON_INDENT = None
+from ganeti import serializer

 
 
 class LogicalUnit(object):
 
 
 class LogicalUnit(object):

@@ -658,7 +650,7 @@ class LUVerifyCluster(NoHooksLU):
   """Verifies the cluster status.
 
   """
   """Verifies the cluster status.
 
   """

-  _OP_REQP = []
+  _OP_REQP = ["skip_checks"]

 
   def _VerifyNode(self, node, file_list, local_cksum, vglist, node_result,
                   remote_version, feedback_fn):
 
   def _VerifyNode(self, node, file_list, local_cksum, vglist, node_result,
                   remote_version, feedback_fn):

@@ -679,7 +671,7 @@ class LUVerifyCluster(NoHooksLU):
     # compares ganeti version
     local_version = constants.PROTOCOL_VERSION
     if not remote_version:
     # compares ganeti version
     local_version = constants.PROTOCOL_VERSION
     if not remote_version:

-      feedback_fn(" - ERROR: connection to %s failed" % (node))
+      feedback_fn("  - ERROR: connection to %s failed" % (node))

       return True
 
     if local_version != remote_version:
       return True
 
     if local_version != remote_version:

@@ -730,7 +722,8 @@ class LUVerifyCluster(NoHooksLU):
       feedback_fn("  - ERROR: hypervisor verify failure: '%s'" % hyp_result)
     return bad
 
       feedback_fn("  - ERROR: hypervisor verify failure: '%s'" % hyp_result)
     return bad
 

-  def _VerifyInstance(self, instance, node_vol_is, node_instance, feedback_fn):
+  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
+                      node_instance, feedback_fn):

     """Verify an instance.
 
     This function checks to see if the required block devices are
     """Verify an instance.
 
     This function checks to see if the required block devices are

@@ -739,13 +732,6 @@ class LUVerifyCluster(NoHooksLU):
     """
     bad = False
 
     """
     bad = False
 

-    instancelist = self.cfg.GetInstanceList()
-    if not instance in instancelist:
-      feedback_fn("  - ERROR: instance %s not in instance list %s" %
-                      (instance, instancelist))
-      bad = True
-
-    instanceconfig = self.cfg.GetInstanceInfo(instance)

     node_current = instanceconfig.primary_node
 
     node_vol_should = {}
     node_current = instanceconfig.primary_node
 
     node_vol_should = {}

@@ -759,7 +745,8 @@ class LUVerifyCluster(NoHooksLU):
           bad = True
 
     if not instanceconfig.status == 'down':
           bad = True
 
     if not instanceconfig.status == 'down':

-      if not instance in node_instance[node_current]:
+      if (node_current not in node_instance or
+          not instance in node_instance[node_current]):

         feedback_fn("  - ERROR: instance %s not running on node %s" %
                         (instance, node_current))
         bad = True
         feedback_fn("  - ERROR: instance %s not running on node %s" %
                         (instance, node_current))
         bad = True

@@ -805,13 +792,44 @@ class LUVerifyCluster(NoHooksLU):
           bad = True
     return bad
 
           bad = True
     return bad
 

+  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
+    """Verify N+1 Memory Resilience.
+
+    Check that if one single node dies we can still start all the instances it
+    was primary for.
+
+    """
+    bad = False
+
+    for node, nodeinfo in node_info.iteritems():
+      # This code checks that every node which is now listed as secondary has
+      # enough memory to host all instances it is supposed to should a single
+      # other node in the cluster fail.
+      # FIXME: not ready for failover to an arbitrary node
+      # FIXME: does not support file-backed instances
+      # WARNING: we currently take into account down instances as well as up
+      # ones, considering that even if they're down someone might want to start
+      # them even in the event of a node failure.
+      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
+        needed_mem = 0
+        for instance in instances:
+          needed_mem += instance_cfg[instance].memory
+        if nodeinfo['mfree'] < needed_mem:
+          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
+                      " failovers should node %s fail" % (node, prinode))
+          bad = True
+    return bad
+

   def CheckPrereq(self):
     """Check prerequisites.
 
   def CheckPrereq(self):
     """Check prerequisites.
 

-    This has no prerequisites.
+    Transform the list of checks we're going to skip into a set and check that
+    all its members are valid.

 
     """
 
     """

-    pass
+    self.skip_set = frozenset(self.op.skip_checks)
+    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
+      raise errors.OpPrereqError("Invalid checks to be skipped specified")

 
   def Exec(self, feedback_fn):
     """Verify integrity of cluster, performing various test on nodes.
 
   def Exec(self, feedback_fn):
     """Verify integrity of cluster, performing various test on nodes.

@@ -825,8 +843,11 @@ class LUVerifyCluster(NoHooksLU):
     vg_name = self.cfg.GetVGName()
     nodelist = utils.NiceSort(self.cfg.GetNodeList())
     instancelist = utils.NiceSort(self.cfg.GetInstanceList())
     vg_name = self.cfg.GetVGName()
     nodelist = utils.NiceSort(self.cfg.GetNodeList())
     instancelist = utils.NiceSort(self.cfg.GetInstanceList())

+    i_non_redundant = [] # Non redundant instances

     node_volume = {}
     node_instance = {}
     node_volume = {}
     node_instance = {}

+    node_info = {}
+    instance_cfg = {}

 
     # FIXME: verify OS list
     # do local checksums
 
     # FIXME: verify OS list
     # do local checksums

@@ -846,6 +867,7 @@ class LUVerifyCluster(NoHooksLU):
       }
     all_nvinfo = rpc.call_node_verify(nodelist, node_verify_param)
     all_rversion = rpc.call_version(nodelist)
       }
     all_nvinfo = rpc.call_node_verify(nodelist, node_verify_param)
     all_rversion = rpc.call_version(nodelist)

+    all_ninfo = rpc.call_node_info(nodelist, self.cfg.GetVGName())

 
     for node in nodelist:
       feedback_fn("* Verifying node %s" % node)
 
     for node in nodelist:
       feedback_fn("* Verifying node %s" % node)

@@ -878,18 +900,74 @@ class LUVerifyCluster(NoHooksLU):
 
       node_instance[node] = nodeinstance
 
 
       node_instance[node] = nodeinstance
 

+      # node_info
+      nodeinfo = all_ninfo[node]
+      if not isinstance(nodeinfo, dict):
+        feedback_fn("  - ERROR: connection to %s failed" % (node,))
+        bad = True
+        continue
+
+      try:
+        node_info[node] = {
+          "mfree": int(nodeinfo['memory_free']),
+          "dfree": int(nodeinfo['vg_free']),
+          "pinst": [],
+          "sinst": [],
+          # dictionary holding all instances this node is secondary for,
+          # grouped by their primary node. Each key is a cluster node, and each
+          # value is a list of instances which have the key as primary and the
+          # current node as secondary.  this is handy to calculate N+1 memory
+          # availability if you can only failover from a primary to its
+          # secondary.
+          "sinst-by-pnode": {},
+        }
+      except ValueError:
+        feedback_fn("  - ERROR: invalid value returned from node %s" % (node,))
+        bad = True
+        continue
+

     node_vol_should = {}
 
     for instance in instancelist:
       feedback_fn("* Verifying instance %s" % instance)
     node_vol_should = {}
 
     for instance in instancelist:
       feedback_fn("* Verifying instance %s" % instance)

-      result =  self._VerifyInstance(instance, node_volume, node_instance,
-                                     feedback_fn)
-      bad = bad or result
-

       inst_config = self.cfg.GetInstanceInfo(instance)
       inst_config = self.cfg.GetInstanceInfo(instance)

+      result =  self._VerifyInstance(instance, inst_config, node_volume,
+                                     node_instance, feedback_fn)
+      bad = bad or result

 
       inst_config.MapLVsByNode(node_vol_should)
 
 
       inst_config.MapLVsByNode(node_vol_should)
 

+      instance_cfg[instance] = inst_config
+
+      pnode = inst_config.primary_node
+      if pnode in node_info:
+        node_info[pnode]['pinst'].append(instance)
+      else:
+        feedback_fn("  - ERROR: instance %s, connection to primary node"
+                    " %s failed" % (instance, pnode))
+        bad = True
+
+      # If the instance is non-redundant we cannot survive losing its primary
+      # node, so we are not N+1 compliant. On the other hand we have no disk
+      # templates with more than one secondary so that situation is not well
+      # supported either.
+      # FIXME: does not support file-backed instances
+      if len(inst_config.secondary_nodes) == 0:
+        i_non_redundant.append(instance)
+      elif len(inst_config.secondary_nodes) > 1:
+        feedback_fn("  - WARNING: multiple secondaries for instance %s"
+                    % instance)
+
+      for snode in inst_config.secondary_nodes:
+        if snode in node_info:
+          node_info[snode]['sinst'].append(instance)
+          if pnode not in node_info[snode]['sinst-by-pnode']:
+            node_info[snode]['sinst-by-pnode'][pnode] = []
+          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
+        else:
+          feedback_fn("  - ERROR: instance %s, connection to secondary node"
+                      " %s failed" % (instance, snode))
+

     feedback_fn("* Verifying orphan volumes")
     result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
                                        feedback_fn)
     feedback_fn("* Verifying orphan volumes")
     result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
                                        feedback_fn)

@@ -900,6 +978,16 @@ class LUVerifyCluster(NoHooksLU):
                                          feedback_fn)
     bad = bad or result
 
                                          feedback_fn)
     bad = bad or result
 

+    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
+      feedback_fn("* Verifying N+1 Memory redundancy")
+      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
+      bad = bad or result
+
+    feedback_fn("* Other Notes")
+    if i_non_redundant:
+      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
+                  % len(i_non_redundant))
+

     return int(bad)
 
 
     return int(bad)
 
 

@@ -2337,7 +2425,7 @@ class LURenameInstance(LogicalUnit):
     instance_list = self.cfg.GetInstanceList()
     if new_name in instance_list:
       raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
     instance_list = self.cfg.GetInstanceList()
     if new_name in instance_list:
       raise errors.OpPrereqError("Instance '%s' is already in the cluster" %

-                                 instance_name)
+                                 new_name)

 
     if not getattr(self.op, "ignore_ip", False):
       command = ["fping", "-q", name_info.ip]
 
     if not getattr(self.op, "ignore_ip", False):
       command = ["fping", "-q", name_info.ip]

@@ -2903,16 +2991,80 @@ def _RemoveDisks(instance, cfg):
   return result
 
 
   return result
 
 

+def _ComputeDiskSize(disk_template, disk_size, swap_size):
+  """Compute disk size requirements in the volume group
+
+  This is currently hard-coded for the two-drive layout.
+
+  """
+  # Required free disk space as a function of disk and swap space
+  req_size_dict = {
+    constants.DT_DISKLESS: None,
+    constants.DT_PLAIN: disk_size + swap_size,
+    constants.DT_LOCAL_RAID1: (disk_size + swap_size) * 2,
+    # 256 MB are added for drbd metadata, 128MB for each drbd device
+    constants.DT_REMOTE_RAID1: disk_size + swap_size + 256,
+    constants.DT_DRBD8: disk_size + swap_size + 256,
+  }
+
+  if disk_template not in req_size_dict:
+    raise errors.ProgrammerError("Disk template '%s' size requirement"
+                                 " is unknown" %  disk_template)
+
+  return req_size_dict[disk_template]
+
+

 class LUCreateInstance(LogicalUnit):
   """Create an instance.
 
   """
   HPATH = "instance-add"
   HTYPE = constants.HTYPE_INSTANCE
 class LUCreateInstance(LogicalUnit):
   """Create an instance.
 
   """
   HPATH = "instance-add"
   HTYPE = constants.HTYPE_INSTANCE

-  _OP_REQP = ["instance_name", "mem_size", "disk_size", "pnode",
+  _OP_REQP = ["instance_name", "mem_size", "disk_size",

               "disk_template", "swap_size", "mode", "start", "vcpus",
               "wait_for_sync", "ip_check", "mac"]
 
               "disk_template", "swap_size", "mode", "start", "vcpus",
               "wait_for_sync", "ip_check", "mac"]
 

+  def _RunAllocator(self):
+    """Run the allocator based on input opcode.
+
+    """
+    disks = [{"size": self.op.disk_size, "mode": "w"},
+             {"size": self.op.swap_size, "mode": "w"}]
+    nics = [{"mac": self.op.mac, "ip": getattr(self.op, "ip", None),
+             "bridge": self.op.bridge}]
+    ial = IAllocator(self.cfg, self.sstore,
+                     name=self.op.instance_name,
+                     disk_template=self.op.disk_template,
+                     tags=[],
+                     os=self.op.os_type,
+                     vcpus=self.op.vcpus,
+                     mem_size=self.op.mem_size,
+                     disks=disks,
+                     nics=nics,
+                     mode=constants.IALLOCATOR_MODE_ALLOC)
+
+    ial.Run(self.op.iallocator)
+
+    if not ial.success:
+      raise errors.OpPrereqError("Can't compute nodes using"
+                                 " iallocator '%s': %s" % (self.op.iallocator,
+                                                           ial.info))
+    req_nodes = 1
+    if self.op.disk_template in constants.DTS_NET_MIRROR:
+      req_nodes += 1
+
+    if len(ial.nodes) != req_nodes:
+      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
+                                 " of nodes (%s), required %s" %
+                                 (len(ial.nodes), req_nodes))
+    self.op.pnode = ial.nodes[0]
+    logger.ToStdout("Selected nodes for the instance: %s" %
+                    (", ".join(ial.nodes),))
+    logger.Info("Selected nodes for instance %s via iallocator %s: %s" %
+                (self.op.instance_name, self.op.iallocator, ial.nodes))
+    if req_nodes == 2:
+      self.op.snode = ial.nodes[1]
+

   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
 

@@ -2949,7 +3101,9 @@ class LUCreateInstance(LogicalUnit):
     """Check prerequisites.
 
     """
     """Check prerequisites.
 
     """

-    for attr in ["kernel_path", "initrd_path", "hvm_boot_order"]:
+    # set optional parameters to none if they don't exist
+    for attr in ["kernel_path", "initrd_path", "hvm_boot_order", "pnode",
+                 "iallocator"]:

       if not hasattr(self.op, attr):
         setattr(self.op, attr, None)
 
       if not hasattr(self.op, attr):
         setattr(self.op, attr, None)
 

@@ -2998,6 +3152,72 @@ class LUCreateInstance(LogicalUnit):
       if getattr(self.op, "os_type", None) is None:
         raise errors.OpPrereqError("No guest OS specified")
 
       if getattr(self.op, "os_type", None) is None:
         raise errors.OpPrereqError("No guest OS specified")
 

+    #### instance parameters check
+
+    # disk template and mirror node verification
+    if self.op.disk_template not in constants.DISK_TEMPLATES:
+      raise errors.OpPrereqError("Invalid disk template name")
+
+    # instance name verification
+    hostname1 = utils.HostInfo(self.op.instance_name)
+
+    self.op.instance_name = instance_name = hostname1.name
+    instance_list = self.cfg.GetInstanceList()
+    if instance_name in instance_list:
+      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
+                                 instance_name)
+
+    # ip validity checks
+    ip = getattr(self.op, "ip", None)
+    if ip is None or ip.lower() == "none":
+      inst_ip = None
+    elif ip.lower() == "auto":
+      inst_ip = hostname1.ip
+    else:
+      if not utils.IsValidIP(ip):
+        raise errors.OpPrereqError("given IP address '%s' doesn't look"
+                                   " like a valid IP" % ip)
+      inst_ip = ip
+    self.inst_ip = self.op.ip = inst_ip
+
+    if self.op.start and not self.op.ip_check:
+      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
+                                 " adding an instance in start mode")
+
+    if self.op.ip_check:
+      if utils.TcpPing(hostname1.ip, constants.DEFAULT_NODED_PORT):
+        raise errors.OpPrereqError("IP %s of instance %s already in use" %
+                                   (hostname1.ip, instance_name))
+
+    # MAC address verification
+    if self.op.mac != "auto":
+      if not utils.IsValidMac(self.op.mac.lower()):
+        raise errors.OpPrereqError("invalid MAC address specified: %s" %
+                                   self.op.mac)
+
+    # bridge verification
+    bridge = getattr(self.op, "bridge", None)
+    if bridge is None:
+      self.op.bridge = self.cfg.GetDefBridge()
+    else:
+      self.op.bridge = bridge
+
+    # boot order verification
+    if self.op.hvm_boot_order is not None:
+      if len(self.op.hvm_boot_order.strip("acdn")) != 0:
+        raise errors.OpPrereqError("invalid boot order specified,"
+                                   " must be one or more of [acdn]")
+    #### allocator run
+
+    if [self.op.iallocator, self.op.pnode].count(None) != 1:
+      raise errors.OpPrereqError("One and only one of iallocator and primary"
+                                 " node must be given")
+
+    if self.op.iallocator is not None:
+      self._RunAllocator()
+
+    #### node related checks
+

     # check primary node
     pnode = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.pnode))
     if pnode is None:
     # check primary node
     pnode = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.pnode))
     if pnode is None:

@@ -3006,10 +3226,8 @@ class LUCreateInstance(LogicalUnit):
     self.op.pnode = pnode.name
     self.pnode = pnode
     self.secondaries = []
     self.op.pnode = pnode.name
     self.pnode = pnode
     self.secondaries = []

-    # disk template and mirror node verification
-    if self.op.disk_template not in constants.DISK_TEMPLATES:
-      raise errors.OpPrereqError("Invalid disk template name")

 
 

+    # mirror node verification

     if self.op.disk_template in constants.DTS_NET_MIRROR:
       if getattr(self.op, "snode", None) is None:
         raise errors.OpPrereqError("The networked disk templates need"
     if self.op.disk_template in constants.DTS_NET_MIRROR:
       if getattr(self.op, "snode", None) is None:
         raise errors.OpPrereqError("The networked disk templates need"

@@ -3024,21 +3242,8 @@ class LUCreateInstance(LogicalUnit):
                                    " the primary node.")
       self.secondaries.append(snode_name)
 
                                    " the primary node.")
       self.secondaries.append(snode_name)
 

-    # Required free disk space as a function of disk and swap space
-    req_size_dict = {
-      constants.DT_DISKLESS: None,
-      constants.DT_PLAIN: self.op.disk_size + self.op.swap_size,
-      constants.DT_LOCAL_RAID1: (self.op.disk_size + self.op.swap_size) * 2,
-      # 256 MB are added for drbd metadata, 128MB for each drbd device
-      constants.DT_REMOTE_RAID1: self.op.disk_size + self.op.swap_size + 256,
-      constants.DT_DRBD8: self.op.disk_size + self.op.swap_size + 256,
-    }
-
-    if self.op.disk_template not in req_size_dict:
-      raise errors.ProgrammerError("Disk template '%s' size requirement"
-                                   " is unknown" %  self.op.disk_template)
-
-    req_size = req_size_dict[self.op.disk_template]
+    req_size = _ComputeDiskSize(self.op.disk_template,
+                                self.op.disk_size, self.op.swap_size)

 
     # Check lv size requirements
     if req_size is not None:
 
     # Check lv size requirements
     if req_size is not None:

@@ -3067,60 +3272,13 @@ class LUCreateInstance(LogicalUnit):
     if self.op.kernel_path == constants.VALUE_NONE:
       raise errors.OpPrereqError("Can't set instance kernel to none")
 
     if self.op.kernel_path == constants.VALUE_NONE:
       raise errors.OpPrereqError("Can't set instance kernel to none")
 

-    # instance verification
-    hostname1 = utils.HostInfo(self.op.instance_name)
-
-    self.op.instance_name = instance_name = hostname1.name
-    instance_list = self.cfg.GetInstanceList()
-    if instance_name in instance_list:
-      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
-                                 instance_name)
-
-    ip = getattr(self.op, "ip", None)
-    if ip is None or ip.lower() == "none":
-      inst_ip = None
-    elif ip.lower() == "auto":
-      inst_ip = hostname1.ip
-    else:
-      if not utils.IsValidIP(ip):
-        raise errors.OpPrereqError("given IP address '%s' doesn't look"
-                                   " like a valid IP" % ip)
-      inst_ip = ip
-    self.inst_ip = inst_ip
-
-    if self.op.start and not self.op.ip_check:
-      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
-                                 " adding an instance in start mode")
-
-    if self.op.ip_check:
-      if utils.TcpPing(hostname1.ip, constants.DEFAULT_NODED_PORT):
-        raise errors.OpPrereqError("IP %s of instance %s already in use" %
-                                   (hostname1.ip, instance_name))
-
-    # MAC address verification
-    if self.op.mac != "auto":
-      if not utils.IsValidMac(self.op.mac.lower()):
-        raise errors.OpPrereqError("invalid MAC address specified: %s" %
-                                   self.op.mac)
-
-    # bridge verification
-    bridge = getattr(self.op, "bridge", None)
-    if bridge is None:
-      self.op.bridge = self.cfg.GetDefBridge()
-    else:
-      self.op.bridge = bridge

 
 

+    # bridge check on primary node

     if not rpc.call_bridges_exist(self.pnode.name, [self.op.bridge]):
       raise errors.OpPrereqError("target bridge '%s' does not exist on"
                                  " destination node '%s'" %
                                  (self.op.bridge, pnode.name))
 
     if not rpc.call_bridges_exist(self.pnode.name, [self.op.bridge]):
       raise errors.OpPrereqError("target bridge '%s' does not exist on"
                                  " destination node '%s'" %
                                  (self.op.bridge, pnode.name))
 

-    # boot order verification
-    if self.op.hvm_boot_order is not None:
-      if len(self.op.hvm_boot_order.strip("acdn")) != 0:
-        raise errors.OpPrereqError("invalid boot order specified,"
-                                   " must be one or more of [acdn]")
-

     if self.op.start:
       self.instance_status = 'up'
     else:
     if self.op.start:
       self.instance_status = 'up'
     else:

@@ -4625,115 +4783,229 @@ class LUTestDelay(NoHooksLU):
                                    " result: %s" % (node, node_result))
 
 
                                    " result: %s" % (node, node_result))
 
 

-def _AllocatorGetClusterData(cfg, sstore):
-  """Compute the generic allocator input data.
+class IAllocator(object):
+  """IAllocator framework.

 
 

-  This is the data that is independent of the actual operation.
+  An IAllocator instance has three sets of attributes:
+    - cfg/sstore that are needed to query the cluster
+    - input data (all members of the _KEYS class attribute are required)
+    - four buffer attributes (in|out_data|text), that represent the
+      input (to the external script) in text and data structure format,
+      and the output from it, again in two formats
+    - the result variables from the script (success, info, nodes) for
+      easy usage

 
   """
 
   """

-  # cluster data
-  data = {
-    "version": 1,
-    "cluster_name": sstore.GetClusterName(),
-    "cluster_tags": list(cfg.GetClusterInfo().GetTags()),
-    # we don't have job IDs
-    }
-
-  # node data
-  node_results = {}
-  node_list = cfg.GetNodeList()
-  node_data = rpc.call_node_info(node_list, cfg.GetVGName())
-  for nname in node_list:
-    ninfo = cfg.GetNodeInfo(nname)
-    if nname not in node_data or not isinstance(node_data[nname], dict):
-      raise errors.OpExecError("Can't get data for node %s" % nname)
-    remote_info = node_data[nname]
-    for attr in ['memory_total', 'memory_free',
-                 'vg_size', 'vg_free']:
-      if attr not in remote_info:
-        raise errors.OpExecError("Node '%s' didn't return attribute '%s'" %
-                                 (nname, attr))
-      try:
-        int(remote_info[attr])
-      except ValueError, err:
-        raise errors.OpExecError("Node '%s' returned invalid value for '%s':"
-                                 " %s" % (nname, attr, str(err)))
-    pnr = {
-      "tags": list(ninfo.GetTags()),
-      "total_memory": utils.TryConvert(int, remote_info['memory_total']),
-      "free_memory": utils.TryConvert(int, remote_info['memory_free']),
-      "total_disk": utils.TryConvert(int, remote_info['vg_size']),
-      "free_disk": utils.TryConvert(int, remote_info['vg_free']),
-      "primary_ip": ninfo.primary_ip,
-      "secondary_ip": ninfo.secondary_ip,
+  _KEYS = [
+    "mode", "name",
+    "mem_size", "disks", "disk_template",
+    "os", "tags", "nics", "vcpus",
+    ]
+
+  def __init__(self, cfg, sstore, **kwargs):
+    self.cfg = cfg
+    self.sstore = sstore
+    # init buffer variables
+    self.in_text = self.out_text = self.in_data = self.out_data = None
+    # init all input fields so that pylint is happy
+    self.mode = self.name = None
+    self.mem_size = self.disks = self.disk_template = None
+    self.os = self.tags = self.nics = self.vcpus = None
+    # init result fields
+    self.success = self.info = self.nodes = None
+    for key in kwargs:
+      if key not in self._KEYS:
+        raise errors.ProgrammerError("Invalid input parameter '%s' to"
+                                     " IAllocator" % key)
+      setattr(self, key, kwargs[key])
+    for key in self._KEYS:
+      if key not in kwargs:
+        raise errors.ProgrammerError("Missing input parameter '%s' to"
+                                     " IAllocator" % key)
+    self._BuildInputData()
+
+  def _ComputeClusterData(self):
+    """Compute the generic allocator input data.
+
+    This is the data that is independent of the actual operation.
+
+    """
+    cfg = self.cfg
+    # cluster data
+    data = {
+      "version": 1,
+      "cluster_name": self.sstore.GetClusterName(),
+      "cluster_tags": list(cfg.GetClusterInfo().GetTags()),
+      # we don't have job IDs

       }
       }

-    node_results[nname] = pnr
-  data["nodes"] = node_results
-
-  # instance data
-  instance_data = {}
-  i_list = cfg.GetInstanceList()
-  for iname in i_list:
-    iinfo = cfg.GetInstanceInfo(iname)
-    nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
-                for n in iinfo.nics]
-    pir = {
-      "tags": list(iinfo.GetTags()),
-      "should_run": iinfo.status == "up",
-      "vcpus": iinfo.vcpus,
-      "memory": iinfo.memory,
-      "os": iinfo.os,
-      "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
-      "nics": nic_data,
-      "disks": [{"size": dsk.size, "mode": "w"} for dsk in iinfo.disks],
-      "disk_template": iinfo.disk_template,
+
+    # node data
+    node_results = {}
+    node_list = cfg.GetNodeList()
+    node_data = rpc.call_node_info(node_list, cfg.GetVGName())
+    for nname in node_list:
+      ninfo = cfg.GetNodeInfo(nname)
+      if nname not in node_data or not isinstance(node_data[nname], dict):
+        raise errors.OpExecError("Can't get data for node %s" % nname)
+      remote_info = node_data[nname]
+      for attr in ['memory_total', 'memory_free',
+                   'vg_size', 'vg_free']:
+        if attr not in remote_info:
+          raise errors.OpExecError("Node '%s' didn't return attribute '%s'" %
+                                   (nname, attr))
+        try:
+          int(remote_info[attr])
+        except ValueError, err:
+          raise errors.OpExecError("Node '%s' returned invalid value for '%s':"
+                                   " %s" % (nname, attr, str(err)))
+      pnr = {
+        "tags": list(ninfo.GetTags()),
+        "total_memory": utils.TryConvert(int, remote_info['memory_total']),
+        "free_memory": utils.TryConvert(int, remote_info['memory_free']),
+        "total_disk": utils.TryConvert(int, remote_info['vg_size']),
+        "free_disk": utils.TryConvert(int, remote_info['vg_free']),
+        "primary_ip": ninfo.primary_ip,
+        "secondary_ip": ninfo.secondary_ip,
+        }
+      node_results[nname] = pnr
+    data["nodes"] = node_results
+
+    # instance data
+    instance_data = {}
+    i_list = cfg.GetInstanceList()
+    for iname in i_list:
+      iinfo = cfg.GetInstanceInfo(iname)
+      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
+                  for n in iinfo.nics]
+      pir = {
+        "tags": list(iinfo.GetTags()),
+        "should_run": iinfo.status == "up",
+        "vcpus": iinfo.vcpus,
+        "memory": iinfo.memory,
+        "os": iinfo.os,
+        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
+        "nics": nic_data,
+        "disks": [{"size": dsk.size, "mode": "w"} for dsk in iinfo.disks],
+        "disk_template": iinfo.disk_template,
+        }
+      instance_data[iname] = pir
+
+    data["instances"] = instance_data
+
+    self.in_data = data
+
+  def _AddNewInstance(self):
+    """Add new instance data to allocator structure.
+
+    This in combination with _AllocatorGetClusterData will create the
+    correct structure needed as input for the allocator.
+
+    The checks for the completeness of the opcode must have already been
+    done.
+
+    """
+    data = self.in_data
+    if len(self.disks) != 2:
+      raise errors.OpExecError("Only two-disk configurations supported")
+
+    disk_space = _ComputeDiskSize(self.disk_template,
+                                  self.disks[0]["size"], self.disks[1]["size"])
+
+    request = {
+      "type": "allocate",
+      "name": self.name,
+      "disk_template": self.disk_template,
+      "tags": self.tags,
+      "os": self.os,
+      "vcpus": self.vcpus,
+      "memory": self.mem_size,
+      "disks": self.disks,
+      "disk_space_total": disk_space,
+      "nics": self.nics,

       }
       }

-    instance_data[iname] = pir
+    data["request"] = request

 
 

-  data["instances"] = instance_data
+  def _AddRelocateInstance(self):
+    """Add relocate instance data to allocator structure.

 
 

-  return data
+    This in combination with _IAllocatorGetClusterData will create the
+    correct structure needed as input for the allocator.

 
 

+    The checks for the completeness of the opcode must have already been
+    done.

 
 

-def _AllocatorAddNewInstance(data, op):
-  """Add new instance data to allocator structure.
+    """
+    data = self.in_data
+    request = {
+      "type": "replace_secondary",
+      "name": self.name,
+      }
+    data["request"] = request

 
 

-  This in combination with _AllocatorGetClusterData will create the
-  correct structure needed as input for the allocator.
+  def _BuildInputData(self):
+    """Build input data structures.

 
 

-  The checks for the completeness of the opcode must have already been
-  done.
+    """
+    self._ComputeClusterData()

 
 

-  """
-  request = {
-    "type": "allocate",
-    "name": op.name,
-    "disk_template": op.disk_template,
-    "tags": op.tags,
-    "os": op.os,
-    "vcpus": op.vcpus,
-    "memory": op.mem_size,
-    "disks": op.disks,
-    "nics": op.nics,
-    }
-  data["request"] = request
+    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
+      self._AddNewInstance()
+    else:
+      self._AddRelocateInstance()

 
 

+    self.in_text = serializer.Dump(self.in_data)

 
 

-def _AllocatorAddRelocateInstance(data, op):
-  """Add relocate instance data to allocator structure.
+  def Run(self, name, validate=True):
+    """Run an instance allocator and return the results.

 
 

-  This in combination with _AllocatorGetClusterData will create the
-  correct structure needed as input for the allocator.
+    """
+    data = self.in_text

 
 

-  The checks for the completeness of the opcode must have already been
-  done.
+    alloc_script = utils.FindFile(name, constants.IALLOCATOR_SEARCH_PATH,
+                                  os.path.isfile)
+    if alloc_script is None:
+      raise errors.OpExecError("Can't find allocator '%s'" % name)

 
 

-  """
-  request = {
-    "type": "replace_secondary",
-    "name": op.name,
-    }
-  data["request"] = request
+    fd, fin_name = tempfile.mkstemp(prefix="ganeti-iallocator.")
+    try:
+      os.write(fd, data)
+      os.close(fd)
+      result = utils.RunCmd([alloc_script, fin_name])
+      if result.failed:
+        raise errors.OpExecError("Instance allocator call failed: %s,"
+                                 " output: %s" %
+                                 (result.fail_reason, result.stdout))
+    finally:
+      os.unlink(fin_name)
+    self.out_text = result.stdout
+    if validate:
+      self._ValidateResult()
+
+  def _ValidateResult(self):
+    """Process the allocator results.
+
+    This will process and if successful save the result in
+    self.out_data and the other parameters.
+
+    """
+    try:
+      rdict = serializer.Load(self.out_text)
+    except Exception, err:
+      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
+
+    if not isinstance(rdict, dict):
+      raise errors.OpExecError("Can't parse iallocator results: not a dict")
+
+    for key in "success", "info", "nodes":
+      if key not in rdict:
+        raise errors.OpExecError("Can't parse iallocator results:"
+                                 " missing key '%s'" % key)
+      setattr(self, key, rdict[key])
+
+    if not isinstance(rdict["nodes"], list):
+      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
+                               " is not a list")
+    self.out_data = rdict

 
 
 class LUTestAllocator(NoHooksLU):
 
 
 class LUTestAllocator(NoHooksLU):

@@ -4750,7 +5022,7 @@ class LUTestAllocator(NoHooksLU):
     This checks the opcode parameters depending on the director and mode test.
 
     """
     This checks the opcode parameters depending on the director and mode test.
 
     """

-    if self.op.mode == constants.ALF_MODE_ALLOC:
+    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:

       for attr in ["name", "mem_size", "disks", "disk_template",
                    "os", "tags", "nics", "vcpus"]:
         if not hasattr(self.op, attr):
       for attr in ["name", "mem_size", "disks", "disk_template",
                    "os", "tags", "nics", "vcpus"]:
         if not hasattr(self.op, attr):

@@ -4771,6 +5043,8 @@ class LUTestAllocator(NoHooksLU):
                                      " 'nics' parameter")
       if not isinstance(self.op.disks, list):
         raise errors.OpPrereqError("Invalid parameter 'disks'")
                                      " 'nics' parameter")
       if not isinstance(self.op.disks, list):
         raise errors.OpPrereqError("Invalid parameter 'disks'")

+      if len(self.op.disks) != 2:
+        raise errors.OpPrereqError("Only two-disk configurations supported")

       for row in self.op.disks:
         if (not isinstance(row, dict) or
             "size" not in row or
       for row in self.op.disks:
         if (not isinstance(row, dict) or
             "size" not in row or

@@ -4779,7 +5053,7 @@ class LUTestAllocator(NoHooksLU):
             row["mode"] not in ['r', 'w']):
           raise errors.OpPrereqError("Invalid contents of the"
                                      " 'disks' parameter")
             row["mode"] not in ['r', 'w']):
           raise errors.OpPrereqError("Invalid contents of the"
                                      " 'disks' parameter")

-    elif self.op.mode == constants.ALF_MODE_RELOC:
+    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:

       if not hasattr(self.op, "name"):
         raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
       fname = self.cfg.ExpandInstanceName(self.op.name)
       if not hasattr(self.op, "name"):
         raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
       fname = self.cfg.ExpandInstanceName(self.op.name)

@@ -4791,11 +5065,10 @@ class LUTestAllocator(NoHooksLU):
       raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
                                  self.op.mode)
 
       raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
                                  self.op.mode)
 

-    if self.op.direction == constants.ALF_DIR_OUT:
-      if not hasattr(self.op, "allocator"):
+    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
+      if not hasattr(self.op, "allocator") or self.op.allocator is None:

         raise errors.OpPrereqError("Missing allocator name")
         raise errors.OpPrereqError("Missing allocator name")

-      raise errors.OpPrereqError("Allocator out mode not supported yet")
-    elif self.op.direction != constants.ALF_DIR_IN:
+    elif self.op.direction != constants.IALLOCATOR_DIR_IN:

       raise errors.OpPrereqError("Wrong allocator test '%s'" %
                                  self.op.direction)
 
       raise errors.OpPrereqError("Wrong allocator test '%s'" %
                                  self.op.direction)
 

@@ -4803,14 +5076,21 @@ class LUTestAllocator(NoHooksLU):
     """Run the allocator test.
 
     """
     """Run the allocator test.
 
     """

-    data = _AllocatorGetClusterData(self.cfg, self.sstore)
-    if self.op.mode == constants.ALF_MODE_ALLOC:
-      _AllocatorAddNewInstance(data, self.op)
+    ial = IAllocator(self.cfg, self.sstore,
+                     mode=self.op.mode,
+                     name=self.op.name,
+                     mem_size=self.op.mem_size,
+                     disks=self.op.disks,
+                     disk_template=self.op.disk_template,
+                     os=self.op.os,
+                     tags=self.op.tags,
+                     nics=self.op.nics,
+                     vcpus=self.op.vcpus,
+                     )
+
+    if self.op.direction == constants.IALLOCATOR_DIR_IN:
+      result = ial.in_text

     else:
     else:

-      _AllocatorAddRelocateInstance(data, self.op)
-
-    if _JSON_INDENT is None:
-      text = simplejson.dumps(data)
-    else:
-      text = simplejson.dumps(data, indent=_JSON_INDENT)
-    return text
+      ial.Run(self.op.allocator, validate=False)
+      result = ial.out_text
+    return result