Process conntrack updates in worker threads

With a large number of instances and/or security group rules,
conntrack updates when ports are removed or rules are changed
can take a long time to process.  By enqueuing these to a set
or worker threads, the agent can continue with other work while
they are processed in the background.

This is a change in behavior in the agent since it could
program a new set of security group rules before all existing
conntrack entries are deleted, but since the iptables or OVSfw
NAT rules will have been removed, it should not pose a
security issue.

Change-Id: Ibf858c7fdf7a822a30e4a0c4722d70fd272741b6
Closes-bug: #1745468

neutron/agent/linux/ip_conntrack.py

@@ -13,9 +13,11 @@
 
 import re
 
+import eventlet
 import netaddr
 from oslo_concurrency import lockutils
 from oslo_log import log as logging
+from six.moves import queue as Queue
 
 from neutron.agent.linux import utils as linux_utils
 from neutron.common import constants as n_const
@@ -27,6 +29,33 @@ MAX_CONNTRACK_ZONES = 65535
 ZONE_START = 4097
 
 
+class IpConntrackUpdate(object):
+    """Encapsulates a conntrack update
+
+    An instance of this object carries the information necessary to
+    process a request to update the conntrack table.
+    """
+    def __init__(self, device_info_list, rule, remote_ips):
+        self.device_info_list = device_info_list
+        self.rule = rule
+        self.remote_ips = remote_ips
+
+
+class IpConntrackProcessingQueue(object):
+    """Manager of the queue of conntrack updates to process."""
+    def __init__(self):
+        self._queue = Queue.Queue()
+
+    def add(self, update):
+        self._queue.put(update)
+
+    def updates(self):
+        """Grabs the next conntrack update from the queue and processes."""
+        while not self._queue.empty():
+            update = self._queue.get()
+            yield update
+
+
 @lockutils.synchronized('conntrack')
 def get_conntrack(get_rules_for_table_func, filtered_ports, unfiltered_ports,
                   execute=None, namespace=None, zone_per_port=False):
@@ -53,6 +82,32 @@ class IpConntrackManager(object):
         self.unfiltered_ports = unfiltered_ports
         self.zone_per_port = zone_per_port  # zone per port vs per network
         self._populate_initial_zone_map()
+        self._queue = IpConntrackProcessingQueue()
+        self.start_process_queue()
+
+    def start_process_queue(self):
+        eventlet.spawn_n(self._process_queue_loop)
+
+    def _process_queue_loop(self):
+        LOG.debug("Starting ipconntrack _process_queue_loop()")
+        pool = eventlet.GreenPool(size=8)
+        while True:
+            pool.spawn_n(self._process_queue)
+
+    def _process_queue(self):
+        for update in self._queue.updates():
+            if update.remote_ips:
+                for remote_ip in update.remote_ips:
+                    self._delete_conntrack_state(
+                        update.device_info_list, update.rule, remote_ip)
+            else:
+                self._delete_conntrack_state(update.device_info_list,
+                                             update.rule)
+
+    def _process(self, device_info_list, rule, remote_ips=None):
+        # queue the update to allow the caller to resume its work
+        update = IpConntrackUpdate(device_info_list, rule, remote_ips)
+        self._queue.add(update)
 
     @staticmethod
     def _generate_conntrack_cmd_by_rule(rule, namespace):
@@ -110,19 +165,14 @@ class IpConntrackManager(object):
                 LOG.exception("Failed execute conntrack command %s", cmd)
 
     def delete_conntrack_state_by_rule(self, device_info_list, rule):
-        self._delete_conntrack_state(device_info_list, rule)
+        self._process(device_info_list, rule)
 
     def delete_conntrack_state_by_remote_ips(self, device_info_list,
                                              ethertype, remote_ips):
         for direction in ['ingress', 'egress']:
             rule = {'ethertype': str(ethertype).lower(),
                     'direction': direction}
-            if remote_ips:
-                for remote_ip in remote_ips:
-                    self._delete_conntrack_state(
-                        device_info_list, rule, remote_ip)
-            else:
-                self._delete_conntrack_state(device_info_list, rule)
+            self._process(device_info_list, rule, remote_ips)
 
     def _populate_initial_zone_map(self):
         """Setup the map between devices and zones based on current rules."""

neutron/tests/unit/agent/linux/test_iptables_firewall.py

@@ -70,6 +70,7 @@ COMMIT
 class BaseIptablesFirewallTestCase(base.BaseTestCase):
     def setUp(self):
         super(BaseIptablesFirewallTestCase, self).setUp()
+        mock.patch('eventlet.spawn_n').start()
         security_config.register_securitygroups_opts()
         agent_config.register_root_helper(cfg.CONF)
         cfg.CONF.set_override('comment_iptables_rules', False, 'AGENT')
@@ -1247,6 +1248,8 @@ class IptablesFirewallTestCase(BaseIptablesFirewallTestCase):
             if not ct_zone:
                 self.assertFalse(self.utils_exec.called)
                 return
+            # process conntrack updates in the queue
+            self.firewall.ipconntrack._process_queue()
             cmd = ['conntrack', '-D']
             if protocol:
                 cmd.extend(['-p', protocol])
@@ -1335,6 +1338,8 @@ class IptablesFirewallTestCase(BaseIptablesFirewallTestCase):
             if not ct_zone:
                 self.assertFalse(self.utils_exec.called)
                 return
+            # process conntrack updates in the queue
+            self.firewall.ipconntrack._process_queue()
             calls = self._get_expected_conntrack_calls(
                 [('ipv4', '10.0.0.1'), ('ipv6', 'fe80::1')], ct_zone)
             self.utils_exec.assert_has_calls(calls)
@@ -1398,6 +1403,8 @@ class IptablesFirewallTestCase(BaseIptablesFirewallTestCase):
             ips = {"ipv4": ['10.0.0.1', '10.0.0.2'],
                    "ipv6": ['fe80::1', 'fe80::2']}
             calls = []
+            # process conntrack updates in the queue
+            self.firewall.ipconntrack._process_queue()
             for direction in ['ingress', 'egress']:
                 direction = '-d' if direction == 'ingress' else '-s'
                 remote_ip_direction = '-s' if direction == '-d' else '-d'
@@ -1642,6 +1649,8 @@ class IptablesFirewallTestCase(BaseIptablesFirewallTestCase):
         if not ct_zone:
             self.assertFalse(self.utils_exec.called)
             return
+        # process conntrack updates in the queue
+        self.firewall.ipconntrack._process_queue()
         calls = self._get_expected_conntrack_calls(
             [('ipv4', '10.0.0.1'), ('ipv6', 'fe80::1')], ct_zone)
         self.utils_exec.assert_has_calls(calls)

releasenotes/notes/add-conntrack-workers-89d303e9ec3b4963.yaml

@@ -0,0 +1,13 @@
+---
+prelude: >
+    In order to reduce the time spent processing security group updates in
+    the L2 agent, conntrack deletion is now performed in a set of worker
+    threads instead of the main agent thread, so it can return to processing
+    other events quickly.
+upgrade:
+  - |
+    On an upgrade, conntrack entries will now be cleaned-up in a worker
+    thread, instead of in the calling thread.
+fixes:
+  - |
+    Fixes bug `1745468 <https://bugs.launchpad.net/neutron/+bug/1745468>`_.