Update prometheus stuff

mautrix_telegram/abstract_user.py

@@ -37,6 +37,7 @@ from mautrix.types import UserID, PresenceState
 from mautrix.errors import MatrixError
 from mautrix.appservice import AppService
 from mautrix.util.logging import TraceLogger
+from mautrix.util.opt_prometheus import Histogram, Counter
 from alchemysession import AlchemySessionContainer
 
 from . import portal as po, puppet as pu, __version__
@@ -57,14 +58,10 @@ UpdateMessage = Union[UpdateShortChatMessage, UpdateShortMessage, UpdateNewChann
                       UpdateNewMessage, UpdateEditMessage, UpdateEditChannelMessage]
 UpdateMessageContent = Union[UpdateShortMessage, UpdateShortChatMessage, Message, MessageService]
 
-try:
-    from prometheus_client import Histogram
-
-    UPDATE_TIME = Histogram("telegram_update", "Time spent processing Telegram updates",
-                            ["update_type"])
-except ImportError:
-    Histogram = None
-    UPDATE_TIME = None
+UPDATE_TIME = Histogram("bridge_telegram_update", "Time spent processing Telegram updates",
+                        ("update_type",))
+UPDATE_ERRORS = Counter("bridge_telegram_update_error",
+                        "Number of fatal errors while handling Telegram updates", ("update_type",))
 
 
 class AbstractUser(ABC):
@@ -191,13 +188,14 @@ class AbstractUser(ABC):
 
     async def _update_catch(self, update: TypeUpdate) -> None:
         start_time = time.time()
+        update_type = type(update).__name__
         try:
             if not await self.update(update):
                 await self._update(update)
         except Exception:
             self.log.exception(f"Failed to handle Telegram update {update}")
-        if UPDATE_TIME:
-            UPDATE_TIME.labels(update_type=type(update).__name__).observe(time.time() - start_time)
+            UPDATE_ERRORS.labels(update_type=update_type).inc()
+        UPDATE_TIME.labels(update_type=update_type).observe(time.time() - start_time)
 
     @property
     @abstractmethod
@@ -396,7 +394,8 @@ class AbstractUser(ABC):
                 portal = po.Portal.get_by_entity(update.to_id, receiver_id=self.tgid)
             sender = pu.Puppet.get(update.from_id) if update.from_id else None
         else:
-            self.log.warning(f"Unexpected message type in User#get_message_details: {type(update)}")
+            self.log.warning("Unexpected message type in User#get_message_details: "
+                             f"{type(update)}")
             return update, None, None
         return update, sender, portal
 

mautrix_telegram/matrix.py

@@ -30,14 +30,6 @@ if TYPE_CHECKING:
     from .context import Context
     from .bot import Bot
 
-try:
-    from prometheus_client import Histogram
-
-    EVENT_TIME = Histogram("matrix_event", "Time spent processing Matrix events", ["event_type"])
-except ImportError:
-    Histogram = None
-    EVENT_TIME = None
-
 RoomMetaStateEventContent = Union[RoomNameStateEventContent, RoomAvatarStateEventContent,
                                   RoomTopicStateEventContent]
 
@@ -410,7 +402,3 @@ class MatrixHandler(BaseMatrixHandler):
         elif evt.type == EventType.ROOM_TOMBSTONE:
             await self.handle_room_upgrade(evt.room_id, evt.sender, evt.content.replacement_room,
                                            evt.event_id)
-
-    async def log_event_handle_duration(self, evt: Event, duration: float) -> None:
-        if EVENT_TIME:
-            EVENT_TIME.labels(event_type=str(evt.type)).observe(duration)

mautrix_telegram/user.py

@@ -31,6 +31,7 @@ from mautrix.errors import MatrixRequestError
 from mautrix.types import UserID, RoomID
 from mautrix.bridge import BaseUser
 from mautrix.util.logging import TraceLogger
+from mautrix.util.opt_prometheus import Enum
 
 from .types import TelegramID
 from .db import User as DBUser, Portal as DBPortal
@@ -45,6 +46,11 @@ config: Optional['Config'] = None
 
 SearchResult = NamedTuple('SearchResult', puppet='pu.Puppet', similarity=int)
 
+METRIC_LOGGED_IN = Enum('bridge_logged_in', 'Bridge Logged in', states=["true", "false"],
+                        labelnames=("tgid",))
+METRIC_CONNECTED = Enum('bridge_connected', 'Bridge Connected', states=["true", "false"],
+                        labelnames=("tgid",))
+
 
 class User(AbstractUser, BaseUser):
     log: TraceLogger = logging.getLogger("mau.user")
@@ -193,15 +199,21 @@ class User(AbstractUser, BaseUser):
 
     async def start(self, delete_unless_authenticated: bool = False) -> 'User':
         await super().start()
+        METRIC_CONNECTED.labels(tgid=self.tgid).state("true")
         if await self.is_logged_in():
             self.log.debug(f"Ensuring post_login() for {self.name}")
             asyncio.ensure_future(self.post_login(), loop=self.loop)
         elif delete_unless_authenticated:
             self.log.debug(f"Unauthenticated user {self.name} start()ed, deleting session...")
             await self.client.disconnect()
+            METRIC_CONNECTED.labels(tgid=self.tgid).state("false")
             self.client.session.delete()
         return self
 
+    async def stop(self) -> None:
+        await super().stop()
+        METRIC_CONNECTED.labels(tgid=self.tgid).state("true")
+
     async def post_login(self, info: TLUser = None, first_login: bool = False) -> None:
         try:
             await self.update_info(info)
@@ -209,6 +221,8 @@ class User(AbstractUser, BaseUser):
             self.log.exception("Failed to update telegram account info")
             return
 
+        METRIC_LOGGED_IN.labels(tgid=self.tgid).state("true")
+
         try:
             puppet = pu.Puppet.get(self.tgid)
             if puppet.custom_mxid != self.mxid and puppet.can_auto_login(self.mxid):
@@ -300,6 +314,7 @@ class User(AbstractUser, BaseUser):
         if not ok:
             return False
         self.delete()
+        METRIC_LOGGED_IN.labels(tgid=self.tgid).state("false")
         return True
 
     def _search_local(self, query: str, max_results: int = 5, min_similarity: int = 45

requirements.txt

@@ -4,6 +4,6 @@ ruamel.yaml>=0.15.35,<0.17
 python-magic>=0.4,<0.5
 commonmark>=0.8,<0.10
 aiohttp>=3,<4
-mautrix>=0.7.3,<0.8
+mautrix>=0.7.4,<0.8
 telethon>=1.16,<1.17
 telethon-session-sqlalchemy>=0.2.14,<0.3