matrix-alertbot/matrix_alertbot/command.py

314 lines
10 KiB
Python
Raw Normal View History

import logging
from typing import List, Optional
import pytimeparse2
2022-07-06 00:54:13 +02:00
from diskcache import Cache
2022-07-10 14:06:36 +02:00
from nio import AsyncClient, MatrixRoom
2019-09-25 14:26:29 +02:00
from matrix_alertbot.alertmanager import AlertmanagerClient
2022-07-10 14:06:36 +02:00
from matrix_alertbot.chat_functions import send_text_to_room
2022-06-13 20:55:01 +02:00
from matrix_alertbot.config import Config
from matrix_alertbot.errors import (
AlertmanagerError,
AlertNotFoundError,
SilenceNotFoundError,
)
2022-07-10 12:51:49 +02:00
from matrix_alertbot.matcher import AlertMatcher, AlertRegexMatcher
logger = logging.getLogger(__name__)
2021-01-10 04:30:07 +01:00
class BaseCommand:
2021-01-10 04:30:07 +01:00
def __init__(
self,
client: AsyncClient,
cache: Cache,
alertmanager: AlertmanagerClient,
2021-01-10 04:30:07 +01:00
config: Config,
cmd: str,
2021-01-10 04:30:07 +01:00
room: MatrixRoom,
2022-07-10 14:06:36 +02:00
sender: str,
event_id: str,
) -> None:
"""A command made by a user.
2019-09-25 14:26:29 +02:00
Args:
client: The client to communicate with Matrix.
2019-09-25 14:26:29 +02:00
cache: Bot cache.
2019-09-25 14:26:29 +02:00
alertmanager: The client to communicate with Alertmanager.
config: Bot configuration parameters.
cmd: The command and arguments.
2019-09-25 14:26:29 +02:00
room: The room the command was sent in.
2019-09-25 14:26:29 +02:00
sender: The sender of the event
event_id: The ID of the event describing the command.
2019-09-25 14:26:29 +02:00
"""
self.client = client
self.cache = cache
self.alertmanager = alertmanager
2019-10-26 01:40:05 +02:00
self.config = config
self.cmd = cmd
self.args = cmd.split()[1:]
2019-09-25 14:26:29 +02:00
self.room = room
2022-07-10 14:06:36 +02:00
self.sender = sender
self.event_id = event_id
2019-09-25 14:26:29 +02:00
2022-06-14 23:37:54 +02:00
async def process(self) -> None:
raise NotImplementedError
2019-09-25 14:26:29 +02:00
class BaseAlertCommand(BaseCommand):
def __init__(
self,
client: AsyncClient,
cache: Cache,
alertmanager: AlertmanagerClient,
config: Config,
cmd: str,
room: MatrixRoom,
sender: str,
event_id: str,
alert_event_id: str,
) -> None:
super().__init__(
client, cache, alertmanager, config, cmd, room, sender, event_id
)
self.alert_event_id = alert_event_id
class AckAlertCommand(BaseAlertCommand):
async def process(self) -> None:
"""Acknowledge an alert and silence it for a certain duration in Alertmanager"""
2022-07-10 12:51:49 +02:00
matchers: List[AlertMatcher] = []
2022-07-10 02:40:04 +02:00
durations = []
for arg in self.args:
if "=~" in arg:
label, regex = arg.split("=~")
regex_matcher = AlertRegexMatcher(label, regex)
matchers.append(regex_matcher)
elif "=" in arg:
label, value = arg.split("=")
matcher = AlertMatcher(label, value)
matchers.append(matcher)
else:
durations.append(arg)
if len(durations) > 0:
duration = " ".join(durations)
else:
duration = "1d"
2022-07-10 02:40:04 +02:00
logger.debug(
2022-07-10 14:06:36 +02:00
f"Receiving a command to create a silence for a duration of {duration}"
)
duration_seconds = pytimeparse2.parse(duration)
if duration_seconds is None:
logger.error(f"Unable to create silence: Invalid duration '{duration}'")
await send_text_to_room(
self.client,
self.room.room_id,
2022-07-12 00:29:46 +02:00
f"I tried really hard, but I can't convert the duration '{duration}' to a number of seconds.",
)
return
logger.debug(
f"Read alert fingerprints for alert event {self.alert_event_id} from cache"
)
2022-07-05 23:35:19 +02:00
if self.alert_event_id not in self.cache:
logger.error(
f"Cannot find fingerprints for alert event {self.alert_event_id} in cache"
)
return
2022-07-10 15:11:25 +02:00
alert_fingerprints = self.cache[self.alert_event_id]
2022-07-10 15:11:25 +02:00
logger.debug(f"Found {len(alert_fingerprints)} in cache")
count_alert_not_found = 0
created_silences = []
2022-07-05 23:35:19 +02:00
for alert_fingerprint in alert_fingerprints:
logger.debug(
f"Create silence for alert with fingerprint {alert_fingerprint} for a duration of {duration}"
)
2022-07-06 00:54:13 +02:00
try:
silence_id = await self.alertmanager.create_silence(
2022-07-10 02:40:04 +02:00
alert_fingerprint,
duration_seconds,
2022-07-10 14:06:36 +02:00
self.room.user_name(self.sender),
2022-07-10 02:40:04 +02:00
matchers,
2022-07-06 00:54:13 +02:00
)
created_silences.append(silence_id)
except AlertNotFoundError as e:
logger.warning(f"Unable to create silence: {e}")
count_alert_not_found += 1
2022-07-09 10:38:40 +02:00
except AlertmanagerError as e:
2022-07-08 23:22:31 +02:00
logger.exception(f"Unable to create silence: {e}", exc_info=e)
2022-07-06 00:54:13 +02:00
self.cache.set(self.event_id, tuple(created_silences), expire=duration_seconds)
if count_alert_not_found > 0:
await send_text_to_room(
self.client,
self.room.room_id,
2022-07-11 23:45:15 +02:00
f"Sorry, I couldn't find {count_alert_not_found} alerts, therefore I couldn't create their silence.",
)
if len(created_silences) > 0:
2022-07-11 23:43:27 +02:00
await send_text_to_room(
self.client,
self.room.room_id,
f"Created {len(created_silences)} silences with a duration of {duration}.",
2022-07-11 23:43:27 +02:00
)
2022-07-06 00:54:13 +02:00
class UnackAlertCommand(BaseAlertCommand):
async def process(self) -> None:
2022-07-06 00:54:13 +02:00
"""Delete an alert's acknowledgement of an alert and remove corresponding silence in Alertmanager"""
2022-07-10 12:51:49 +02:00
matchers: List[AlertMatcher] = []
for arg in self.args:
if "=~" in arg:
label, regex = arg.split("=~")
regex_matcher = AlertRegexMatcher(label, regex)
matchers.append(regex_matcher)
elif "=" in arg:
label, value = arg.split("=")
matcher = AlertMatcher(label, value)
matchers.append(matcher)
2022-07-10 14:06:36 +02:00
logger.debug("Receiving a command to delete a silence")
logger.debug(
f"Read alert fingerprints for alert event {self.alert_event_id} from cache"
)
2022-07-06 00:54:13 +02:00
if self.alert_event_id not in self.cache:
logger.error(
f"Cannot find fingerprints for event {self.alert_event_id} in cache"
)
return
2022-07-10 15:11:25 +02:00
alert_fingerprints = self.cache[self.alert_event_id]
2022-07-10 15:11:25 +02:00
logger.debug(f"Found {len(alert_fingerprints)} in cache")
count_alert_not_found = 0
2022-07-10 15:11:25 +02:00
count_removed_silences = 0
2022-07-06 00:54:13 +02:00
for alert_fingerprint in alert_fingerprints:
logger.debug(
f"Delete silence for alert with fingerprint {alert_fingerprint}"
2022-07-05 23:35:19 +02:00
)
2022-07-06 00:54:13 +02:00
try:
2022-07-09 09:56:28 +02:00
removed_silences = await self.alertmanager.delete_silences(
2022-07-10 12:51:49 +02:00
alert_fingerprint, matchers
2022-07-09 00:08:51 +02:00
)
2022-07-09 09:56:28 +02:00
count_removed_silences += len(removed_silences)
except (AlertNotFoundError, SilenceNotFoundError) as e:
2022-07-11 23:41:30 +02:00
logger.error(f"Unable to delete silence: {e}")
count_alert_not_found += 1
2022-07-09 10:38:40 +02:00
except AlertmanagerError as e:
2022-07-08 23:22:31 +02:00
logger.exception(f"Unable to delete silence: {e}", exc_info=e)
2022-07-06 00:54:13 +02:00
if count_alert_not_found > 0:
await send_text_to_room(
self.client,
self.room.room_id,
2022-07-11 23:45:15 +02:00
f"Sorry, I couldn't find {count_alert_not_found} alerts, therefore I couldn't remove their silences.",
)
2022-07-11 23:43:27 +02:00
if count_removed_silences > 0:
await send_text_to_room(
self.client,
self.room.room_id,
f"Removed {count_removed_silences} silences.",
)
2019-09-25 14:26:29 +02:00
class HelpCommand(BaseCommand):
async def process(self) -> None:
2019-09-25 14:26:29 +02:00
"""Show the help text"""
2022-07-10 14:06:36 +02:00
logger.debug(f"Displaying help to room {self.room.display_name}")
2019-09-25 14:26:29 +02:00
if not self.args:
2020-08-10 00:02:07 +02:00
text = (
"Hello, I am a bot made with matrix-nio! Use `help commands` to view "
"available commands."
)
2019-09-25 14:26:29 +02:00
await send_text_to_room(self.client, self.room.room_id, text)
return
topic = self.args[0]
if topic == "rules":
text = "These are the rules!"
elif topic == "commands":
text = "Available commands: ..."
2019-09-25 14:26:29 +02:00
else:
text = "Unknown help topic!"
await send_text_to_room(self.client, self.room.room_id, text)
class UnknownCommand(BaseCommand):
async def process(self) -> None:
logger.debug(
2022-07-10 14:06:36 +02:00
f"Sending unknown command response to room {self.room.display_name}"
)
2019-09-25 14:26:29 +02:00
await send_text_to_room(
self.client,
self.room.room_id,
f"Unknown command '{self.cmd}'. Try the 'help' command for more information.",
2019-09-25 14:26:29 +02:00
)
class CommandFactory:
@staticmethod
def create(
cmd: str,
client: AsyncClient,
cache: Cache,
alertmanager: AlertmanagerClient,
config: Config,
room: MatrixRoom,
sender: str,
event_id: str,
reacted_to_event_id: Optional[str] = None,
) -> BaseCommand:
if cmd.startswith("ack"):
assert reacted_to_event_id is not None
return AckAlertCommand(
client,
cache,
alertmanager,
config,
cmd,
room,
sender,
event_id,
reacted_to_event_id,
)
elif cmd.startswith("unack") or cmd.startswith("nack"):
assert reacted_to_event_id is not None
return UnackAlertCommand(
client,
cache,
alertmanager,
config,
cmd,
room,
sender,
event_id,
reacted_to_event_id,
)
elif cmd.startswith("help"):
return HelpCommand(
client, cache, alertmanager, config, cmd, room, sender, event_id
)
else:
return UnknownCommand(
client, cache, alertmanager, config, cmd, room, sender, event_id
)