Multi host DHCP networking and local DNS resolving

Configure all dnsmasq instances to use the hosts file instead of the
default local file '/etc/host' when the network is configured in
'multi_host' mode.

This hosts file contains all DNS entries of the network and it is
replicated on all host where a 'nova-network' daemon is instantiated.
There is a hosts file for each networks.

It needs to add a new network API to update the host file on all
network hosts.

When a network host is called to (de)allocate a VM, it (de)allocate the
instance fixed IP adress(es) on the concerned network host and when it's
done, it calls all networks to update their DNS db entries if it's
needed through a fanout cast.

DocImpact: new config options
Fixes LP bug #1078808

Change-Id: I5c13bd895ebd31f276eb14e996025ddcfb67c3d9

7 files changed, 103 insertions, 8 deletions

diff --git a/nova/network/linux_net.py b/nova/network/linux_net.py
index f5d3b19ac..7929c235c 100644
--- a/nova/network/linux_net.py
+++ b/nova/network/linux_net.py
@@ -729,6 +729,15 @@ def get_dhcp_hosts(context, network_ref):
     return '\n'.join(hosts)
 
 
+def get_dns_hosts(context, network_ref):
+    """Get network's DNS hosts in hosts format."""
+    hosts = []
+    for data in db.network_get_associated_fixed_ips(context,
+                                                    network_ref['id']):
+        hosts.append(_host_dns(data))
+    return '\n'.join(hosts)
+
+
 def _add_dnsmasq_accept_rules(dev):
     """Allow DHCP and DNS traffic through to dnsmasq."""
     table = iptables_manager.ipv4['filter']
@@ -780,6 +789,12 @@ def update_dhcp(context, dev, network_ref):
     restart_dhcp(context, dev, network_ref)
 
 
+def update_dns(context, dev, network_ref):
+    hostsfile = _dhcp_file(dev, 'hosts')
+    write_to_file(hostsfile, get_dns_hosts(context, network_ref))
+    restart_dhcp(context, dev, network_ref)
+
+
 def update_dhcp_hostfile_with_text(dev, hosts_text):
     conffile = _dhcp_file(dev, 'conf')
     write_to_file(conffile, hosts_text)
@@ -858,6 +873,8 @@ def restart_dhcp(context, dev, network_ref):
            '--dhcp-hostsfile=%s' % _dhcp_file(dev, 'conf'),
            '--dhcp-script=%s' % CONF.dhcpbridge,
            '--leasefile-ro']
+    if network_ref['multi_host'] and not CONF.dns_server:
+        cmd += ['--no-hosts', '--addn-hosts=%s' % _dhcp_file(dev, 'hosts')]
     if CONF.dns_server:
         cmd += ['-h', '-R', '--server=%s' % CONF.dns_server]
 
@@ -945,6 +962,12 @@ def _host_dhcp(data):
                                data['address'])
 
 
+def _host_dns(data):
+    return '%s\t%s.%s' % (data['address'],
+                          data['instance_hostname'],
+                          CONF.dhcp_domain)
+
+
 def _host_dhcp_opts(data):
     """Return an empty gateway option."""
     return '%s,%s' % (_host_dhcp_network(data), 3)
@@ -967,7 +990,7 @@ def device_exists(device):
 
 
 def _dhcp_file(dev, kind):
-    """Return path to a pid, leases or conf file for a bridge/device."""
+    """Return path to a pid, leases, hosts or conf file for a bridge/device."""
     fileutils.ensure_tree(CONF.networks_path)
     return os.path.abspath('%s/nova-%s.%s' % (CONF.networks_path,
                                               dev,
diff --git a/nova/network/manager.py b/nova/network/manager.py
index 92b0ec0ae..a7934159b 100644
--- a/nova/network/manager.py
+++ b/nova/network/manager.py
@@ -157,6 +157,11 @@ network_opts = [
                 default=False,
                 help='If True in multi_host mode, all compute hosts share '
                      'the same dhcp address.'),
+    cfg.BoolOpt('update_dns_entries',
+                default=False,
+                help='If True, when a DNS entry must be updated, it sends a '
+                     'fanout cast to all network hosts to update their DNS '
+                     'entries in multi host mode'),
     cfg.StrOpt('dhcp_domain',
                default='novalocal',
                help='domain to use for building the hostnames'),
@@ -871,7 +876,7 @@ class NetworkManager(manager.SchedulerDependentManager):
         The one at a time part is to flatten the layout to help scale
     """
 
-    RPC_API_VERSION = '1.2'
+    RPC_API_VERSION = '1.3'
 
     # If True, this manager requires VIF to create a bridge.
     SHOULD_CREATE_BRIDGE = False
@@ -949,6 +954,9 @@ class NetworkManager(manager.SchedulerDependentManager):
         ctxt = context.get_admin_context()
         for network in self.db.network_get_all_by_host(ctxt, self.host):
             self._setup_network_on_host(ctxt, network)
+            if CONF.update_dns_entries:
+                dev = self.driver.get_dev(network)
+                self.driver.update_dns(ctxt, dev, network)
 
     @manager.periodic_task
     def _disassociate_stale_fixed_ips(self, context):
@@ -1102,6 +1110,11 @@ class NetworkManager(manager.SchedulerDependentManager):
         self._allocate_fixed_ips(admin_context, instance_id,
                                  host, networks, vpn=vpn,
                                  requested_networks=requested_networks)
+
+        if CONF.update_dns_entries:
+            network_ids = [network['id'] for network in networks]
+            self.network_rpcapi.update_dns(context, network_ids)
+
         return self.get_instance_nw_info(context, instance_id, instance_uuid,
                                          rxtx_factor, host)
 
@@ -1119,6 +1132,7 @@ class NetworkManager(manager.SchedulerDependentManager):
 
         instance_id = kwargs.pop('instance_id')
         instance = self.db.instance_get(read_deleted_context, instance_id)
+        host = kwargs.get('host')
 
         try:
             fixed_ips = (kwargs.get('fixed_ips') or
@@ -1130,8 +1144,11 @@ class NetworkManager(manager.SchedulerDependentManager):
                   context=read_deleted_context)
         # deallocate fixed ips
         for fixed_ip in fixed_ips:
-            self.deallocate_fixed_ip(context, fixed_ip['address'],
-                host=kwargs.get('host'))
+            self.deallocate_fixed_ip(context, fixed_ip['address'], host=host)
+
+        if CONF.update_dns_entries:
+            network_ids = [fixed_ip['network_id'] for fixed_ip in fixed_ips]
+            self.network_rpcapi.update_dns(context, network_ids)
 
         # deallocate vifs (mac addresses)
         self.db.virtual_interface_delete_by_instance(read_deleted_context,
@@ -1910,6 +1927,21 @@ class NetworkManager(manager.SchedulerDependentManager):
         return self.db.virtual_interface_get_by_address(context,
                                                         mac_address)
 
+    def update_dns(self, context, network_ids):
+        """Called when fixed IP is allocated or deallocated"""
+        if CONF.fake_network:
+            return
+
+        for network_id in network_ids:
+            network = self.db.network_get(context, network_id)
+            if not network['multi_host']:
+                continue
+            host_networks = self.db.network_get_all_by_host(context, self.host)
+            for host_network in host_networks:
+                if host_network['id'] == network_id:
+                    dev = self.driver.get_dev(network)
+                    self.driver.update_dns(context, dev, network)
+
 
 class FlatManager(NetworkManager):
     """Basic network where no vlans are used.
@@ -2017,6 +2049,10 @@ class FlatManager(NetworkManager):
                                 source=None, dest=None):
         pass
 
+    def update_dns(self, context, network_ids):
+        """Called when fixed IP is allocated or deallocated"""
+        pass
+
 
 class FlatDHCPManager(RPCAllocateFixedIP, FloatingIP, NetworkManager):
     """Flat networking with dhcp.
diff --git a/nova/network/rpcapi.py b/nova/network/rpcapi.py
index 8dd7ca9ae..dc2ae6698 100644
--- a/nova/network/rpcapi.py
+++ b/nova/network/rpcapi.py
@@ -35,6 +35,7 @@ class NetworkAPI(rpc_proxy.RpcProxy):
         1.0 - Initial version.
         1.1 - Adds migrate_instance_[start|finish]
         1.2 - Make migrate_instance_[start|finish] a little more flexible
+        1.3 - Adds fanout cast update_dns for multi_host networks
     '''
 
     #
@@ -240,6 +241,10 @@ class NetworkAPI(rpc_proxy.RpcProxy):
                 address=address, host=host),
                 topic=rpc.queue_get_for(ctxt, self.topic, host))
 
+    def update_dns(self, ctxt, network_ids):
+        return self.fanout_cast(ctxt, self.make_msg('update_dns',
+                network_ids=network_ids), version='1.3')
+
     # NOTE(russellb): Ideally this would not have a prefix of '_' since it is
     # a part of the rpc API. However, this is how it was being called when the
     # 1.0 API was being documented using this client proxy class.  It should be
diff --git a/nova/tests/fake_network.py b/nova/tests/fake_network.py
index fc8e4c249..1b96f95dd 100644
--- a/nova/tests/fake_network.py
+++ b/nova/tests/fake_network.py
@@ -24,6 +24,7 @@ from nova.network import api as network_api
 from nova.network import manager as network_manager
 from nova.network import model as network_model
 from nova.network import nova_ipam_lib
+from nova.network import rpcapi as network_rpcapi
 from nova.openstack.common import cfg
 from nova import utils
 from nova.virt.libvirt import config as libvirt_config
@@ -146,6 +147,7 @@ class FakeNetworkManager(network_manager.NetworkManager):
         self.db = self.FakeDB()
         self.deallocate_called = None
         self.deallocate_fixed_ip_calls = []
+        self.network_rpcapi = network_rpcapi.NetworkAPI()
 
     # TODO(matelakat) method signature should align with the faked one's
     def deallocate_fixed_ip(self, context, address=None, host=None):
diff --git a/nova/tests/network/test_linux_net.py b/nova/tests/network/test_linux_net.py
index 4f94eec95..55a9c7777 100644
--- a/nova/tests/network/test_linux_net.py
+++ b/nova/tests/network/test_linux_net.py
@@ -306,6 +306,24 @@ class LinuxNetworkTestCase(test.TestCase):
 
         self.assertEquals(actual_hosts, expected)
 
+    def test_get_dns_hosts_for_nw00(self):
+        expected = (
+                "192.168.0.100\tfake_instance00.novalocal\n"
+                "192.168.1.101\tfake_instance01.novalocal\n"
+                "192.168.0.102\tfake_instance00.novalocal"
+        )
+        actual_hosts = self.driver.get_dns_hosts(self.context, networks[0])
+        self.assertEquals(actual_hosts, expected)
+
+    def test_get_dns_hosts_for_nw01(self):
+        expected = (
+                "192.168.1.100\tfake_instance00.novalocal\n"
+                "192.168.0.101\tfake_instance01.novalocal\n"
+                "192.168.1.102\tfake_instance01.novalocal"
+        )
+        actual_hosts = self.driver.get_dns_hosts(self.context, networks[1])
+        self.assertEquals(actual_hosts, expected)
+
     def test_get_dhcp_opts_for_nw00(self):
         expected_opts = 'NW-3,3\nNW-4,3'
         actual_opts = self.driver.get_dhcp_opts(self.context, networks[0])
@@ -333,6 +351,12 @@ class LinuxNetworkTestCase(test.TestCase):
         actual = self.driver._host_dhcp(data)
         self.assertEquals(actual, expected)
 
+    def test_host_dns_without_default_gateway_network(self):
+        expected = "192.168.0.100\tfake_instance00.novalocal"
+        data = get_associated(self.context, 0)[0]
+        actual = self.driver._host_dns(data)
+        self.assertEquals(actual, expected)
+
     def test_linux_bridge_driver_plug(self):
         """Makes sure plug doesn't drop FORWARD by default.
 
diff --git a/nova/tests/network/test_manager.py b/nova/tests/network/test_manager.py
index d33b0e582..8d4a511b6 100644
--- a/nova/tests/network/test_manager.py
+++ b/nova/tests/network/test_manager.py
@@ -1120,7 +1120,8 @@ class CommonNetworkTestCase(test.TestCase):
         db.virtual_interface_delete_by_instance = lambda _x, _y: None
         ctx = context.RequestContext('igonre', 'igonre')
 
-        db.fixed_ip_get_by_instance = lambda x, y: [dict(address='1.2.3.4')]
+        db.fixed_ip_get_by_instance = lambda x, y: [dict(address='1.2.3.4',
+                                                    network_id='ignoredid')]
 
         manager.deallocate_for_instance(
             ctx, instance_id='ignore', host='somehost')
diff --git a/nova/tests/network/test_rpcapi.py b/nova/tests/network/test_rpcapi.py
index 760db7ecb..c31b34a51 100644
--- a/nova/tests/network/test_rpcapi.py
+++ b/nova/tests/network/test_rpcapi.py
@@ -43,10 +43,10 @@ class NetworkRpcAPITestCase(test.TestCase):
             args['dest'] = args.pop('dest_compute')
         targeted_methods = [
             'lease_fixed_ip', 'release_fixed_ip', 'rpc_setup_network_on_host',
-            '_rpc_allocate_fixed_ip', 'deallocate_fixed_ip',
+            '_rpc_allocate_fixed_ip', 'deallocate_fixed_ip', 'update_dns',
             '_associate_floating_ip', '_disassociate_floating_ip',
-            'lease_fixed_ip', 'release_fixed_ip',
-            'migrate_instance_start', 'migrate_instance_finish',
+            'lease_fixed_ip', 'release_fixed_ip', 'migrate_instance_start',
+            'migrate_instance_finish',
         ]
         if method in targeted_methods and 'host' in kwargs:
             if method != 'deallocate_fixed_ip':
@@ -250,6 +250,10 @@ class NetworkRpcAPITestCase(test.TestCase):
         self._test_network_api('deallocate_fixed_ip', rpc_method='call',
                 address='fake_addr', host='fake_host')
 
+    def test_update_dns(self):
+        self._test_network_api('update_dns', rpc_method='fanout_cast',
+                network_ids='fake_id', version='1.3')
+
     def test__associate_floating_ip(self):
         self._test_network_api('_associate_floating_ip', rpc_method='call',
                 floating_address='fake_addr', fixed_address='fixed_address',