Merge branch 'dev' into jorlando/splunk-migration-tool

.drone.yml

@@ -70,7 +70,7 @@ steps:
  - apt-get update && apt-get install -y netcat-traditional
  - cd engine/
  - pip install uv
- - uv pip install -r requirements.txt -r requirements-dev.txt
+ - uv pip install --system -r requirements.txt -r requirements-dev.txt
  - ./wait_for_test_mysql_start.sh && pytest
  depends_on:
  - rabbit_test
@@ -386,6 +386,6 @@ name: cloud_access_policy_token
 
 ---
 kind: signature
-hmac: d541ed21fc2472272c6772e246aaf1a2606db112b4e72a44bc4530831e9ca4d3
+hmac: a045d72f3f3510895da049f4bf8f5ae4ac21f3ffa3d24cda047152c286df5bc2
 
 ...

Tiltfile

@@ -59,13 +59,27 @@ docker_build_sub(
  ],
 )
 
-# Build the plugin in the background
-local_resource(
- "build-ui",
- labels=["OnCallUI"],
- serve_cmd="cd grafana-plugin && yarn watch",
- allow_parallel=True,
-)
+# On CI dependencies are installed separately so we just build prod bundle to be consumed by Grafana dev server
+if is_ci:
+ local_resource(
+ "build-ui",
+ labels=["OnCallUI"],
+ dir="grafana-plugin",
+ cmd="yarn build",
+ allow_parallel=True,
+ )
+
+# Locally we install dependencies and we run watch mode
+if not is_ci:
+ local_resource(
+ "build-ui",
+ labels=["OnCallUI"],
+ dir="grafana-plugin",
+ cmd="yarn install",
+ serve_dir="grafana-plugin",
+ serve_cmd="yarn watch",
+ allow_parallel=True,
+ )
 
 local_resource(
  "e2e-tests",
@@ -130,7 +144,7 @@ configmap_create(
 k8s_resource(
  objects=["grafana-oncall-app-provisioning:configmap"],
  new_name="grafana-oncall-app-provisioning-configmap",
- resource_deps=["build-ui", "engine"],
+ resource_deps=["build-ui"],
  labels=["Grafana"],
 )
 
@@ -141,7 +155,7 @@ if not running_under_parent_tiltfile:
  context="grafana-plugin",
  plugin_files=["grafana-plugin/src/plugin.json"],
  namespace="default",
- deps=["grafana-oncall-app-provisioning-configmap", "build-ui", "engine"],
+ deps=["grafana-oncall-app-provisioning-configmap", "build-ui"],
  extra_env={
  "GF_SECURITY_ADMIN_PASSWORD": "oncall",
  "GF_SECURITY_ADMIN_USER": "oncall",

dev/README.md

@@ -11,6 +11,7 @@ Related: [How to develop integrations](/engine/config_integrations/README.md)
 - [Tilt | Kubernetes for Prod, Tilt for Dev](https://tilt.dev/)
 - [tilt-dev/ctlptl: Making local Kubernetes clusters fun and easy to set up](https://github.com/tilt-dev/ctlptl)
 - [Kind](https://kind.sigs.k8s.io)
+- [Node.js v18.x](https://nodejs.org/en/download)
 - [Yarn](https://classic.yarnpkg.com/lang/en/docs/install/#mac-stable)
 
 ### Launch the environment

engine/apps/alerts/escalation_snapshot/snapshot_classes/escalation_policy_snapshot.py

@@ -472,24 +472,32 @@ def _escalation_step_notify_if_num_alerts_in_time_window(
  def _escalation_step_trigger_custom_webhook(self, alert_group: "AlertGroup", _reason: str) -> None:
  tasks = []
  webhook = self.custom_webhook
+ failure_reason = None
  if webhook is not None:
- custom_webhook_task = custom_webhook_result.signature(
- (webhook.pk, alert_group.pk),
- {
- "escalation_policy_pk": self.id,
- },
- immutable=True,
- )
- tasks.append(custom_webhook_task)
+ if webhook.is_webhook_enabled:
+ custom_webhook_task = custom_webhook_result.signature(
+ (webhook.pk, alert_group.pk),
+ {
+ "escalation_policy_pk": self.id,
+ },
+ immutable=True,
+ )
+ tasks.append(custom_webhook_task)
+ else:
+ failure_reason = AlertGroupLogRecord.ERROR_ESCALATION_TRIGGER_WEBHOOK_IS_DISABLED
  else:
+ failure_reason = AlertGroupLogRecord.ERROR_ESCALATION_TRIGGER_WEBHOOK_STEP_IS_NOT_CONFIGURED
+
+ if failure_reason:
  log_record = AlertGroupLogRecord(
  type=AlertGroupLogRecord.TYPE_ESCALATION_FAILED,
  alert_group=alert_group,
  escalation_policy=self.escalation_policy,
- escalation_error_code=AlertGroupLogRecord.ERROR_ESCALATION_TRIGGER_WEBHOOK_STEP_IS_NOT_CONFIGURED,
+ escalation_error_code=failure_reason,
  escalation_policy_step=self.step,
  )
  log_record.save()
+
  self._execute_tasks(tasks)
 
  def _escalation_step_repeat_escalation_n_times(

engine/apps/alerts/models/alert_group.py

@@ -511,17 +511,8 @@ def slack_app_link(self) -> typing.Optional[str]:
 
  @property
  def telegram_permalink(self) -> typing.Optional[str]:
- """
- This property will attempt to access an attribute, `prefetched_telegram_messages`, representing a list of
- prefetched telegram messages. If this attribute does not exist, it falls back to performing a query.
-
- See `apps.public_api.serializers.incidents.IncidentSerializer.PREFETCH_RELATED` as an example.
- """
  from apps.telegram.models.message import TelegramMessage
 
- if hasattr(self, "prefetched_telegram_messages"):
- return self.prefetched_telegram_messages[0].link if self.prefetched_telegram_messages else None
-
  main_telegram_message = self.telegram_messages.filter(
  chat_id__startswith="-", message_type=TelegramMessage.ALERT_GROUP_MESSAGE
  ).first()

engine/apps/alerts/models/alert_group_log_record.py

@@ -160,7 +160,8 @@ class AlertGroupLogRecord(models.Model):
  ERROR_ESCALATION_NOTIFY_IF_NUM_ALERTS_IN_WINDOW_STEP_IS_NOT_CONFIGURED,
  ERROR_ESCALATION_TRIGGER_CUSTOM_WEBHOOK_ERROR,
  ERROR_ESCALATION_NOTIFY_TEAM_MEMBERS_STEP_IS_NOT_CONFIGURED,
- ) = range(19)
+ ERROR_ESCALATION_TRIGGER_WEBHOOK_IS_DISABLED,
+ ) = range(20)
 
  type = models.IntegerField(choices=TYPE_CHOICES)
 
@@ -590,6 +591,8 @@ def rendered_log_line_action(self, for_slack=False, html=False, substitute_autho
  usergroup_handle = self.escalation_policy.notify_to_group.handle
  usergroup_handle_text = f" @{usergroup_handle}" if usergroup_handle else ""
  result += f"failed to notify User Group{usergroup_handle_text} in Slack"
+ elif self.escalation_error_code == AlertGroupLogRecord.ERROR_ESCALATION_TRIGGER_WEBHOOK_IS_DISABLED:
+ result += 'skipped escalation step "Trigger Outgoing Webhook" because it is disabled'
  return result
 
  def get_step_specific_info(self):

engine/apps/alerts/tests/test_escalation_policy_snapshot.py

@@ -486,6 +486,33 @@ def test_escalation_step_trigger_custom_webhook(
  mock_webhook_escalation_step.assert_called_once_with(alert_group, reason)
 
 
+@patch("apps.alerts.escalation_snapshot.snapshot_classes.EscalationPolicySnapshot._execute_tasks", return_value=None)
+@pytest.mark.django_db
+def test_escalation_step_trigger_disabled_custom_webhook(
+ mocked_execute_tasks,
+ escalation_step_test_setup,
+ make_custom_webhook,
+ make_escalation_policy,
+):
+ organization, _, _, channel_filter, alert_group, reason = escalation_step_test_setup
+
+ custom_webhook = make_custom_webhook(organization=organization, is_webhook_enabled=False)
+
+ trigger_custom_webhook_step = make_escalation_policy(
+ escalation_chain=channel_filter.escalation_chain,
+ escalation_policy_step=EscalationPolicy.STEP_TRIGGER_CUSTOM_WEBHOOK,
+ custom_webhook=custom_webhook,
+ )
+ escalation_policy_snapshot = get_escalation_policy_snapshot_from_model(trigger_custom_webhook_step)
+ escalation_policy_snapshot.execute(alert_group, reason)
+ assert call([]) in mocked_execute_tasks.call_args_list
+
+ log_record = AlertGroupLogRecord.objects.get(
+ alert_group_id=alert_group.id, escalation_policy=trigger_custom_webhook_step
+ )
+ assert log_record.escalation_error_code == AlertGroupLogRecord.ERROR_ESCALATION_TRIGGER_WEBHOOK_IS_DISABLED
+
+
 @patch("apps.alerts.escalation_snapshot.snapshot_classes.EscalationPolicySnapshot._execute_tasks", return_value=None)
 @pytest.mark.django_db
 def test_escalation_step_repeat_escalation_n_times(

engine/apps/google/client.py

@@ -5,6 +5,7 @@
 from django.conf import settings
 from google.oauth2.credentials import Credentials
 from googleapiclient.discovery import build
+from googleapiclient.errors import HttpError
 
 from apps.google import constants, utils
 from apps.google.types import GoogleCalendarEvent as GoogleCalendarEventType
@@ -23,6 +24,11 @@ def __init__(self, event: GoogleCalendarEventType):
  self.end_time_utc = self._end_time.astimezone(datetime.timezone.utc)
 
 
+class GoogleCalendarHTTPError(Exception):
+ def __init__(self, http_error) -> None:
+ self.error = http_error
+
+
 class GoogleCalendarAPIClient:
  MAX_NUMBER_OF_CALENDAR_EVENTS_TO_FETCH = 250
  """
@@ -68,17 +74,22 @@ def fetch_out_of_office_events(self) -> typing.List[GoogleCalendarEvent]:
  now + datetime.timedelta(days=constants.DAYS_IN_FUTURE_TO_CONSIDER_OUT_OF_OFFICE_EVENTS)
  )
 
- events_result = (
- self.service.events()
- .list(
- calendarId=self.CALENDAR_ID,
- timeMin=time_min,
- timeMax=time_max,
- maxResults=self.MAX_NUMBER_OF_CALENDAR_EVENTS_TO_FETCH,
- singleEvents=True,
- orderBy="startTime",
- eventTypes="outOfOffice",
+ try:
+ events_result = (
+ self.service.events()
+ .list(
+ calendarId=self.CALENDAR_ID,
+ timeMin=time_min,
+ timeMax=time_max,
+ maxResults=self.MAX_NUMBER_OF_CALENDAR_EVENTS_TO_FETCH,
+ singleEvents=True,
+ orderBy="startTime",
+ eventTypes="outOfOffice",
+ )
+ .execute()
  )
- .execute()
- )
+ except HttpError as e:
+ logger.error(f"GoogleCalendarAPIClient - Error fetching out of office events: {e}")
+ raise GoogleCalendarHTTPError(e)
+
  return [GoogleCalendarEvent(event) for event in events_result.get("items", [])]

engine/apps/google/tasks.py

@@ -3,7 +3,7 @@
 from celery.utils.log import get_task_logger
 
 from apps.google import constants
-from apps.google.client import GoogleCalendarAPIClient
+from apps.google.client import GoogleCalendarAPIClient, GoogleCalendarHTTPError
 from apps.google.models import GoogleOAuth2User
 from apps.schedules.models import OnCallSchedule, ShiftSwapRequest
 from common.custom_celery_tasks import shared_dedicated_queue_retry_task
@@ -31,7 +31,13 @@ def sync_out_of_office_calendar_events_for_user(google_oauth2_user_pk: int) -> N
  if oncall_schedules_to_consider_for_shift_swaps:
  users_schedules = users_schedules.filter(public_primary_key__in=oncall_schedules_to_consider_for_shift_swaps)
 
- for out_of_office_event in google_api_client.fetch_out_of_office_events():
+ try:
+ out_of_office_events = google_api_client.fetch_out_of_office_events()
+ except GoogleCalendarHTTPError:
+ logger.info(f"Failed to fetch out of office events for user {user_id}")
+ return
+
+ for out_of_office_event in out_of_office_events:
  raw_event = out_of_office_event.raw_event
 
  event_title = raw_event["summary"]

engine/apps/google/tests/test_sync_out_of_office_calendar_events_for_user.py

@@ -3,6 +3,7 @@
 
 import pytest
 from django.utils import timezone
+from googleapiclient.errors import HttpError
 
 from apps.google import constants, tasks
 from apps.schedules.models import CustomOnCallShift, OnCallScheduleWeb, ShiftSwapRequest
@@ -140,6 +141,28 @@ def _test_setup(out_of_office_events):
  return _test_setup
 
 
+class MockResponse:
+ def __init__(self, reason=None, status=200) -> None:
+ self.reason = reason or ""
+ self.status = status
+
+
+@patch("apps.google.client.build")
+@pytest.mark.django_db
+def test_sync_out_of_office_calendar_events_for_user_httperror(mock_google_api_client_build, test_setup):
+ mock_response = MockResponse(reason="forbidden", status=403)
+ mock_google_api_client_build.return_value.events.return_value.list.return_value.execute.side_effect = HttpError(
+ resp=mock_response, content=b"error"
+ )
+
+ google_oauth2_user, schedule = test_setup([])
+ user = google_oauth2_user.user
+
+ tasks.sync_out_of_office_calendar_events_for_user(google_oauth2_user.pk)
+
+ assert ShiftSwapRequest.objects.filter(beneficiary=user, schedule=schedule).count() == 0
+
+
 @patch("apps.google.client.build")
 @pytest.mark.django_db
 def test_sync_out_of_office_calendar_events_for_user_no_ooo_events(mock_google_api_client_build, test_setup):

engine/apps/public_api/serializers/incidents.py

@@ -1,8 +1,6 @@
-from django.db.models import Prefetch
 from rest_framework import serializers
 
 from apps.alerts.models import AlertGroup
-from apps.telegram.models.message import TelegramMessage
 from common.api_helpers.custom_fields import UserIdField
 from common.api_helpers.mixins import EagerLoadingMixin
 
@@ -19,14 +17,6 @@ class IncidentSerializer(EagerLoadingMixin, serializers.ModelSerializer):
  resolved_by = UserIdField(read_only=True, source="resolved_by_user")
 
  SELECT_RELATED = ["channel", "channel_filter", "slack_message", "channel__organization"]
- PREFETCH_RELATED = [
- "alerts",
- Prefetch(
- "telegram_messages",
- TelegramMessage.objects.filter(chat_id__startswith="-", message_type=TelegramMessage.ALERT_GROUP_MESSAGE),
- to_attr="prefetched_telegram_messages",
- ),
- ]
 
  class Meta:
  model = AlertGroup
@@ -50,7 +40,7 @@ def get_title(self, obj):
  return obj.web_title_cache
 
  def get_alerts_count(self, obj):
- return len(obj.alerts.all())
+ return obj.alerts.count()
 
  def get_state(self, obj):
  return obj.state

engine/apps/slack/errors.py

@@ -68,6 +68,10 @@ class SlackAPIMessageNotFoundError(SlackAPIError):
  errors = ("message_not_found",)
 
 
+class SlackAPICantUpdateMessageError(SlackAPIError):
+ errors = ("cant_update_message",)
+
+
 class SlackAPIUserNotFoundError(SlackAPIError):
  errors = ("user_not_found",)
 

engine/apps/slack/models/slack_message.py

@@ -98,7 +98,7 @@ def permalink(self) -> typing.Optional[str]:
 
  @property
  def deep_link(self) -> str:
- return f"slack://channel?team={self.slack_team_identity.slack_id}&id={self.channel_id}&message={self.slack_id}"
+ return f"https://slack.com/app_redirect?channel={self.channel_id}&team={self.slack_team_identity.slack_id}&message={self.slack_id}"
 
  def send_slack_notification(self, user, alert_group, notification_policy):
  from apps.base.models import UserNotificationPolicyLogRecord

engine/apps/slack/scenarios/distribute_alerts.py

@@ -13,6 +13,7 @@
 from apps.api.permissions import RBACPermission
 from apps.slack.constants import CACHE_UPDATE_INCIDENT_SLACK_MESSAGE_LIFETIME
 from apps.slack.errors import (
+ SlackAPICantUpdateMessageError,
  SlackAPIChannelArchivedError,
  SlackAPIChannelInactiveError,
  SlackAPIChannelNotFoundError,
@@ -947,6 +948,7 @@ def update_log_message(self, alert_group: AlertGroup) -> None:
  SlackAPIChannelArchivedError,
  SlackAPIChannelInactiveError,
  SlackAPIInvalidAuthError,
+ SlackAPICantUpdateMessageError,
  ):
  pass
  else: