allow silence deletion

matrix_alertbot/alertmanager.py

@@ -1,29 +1,40 @@
 import datetime
 from typing import Dict, List
 
+import diskcache
 import pytimeparse
 import requests
+from requests import RequestException
 
-from matrix_alertbot.cache import Cache
-from matrix_alertbot.errors import AlertNotFoundError
+from matrix_alertbot.config import Config
+from matrix_alertbot.errors import (
+    AlertmanagerError,
+    AlertNotFoundError,
+    SilenceNotFoundError,
+)
 
 
 class AlertmanagerClient:
-    def __init__(self, url: str, cache: Cache) -> None:
+    def __init__(self, config: Config) -> None:
+        url = config.alertmanager_url
         self.api_url = f"{url}/api/v2"
-        self.cache = cache
+        self.cache = diskcache.Cache(config.cache_dir)
 
     def get_alerts(self) -> List[Dict]:
-        response = requests.get(f"{self.api_url}/alerts")
-        response.raise_for_status()
+        try:
+            response = requests.get(f"{self.api_url}/alerts")
+            response.raise_for_status()
+        except RequestException as e:
+            raise AlertmanagerError(f"Cannot fetch alerts from Alertmanager") from e
         return response.json()
 
+    def get_alert(self, fingerprint: str) -> Dict:
+        alerts = self.get_alerts()
+        return self._find_alert(fingerprint, alerts)
+
     def get_alert_labels(self, fingerprint: str) -> Dict[str, str]:
-        if fingerprint not in self.cache:
-            alerts = self.get_alerts()
-            alert = self._find_alert(alerts, fingerprint)
-            self.cache[fingerprint] = alert["labels"]
-        return self.cache[fingerprint]
+        alert = self.get_alert(fingerprint)
+        return alert["labels"]
 
     def create_silence(self, fingerprint: str, duration: str, user: str) -> str:
         labels = self.get_alert_labels(fingerprint)
@@ -45,13 +56,37 @@ class AlertmanagerClient:
             "createdBy": user,
             "comment": "Acknowledge alert from Matrix",
         }
-        response = requests.post(f"{self.api_url}/silences", json=silence)
-        response.raise_for_status()
+        try:
+            response = requests.post(f"{self.api_url}/silences", json=silence)
+            response.raise_for_status()
+        except RequestException as e:
+            raise AlertmanagerError(
+                f"Cannot create silence for alert fingerprint {fingerprint}"
+            ) from e
         data = response.json()
         return data["silenceID"]
 
+    def delete_silence(self, fingerprint: str) -> None:
+        alert = self.get_alert(fingerprint)
+
+        alert_state = alert["status"]["state"]
+        if alert_state != "suppressed":
+            raise SilenceNotFoundError(
+                f"Cannot find silences for alert fingerprint {fingerprint} in state {alert_state}"
+            )
+
+        silences = alert["status"]["silencedBy"]
+        for silence in silences:
+            try:
+                response = requests.delete(f"{self.api_url}/silence/{silence}")
+                response.raise_for_status()
+            except RequestException as e:
+                raise AlertmanagerError(
+                    f"Cannot delete silence with ID {silence}"
+                ) from e
+
     @staticmethod
-    def _find_alert(alerts: List[Dict], fingerprint: str) -> Dict:
+    def _find_alert(fingerprint: str, alerts: List[Dict]) -> Dict:
         for alert in alerts:
             if alert["fingerprint"] == fingerprint:
                 return alert

matrix_alertbot/bot_commands.py

@@ -1,11 +1,12 @@
 import logging
 
+from diskcache import Cache
 from nio import AsyncClient, MatrixRoom, RoomMessageText
 
 from matrix_alertbot.alertmanager import AlertmanagerClient
-from matrix_alertbot.cache import Cache
 from matrix_alertbot.chat_functions import react_to_event, send_text_to_room
 from matrix_alertbot.config import Config
+from matrix_alertbot.errors import AlertmanagerError, AlertNotFoundError
 
 logger = logging.getLogger(__name__)
 
@@ -49,6 +50,8 @@ class Command:
         """Process the command"""
         if self.command.startswith("ack"):
             await self._ack()
+        if self.command.startswith("unack") or self.command.startswith("nack"):
+            await self._unack()
         elif self.command.startswith("react"):
             await self._react()
         elif self.command.startswith("help"):
@@ -63,7 +66,7 @@ class Command:
         else:
             duration = "1d"
         logger.debug(
-            f"Acknowledging alert with fingerprint {self.room.display_name} for a duration of {duration} | "
+            f"Receiving a command to create a silence for a duration of {duration} | "
             f"{self.room.user_name(self.event.sender)}: {self.event.body}"
         )
 
@@ -75,21 +78,59 @@ class Command:
             return
         logger.debug(f"Read alert fingerprints for event {alert_event_id} from cache")
 
-        silence_ids = []
+        count_created_silences = 0
         alert_fingerprints = self.cache[alert_event_id]
         for alert_fingerprint in alert_fingerprints:
             logger.debug(
                 f"Create silence for alert with fingerprint {alert_fingerprint} for a duration of {duration}"
             )
-            silence_id = self.alertmanager.create_silence(
-                alert_fingerprint, duration, self.room.user_name(self.event.sender)
-            )
-            silence_ids.append(silence_id)
-        silences = ", ".join(silence_ids)
+            try:
+                silence_id = self.alertmanager.create_silence(
+                    alert_fingerprint, duration, self.room.user_name(self.event.sender)
+                )
+            except (AlertNotFoundError, AlertmanagerError) as e:
+                logger.error(f"Unable to create silence: {e}")
+                continue
+            count_created_silences += 1
+
         await send_text_to_room(
             self.client,
             self.room.room_id,
-            f"Created silences {silences} for a duration of {duration}",
+            f"Created {count_created_silences} silences with a duration of {duration}",
+        )
+
+    async def _unack(self) -> None:
+        """Delete an alert's acknowledgement of an alert and remove corresponding silence in Alertmanager"""
+        logger.debug(
+            f"Receiving a command to delete a silence | "
+            f"{self.room.user_name(self.event.sender)}: {self.event.body}"
+        )
+
+        source_content = self.event.source["content"]
+        try:
+            alert_event_id = source_content["m.relates_to"]["m.in_reply_to"]["event_id"]
+        except KeyError:
+            logger.debug("Unable to find the event ID of the alert")
+            return
+        logger.debug(f"Read alert fingerprints for event {alert_event_id} from cache")
+
+        count_removed_silences = 0
+        alert_fingerprints = self.cache[alert_event_id]
+        for alert_fingerprint in alert_fingerprints:
+            logger.debug(
+                f"Delete silence for alert with fingerprint {alert_fingerprint}"
+            )
+            try:
+                self.alertmanager.delete_silence(alert_fingerprint)
+            except (AlertNotFoundError, AlertmanagerError) as e:
+                logger.error(f"Unable to delete silence: {e}")
+                continue
+            count_removed_silences += 1
+
+        await send_text_to_room(
+            self.client,
+            self.room.room_id,
+            f"Removed {count_removed_silences} silences",
         )
 
     async def _react(self) -> None:

matrix_alertbot/cache.py

@@ -1,18 +0,0 @@
-from typing import Any
-
-import diskcache
-
-
-class Cache:
-    def __init__(self, directory: str, expire: int):
-        self.cache = diskcache.Cache(directory)
-        self.expire = expire
-
-    def __getitem__(self, key: str) -> Any:
-        return self.cache[key]
-
-    def __setitem__(self, key: str, value: Any) -> None:
-        self.cache.set(key, value, expire=self.expire)
-
-    def __contains__(self, key: str) -> bool:
-        return key in self.cache

matrix_alertbot/callbacks.py

@@ -1,5 +1,6 @@
 import logging
 
+from diskcache import Cache
 from nio import (
     AsyncClient,
     InviteMemberEvent,
@@ -13,13 +14,7 @@ from nio import (
 
 from matrix_alertbot.alertmanager import AlertmanagerClient
 from matrix_alertbot.bot_commands import Command
-from matrix_alertbot.cache import Cache
-from matrix_alertbot.chat_functions import (
-    make_pill,
-    react_to_event,
-    send_text_to_room,
-    strip_fallback,
-)
+from matrix_alertbot.chat_functions import make_pill, send_text_to_room, strip_fallback
 from matrix_alertbot.config import Config
 
 logger = logging.getLogger(__name__)
@@ -29,7 +24,6 @@ class Callbacks:
     def __init__(
         self,
         client: AsyncClient,
-        cache: Cache,
         alertmanager: AlertmanagerClient,
         config: Config,
     ):
@@ -44,7 +38,7 @@ class Callbacks:
             config: Bot configuration parameters.
         """
         self.client = client
-        self.cache = cache
+        self.cache = Cache(config.cache_dir)
         self.alertmanager = alertmanager
         self.config = config
         self.command_prefix = config.command_prefix

matrix_alertbot/chat_functions.py

@@ -1,15 +1,7 @@
 import logging
 from typing import Optional, Union
 
-from nio import (
-    AsyncClient,
-    ErrorResponse,
-    MatrixRoom,
-    MegolmEvent,
-    Response,
-    RoomSendResponse,
-    SendRetryError,
-)
+from nio import AsyncClient, ErrorResponse, Response, RoomSendResponse, SendRetryError
 
 logger = logging.getLogger(__name__)
 

matrix_alertbot/errors.py

@@ -19,3 +19,11 @@ class AlertNotFoundError(RuntimeError):
     """
 
     pass
+
+
+class SilenceNotFoundError(RuntimeError):
+    pass
+
+
+class AlertmanagerError(RuntimeError):
+    pass

matrix_alertbot/main.py

@@ -5,7 +5,8 @@ import sys
 from asyncio import TimeoutError
 from time import sleep
 
-from aiohttp import ClientConnectionError, ServerDisconnectedError, web
+import diskcache
+from aiohttp import ClientConnectionError, ServerDisconnectedError
 from nio import (
     AsyncClient,
     AsyncClientConfig,
@@ -18,7 +19,6 @@ from nio import (
 )
 
 from matrix_alertbot.alertmanager import AlertmanagerClient
-from matrix_alertbot.cache import Cache
 from matrix_alertbot.callbacks import Callbacks
 from matrix_alertbot.config import Config
 from matrix_alertbot.webhook import Webhook
@@ -88,11 +88,8 @@ def main() -> None:
     # Read the parsed config file and create a Config object
     config = Config(config_path)
 
-    # Configure the cache
-    cache = Cache(config.cache_dir, config.cache_expire_time)
-
     # Configure Alertmanager client
-    alertmanager = AlertmanagerClient(config.alertmanager_url, cache)
+    alertmanager = AlertmanagerClient(config)
 
     # Configuration options for the AsyncClient
     client_config = AsyncClientConfig(
@@ -116,7 +113,7 @@ def main() -> None:
         client.user_id = config.user_id
 
     # Set up event callbacks
-    callbacks = Callbacks(client, cache, alertmanager, config)
+    callbacks = Callbacks(client, alertmanager, config)
     client.add_event_callback(callbacks.message, (RoomMessageText,))
     client.add_event_callback(
         callbacks.invite_event_filtered_callback, (InviteMemberEvent,)
@@ -124,7 +121,7 @@ def main() -> None:
     client.add_event_callback(callbacks.decryption_failure, (MegolmEvent,))
     client.add_event_callback(callbacks.unknown, (UnknownEvent,))
 
-    webhook_server = Webhook(client, cache, config)
+    webhook_server = Webhook(client, config)
 
     loop = asyncio.get_event_loop()
     loop.create_task(webhook_server.start())

matrix_alertbot/webhook.py

@@ -1,9 +1,8 @@
 import logging
-from typing import List
 
 from aiohttp import web, web_request
+from diskcache import Cache
 from nio import AsyncClient, SendRetryError
-from matrix_alertbot.cache import Cache
 
 from matrix_alertbot.chat_functions import send_text_to_room
 from matrix_alertbot.config import Config
@@ -18,8 +17,9 @@ routes = web.RouteTableDef()
 async def create_alert(request: web_request.Request) -> web.Response:
     data = await request.json()
     logger.info(f"Received alert: {data}")
-    client = request.app["client"]
-    cache = request.app["cache"]
+    client: AsyncClient = request.app["client"]
+    config: Config = request.app["config"]
+    cache: Cache = request.app["cache"]
 
     plaintext = ""
     html = ""
@@ -34,21 +34,26 @@ async def create_alert(request: web_request.Request) -> web.Response:
 
     try:
         event = await send_text_to_room(
-            client, request.app["room_id"], plaintext, html, notice=False
+            client, config.room, plaintext, html, notice=False
         )
     except SendRetryError as e:
         logger.error(e)
         return web.Response(status=500)
 
-    cache[event.event_id] = tuple(alert["fingerprint"] for alert in data["alerts"])
+    fingerprints = tuple(alert["fingerprint"] for alert in data["alerts"])
+    cache.set(
+        event.event_id, fingerprints, expire=config.cache_expire_time, tag="event"
+    )
     return web.Response(status=200)
 
 
 class Webhook:
-    def __init__(self, client: AsyncClient, cache: Cache, config: Config) -> None:
+    def __init__(self, client: AsyncClient, config: Config) -> None:
+        cache = Cache(config.cache_dir)
+
         self.app = web.Application(logger=logger)
         self.app["client"] = client
-        self.app["room_id"] = config.room
+        self.app["config"] = config
         self.app["cache"] = cache
         self.app.add_routes(routes)
         self.runner = web.AppRunner(self.app)

tests/test_callbacks.py

@@ -4,7 +4,6 @@ from unittest.mock import Mock
 import nio
 
 from matrix_alertbot.alertmanager import AlertmanagerClient
-from matrix_alertbot.cache import Cache
 from matrix_alertbot.callbacks import Callbacks
 
 from tests.utils import make_awaitable, run_coroutine
@@ -16,14 +15,13 @@ class CallbacksTestCase(unittest.TestCase):
         self.fake_client = Mock(spec=nio.AsyncClient)
         self.fake_client.user = "@fake_user:example.com"
 
-        self.fake_cache = Mock(spec=Cache)
         self.fake_alertmanager = Mock(spec=AlertmanagerClient)
 
         # We don't spec config, as it doesn't currently have well defined attributes
         self.fake_config = Mock()
 
         self.callbacks = Callbacks(
-            self.fake_client, self.fake_cache, self.fake_alertmanager, self.fake_config
+            self.fake_client, self.fake_alertmanager, self.fake_config
         )
 
     def test_invite(self) -> None: