Allow 'add instance' to not start the new instance
[ganeti-local] / lib / cmdlib.py
index 014290f..d3f583c 100644 (file)
@@ -165,27 +165,51 @@ class NoHooksLU(LogicalUnit):
 
 
 def _GetWantedNodes(lu, nodes):
-  """Returns list of checked and expanded nodes.
+  """Returns list of checked and expanded node names.
 
   Args:
     nodes: List of nodes (strings) or None for all
 
   """
-  if nodes is not None and not isinstance(nodes, list):
+  if not isinstance(nodes, list):
     raise errors.OpPrereqError("Invalid argument type 'nodes'")
 
   if nodes:
-    wanted_nodes = []
+    wanted = []
 
     for name in nodes:
-      node = lu.cfg.GetNodeInfo(lu.cfg.ExpandNodeName(name))
+      node = lu.cfg.ExpandNodeName(name)
       if node is None:
         raise errors.OpPrereqError("No such node name '%s'" % name)
-    wanted_nodes.append(node)
+      wanted.append(node)
 
-    return wanted_nodes
   else:
-    return [lu.cfg.GetNodeInfo(name) for name in lu.cfg.GetNodeList()]
+    wanted = lu.cfg.GetNodeList()
+  return utils.NiceSort(wanted)
+
+
+def _GetWantedInstances(lu, instances):
+  """Returns list of checked and expanded instance names.
+
+  Args:
+    instances: List of instances (strings) or None for all
+
+  """
+  if not isinstance(instances, list):
+    raise errors.OpPrereqError("Invalid argument type 'instances'")
+
+  if instances:
+    wanted = []
+
+    for name in instances:
+      instance = lu.cfg.ExpandInstanceName(name)
+      if instance is None:
+        raise errors.OpPrereqError("No such instance name '%s'" % name)
+      wanted.append(instance)
+
+  else:
+    wanted = lu.cfg.GetInstanceList()
+  return utils.NiceSort(wanted)
 
 
 def _CheckOutputFields(static, dynamic, selected):
@@ -523,9 +547,9 @@ class LUInitCluster(LogicalUnit):
     """
     env = {
       "CLUSTER": self.op.cluster_name,
-      "MASTER": self.hostname['hostname_full'],
+      "MASTER": self.hostname.name,
       }
-    return env, [], [self.hostname['hostname_full']]
+    return env, [], [self.hostname.name]
 
   def CheckPrereq(self):
     """Verify that the passed name is a valid one.
@@ -540,33 +564,33 @@ class LUInitCluster(LogicalUnit):
       raise errors.OpPrereqError("Cannot resolve my own hostname ('%s')" %
                                  hostname_local)
 
-    if hostname["hostname_full"] != hostname_local:
+    if hostname.name != hostname_local:
       raise errors.OpPrereqError("My own hostname (%s) does not match the"
                                  " resolver (%s): probably not using FQDN"
                                  " for hostname." %
-                                 (hostname_local, hostname["hostname_full"]))
+                                 (hostname_local, hostname.name))
 
-    if hostname["ip"].startswith("127."):
+    if hostname.ip.startswith("127."):
       raise errors.OpPrereqError("This host's IP resolves to the private"
                                  " range (%s). Please fix DNS or /etc/hosts." %
-                                 (hostname["ip"],))
+                                 (hostname.ip,))
 
     self.clustername = clustername = utils.LookupHostname(self.op.cluster_name)
     if not clustername:
       raise errors.OpPrereqError("Cannot resolve given cluster name ('%s')"
                                  % self.op.cluster_name)
 
-    result = utils.RunCmd(["fping", "-S127.0.0.1", "-q", hostname['ip']])
+    result = utils.RunCmd(["fping", "-S127.0.0.1", "-q", hostname.ip])
     if result.failed:
       raise errors.OpPrereqError("Inconsistency: this host's name resolves"
                                  " to %s,\nbut this ip address does not"
                                  " belong to this host."
-                                 " Aborting." % hostname['ip'])
+                                 " Aborting." % hostname.ip)
 
     secondary_ip = getattr(self.op, "secondary_ip", None)
     if secondary_ip and not utils.IsValidIP(secondary_ip):
       raise errors.OpPrereqError("Invalid secondary ip given")
-    if secondary_ip and secondary_ip != hostname['ip']:
+    if secondary_ip and secondary_ip != hostname.ip:
       result = utils.RunCmd(["fping", "-S127.0.0.1", "-q", secondary_ip])
       if result.failed:
         raise errors.OpPrereqError("You gave %s as secondary IP,\n"
@@ -605,16 +629,16 @@ class LUInitCluster(LogicalUnit):
     # set up the simple store
     ss = ssconf.SimpleStore()
     ss.SetKey(ss.SS_HYPERVISOR, self.op.hypervisor_type)
-    ss.SetKey(ss.SS_MASTER_NODE, hostname['hostname_full'])
-    ss.SetKey(ss.SS_MASTER_IP, clustername['ip'])
+    ss.SetKey(ss.SS_MASTER_NODE, hostname.name)
+    ss.SetKey(ss.SS_MASTER_IP, clustername.ip)
     ss.SetKey(ss.SS_MASTER_NETDEV, self.op.master_netdev)
-    ss.SetKey(ss.SS_CLUSTER_NAME, clustername['hostname'])
+    ss.SetKey(ss.SS_CLUSTER_NAME, clustername.name)
 
     # set up the inter-node password and certificate
     _InitGanetiServerSetup(ss)
 
     # start the master ip
-    rpc.call_node_start_master(hostname['hostname_full'])
+    rpc.call_node_start_master(hostname.name)
 
     # set up ssh config and /etc/hosts
     f = open('/etc/ssh/ssh_host_rsa_key.pub', 'r')
@@ -624,20 +648,15 @@ class LUInitCluster(LogicalUnit):
       f.close()
     sshkey = sshline.split(" ")[1]
 
-    _UpdateEtcHosts(hostname['hostname_full'],
-                    hostname['ip'],
-                    )
+    _UpdateEtcHosts(hostname.name, hostname.ip)
 
-    _UpdateKnownHosts(hostname['hostname_full'],
-                      hostname['ip'],
-                      sshkey,
-                      )
+    _UpdateKnownHosts(hostname.name, hostname.ip, sshkey)
 
-    _InitSSHSetup(hostname['hostname'])
+    _InitSSHSetup(hostname.name)
 
     # init of cluster config file
     cfgw = config.ConfigWriter()
-    cfgw.InitConfig(hostname['hostname'], hostname['ip'], self.secondary_ip,
+    cfgw.InitConfig(hostname.name, hostname.ip, self.secondary_ip,
                     sshkey, self.op.mac_prefix,
                     self.op.vg_name, self.op.def_bridge)
 
@@ -920,6 +939,87 @@ class LUVerifyCluster(NoHooksLU):
     return int(bad)
 
 
+class LURenameCluster(LogicalUnit):
+  """Rename the cluster.
+
+  """
+  HPATH = "cluster-rename"
+  HTYPE = constants.HTYPE_CLUSTER
+  _OP_REQP = ["name"]
+
+  def BuildHooksEnv(self):
+    """Build hooks env.
+
+    """
+    env = {
+      "NEW_NAME": self.op.name,
+      }
+    mn = self.sstore.GetMasterNode()
+    return env, [mn], [mn]
+
+  def CheckPrereq(self):
+    """Verify that the passed name is a valid one.
+
+    """
+    hostname = utils.LookupHostname(self.op.name)
+    if not hostname:
+      raise errors.OpPrereqError("Cannot resolve the new cluster name ('%s')" %
+                                 self.op.name)
+
+    new_name = hostname.name
+    self.ip = new_ip = hostname.ip
+    old_name = self.sstore.GetClusterName()
+    old_ip = self.sstore.GetMasterIP()
+    if new_name == old_name and new_ip == old_ip:
+      raise errors.OpPrereqError("Neither the name nor the IP address of the"
+                                 " cluster has changed")
+    if new_ip != old_ip:
+      result = utils.RunCmd(["fping", "-q", new_ip])
+      if not result.failed:
+        raise errors.OpPrereqError("The given cluster IP address (%s) is"
+                                   " reachable on the network. Aborting." %
+                                   new_ip)
+
+    self.op.name = new_name
+
+  def Exec(self, feedback_fn):
+    """Rename the cluster.
+
+    """
+    clustername = self.op.name
+    ip = self.ip
+    ss = self.sstore
+
+    # shutdown the master IP
+    master = ss.GetMasterNode()
+    if not rpc.call_node_stop_master(master):
+      raise errors.OpExecError("Could not disable the master role")
+
+    try:
+      # modify the sstore
+      ss.SetKey(ss.SS_MASTER_IP, ip)
+      ss.SetKey(ss.SS_CLUSTER_NAME, clustername)
+
+      # Distribute updated ss config to all nodes
+      myself = self.cfg.GetNodeInfo(master)
+      dist_nodes = self.cfg.GetNodeList()
+      if myself.name in dist_nodes:
+        dist_nodes.remove(myself.name)
+
+      logger.Debug("Copying updated ssconf data to all nodes")
+      for keyname in [ss.SS_CLUSTER_NAME, ss.SS_MASTER_IP]:
+        fname = ss.KeyToFilename(keyname)
+        result = rpc.call_upload_file(dist_nodes, fname)
+        for to_node in dist_nodes:
+          if not result[to_node]:
+            logger.Error("copy of file %s to node %s failed" %
+                         (fname, to_node))
+    finally:
+      if not rpc.call_node_start_master(master):
+        logger.Error("Could not re-enable the master role on the master,\n"
+                     "please restart manually.")
+
+
 def _WaitForSync(cfgw, instance, oneshot=False, unlock=False):
   """Sleep and poll for an instance's disk to sync.
 
@@ -1105,7 +1205,7 @@ class LUQueryNodes(NoHooksLU):
   """Logical unit for querying nodes.
 
   """
-  _OP_REQP = ["output_fields"]
+  _OP_REQP = ["output_fields", "names"]
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -1116,19 +1216,21 @@ class LUQueryNodes(NoHooksLU):
     self.dynamic_fields = frozenset(["dtotal", "dfree",
                                      "mtotal", "mnode", "mfree"])
 
-    _CheckOutputFields(static=["name", "pinst", "sinst", "pip", "sip"],
+    _CheckOutputFields(static=["name", "pinst_cnt", "sinst_cnt",
+                               "pinst_list", "sinst_list",
+                               "pip", "sip"],
                        dynamic=self.dynamic_fields,
                        selected=self.op.output_fields)
 
+    self.wanted = _GetWantedNodes(self, self.op.names)
 
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
 
     """
-    nodenames = utils.NiceSort(self.cfg.GetNodeList())
+    nodenames = self.wanted
     nodelist = [self.cfg.GetNodeInfo(name) for name in nodenames]
 
-
     # begin data gathering
 
     if self.dynamic_fields.intersection(self.op.output_fields):
@@ -1149,17 +1251,21 @@ class LUQueryNodes(NoHooksLU):
     else:
       live_data = dict.fromkeys(nodenames, {})
 
-    node_to_primary = dict.fromkeys(nodenames, 0)
-    node_to_secondary = dict.fromkeys(nodenames, 0)
+    node_to_primary = dict([(name, set()) for name in nodenames])
+    node_to_secondary = dict([(name, set()) for name in nodenames])
 
-    if "pinst" in self.op.output_fields or "sinst" in self.op.output_fields:
+    inst_fields = frozenset(("pinst_cnt", "pinst_list",
+                             "sinst_cnt", "sinst_list"))
+    if inst_fields & frozenset(self.op.output_fields):
       instancelist = self.cfg.GetInstanceList()
 
-      for instance in instancelist:
-        instanceinfo = self.cfg.GetInstanceInfo(instance)
-        node_to_primary[instanceinfo.primary_node] += 1
-        for secnode in instanceinfo.secondary_nodes:
-          node_to_secondary[secnode] += 1
+      for instance_name in instancelist:
+        inst = self.cfg.GetInstanceInfo(instance_name)
+        if inst.primary_node in node_to_primary:
+          node_to_primary[inst.primary_node].add(inst.name)
+        for secnode in inst.secondary_nodes:
+          if secnode in node_to_secondary:
+            node_to_secondary[secnode].add(inst.name)
 
     # end data gathering
 
@@ -1169,19 +1275,22 @@ class LUQueryNodes(NoHooksLU):
       for field in self.op.output_fields:
         if field == "name":
           val = node.name
-        elif field == "pinst":
-          val = node_to_primary[node.name]
-        elif field == "sinst":
-          val = node_to_secondary[node.name]
+        elif field == "pinst_list":
+          val = list(node_to_primary[node.name])
+        elif field == "sinst_list":
+          val = list(node_to_secondary[node.name])
+        elif field == "pinst_cnt":
+          val = len(node_to_primary[node.name])
+        elif field == "sinst_cnt":
+          val = len(node_to_secondary[node.name])
         elif field == "pip":
           val = node.primary_ip
         elif field == "sip":
           val = node.secondary_ip
         elif field in self.dynamic_fields:
-          val = live_data[node.name].get(field, "?")
+          val = live_data[node.name].get(field, None)
         else:
           raise errors.ParameterError(field)
-        val = str(val)
         node_output.append(val)
       output.append(node_output)
 
@@ -1211,7 +1320,7 @@ class LUQueryNodeVolumes(NoHooksLU):
     """Computes the list of nodes and their attributes.
 
     """
-    nodenames = utils.NiceSort([node.name for node in self.nodes])
+    nodenames = self.nodes
     volumes = rpc.call_node_volumes(nodenames)
 
     ilist = [self.cfg.GetInstanceInfo(iname) for iname
@@ -1299,8 +1408,8 @@ class LUAddNode(LogicalUnit):
     if not dns_data:
       raise errors.OpPrereqError("Node %s is not resolvable" % node_name)
 
-    node = dns_data['hostname']
-    primary_ip = self.op.primary_ip = dns_data['ip']
+    node = dns_data.name
+    primary_ip = self.op.primary_ip = dns_data.ip
     secondary_ip = getattr(self.op, "secondary_ip", None)
     if secondary_ip is None:
       secondary_ip = primary_ip
@@ -1609,7 +1718,7 @@ class LUClusterCopyFile(NoHooksLU):
 
     myname = socket.gethostname()
 
-    for node in [node.name for node in self.nodes]:
+    for node in self.nodes:
       if node == myname:
         continue
       if not ssh.CopyFileToNode(node, filename):
@@ -1655,8 +1764,8 @@ class LURunClusterCommand(NoHooksLU):
     """
     data = []
     for node in self.nodes:
-      result = ssh.SSHCall(node.name, "root", self.op.command)
-      data.append((node.name, result.output, result.exit_code))
+      result = ssh.SSHCall(node, "root", self.op.command)
+      data.append((node, result.output, result.exit_code))
 
     return data
 
@@ -2001,6 +2110,87 @@ class LUReinstallInstance(LogicalUnit):
       _ShutdownInstanceDisks(inst, self.cfg)
 
 
+class LURenameInstance(LogicalUnit):
+  """Rename an instance.
+
+  """
+  HPATH = "instance-rename"
+  HTYPE = constants.HTYPE_INSTANCE
+  _OP_REQP = ["instance_name", "new_name"]
+
+  def BuildHooksEnv(self):
+    """Build hooks env.
+
+    This runs on master, primary and secondary nodes of the instance.
+
+    """
+    env = _BuildInstanceHookEnvByObject(self.instance)
+    env["INSTANCE_NEW_NAME"] = self.op.new_name
+    nl = ([self.sstore.GetMasterNode(), self.instance.primary_node] +
+          list(self.instance.secondary_nodes))
+    return env, nl, nl
+
+  def CheckPrereq(self):
+    """Check prerequisites.
+
+    This checks that the instance is in the cluster and is not running.
+
+    """
+    instance = self.cfg.GetInstanceInfo(
+      self.cfg.ExpandInstanceName(self.op.instance_name))
+    if instance is None:
+      raise errors.OpPrereqError("Instance '%s' not known" %
+                                 self.op.instance_name)
+    if instance.status != "down":
+      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
+                                 self.op.instance_name)
+    remote_info = rpc.call_instance_info(instance.primary_node, instance.name)
+    if remote_info:
+      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
+                                 (self.op.instance_name,
+                                  instance.primary_node))
+    self.instance = instance
+
+    # new name verification
+    hostname1 = utils.LookupHostname(self.op.new_name)
+    if not hostname1:
+      raise errors.OpPrereqError("New instance name '%s' not found in dns" %
+                                 self.op.new_name)
+
+    self.op.new_name = new_name = hostname1.name
+    if not getattr(self.op, "ignore_ip", False):
+      command = ["fping", "-q", hostname1.ip]
+      result = utils.RunCmd(command)
+      if not result.failed:
+        raise errors.OpPrereqError("IP %s of instance %s already in use" %
+                                   (hostname1.ip, new_name))
+
+
+  def Exec(self, feedback_fn):
+    """Reinstall the instance.
+
+    """
+    inst = self.instance
+    old_name = inst.name
+
+    self.cfg.RenameInstance(inst.name, self.op.new_name)
+
+    # re-read the instance from the configuration after rename
+    inst = self.cfg.GetInstanceInfo(self.op.new_name)
+
+    _StartInstanceDisks(self.cfg, inst, None)
+    try:
+      if not rpc.call_instance_run_rename(inst.primary_node, inst, old_name,
+                                          "sda", "sdb"):
+        msg = ("Could run OS rename script for instance %s\n"
+               "on node %s\n"
+               "(but the instance has been renamed in Ganeti)" %
+               (inst.name, inst.primary_node))
+        logger.Error(msg)
+    finally:
+      _ShutdownInstanceDisks(inst, self.cfg)
+
+
 class LURemoveInstance(LogicalUnit):
   """Remove an instance.
 
@@ -2058,7 +2248,7 @@ class LUQueryInstances(NoHooksLU):
   """Logical unit for querying instances.
 
   """
-  _OP_REQP = ["output_fields"]
+  _OP_REQP = ["output_fields", "names"]
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -2074,11 +2264,13 @@ class LUQueryInstances(NoHooksLU):
                        dynamic=self.dynamic_fields,
                        selected=self.op.output_fields)
 
+    self.wanted = _GetWantedInstances(self, self.op.names)
+
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
 
     """
-    instance_names = utils.NiceSort(self.cfg.GetInstanceList())
+    instance_names = self.wanted
     instance_list = [self.cfg.GetInstanceInfo(iname) for iname
                      in instance_names]
 
@@ -2113,25 +2305,19 @@ class LUQueryInstances(NoHooksLU):
         elif field == "pnode":
           val = instance.primary_node
         elif field == "snodes":
-          val = ",".join(instance.secondary_nodes) or "-"
+          val = list(instance.secondary_nodes)
         elif field == "admin_state":
-          if instance.status == "down":
-            val = "no"
-          else:
-            val = "yes"
+          val = (instance.status != "down")
         elif field == "oper_state":
           if instance.primary_node in bad_nodes:
-            val = "(node down)"
+            val = None
           else:
-            if live_data.get(instance.name):
-              val = "running"
-            else:
-              val = "stopped"
+            val = bool(live_data.get(instance.name))
         elif field == "admin_ram":
           val = instance.memory
         elif field == "oper_ram":
           if instance.primary_node in bad_nodes:
-            val = "(node down)"
+            val = None
           elif instance.name in live_data:
             val = live_data[instance.name].get("memory", "?")
           else:
@@ -2147,12 +2333,11 @@ class LUQueryInstances(NoHooksLU):
         elif field == "sda_size" or field == "sdb_size":
           disk = instance.FindDisk(field[:3])
           if disk is None:
-            val = "N/A"
+            val = None
           else:
             val = disk.size
         else:
           raise errors.ParameterError(field)
-        val = str(val)
         iout.append(val)
       output.append(iout)
 
@@ -2509,7 +2694,7 @@ class LUCreateInstance(LogicalUnit):
   HTYPE = constants.HTYPE_INSTANCE
   _OP_REQP = ["instance_name", "mem_size", "disk_size", "pnode",
               "disk_template", "swap_size", "mode", "start", "vcpus",
-              "wait_for_sync"]
+              "wait_for_sync", "ip_check"]
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -2659,7 +2844,7 @@ class LUCreateInstance(LogicalUnit):
       raise errors.OpPrereqError("Instance name '%s' not found in dns" %
                                  self.op.instance_name)
 
-    self.op.instance_name = instance_name = hostname1['hostname']
+    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" %
@@ -2669,7 +2854,7 @@ class LUCreateInstance(LogicalUnit):
     if ip is None or ip.lower() == "none":
       inst_ip = None
     elif ip.lower() == "auto":
-      inst_ip = hostname1['ip']
+      inst_ip = hostname1.ip
     else:
       if not utils.IsValidIP(ip):
         raise errors.OpPrereqError("given IP address '%s' doesn't look"
@@ -2677,11 +2862,16 @@ class LUCreateInstance(LogicalUnit):
       inst_ip = ip
     self.inst_ip = inst_ip
 
-    command = ["fping", "-q", hostname1['ip']]
-    result = utils.RunCmd(command)
-    if not result.failed:
-      raise errors.OpPrereqError("IP %s of instance %s already in use" %
-                                 (hostname1['ip'], instance_name))
+    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:
+      command = ["fping", "-q", hostname1.ip]
+      result = utils.RunCmd(command)
+      if not result.failed:
+        raise errors.OpPrereqError("IP address %s of instance %s already"
+                                   " in use" % (hostname1.ip, instance_name))
 
     # bridge verification
     bridge = getattr(self.op, "bridge", None)
@@ -3282,37 +3472,6 @@ class LUQueryInstanceData(NoHooksLU):
     return result
 
 
-class LUQueryNodeData(NoHooksLU):
-  """Logical unit for querying node data.
-
-  """
-  _OP_REQP = ["nodes"]
-
-  def CheckPrereq(self):
-    """Check prerequisites.
-
-    This only checks the optional node list against the existing names.
-
-    """
-    self.wanted_nodes = _GetWantedNodes(self, self.op.nodes)
-
-  def Exec(self, feedback_fn):
-    """Compute and return the list of nodes.
-
-    """
-    ilist = [self.cfg.GetInstanceInfo(iname) for iname
-             in self.cfg.GetInstanceList()]
-    result = []
-    for node in self.wanted_nodes:
-      result.append((node.name, node.primary_ip, node.secondary_ip,
-                     [inst.name for inst in ilist
-                      if inst.primary_node == node.name],
-                     [inst.name for inst in ilist
-                      if node.name in inst.secondary_nodes],
-                     ))
-    return result
-
-
 class LUSetInstanceParms(LogicalUnit):
   """Modifies an instances's parameters.
 
@@ -3435,7 +3594,7 @@ class LUQueryExports(NoHooksLU):
       that node.
 
     """
-    return rpc.call_export_list([node.name for node in self.nodes])
+    return rpc.call_export_list(self.nodes)
 
 
 class LUExportInstance(LogicalUnit):