Support ChatBot health checks

.github/workflows/license_tests.yml

@@ -9,4 +9,4 @@ jobs:
   license_tests:
     uses: neongeckocom/.github/.github/workflows/license_tests.yml@master
     with:
-      packages-exclude: '^(bs4|nvidia|bitstruct|audioread|klat-connector|neon-chatbot-core|tqdm|dnspython|RapidFuzz|typing_extensions|wheel|urllib).*'
+      packages-exclude: '^(bs4|nvidia|bitstruct|audioread|klat-connector|neon-chatbot-core|tqdm|dnspython|RapidFuzz|typing_extensions|wheel|urllib|click|typing-inspection).*'

chatbot_core/chatbot_abc.py

@@ -28,6 +28,7 @@
 
 from neon_utils.log_utils import init_log
 from ovos_utils.log import LOG
+from ovos_utils.process_utils import ProcessStatus
 
 
 class ChatBotABC(ABC):
@@ -45,6 +46,13 @@ def __init__(self, bot_id: str, config: dict = None):
         self.shout_queue = Queue(maxsize=256)
         self.__log = None
 
+    @property
+    def status(self) -> Optional[ProcessStatus]:
+        """"Get a ProcessStatus object for this bot"""
+        if hasattr(self, '_status'):
+            return self._status
+        return None
+
     @property
     def log(self):
         if not self.__log:

chatbot_core/cli.py

@@ -21,6 +21,7 @@
 import argparse
 import click
 
+from typing import Optional
 from os.path import expanduser, relpath
 
 from ovos_utils import wait_for_exit_signal
@@ -40,11 +41,17 @@ def chatbot_core_cli(version: bool = False):
 
 
 @chatbot_core_cli.command(help="Start an MQ chatbot")
+@click.option("--health-check-server-port", "-hp", type=int, default=None,
+              help="Port for health check server to listen on")
 @click.argument("bot_entrypoint")
-def start_mq_bot(bot_entrypoint):
+def start_mq_bot(bot_entrypoint, health_check_server_port: Optional[int] = None):
     os.environ['CHATBOT_VERSION'] = 'v2'
     from chatbot_core.utils.bot_utils import run_mq_bot
     bot = run_mq_bot(bot_entrypoint)
+    if health_check_server_port:
+        from neon_utils.process_utils import start_health_check_server
+        start_health_check_server(bot.status, health_check_server_port, 
+                                  bot.check_health)
     wait_for_exit_signal()
     bot.stop()
 

chatbot_core/v1/__init__.py

@@ -26,7 +26,7 @@
 from threading import Thread
 from klat_connector.klat_api import KlatApi
 from klat_connector import start_socket
-from ovos_utils.log import LOG
+from ovos_utils.process_utils import ProcessStatus
 
 from chatbot_core.utils.enum import ConversationState, ConversationControls, BotTypes
 from chatbot_core.utils.string_utils import remove_prefix
@@ -35,6 +35,8 @@
 
 class ChatBot(KlatApi, ChatBotABC):
     def __init__(self, *args, **kwargs):
+        self._status = ProcessStatus("chatbot")
+        self._status.set_alive()
         socket, domain, username, password, on_server, is_prompter = \
             self.parse_init(*args, **kwargs)
         ChatBotABC.__init__(self, username)
@@ -94,6 +96,11 @@ def __init__(self, *args, **kwargs):
 
         self.shout_thread = Thread(target=self._handle_next_shout, daemon=True)
         self.shout_thread.start()
+        self._status.set_ready()
+
+    def check_health(self) -> bool:
+        # Unimplemented health check, assume service is healthy
+        return True
 
     def parse_init(self, *args, **kwargs) -> tuple:
         """Parses dynamic params input to ChatBot v1"""
@@ -637,6 +644,7 @@ def exit(self):
         from chatbot_core.utils.bot_utils import clean_up_bot
         # import sys
         # self.socket.disconnect()
+        self._status.set_stopping()
         while not self.shout_queue.empty():
             self.shout_queue.get(timeout=1)
         clean_up_bot(self)

chatbot_core/v2/__init__.py

@@ -23,11 +23,13 @@
 
 from neon_mq_connector.utils import RepeatingTimer
 from neon_mq_connector.utils.rabbit_utils import create_mq_callback
+from neon_mq_connector.connector import MQConnector
 from klat_connector.mq_klat_api import KlatAPIMQ
 from pika.exchange_type import ExchangeType
 from neon_data_models.models.api.mq.chatbots import ChatbotsMqRequest, \
     ChatbotsMqSubmindResponse
 from neon_data_models.enum import CcaiState as ConversationState
+from ovos_utils.process_utils import ProcessStatus
 
 from chatbot_core.utils.enum import BotTypes
 from chatbot_core.chatbot_abc import ChatBotABC
@@ -41,6 +43,8 @@ class ChatBot(KlatAPIMQ, ChatBotABC):
 
     def __init__(self, *args, **kwargs):
         config, service_name, vhost, bot_type = self.parse_init(*args, **kwargs)
+        self._status = ProcessStatus(service_name)
+        self._status.set_alive()
         mq_config = config.get("MQ") or config
         bot_config = config.get("chatbots", {}).get(service_name)
         KlatAPIMQ.__init__(self, mq_config, service_name, vhost)
@@ -57,6 +61,18 @@ def __init__(self, *args, **kwargs):
         self.shout_thread = RepeatingTimer(function=self._handle_next_shout,
                                            interval=kwargs.get('shout_thread_interval', 10))
         self.shout_thread.start()
+        self._status.set_ready()
+
+    def check_health(self) -> bool:
+        if not MQConnector.check_health(self):
+            self.log.error("MQ connection is not healthy")
+            self._status.set_error("MQ connection is not healthy")
+            return False
+        if not self.shout_thread.is_alive():
+            self.log.error("Shout thread is not alive")
+            self._status.set_error("Shout thread is not alive")
+            return False
+        return True
 
     def parse_init(self, *args, **kwargs) -> tuple:
         """Parses dynamic params input to ChatBot v2"""
@@ -620,10 +636,12 @@ def stop_shout_thread(self):
             self.shout_thread = None
 
     def shutdown(self):
+        # TODO: not used?
         self.shout_thread.cancel()
         self.shout_thread.join()
 
     def stop(self):
+        self._status.set_stopping()
         self.stop_shout_thread()
         KlatAPIMQ.stop(self)
 

requirements/requirements.txt

@@ -1,7 +1,8 @@
 click~=8.0
 klat-connector~=0.8
-neon-mq-connector~=0.8
-neon-utils[network,sentry]~=1.12
+neon-mq-connector~=0.8,>=0.9.1a1
+neon-utils[network,sentry]~=1.12,>=1.12.2a2
+ovos-utils~=0.0
 ovos-bus-client~=0.0,>=0.0.5
 psutil~=5.7
-neon-data-models~=0.0,>=0.0.2a2
+neon-data-models~=0.0,>=0.0.2a2