fix: Add support to customize ignorable statuses for maintenance windows

keep-ui/app/(keep)/maintenance/create-or-update-maintenance-rule.tsx

@@ -9,6 +9,8 @@ import {
   NumberInput,
   Select,
   SelectItem,
+  MultiSelect,
+  MultiSelectItem,
 } from "@tremor/react";
 import { FormEvent, useEffect, useState } from "react";
 import { toast } from "react-toastify";
@@ -20,12 +22,19 @@ import "react-datepicker/dist/react-datepicker.css";
 import { useRouter } from "next/navigation";
 import { useApi } from "@/shared/lib/hooks/useApi";
 import { showErrorToast } from "@/shared/ui";
+import { Status } from "@/entities/alerts/model";
+import { capitalize } from "@/utils/helpers";
 
 interface Props {
   maintenanceToEdit: MaintenanceRule | null;
   editCallback: (rule: MaintenanceRule | null) => void;
 }
 
+const DEFAULT_IGNORE_STATUSES = [
+    "resolved",
+    "acknowledged",
+]
+
 export default function CreateOrUpdateMaintenanceRule({
   maintenanceToEdit,
   editCallback,
@@ -40,6 +49,7 @@ export default function CreateOrUpdateMaintenanceRule({
   const [intervalType, setIntervalType] = useState<string>("minutes");
   const [enabled, setEnabled] = useState<boolean>(true);
   const [suppress, setSuppress] = useState<boolean>(false);
+  const [ignoreStatuses, setIgnoreStatuses] = useState<string[]>(DEFAULT_IGNORE_STATUSES);
   const editMode = maintenanceToEdit !== null;
   const router = useRouter();
   useEffect(() => {
@@ -50,6 +60,7 @@ export default function CreateOrUpdateMaintenanceRule({
       setStartTime(new Date(maintenanceToEdit.start_time));
       setSuppress(maintenanceToEdit.suppress);
       setEnabled(maintenanceToEdit.enabled);
+      setIgnoreStatuses(maintenanceToEdit.ignore_statuses);
       if (maintenanceToEdit.duration_seconds) {
         setEndInterval(maintenanceToEdit.duration_seconds / 60);
       }
@@ -64,6 +75,7 @@ export default function CreateOrUpdateMaintenanceRule({
     setEndInterval(5);
     setSuppress(false);
     setEnabled(true);
+    setIgnoreStatuses([]);
     router.replace("/maintenance");
   };
 
@@ -99,6 +111,7 @@ export default function CreateOrUpdateMaintenanceRule({
         duration_seconds: calculateDurationInSeconds(),
         suppress: suppress,
         enabled: enabled,
+        ignore_statuses: ignoreStatuses,
       });
       clearForm();
       mutate();
@@ -123,6 +136,7 @@ export default function CreateOrUpdateMaintenanceRule({
         duration_seconds: calculateDurationInSeconds(),
         suppress: suppress,
         enabled: enabled,
+        ignore_statuses: ignoreStatuses,
       });
       exitEditMode();
       mutate();
@@ -178,6 +192,14 @@ export default function CreateOrUpdateMaintenanceRule({
           showSqlImport={false}
         />
       </div>
+
+      <div className="mt-2.5">
+        <MultiSelect value={ignoreStatuses} onValueChange={setIgnoreStatuses}>
+          {Object.values(Status).map((value) => {
+            return <MultiSelectItem key={value} value={value}>{capitalize(value)}</MultiSelectItem>
+          })}
+        </MultiSelect>
+      </div>
       <div className="mt-2.5">
         <Text>
           Start At<span className="text-red-500 text-xs">*</span>

keep-ui/app/(keep)/maintenance/model.ts

@@ -10,6 +10,7 @@ export interface MaintenanceRule {
   updated_at?: Date;
   suppress: boolean;
   enabled: boolean;
+  ignore_statuses: string[];
 }
 
 export interface MaintenanceRuleCreate {
@@ -20,4 +21,5 @@ export interface MaintenanceRuleCreate {
   end_time?: Date;
   duration_seconds?: number;
   enabled: boolean;
+  ignore_statuses: string[];
 }

keep/api/bl/maintenance_windows_bl.py

@@ -15,11 +15,6 @@
 
 class MaintenanceWindowsBl:
 
-    ALERT_STATUSES_TO_IGNORE = [
-        AlertStatus.RESOLVED.value,
-        AlertStatus.ACKNOWLEDGED.value,
-    ]
-
     def __init__(self, tenant_id: str, session: Session | None) -> None:
         self.logger = logging.getLogger(__name__)
         self.tenant_id = tenant_id
@@ -42,17 +37,17 @@
             )
             return False
 
-        if alert.status in self.ALERT_STATUSES_TO_IGNORE:
-            self.logger.debug(
-                "Alert status is set to be ignored, ignoring maintenance windows",
-                extra={"tenant_id": self.tenant_id},
-            )
-            return False
-
         self.logger.info("Checking maintenance window for alert", extra=extra)
         env = celpy.Environment()
 
         for maintenance_rule in self.maintenance_rules:
+            if alert.status in maintenance_rule.ignore_statuses:
+                self.logger.debug(
+                    "Alert status is set to be ignored, ignoring maintenance windows",
+                    extra={"tenant_id": self.tenant_id},
+                )
+                continue
+
             if maintenance_rule.end_time <= datetime.datetime.now():
                 # this is wtf error, should not happen because of query in init
                 self.logger.error(

keep/api/models/db/maintenance_window.py

@@ -3,11 +3,17 @@
 from typing import Optional
 
 from pydantic import BaseModel
-from sqlalchemy import DateTime
+from sqlalchemy import DateTime, JSON
 
 # third-parties
 from sqlmodel import Column, Field, Index, SQLModel, func
 
+from keep.api.models.alert import AlertStatus
+
+DEFAULT_ALERT_STATUSES_TO_IGNORE = [
+    AlertStatus.RESOLVED.value,
+    AlertStatus.ACKNOWLEDGED.value,
+]
 
 class MaintenanceWindowRule(SQLModel, table=True):
     id: Optional[int] = Field(default=None, primary_key=True)
@@ -29,6 +35,7 @@ class MaintenanceWindowRule(SQLModel, table=True):
     )
     suppress: bool = False
     enabled: bool = True
+    ignore_statuses: list = Field(sa_column=Column(JSON), default_factory=list)
 
     __table_args__ = (
         Index("ix_maintenance_rule_tenant_id", "tenant_id"),
@@ -44,6 +51,7 @@ class MaintenanceRuleCreate(BaseModel):
     duration_seconds: Optional[int] = None
     suppress: bool = False
     enabled: bool = True
+    ignore_statuses: list[str] = DEFAULT_ALERT_STATUSES_TO_IGNORE
 
 
 class MaintenanceRuleRead(BaseModel):
@@ -58,3 +66,4 @@ class MaintenanceRuleRead(BaseModel):
     updated_at: Optional[datetime]
     suppress: bool = False
     enabled: bool = True
+    ignore_statuses: list[str] = DEFAULT_ALERT_STATUSES_TO_IGNORE

keep/api/models/db/migrations/versions/2025-05-16-14-33_aa167915c4d6.py

@@ -0,0 +1,46 @@
+"""Add ignore_statuses to MaintenanceWindowRule
+
+Revision ID: aa167915c4d6
+Revises: bedb5f07417b
+Create Date: 2025-05-16 14:33:29.828572
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+from sqlmodel import Session
+
+from keep.api.models.db.maintenance_window import DEFAULT_ALERT_STATUSES_TO_IGNORE
+
+# revision identifiers, used by Alembic.
+revision = "aa167915c4d6"
+down_revision = "bedb5f07417b"
+branch_labels = None
+depends_on = None
+
+migration_metadata = sa.MetaData()
+
+mwr_table = sa.Table(
+    'maintenancewindowrule',
+    migration_metadata,
+    sa.Column('id', sa.Integer, primary_key=True),
+    sa.Column('ignore_statuses', sa.JSON)
+)
+
+def populate_db():
+    session = Session(op.get_bind())
+    session.execute(sa.update(mwr_table).values(ignore_statuses=DEFAULT_ALERT_STATUSES_TO_IGNORE))
+
+
+def upgrade() -> None:
+
+    with op.batch_alter_table("maintenancewindowrule", schema=None) as batch_op:
+        batch_op.add_column(sa.Column("ignore_statuses", sa.JSON(), nullable=True))
+
+    populate_db()
+
+
+def downgrade() -> None:
+
+    with op.batch_alter_table("maintenancewindowrule", schema=None) as batch_op:
+        batch_op.drop_column("ignore_statuses")

tests/test_maintenance_windows_bl.py

@@ -13,6 +13,20 @@ def mock_session():
     return MagicMock()
 
 
+@pytest.fixture
+def active_maintenance_window_rule_custom_ignore():
+    return MaintenanceWindowRule(
+        id=1,
+        name="Active maintenance_window",
+        tenant_id="test-tenant",
+        cel_query='source == "test-source"',
+        start_time=datetime.utcnow() - timedelta(hours=1),
+        end_time=datetime.utcnow() + timedelta(days=1),
+        enabled=True,
+        ignore_statuses=[AlertStatus.FIRING.value,],
+    )
+
+
 @pytest.fixture
 def active_maintenance_window_rule():
     return MaintenanceWindowRule(
@@ -23,6 +37,7 @@ def active_maintenance_window_rule():
         start_time=datetime.utcnow() - timedelta(hours=1),
         end_time=datetime.utcnow() + timedelta(days=1),
         enabled=True,
+        ignore_statuses=[AlertStatus.RESOLVED.value, AlertStatus.ACKNOWLEDGED.value],
     )
 
 
@@ -198,3 +213,26 @@ def test_alert_with_missing_cel_field(mock_session, active_maintenance_window_ru
 
     # Should return False because the field doesn't exist
     assert result is False
+
+
+def test_alert_not_ignored_due_to_custom_status(
+    mock_session, active_maintenance_window_rule_custom_ignore, alert_dto
+):
+    # Set the alert status to RESOLVED
+
+    mock_session.query.return_value.filter.return_value.filter.return_value.filter.return_value.all.return_value = [
+        active_maintenance_window_rule_custom_ignore
+    ]
+
+    maintenance_window_bl = MaintenanceWindowsBl(
+        tenant_id="test-tenant", session=mock_session
+    )
+
+    # Should return False because the alert status is FIRING
+    alert_dto.status = AlertStatus.FIRING.value
+    result = maintenance_window_bl.check_if_alert_in_maintenance_windows(alert_dto)
+    assert result is False
+
+    alert_dto.status = AlertStatus.RESOLVED.value
+    result = maintenance_window_bl.check_if_alert_in_maintenance_windows(alert_dto)
+    assert result is True