initial PoC for wsrelay

Checkpoint

Makefile

@@ -223,11 +223,11 @@ daphne:
 	fi; \
 	daphne -b 127.0.0.1 -p 8051 awx.asgi:channel_layer
 
-wsbroadcast:
+wsrelay:
 	@if [ "$(VENV_BASE)" ]; then \
 		. $(VENV_BASE)/awx/bin/activate; \
 	fi; \
-	$(PYTHON) manage.py run_wsbroadcast
+	$(PYTHON) manage.py run_wsrelay
 
 ## Run to start the background task dispatcher for development.
 dispatcher:

awx/conf/settings.py

@@ -5,6 +5,8 @@ import threading
 import time
 import os
 
+from concurrent.futures import ThreadPoolExecutor
+
 # Django
 from django.conf import LazySettings
 from django.conf import settings, UserSettingsHolder
@@ -157,7 +159,7 @@ class EncryptedCacheProxy(object):
             obj_id = self.cache.get(Setting.get_cache_id_key(key), default=empty)
             if obj_id is empty:
                 logger.info('Efficiency notice: Corresponding id not stored in cache %s', Setting.get_cache_id_key(key))
-                obj_id = getattr(self._get_setting_from_db(key), 'pk', None)
+                obj_id = getattr(_get_setting_from_db(self.registry, key), 'pk', None)
             elif obj_id == SETTING_CACHE_NONE:
                 obj_id = None
             return method(TransientSetting(pk=obj_id, value=value), 'value')
@@ -166,11 +168,6 @@ class EncryptedCacheProxy(object):
         # a no-op; it just returns the provided value
         return value
 
-    def _get_setting_from_db(self, key):
-        field = self.registry.get_setting_field(key)
-        if not field.read_only:
-            return Setting.objects.filter(key=key, user__isnull=True).order_by('pk').first()
-
     def __getattr__(self, name):
         return getattr(self.cache, name)
 
@@ -186,6 +183,18 @@ def get_settings_to_cache(registry):
     return dict([(key, SETTING_CACHE_NOTSET) for key in get_writeable_settings(registry)])
 
 
+# HACK: runs in thread in order to work in an asyncio context
+def _get_setting_from_db(registry, key):
+    def wrapped(registry, key):
+        field = registry.get_setting_field(key)
+        if not field.read_only:
+            return Setting.objects.filter(key=key, user__isnull=True).order_by('pk').first()
+
+    with ThreadPoolExecutor(max_workers=1) as executor:
+        future = executor.submit(wrapped, registry, key)
+        return future.result()
+
+
 def get_cache_value(value):
     """Returns the proper special cache setting for a value
     based on instance type.
@@ -345,7 +354,7 @@ class SettingsWrapper(UserSettingsHolder):
             setting_id = None
             # this value is read-only, however we *do* want to fetch its value from the database
             if not field.read_only or name == 'INSTALL_UUID':
-                setting = Setting.objects.filter(key=name, user__isnull=True).order_by('pk').first()
+                setting = _get_setting_from_db(self.registry, name)
             if setting:
                 if getattr(field, 'encrypted', False):
                     value = decrypt_field(setting, 'value')

awx/main/analytics/broadcast_websocket.py

@@ -65,7 +65,7 @@ class FixedSlidingWindow:
         return sum(self.buckets.values()) or 0
 
 
-class BroadcastWebsocketStatsManager:
+class RelayWebsocketStatsManager:
     def __init__(self, event_loop, local_hostname):
         self._local_hostname = local_hostname
 
@@ -74,7 +74,7 @@ class BroadcastWebsocketStatsManager:
         self._redis_key = BROADCAST_WEBSOCKET_REDIS_KEY_NAME
 
     def new_remote_host_stats(self, remote_hostname):
-        self._stats[remote_hostname] = BroadcastWebsocketStats(self._local_hostname, remote_hostname)
+        self._stats[remote_hostname] = RelayWebsocketStats(self._local_hostname, remote_hostname)
         return self._stats[remote_hostname]
 
     def delete_remote_host_stats(self, remote_hostname):
@@ -107,7 +107,7 @@ class BroadcastWebsocketStatsManager:
         return parser.text_string_to_metric_families(stats_str.decode('UTF-8'))
 
 
-class BroadcastWebsocketStats:
+class RelayWebsocketStats:
     def __init__(self, local_hostname, remote_hostname):
         self._local_hostname = local_hostname
         self._remote_hostname = remote_hostname

awx/main/analytics/subsystem_metrics.py

@@ -311,7 +311,14 @@ class Metrics:
                 }
                 # store a local copy as well
                 self.store_metrics(json.dumps(payload))
-                emit_channel_notification("metrics", payload)
+
+                # 🚨🚨🚨🚨🚨🚨🚨🚨
+                # TODO: rework how metrics are emitted and recorded. we used to exploit wsbroadcast's behavior of
+                # sending the same data out to  every other node.
+                # Should we increment this data in redis but ultimately just store it in the database?
+                # emit_channel_notification("metrics", payload)
+                # 🚨🚨🚨🚨🚨🚨🚨🚨
+
                 self.previous_send_metrics.set(current_time)
                 self.previous_send_metrics.store_value(self.conn)
         finally:

awx/main/consumers.py

@@ -80,7 +80,7 @@ class WebsocketSecretAuthHelper:
         WebsocketSecretAuthHelper.verify_secret(secret)
 
 
-class BroadcastConsumer(AsyncJsonWebsocketConsumer):
+class RelayConsumer(AsyncJsonWebsocketConsumer):
     async def connect(self):
         try:
             WebsocketSecretAuthHelper.is_authorized(self.scope)
@@ -100,6 +100,16 @@ class BroadcastConsumer(AsyncJsonWebsocketConsumer):
     async def internal_message(self, event):
         await self.send(event['text'])
 
+    async def receive_json(self, data):
+        (group, message) = unwrap_broadcast_msg(data)
+        await self.channel_layer.group_send(group, message)
+
+    async def consumer_subscribe(self, event):
+        await self.send_json(event)
+
+    async def consumer_unsubscribe(self, event):
+        await self.send_json(event)
+
 
 class EventConsumer(AsyncJsonWebsocketConsumer):
     async def connect(self):
@@ -128,6 +138,11 @@ class EventConsumer(AsyncJsonWebsocketConsumer):
                 self.channel_name,
             )
 
+        await self.channel_layer.group_send(
+            settings.BROADCAST_WEBSOCKET_GROUP_NAME,
+            {"type": "consumer.unsubscribe", "groups": list(current_groups), "origin_channel": self.channel_name},
+        )
+
     @database_sync_to_async
     def user_can_see_object_id(self, user_access, oid):
         # At this point user is a channels.auth.UserLazyObject object
@@ -176,9 +191,20 @@ class EventConsumer(AsyncJsonWebsocketConsumer):
                     self.channel_name,
                 )
 
+            if len(old_groups):
+                await self.channel_layer.group_send(
+                    settings.BROADCAST_WEBSOCKET_GROUP_NAME,
+                    {"type": "consumer.unsubscribe", "groups": list(old_groups), "origin_channel": self.channel_name},
+                )
+
             new_groups_exclusive = new_groups - current_groups
             for group_name in new_groups_exclusive:
                 await self.channel_layer.group_add(group_name, self.channel_name)
+
+                await self.channel_layer.group_send(
+                    settings.BROADCAST_WEBSOCKET_GROUP_NAME,
+                    {"type": "consumer.subscribe", "groups": list(new_groups), "origin_channel": self.channel_name},
+                )
             self.scope['session']['groups'] = new_groups
             await self.send_json({"groups_current": list(new_groups), "groups_left": list(old_groups), "groups_joined": list(new_groups_exclusive)})
 
@@ -200,9 +226,11 @@ def _dump_payload(payload):
         return None
 
 
-def emit_channel_notification(group, payload):
+def unwrap_broadcast_msg(payload: dict):
-    from awx.main.wsbroadcast import wrap_broadcast_msg  # noqa
+    return (payload['group'], payload['message'])
 
+
+def emit_channel_notification(group, payload):
     payload_dumped = _dump_payload(payload)
     if payload_dumped is None:
         return
@@ -215,13 +243,3 @@ def emit_channel_notification(group, payload):
             {"type": "internal.message", "text": payload_dumped},
         )
     )
-
-    run_sync(
-        channel_layer.group_send(
-            settings.BROADCAST_WEBSOCKET_GROUP_NAME,
-            {
-                "type": "internal.message",
-                "text": wrap_broadcast_msg(group, payload_dumped),
-            },
-        )
-    )

awx/main/db/profiled_pg/base.py

@@ -63,7 +63,7 @@ class RecordedQueryLog(object):
             if not os.path.isdir(self.dest):
                 os.makedirs(self.dest)
             progname = ' '.join(sys.argv)
-            for match in ('uwsgi', 'dispatcher', 'callback_receiver', 'wsbroadcast'):
+            for match in ('uwsgi', 'dispatcher', 'callback_receiver', 'wsrelay'):
                 if match in progname:
                     progname = match
                     break

awx/main/management/commands/run_wsrelay.py

@@ -13,13 +13,13 @@ from django.db import connection
 from django.db.migrations.executor import MigrationExecutor
 
 from awx.main.analytics.broadcast_websocket import (
-    BroadcastWebsocketStatsManager,
+    RelayWebsocketStatsManager,
     safe_name,
 )
-from awx.main.wsbroadcast import BroadcastWebsocketManager
+from awx.main.wsrelay import WebSocketRelayManager
 
 
-logger = logging.getLogger('awx.main.wsbroadcast')
+logger = logging.getLogger('awx.main.wsrelay')
 
 
 class Command(BaseCommand):
@@ -99,7 +99,7 @@ class Command(BaseCommand):
             executor = MigrationExecutor(connection)
             migrating = bool(executor.migration_plan(executor.loader.graph.leaf_nodes()))
         except Exception as exc:
-            logger.info(f'Error on startup of run_wsbroadcast (error: {exc}), retry in 10s...')
+            logger.info(f'Error on startup of run_wsrelay (error: {exc}), retry in 10s...')
             time.sleep(10)
             return
 
@@ -130,9 +130,9 @@ class Command(BaseCommand):
 
         if options.get('status'):
             try:
-                stats_all = BroadcastWebsocketStatsManager.get_stats_sync()
+                stats_all = RelayWebsocketStatsManager.get_stats_sync()
             except redis.exceptions.ConnectionError as e:
-                print(f"Unable to get Broadcast Websocket Status. Failed to connect to redis {e}")
+                print(f"Unable to get Relay Websocket Status. Failed to connect to redis {e}")
                 return
 
             data = {}
@@ -151,22 +151,19 @@ class Command(BaseCommand):
             host_stats = Command.get_connection_status(hostnames, data)
             lines = Command._format_lines(host_stats)
 
-            print(f'Broadcast websocket connection status from "{my_hostname}" to:')
+            print(f'Relay websocket connection status from "{my_hostname}" to:')
             print('\n'.join(lines))
 
             host_stats = Command.get_connection_stats(hostnames, data)
             lines = Command._format_lines(host_stats)
 
-            print(f'\nBroadcast websocket connection stats from "{my_hostname}" to:')
+            print(f'\nRelay websocket connection stats from "{my_hostname}" to:')
             print('\n'.join(lines))
 
             return
 
         try:
-            broadcast_websocket_mgr = BroadcastWebsocketManager()
+            websocket_relay_manager = WebSocketRelayManager()
-            task = broadcast_websocket_mgr.start()
+            asyncio.run(websocket_relay_manager.run())
-
-            loop = asyncio.get_event_loop()
-            loop.run_until_complete(task)
         except KeyboardInterrupt:
-            logger.debug('Terminating Websocket Broadcaster')
+            logger.debug('Terminating Websocket Relayer')

awx/main/routing.py

@@ -27,8 +27,8 @@ class AWXProtocolTypeRouter(ProtocolTypeRouter):
 
 
 websocket_urlpatterns = [
-    re_path(r'websocket/$', consumers.EventConsumer.as_asgi()),
+    re_path(r'websocket/$', consumers.EventConsumer),
-    re_path(r'websocket/broadcast/$', consumers.BroadcastConsumer.as_asgi()),
+    re_path(r'websocket/relay/$', consumers.RelayConsumer),
 ]
 
 application = AWXProtocolTypeRouter(

awx/main/wsrelay.py

@@ -7,28 +7,24 @@ from aiohttp import client_exceptions
 from asgiref.sync import sync_to_async
 
 from channels.layers import get_channel_layer
+from channels.db import database_sync_to_async
 
 from django.conf import settings
 from django.apps import apps
-from django.core.serializers.json import DjangoJSONEncoder
 
 from awx.main.analytics.broadcast_websocket import (
-    BroadcastWebsocketStats,
+    RelayWebsocketStats,
-    BroadcastWebsocketStatsManager,
+    RelayWebsocketStatsManager,
 )
 import awx.main.analytics.subsystem_metrics as s_metrics
 
-logger = logging.getLogger('awx.main.wsbroadcast')
+logger = logging.getLogger('awx.main.wsrelay')
 
 
 def wrap_broadcast_msg(group, message: str):
     # TODO: Maybe wrap as "group","message" so that we don't need to
     # encode/decode as json.
-    return json.dumps(dict(group=group, message=message), cls=DjangoJSONEncoder)
+    return dict(group=group, message=message)
-
-
-def unwrap_broadcast_msg(payload: dict):
-    return (payload['group'], payload['message'])
 
 
 @sync_to_async
@@ -50,28 +46,26 @@ def get_local_host():
     return Instance.objects.my_hostname()
 
 
-class WebsocketTask:
+class WebsocketRelayConnection:
     def __init__(
         self,
         name,
-        event_loop,
+        stats: RelayWebsocketStats,
-        stats: BroadcastWebsocketStats,
         remote_host: str,
         remote_port: int = settings.BROADCAST_WEBSOCKET_PORT,
         protocol: str = settings.BROADCAST_WEBSOCKET_PROTOCOL,
         verify_ssl: bool = settings.BROADCAST_WEBSOCKET_VERIFY_CERT,
-        endpoint: str = 'broadcast',
     ):
         self.name = name
-        self.event_loop = event_loop
+        self.event_loop = asyncio.get_event_loop()
         self.stats = stats
         self.remote_host = remote_host
         self.remote_port = remote_port
-        self.endpoint = endpoint
         self.protocol = protocol
         self.verify_ssl = verify_ssl
         self.channel_layer = None
         self.subsystem_metrics = s_metrics.Metrics(instance_name=name)
+        self.producers = dict()
 
     async def run_loop(self, websocket: aiohttp.ClientWebSocketResponse):
         raise RuntimeError("Implement me")
@@ -95,7 +89,7 @@ class WebsocketTask:
             logger.warning(f"Connection from {self.name} to {self.remote_host} cancelled")
             raise
 
-        uri = f"{self.protocol}://{self.remote_host}:{self.remote_port}/websocket/{self.endpoint}/"
+        uri = f"{self.protocol}://{self.remote_host}:{self.remote_port}/websocket/relay/"
         timeout = aiohttp.ClientTimeout(total=10)
 
         secret_val = WebsocketSecretAuthHelper.construct_secret()
@@ -105,7 +99,7 @@ class WebsocketTask:
                     logger.info(f"Connection from {self.name} to {self.remote_host} established.")
                     self.stats.record_connection_established()
                     attempt = 0
-                    await self.run_loop(websocket)
+                    await self.run_connection(websocket)
         except asyncio.CancelledError:
             # TODO: Check if connected and disconnect
             # Possibly use run_until_complete() if disconnect is async
@@ -128,12 +122,12 @@ class WebsocketTask:
     def start(self, attempt=0):
         self.async_task = self.event_loop.create_task(self.connect(attempt=attempt))
 
+        return self.async_task
+
     def cancel(self):
         self.async_task.cancel()
 
-
+    async def run_connection(self, websocket: aiohttp.ClientWebSocketResponse):
-class BroadcastWebsocketTask(WebsocketTask):
-    async def run_loop(self, websocket: aiohttp.ClientWebSocketResponse):
         async for msg in websocket:
             self.stats.record_message_received()
 
@@ -148,38 +142,79 @@ class BroadcastWebsocketTask(WebsocketTask):
                         logmsg = "{} {}".format(logmsg, payload)
                     logger.warning(logmsg)
                     continue
-                (group, message) = unwrap_broadcast_msg(payload)
+
-                if group == "metrics":
+            from remote_pdb import RemotePdb
-                    self.subsystem_metrics.store_metrics(message)
+
+            RemotePdb('127.0.0.1', 4444).set_trace()
+
+            if payload.get("type") == "consumer.subscribe":
+                for group in payload['groups']:
+                    name = f"{self.remote_host}-{group}"
+                    origin_channel = payload['origin_channel']
+                    if not self.producers.get(name):
+                        producer = self.event_loop.create_task(self.run_producer(name, websocket, group))
+
+                        self.producers[name] = {"task": producer, "subscriptions": {origin_channel}}
+                    else:
+                        self.producers[name]["subscriptions"].add(origin_channel)
+
+            if payload.get("type") == "consumer.unsubscribe":
+                for group in payload['groups']:
+                    name = f"{self.remote_host}-{group}"
+                    origin_channel = payload['origin_channel']
+                    self.producers[name]["subscriptions"].remove(origin_channel)
+
+    async def run_producer(self, name, websocket, group):
+        try:
+            logger.info(f"Starting producer for {name}")
+
+            consumer_channel = await self.channel_layer.new_channel()
+            await self.channel_layer.group_add(group, consumer_channel)
+
+            while True:
+                try:
+                    msg = await asyncio.wait_for(self.channel_layer.receive(consumer_channel), timeout=10)
+                except asyncio.TimeoutError:
+                    current_subscriptions = self.producers[name]["subscriptions"]
+                    if len(current_subscriptions) == 0:
+                        logger.info(f"Producer {name} has no subscribers, shutting down.")
+                        return
+
                     continue
-                await self.channel_layer.group_send(group, {"type": "internal.message", "text": message})
+
+                await websocket.send_json(wrap_broadcast_msg(group, msg))
+        except Exception:
+            # Note, this is very intentional and important since we do not otherwise
+            # ever check the result of this future. Without this line you will not see an error if
+            # something goes wrong in here.
+            logger.exception(f"Event relay producer {name} crashed")
+        finally:
+            await self.channel_layer.group_discard(group, consumer_channel)
+            del self.producers[name]
 
 
-class BroadcastWebsocketManager(object):
+class WebSocketRelayManager(object):
     def __init__(self):
-        self.event_loop = asyncio.get_event_loop()
-        '''
-        {
-            'hostname1': BroadcastWebsocketTask(),
-            'hostname2': BroadcastWebsocketTask(),
-            'hostname3': BroadcastWebsocketTask(),
-        }
-        '''
-        self.broadcast_tasks = dict()
-        self.local_hostname = get_local_host()
-        self.stats_mgr = BroadcastWebsocketStatsManager(self.event_loop, self.local_hostname)
 
-    async def run_per_host_websocket(self):
+        self.relay_connections = dict()
+        self.local_hostname = get_local_host()
+        self.event_loop = asyncio.get_event_loop()
+        self.stats_mgr = RelayWebsocketStatsManager(self.event_loop, self.local_hostname)
+
+    async def run(self):
+        self.stats_mgr.start()
+
+        # Establishes a websocket connection to /websocket/relay on all API servers
         while True:
             known_hosts = await get_broadcast_hosts()
             future_remote_hosts = known_hosts.keys()
-            current_remote_hosts = self.broadcast_tasks.keys()
+            current_remote_hosts = self.relay_connections.keys()
             deleted_remote_hosts = set(current_remote_hosts) - set(future_remote_hosts)
             new_remote_hosts = set(future_remote_hosts) - set(current_remote_hosts)
 
-            remote_addresses = {k: v.remote_host for k, v in self.broadcast_tasks.items()}
+            remote_addresses = {k: v.remote_host for k, v in self.relay_connections.items()}
             for hostname, address in known_hosts.items():
-                if hostname in self.broadcast_tasks and address != remote_addresses[hostname]:
+                if hostname in self.relay_connections and address != remote_addresses[hostname]:
                     deleted_remote_hosts.add(hostname)
                     new_remote_hosts.add(hostname)
 
@@ -189,20 +224,17 @@ class BroadcastWebsocketManager(object):
                 logger.warning(f"Adding {new_remote_hosts} to websocket broadcast list")
 
             for h in deleted_remote_hosts:
-                self.broadcast_tasks[h].cancel()
+                self.relay_connections[h].cancel()
-                del self.broadcast_tasks[h]
+                del self.relay_connections[h]
                 self.stats_mgr.delete_remote_host_stats(h)
 
             for h in new_remote_hosts:
                 stats = self.stats_mgr.new_remote_host_stats(h)
-                broadcast_task = BroadcastWebsocketTask(name=self.local_hostname, event_loop=self.event_loop, stats=stats, remote_host=known_hosts[h])
+                relay_connection = WebsocketRelayConnection(name=self.local_hostname, stats=stats, remote_host=known_hosts[h])
-                broadcast_task.start()
+                relay_connection.start()
-                self.broadcast_tasks[h] = broadcast_task
+                self.relay_connections[h] = relay_connection
+
+            # for host, conn in self.relay_connections.items():
+            #     logger.info(f"Current producers for {host}: {conn.producers}")
 
             await asyncio.sleep(settings.BROADCAST_WEBSOCKET_NEW_INSTANCE_POLL_RATE_SECONDS)
-
-    def start(self):
-        self.stats_mgr.start()
-
-        self.async_task = self.event_loop.create_task(self.run_per_host_websocket())
-        return self.async_task

awx/settings/defaults.py

@@ -866,9 +866,9 @@ LOGGING = {
         'awx.main.commands.run_callback_receiver': {'handlers': ['callback_receiver']},  # level handled by dynamic_level_filter
         'awx.main.dispatch': {'handlers': ['dispatcher']},
         'awx.main.consumers': {'handlers': ['console', 'file', 'tower_warnings'], 'level': 'INFO'},
-        'awx.main.wsbroadcast': {'handlers': ['wsbroadcast']},
         'awx.main.rsyslog_configurer': {'handlers': ['rsyslog_configurer']},
         'awx.main.cache_clear': {'handlers': ['cache_clear']},
+        'awx.main.wsrelay': {'handlers': ['wsrelay']},
         'awx.main.commands.inventory_import': {'handlers': ['inventory_import'], 'propagate': False},
         'awx.main.tasks': {'handlers': ['task_system', 'external_logger'], 'propagate': False},
         'awx.main.analytics': {'handlers': ['task_system', 'external_logger'], 'level': 'INFO', 'propagate': False},
@@ -895,7 +895,7 @@ handler_config = {
     'tower_warnings': {'filename': 'tower.log'},
     'callback_receiver': {'filename': 'callback_receiver.log'},
     'dispatcher': {'filename': 'dispatcher.log', 'formatter': 'dispatcher'},
-    'wsbroadcast': {'filename': 'wsbroadcast.log'},
+    'wsrelay': {'filename': 'wsrelay.log'},
     'task_system': {'filename': 'task_system.log'},
     'rbac_migrations': {'filename': 'tower_rbac_migrations.log'},
     'job_lifecycle': {'filename': 'job_lifecycle.log', 'formatter': 'job_lifecycle'},

docs/websockets.md

@@ -10,13 +10,13 @@ To communicate between our different services we use websockets. Every AWX node
 
 Inside AWX we use the `emit_channel_notification` function which places messages onto the queue. The messages are given an explicit event group and event type which we later use in our wire protocol to control message delivery to the client.
 
-### Broadcast Backplane
+### Relay Backplane
 
-Previously, AWX leveraged RabbitMQ to deliver Ansible events that emanated from one AWX node to all other AWX nodes so that any client listening and subscribed to the Websockets could get events from any running playbook. We are since moved off of RabbitMQ and onto a per-node local Redis instance. To maintain the requirement that any Websocket connection can receive events from any playbook running on any AWX node we still need to deliver every event to every AWX node. AWX does this via a fully connected Websocket backplane. 
+Previously, AWX leveraged RabbitMQ to deliver Ansible events that emanated from one AWX node to all other AWX nodes so that any client listening and subscribed to the Websockets could get events from any running playbook. We are since moved off of RabbitMQ and onto a per-node local Redis instance. To maintain the requirement that any Websocket connection can receive events from any playbook running on any AWX node we still need to deliver every event to every AWX node. AWX does this via a fully connected Websocket backplane.
 
-#### Broadcast Backplane Token
+#### Relay Backplane Token
 
-AWX node(s) connect to every other node via the Websocket backplane. The backplane websockets initiate from the `wsbroadcast` process and connect to other nodes via the same nginx process that serves webpage websocket connections and marshalls incoming web/API requests. If you have configured AWX to run with an ssl terminated connection in front of nginx then you likely will have nginx configured to handle http traffic and thus the websocket connection will flow unencrypted over http. If you have nginx configured with ssl enabled, then the websocket traffic will flow encrypted.
+AWX node(s) connect to every other node via the Websocket backplane. The backplane websockets initiate from the `wsrelay` process and connect to other nodes via the same nginx process that serves webpage websocket connections and marshalls incoming web/API requests. If you have configured AWX to run with an ssl terminated connection in front of nginx then you likely will have nginx configured to handle http traffic and thus the websocket connection will flow unencrypted over http. If you have nginx configured with ssl enabled, then the websocket traffic will flow encrypted.
 
 Authentication is accomplished via a shared secret that is generated and set at playbook install time. The shared secret is used to derive a payload that is exchanged via the http(s) header `secret`. The shared secret payload consists of a a `secret`, containing the shared secret, and a `nonce` which is used to mitigate replay attack windows.
 
@@ -65,14 +65,14 @@ This section will specifically discuss deployment in the context of websockets a
 | `nginx`     | listens on ports 80/443, handles HTTPS proxying, serves static assets, routes requests for `daphne` and `uwsgi` |
 | `uwsgi`      | listens on port 8050, handles API requests |
 | `daphne`      | listens on port 8051, handles websocket requests |
-| `wsbroadcast`   | no listening port, forwards all group messages to all cluster nodes |
+| `wsrelay`   | no listening port, forwards all group messages to all cluster nodes |
 | `supervisord` | (production-only) handles the process management of all the services except `nginx` |
 
 When a request comes in to `nginx` and has the `Upgrade` header and is for the path `/websocket`, then `nginx` knows that it should be routing that request to our `daphne` service.
 
 `daphne` handles websocket connections proxied by nginx.
 
-`wsbroadcast` fully connects all cluster nodes via the `/websocket/broadcast/` endpoint to every other cluster nodes. Sends a copy of all group websocket messages to all other cluster nodes (i.e. job event type messages).
+`wsrelay` fully connects all cluster nodes via the `/websocket/broadcast/` endpoint to every other cluster nodes. Sends a copy of all group websocket messages to all other cluster nodes (i.e. job event type messages).
 
 ### Development
  - `nginx` listens on 8013/8043 instead of 80/443

tools/ansible/roles/dockerfile/templates/supervisor.conf.j2

@@ -22,12 +22,11 @@ stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 
 [program:uwsgi]
-
 {% if kube_dev | bool %}
 command = make uwsgi
 directory = /awx_devel
 environment =
-    DEV_RELOAD_COMMAND='supervisorctl -c /etc/supervisord_task.conf restart all; supervisorctl restart tower-processes:daphne tower-processes:wsbroadcast'
+    DEV_RELOAD_COMMAND='supervisorctl -c /etc/supervisord_task.conf restart all; supervisorctl restart tower-processes:daphne'
 {% else %}
 command = /var/lib/awx/venv/awx/bin/uwsgi /etc/tower/uwsgi.ini
 {% endif %}
@@ -58,12 +57,12 @@ stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 
-[program:wsbroadcast]
+[program:heartbeet]
 {% if kube_dev | bool %}
-command = make wsbroadcast
+command = make heartbeet
 directory = /awx_devel
 {% else %}
-command = awx-manage run_wsbroadcast
+command = awx-manage run_heartbeet
 directory = /var/lib/awx
 {% endif %}
 autorestart = true
@@ -87,7 +86,7 @@ stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 
 [group:tower-processes]
-programs=nginx,uwsgi,daphne,wsbroadcast,awx-cache-clear
+programs=nginx,uwsgi,daphne,awx-cache-clear,heartbeet
 priority=5
 
 [eventlistener:superwatcher]

tools/ansible/roles/dockerfile/templates/supervisor_task.conf.j2

@@ -22,6 +22,23 @@ stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 
+[program:wsrelay]
+{% if kube_dev | bool %}
+command = make wsrelay
+directory = /awx_devel
+{% else %}
+command = awx-manage run_wsrelay
+directory = /var/lib/awx
+{% endif %}
+autorestart = true
+startsecs = 30
+stopasgroup=true
+killasgroup=true
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+
 [program:callback-receiver]
 {% if kube_dev | bool %}
 command = make receiver

tools/docker-compose/README.md

@@ -293,7 +293,7 @@ Certain features or bugs are only applicable when running a cluster of AWX nodes
 
 `CONTROL_PLANE_NODE_COUNT` is configurable and defaults to 1, effectively a non-clustered AWX.
 
-Note that you may see multiple messages of the form `2021-03-04 20:11:47,666 WARNING [-] awx.main.wsbroadcast Connection from awx_2 to awx_5 failed: 'Cannot connect to host awx_5:8013 ssl:False [Name or service not known]'.`. This can happen when you bring up a cluster of many nodes, say 10, then you bring up a cluster of less nodes, say 3. In this example, there will be 7 `Instance` records in the database that represent AWX instances. The AWX development environment mimics the VM deployment (vs. kubernetes) and expects the missing nodes to be brought back to healthy by the admin. The warning message you are seeing is all of the AWX nodes trying to connect the websocket backplane. You can manually delete the `Instance` records from the database i.e. `Instance.objects.get(hostname='awx_9').delete()` to stop the warnings.
+Note that you may see multiple messages of the form `2021-03-04 20:11:47,666 WARNING [-] awx.main.wsrelay Connection from awx_2 to awx_5 failed: 'Cannot connect to host awx_5:8013 ssl:False [Name or service not known]'.`. This can happen when you bring up a cluster of many nodes, say 10, then you bring up a cluster of less nodes, say 3. In this example, there will be 7 `Instance` records in the database that represent AWX instances. The AWX development environment mimics the VM deployment (vs. kubernetes) and expects the missing nodes to be brought back to healthy by the admin. The warning message you are seeing is all of the AWX nodes trying to connect the websocket backplane. You can manually delete the `Instance` records from the database i.e. `Instance.objects.get(hostname='awx_9').delete()` to stop the warnings.
 
 ### Start with Minikube
 

tools/docker-compose/ansible/roles/sources/templates/haproxy.cfg.j2

@@ -32,7 +32,7 @@ backend nodes
     option httpchk HEAD / HTTP/1.1\r\nHost:localhost
 {% for i in range(control_plane_node_count|int) %}
     {% set container_postfix = loop.index %}
-    server tools_awx_{{ container_postfix }} tools_awx_{{ container_postfix }}:8013 check
+    server tools_awx_{{ container_postfix }} tools_awx_{{ container_postfix }}:8013 check inter 10s
 {% endfor %}
 
 backend nodes_ssl
@@ -40,7 +40,7 @@ backend nodes_ssl
     balance roundrobin
 {% for i in range(control_plane_node_count|int) %}
     {% set container_postfix = loop.index %}
-    server tools_awx_{{ container_postfix }} tools_awx_{{ container_postfix }}:8043 check
+    server tools_awx_{{ container_postfix }} tools_awx_{{ container_postfix }}:8043 check inter 10s
 {% endfor %}
 
 listen stats

tools/docker-compose/supervisor.conf

@@ -8,31 +8,25 @@ command = make dispatcher
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [program:awx-receiver]
 command = make receiver
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
-[program:awx-wsbroadcast]
+[program:awx-wsrelay]
-command = make wsbroadcast
+command = make wsrelay
 autorestart = true
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [program:awx-rsyslog-configurer]
 command = make rsyslog-configurer
@@ -61,30 +55,24 @@ stopwaitsecs = 1
 stopsignal=KILL
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [program:awx-daphne]
 command = make daphne
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [program:awx-nginx]
 command = make nginx
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [program:awx-rsyslogd]
 command = rsyslogd -n -i /var/run/awx-rsyslog/rsyslog.pid -f /var/lib/awx/rsyslog/rsyslog.conf
@@ -100,13 +88,11 @@ command = receptor --config /etc/receptor/receptor.conf
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
+stdout_events_enabled = true
-stdout_logfile_maxbytes=0
+stderr_events_enabled = true
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 
 [group:tower-processes]
-programs=awx-dispatcher,awx-receiver,awx-uwsgi,awx-daphne,awx-nginx,awx-wsbroadcast,awx-rsyslogd, awx-cache-clear
+programs=awx-dispatcher,awx-receiver,awx-uwsgi,awx-daphne,awx-nginx,awx-wsrelay,awx-rsyslogd,awx-cache-clear
 priority=5
 
 [program:awx-autoreload]
@@ -115,10 +101,6 @@ autostart = true
 autorestart = true
 stopasgroup=true
 killasgroup=true
-stdout_logfile=/dev/stdout
-stdout_logfile_maxbytes=0
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
 stdout_events_enabled = true
 stderr_events_enabled = true
 
@@ -127,9 +109,6 @@ command=stop-supervisor
 events=PROCESS_STATE_FATAL
 autorestart = true
 stderr_logfile=/dev/stdout
-stderr_logfile_maxbytes=0
-stdout_logfile=/dev/stdout
-stdout_logfile_maxbytes=0
 
 [unix_http_server]
 file=/var/run/supervisor/supervisor.sock