Added --dns-sufix option

starcluster/cluster.py

@@ -428,6 +428,7 @@ def __init__(self,
                  public_ips=None,
                  plugins_order=[],
                  config_on_master=False,
+                 dns_sufix=None,
                  **kwargs):
         # update class vars with given vars
         _vars = locals().copy()
@@ -454,6 +455,7 @@ def __init__(self,
         self.force_spot_master = force_spot_master
         self.disable_cloudinit = disable_cloudinit
         self.plugins_order = plugins_order
+        self.dns_sufix = dns_sufix and cluster_tag
 
         self._cluster_group = None
         self._placement_group = None
@@ -467,6 +469,7 @@ def __init__(self,
         self._vpc_id = None
         self._subnet_zones_mapping = None
 
+
     def __repr__(self):
         return '<Cluster: %s (%s-node)>' % (self.cluster_tag,
                                             self.cluster_size)
@@ -748,7 +751,8 @@ def _get_settings(self):
                              public_ips=self.public_ips,
                              disable_queue=self.disable_queue,
                              disable_cloudinit=self.disable_cloudinit,
-                             plugins_order=self.plugins_order)
+                             plugins_order=self.plugins_order,
+                             dns_sufix=self.dns_sufix)
         user_settings = dict(cluster_user=self.cluster_user,
                              cluster_shell=self.cluster_shell,
                              keyname=self.keyname, spot_bid=self.spot_bid)
@@ -971,18 +975,17 @@ def _nodes_in_states(self, states):
 
     def _make_alias(self, id=None, master=False):
         if master:
-            if self.dns_prefix:
-                return "%s-master" % self.dns_prefix
-            else:
-                return "master"
+            alias = "master"
         elif id is not None:
-            if self.dns_prefix:
-                alias = '%s-node%.3d' % (self.dns_prefix, id)
-            else:
-                alias = 'node%.3d' % id
+            alias = 'node{:03d}'.format(id)
         else:
             raise AttributeError("_make_alias(...) must receive either"
                                  " master=True or a node id number")
+
+        if self.dns_prefix:
+            alias = "{}-{}".format(self.dns_prefix, alias)
+        if self.dns_sufix:
+            alias = "{}.{}".format(alias, self.dns_sufix)
         return alias
 
     @property
@@ -2140,6 +2143,7 @@ def validate(self):
             self.validate_required_settings()
             self.validate_vpc()
             self.validate_dns_prefix()
+            self.validate_dns_sufix()
             self.validate_spot_bid()
             self.validate_cluster_size()
             self.validate_cluster_user()
@@ -2186,6 +2190,22 @@ def validate_dns_prefix(self):
                     dns_prefix=self.cluster.dns_prefix))
         return True
 
+    def validate_dns_sufix(self):
+        if not self.cluster.dns_sufix:
+            return True
+
+        # check that the dns prefix is a valid hostname
+        is_valid = utils.is_valid_hostname(self.cluster.dns_sufix)
+        if not is_valid:
+            raise exception.ClusterValidationError(
+                "The cluster name you chose, {dns_prefix}, is"
+                " not a valid dns name. "
+                " Since you have chosen to prepend the hostnames"
+                " via the dns_prefix option, {dns_prefix} should only have"
+                " alphanumeric characters and a '-' or '.'".format(
+                    dns_prefix=self.cluster.dns_sufix))
+        return True
+
     def validate_spot_bid(self):
         cluster = self.cluster
         if cluster.spot_bid is not None:

starcluster/commands/start.py

@@ -187,11 +187,19 @@ def addopts(self, parser):
                           action='store_true', help="Store the config on the "
                           "master node rather than into the security group "
                           "tags")
+        parser.add_option("--dns-sufix", action='store_true',
+                          help="Sufix dns names of all nodes in the cluster "
+                          "with the cluster tag.")
 
     def execute(self, args):
         if len(args) != 1:
             self.parser.error("please specify a <cluster_tag>")
         tag = args[0]
+        if tag.find("master") > -1:
+            # Because of Node.is_master
+            raise exception.ClusterValidationError("Cluster name cannot "
+                                                   "contain master")
+
         create = not self.opts.no_create
         scluster = self.cm.get_cluster_group_or_none(tag)
         if scluster and create:
@@ -252,6 +260,8 @@ def execute(self, args):
                 self.warn_experimental(msg, num_secs=5)
         if self.opts.dns_prefix:
             scluster.dns_prefix = tag
+        if self.opts.dns_sufix:
+            scluster.dns_sufix = tag
         if config_on_master:
             scluster.config_on_master = True
             if self.opts.no_create:

starcluster/node.py

@@ -164,6 +164,10 @@ def alias(self):
             self._alias = alias
         return self._alias
 
+    @property
+    def short_alias(self):
+        return self.alias.split(".", 1)[0]
+
     def reset_alias(self):
         """
         Used to reset the name and alias when there is a conflict
@@ -878,8 +882,8 @@ def remove_from_etc_hosts(self, nodes):
         Remove all network names and ips for node in nodes arg from this node's
         /etc/hosts file
         """
-        aliases = map(lambda x: x.alias, nodes)
-        ips = map(lambda x: x.private_ip_address.replace(".", "\."), nodes)
+        aliases = [n.short_alias for n in nodes]
+        ips = [n.private_ip_address.replace(".", "\.") for n in nodes]
         self.ssh.remove_lines_from_file('/etc/hosts', '|'.join(aliases))
         self.ssh.remove_lines_from_file('/etc/hosts', '|'.join(ips))
 
@@ -912,6 +916,7 @@ def network_names(self):
         names['INTERNAL_NAME'] = self.private_dns_name
         names['INTERNAL_NAME_SHORT'] = self.private_dns_name_short
         names['INTERNAL_ALIAS'] = self.alias
+        names['INTERNAL_SHORT_ALIAS'] = self.short_alias
         return names
 
     @property
@@ -968,7 +973,7 @@ def get_spot_request(self):
             return spot[0]
 
     def is_master(self):
-        return self.alias == 'master' or self.alias.endswith("-master")
+        return self.alias.find("master") != -1
 
     def is_instance_store(self):
         return self.instance.root_device_type == "instance-store"
@@ -1240,7 +1245,8 @@ def shell(self, user=None, forward_x11=False, forward_agent=False,
 
     def get_hosts_entry(self):
         """ Returns /etc/hosts entry for this node """
-        etc_hosts_line = "%(INTERNAL_IP)s %(INTERNAL_ALIAS)s"
+        etc_hosts_line = \
+            "%(INTERNAL_IP)s %(INTERNAL_ALIAS)s %(INTERNAL_SHORT_ALIAS)s"
         etc_hosts_line = etc_hosts_line % self.network_names
         return etc_hosts_line
 

starcluster/plugins/sge.py

@@ -197,12 +197,11 @@ def get_nodes_to_recover(self, nodes):
         for node in nodes:
             # nodes missing from qhost
             if node.alias not in parsed_qhosts:
-                if node.alias == "master":
+                if node.alias.is_master():
                     assert(not self.master_is_exec_host)
                 else:
                     missing.append(node)
-            elif parsed_qhosts[node.alias][-1] == "-" \
-                    and node.alias != "master":
+            elif parsed_qhosts[node.alias][-1] == "-" and not node.is_master():
                 # nodes present but w/o stats
                 try:
                     node.ssh.execute("qhost", source_profile=True)