[Alerts] Update entity id to take specific id & fix slack job overview (

#5595)

mlrun/alerts/alert.py

@@ -137,7 +137,7 @@ def _apply_template(self, template):
             template = db.get_alert_template(template)
 
         # Extract parameters from the template and apply them to the AlertConfig object
-        self.description = template.description
+        self.summary = template.summary
         self.severity = template.severity
         self.criteria = template.criteria
         self.trigger = template.trigger

mlrun/common/schemas/alert.py

@@ -162,7 +162,7 @@ class AlertTemplate(
     system_generated: bool = False
 
     # AlertConfig fields that are pre-defined
-    description: Optional[str] = (
+    summary: Optional[str] = (
         "String to be sent in the generated notifications e.g. 'Model {{project}}/{{entity}} is drifting.'"
         "See AlertConfig.summary description"
     )
@@ -175,7 +175,7 @@ class AlertTemplate(
     def templates_differ(self, other):
         return (
             self.template_description != other.template_description
-            or self.description != other.description
+            or self.summary != other.summary
             or self.severity != other.severity
             or self.trigger != other.trigger
             or self.reset_policy != other.reset_policy

mlrun/utils/helpers.py

@@ -973,6 +973,15 @@ def get_ui_url(project, uid=None):
     return url
 
 
+def get_model_endpoint_url(project, model_name, model_endpoint_id):
+    url = ""
+    if mlrun.mlconf.resolve_ui_url():
+        url = f"{mlrun.mlconf.resolve_ui_url()}/{mlrun.mlconf.ui.projects_prefix}/{project}/models"
+        if model_name:
+            url += f"/model-endpoints/{model_name}/{model_endpoint_id}/overview"
+    return url
+
+
 def get_workflow_url(project, id=None):
     url = ""
     if mlrun.mlconf.resolve_ui_url():

mlrun/utils/notifications/notification/slack.py

@@ -136,7 +136,7 @@ def _get_alert_fields(
         line = [
             self._get_slack_row(f":bell: {alert.name} alert has occurred"),
             self._get_slack_row(f"*Project:*\n{alert.project}"),
-            self._get_slack_row(f"*UID:*\n{event_data.entity.ids[0]}"),
+            self._get_slack_row(f"*ID:*\n{event_data.entity.ids[0]}"),
         ]
 
         if alert.summary:
@@ -153,10 +153,21 @@ def _get_alert_fields(
             data_text = "\n".join(data_lines)
             line.append(self._get_slack_row(f"*Event data:*\n{data_text}"))
 
-        if url := mlrun.utils.helpers.get_ui_url(
-            alert.project, event_data.entity.ids[0]
-        ):
-            line.append(self._get_slack_row(f"*Overview:*\n<{url}|*Job overview*>"))
+        if (
+            event_data.entity.kind == mlrun.common.schemas.alert.EventEntityKind.JOB
+        ):  # JOB entity
+            uid = event_data.value_dict.get("uid")
+            url = mlrun.utils.helpers.get_ui_url(alert.project, uid)
+            overview_type = "Job overview"
+        else:  # MODEL entity
+            model_name = event_data.value_dict.get("model")
+            model_endpoint_id = event_data.value_dict.get("model_endpoint_id")
+            url = mlrun.utils.helpers.get_model_endpoint_url(
+                alert.project, model_name, model_endpoint_id
+            )
+            overview_type = "Model endpoint"
+
+        line.append(self._get_slack_row(f"*Overview:*\n<{url}|*{overview_type}*>"))
 
         return line
 

server/api/constants.py

@@ -39,7 +39,7 @@ class MaskOperations(StrEnum):
         template_name="JobFailed",
         template_description="Generic template for job failure alerts",
         system_generated=True,
-        description="A job has failed",
+        summary="A job has failed",
         severity=mlrun.common.schemas.alert.AlertSeverity.MEDIUM,
         trigger={"events": [mlrun.common.schemas.alert.EventKind.FAILED]},
         reset_policy=mlrun.common.schemas.alert.ResetPolicy.MANUAL,
@@ -48,7 +48,7 @@ class MaskOperations(StrEnum):
         template_name="DriftDetected",
         template_description="Generic template for drift detected alerts",
         system_generated=True,
-        description="Model drift has been detected",
+        summary="Model drift has been detected",
         severity=mlrun.common.schemas.alert.AlertSeverity.HIGH,
         trigger={"events": [mlrun.common.schemas.alert.EventKind.DRIFT_DETECTED]},
         reset_policy=mlrun.common.schemas.alert.ResetPolicy.MANUAL,
@@ -57,7 +57,7 @@ class MaskOperations(StrEnum):
         template_name="DriftSuspected",
         template_description="Generic template for drift suspected alerts",
         system_generated=True,
-        description="Model drift is suspected",
+        summary="Model drift is suspected",
         severity=mlrun.common.schemas.alert.AlertSeverity.MEDIUM,
         trigger={"events": [mlrun.common.schemas.alert.EventKind.DRIFT_SUSPECTED]},
         reset_policy=mlrun.common.schemas.alert.ResetPolicy.MANUAL,

server/api/crud/alerts.py

@@ -184,6 +184,8 @@ def process_event(
                 obj=state_obj,
                 active=active,
             )
+        else:
+            logger.debug("The entity of the alert does not match the one in event")
 
     @staticmethod
     def _event_entity_matches(alert_entity, event_entity):

server/api/main.py

@@ -777,14 +777,16 @@ def _generate_event_on_failed_runs(
 
     for run in runs:
         project = run["metadata"]["project"]
-        uid = run["metadata"]["uid"]
-        entity = {
-            "kind": alert_objects.EventEntityKind.JOB,
-            "project": project,
-            "ids": [uid],
-        }
+        run_uid = run["metadata"]["uid"]
+        run_name = run["metadata"]["name"]
+        entity = mlrun.common.schemas.alert.EventEntities(
+            kind=alert_objects.EventEntityKind.JOB,
+            project=project,
+            ids=[run_name],
+        )
+        event_value = {"uid": run_uid}
         event_data = mlrun.common.schemas.Event(
-            kind=alert_objects.EventKind.FAILED, entity=entity
+            kind=alert_objects.EventKind.FAILED, entity=entity, value_dict=event_value
         )
         mlrun.get_run_db().generate_event(alert_objects.EventKind.FAILED, event_data)
 

tests/api/api/test_projects.py

@@ -1251,7 +1251,7 @@ def _create_resources_of_all_kinds(
         entities={
             "kind": mlrun.common.schemas.alert.EventEntityKind.MODEL,
             "project": project,
-            "ids": ["*"],
+            "ids": [1234],
         },
         trigger={"events": [mlrun.common.schemas.alert.EventKind.DRIFT_DETECTED]},
         notifications=[{"notification": notification.to_dict()}],

tests/integration/sdk_api/alerts/test_alerts.py

@@ -193,17 +193,15 @@ def test_alert_templates(self):
 
         # generate an alert from a template
         alert_name = "new_alert"
-        alert_summary = "Model is drifting"
         alert_from_template = mlrun.alerts.alert.AlertConfig(
             project=project_name,
             name=alert_name,
-            summary=alert_summary,
             template=drift_template,
         )
 
         # test modifiers on the alert config
         entities = alert_objects.EventEntities(
-            kind=alert_objects.EventEntityKind.MODEL, project=project_name, ids=["*"]
+            kind=alert_objects.EventEntityKind.MODEL, project=project_name, ids=[1234]
         )
         alert_from_template.with_entities(entities=entities)
 
@@ -226,8 +224,7 @@ def test_alert_templates(self):
             alert_from_template,
             project_name=project_name,
             alert_name=alert_name,
-            alert_summary=alert_summary,
-            alert_description=drift_template.description,
+            alert_summary=drift_template.summary,
             alert_severity=drift_template.severity,
             alert_trigger=drift_template.trigger,
             alert_reset_policy=drift_template.reset_policy,
@@ -641,7 +638,7 @@ def _generate_alert_create_request(
             name=name,
             summary=summary,
             severity=severity,
-            entities={"kind": entity_kind, "project": entity_project, "ids": ["*"]},
+            entities={"kind": entity_kind, "project": entity_project, "ids": [1234]},
             trigger={"events": [event_name]},
             criteria=criteria,
             notifications=notifications,

tests/system/alerts/test_alerts.py

@@ -56,11 +56,13 @@ def test_job_failure_alert(self):
         # create an alert with webhook notification
         alert_name = "failure_webhook"
         alert_summary = "Job failed"
+        run_id = "test-func-handler"
         notifications = self._generate_failure_notifications(nuclio_function_url)
         self._create_alert_config(
             self.project_name,
             alert_name,
             alert_objects.EventEntityKind.JOB,
+            run_id,
             alert_summary,
             alert_objects.EventKind.FAILED,
             notifications,
@@ -92,11 +94,13 @@ def test_drift_detection_alert(self):
         # create an alert with two webhook notifications
         alert_name = "drift_webhook"
         alert_summary = "Model is drifting"
+        endpoint_id = "demo-endpoint"
         notifications = self._generate_drift_notifications(nuclio_function_url)
         self._create_alert_config(
             self.project_name,
             alert_name,
             alert_objects.EventEntityKind.MODEL,
+            endpoint_id,
             alert_summary,
             alert_objects.EventKind.DRIFT_DETECTED,
             notifications,
@@ -108,7 +112,6 @@ def test_drift_detection_alert(self):
         )
         writer._wait_for_function_deployment(db=writer._get_db())
 
-        endpoint_id = "demo-endpoint"
         mlrun.model_monitoring.api.get_or_create_model_endpoint(
             project=self.project.metadata.name,
             endpoint_id=endpoint_id,
@@ -220,6 +223,7 @@ def _create_alert_config(
         project,
         name,
         entity_kind,
+        entity_id,
         summary,
         event_name,
         notifications,
@@ -231,7 +235,7 @@ def _create_alert_config(
             summary=summary,
             severity=alert_objects.AlertSeverity.LOW,
             entities=alert_objects.EventEntities(
-                kind=entity_kind, project=project, ids=["*"]
+                kind=entity_kind, project=project, ids=[entity_id]
             ),
             trigger=alert_objects.AlertTrigger(events=[event_name]),
             criteria=criteria,