feat: Grafana feature health provider (#5098)

api/Makefile

@@ -77,12 +77,21 @@ django-make-migrations:
 	poetry run python manage.py waitfordb
 	poetry run python manage.py makemigrations $(opts)
 
+.PHONY: django-squash-migrations
+django-squash-migrations:
+	poetry run python manage.py waitfordb
+	poetry run python manage.py squashmigrations $(opts)
+
 .PHONY: django-migrate
 django-migrate:
 	poetry run python manage.py waitfordb
 	poetry run python manage.py migrate
 	poetry run python manage.py createcachetable
 
+.PHONY: django-shell
+django-shell:
+	poetry run python manage.py shell
+
 .PHONY: django-collect-static
 django-collect-static:
 	poetry run python manage.py collectstatic --noinput
@@ -98,7 +107,7 @@ serve:
 generate-ld-client-types:
 	curl -sSL https://app.launchdarkly.com/api/v2/openapi.json | \
 	npx openapi-format /dev/fd/0 \
-		--filterFile ld-openapi-filter.yaml | \
+		--filterFile openapi-filter-launchdarkly.yaml | \
 	datamodel-codegen \
 		--output integrations/launch_darkly/types.py \
 		--output-model-type typing.TypedDict \
@@ -108,6 +117,20 @@ generate-ld-client-types:
 		--wrap-string-literal \
 		--special-field-name-prefix=
 
+.PHONY: generate-grafana-client-types
+generate-grafana-client-types:
+	curl -sSL https://raw.githubusercontent.com/grafana/grafana/refs/heads/main/public/openapi3.json | \
+	npx openapi-format /dev/fd/0 \
+		--filterFile openapi-filter-grafana.yaml | \
+	datamodel-codegen \
+		--output integrations/grafana/types.py \
+		--output-model-type typing.TypedDict \
+		--target-python-version 3.10 \
+		--use-double-quotes \
+		--use-standard-collections \
+		--wrap-string-literal \
+		--special-field-name-prefix=
+
 .PHONY: integrate-private-tests
 integrate-private-tests:
 	$(eval WORKFLOW_REVISION := $(shell grep -A 1 "\[tool.poetry.group.workflows.dependencies\]" pyproject.toml | awk -F '"' '{printf $$4}'))

api/features/feature_health/mappers.py

@@ -0,0 +1,32 @@
+import json
+import typing
+
+from features.feature_health.models import FeatureHealthEvent
+from features.feature_health.types import FeatureHealthEventData
+
+if typing.TYPE_CHECKING:
+    from environments.models import Environment
+    from features.models import Feature
+
+
+def map_feature_health_event_data_to_feature_health_event(
+    *,
+    feature_health_event_data: FeatureHealthEventData,
+    feature: "Feature",
+    environment: "Environment | None",
+) -> FeatureHealthEvent:
+    if reason := feature_health_event_data.reason:
+        instance_reason = json.dumps(reason)
+    else:
+        instance_reason = None
+    instance = FeatureHealthEvent(
+        feature=feature,
+        environment=environment,
+        type=feature_health_event_data.type.value,
+        reason=instance_reason,
+        external_id=feature_health_event_data.external_id,
+        provider_name=feature_health_event_data.provider_name,
+    )
+    if feature_health_event_data.created_at:
+        instance.created_at = feature_health_event_data.created_at
+    return instance

api/features/feature_health/migrations/0002_featurehealthevent_add_external_id_alter_created_at.py

@@ -0,0 +1,38 @@
+# Generated by Django 4.2.18 on 2025-02-17 15:13
+
+from django.db import migrations, models
+import django.utils.timezone
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ("feature_health", "0001_initial"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="featurehealthevent",
+            name="external_id",
+            field=models.CharField(blank=True, max_length=255, null=True),
+        ),
+        migrations.AddField(
+            model_name="historicalfeaturehealthevent",
+            name="external_id",
+            field=models.CharField(blank=True, max_length=255, null=True),
+        ),
+        migrations.AlterField(
+            model_name="featurehealthevent",
+            name="created_at",
+            field=models.DateTimeField(
+                blank=True, default=django.utils.timezone.now, editable=False
+            ),
+        ),
+        migrations.AlterField(
+            model_name="historicalfeaturehealthevent",
+            name="created_at",
+            field=models.DateTimeField(
+                blank=True, default=django.utils.timezone.now, editable=False
+            ),
+        ),
+    ]

api/features/feature_health/models.py

@@ -5,7 +5,7 @@
     abstract_base_auditable_model_factory,
 )
 from django.db import models
-from django_lifecycle import AFTER_CREATE, LifecycleModelMixin, hook  # type: ignore[import-untyped]
+from django.utils import timezone
 
 from audit.related_object_type import RelatedObjectType
 from features.feature_health.constants import (
@@ -83,8 +83,17 @@ def get_latest_by_feature(
     ) -> "models.QuerySet[FeatureHealthEvent]":
         return (
             self.filter(feature=feature)
-            .order_by("provider_name", "environment_id", "-created_at")
-            .distinct("provider_name", "environment_id")
+            .order_by(
+                "provider_name",
+                "environment_id",
+                "external_id",
+                "-created_at",
+            )
+            .distinct(
+                "provider_name",
+                "environment_id",
+                "external_id",
+            )
         )
 
     def get_latest_by_project(
@@ -93,13 +102,23 @@ def get_latest_by_project(
     ) -> "models.QuerySet[FeatureHealthEvent]":
         return (
             self.filter(feature__project=project)
-            .order_by("provider_name", "environment_id", "feature_id", "-created_at")
-            .distinct("provider_name", "environment_id", "feature_id")
+            .order_by(
+                "provider_name",
+                "environment_id",
+                "external_id",
+                "feature_id",
+                "-created_at",
+            )
+            .distinct(
+                "provider_name",
+                "environment_id",
+                "external_id",
+                "feature_id",
+            )
         )
 
 
 class FeatureHealthEvent(
-    LifecycleModelMixin,  # type: ignore[misc]
     AbstractBaseExportableModel,
     abstract_base_auditable_model_factory(["uuid"]),  # type: ignore[misc]
 ):
@@ -126,16 +145,11 @@ class FeatureHealthEvent(
         null=True,
     )
 
-    created_at = models.DateTimeField(auto_now_add=True)
+    created_at = models.DateTimeField(default=timezone.now, editable=False, blank=True)
     type = models.CharField(max_length=50, choices=FeatureHealthEventType.choices)
     provider_name = models.CharField(max_length=255, null=True, blank=True)
     reason = models.TextField(null=True, blank=True)
-
-    @hook(AFTER_CREATE)
-    def set_feature_health_tag(self):  # type: ignore[no-untyped-def]
-        from features.feature_health.tasks import update_feature_unhealthy_tag
-
-        update_feature_unhealthy_tag.delay(args=(self.feature.id,))
+    external_id = models.CharField(max_length=255, null=True, blank=True)
 
     def get_create_log_message(
         self,

api/features/feature_health/providers/grafana/__init__.py

@@ -0,0 +1,5 @@
+from features.feature_health.providers.grafana.services import (
+    get_provider_response,
+)
+
+__all__ = ("get_provider_response",)

api/features/feature_health/providers/grafana/constants.py

@@ -0,0 +1,2 @@
+GRAFANA_FEATURE_LABEL_NAME = "flagsmith_feature"
+GRAFANA_ENVIRONMENT_LABEL_NAME = "flagsmith_environment"

api/features/feature_health/providers/grafana/mappers.py

@@ -0,0 +1,83 @@
+from features.feature_health.models import (
+    FeatureHealthEventType,
+    FeatureHealthProviderName,
+)
+from features.feature_health.providers.grafana.constants import (
+    GRAFANA_ENVIRONMENT_LABEL_NAME,
+    GRAFANA_FEATURE_LABEL_NAME,
+)
+from features.feature_health.providers.grafana.types import (
+    GrafanaAlertInstance,
+    GrafanaWebhookData,
+)
+from features.feature_health.types import (
+    FeatureHealthEventData,
+    FeatureHealthEventReason,
+)
+
+
+def map_payload_to_alert_instances(payload: str) -> list[GrafanaAlertInstance]:
+    webhook_data = GrafanaWebhookData.model_validate_json(payload)
+
+    return webhook_data.alerts
+
+
+def map_alert_instance_to_feature_health_event_reason(
+    alert_instance: GrafanaAlertInstance,
+) -> FeatureHealthEventReason:
+    reason_data: FeatureHealthEventReason = {"text_blocks": [], "url_blocks": []}
+
+    annotations = alert_instance.annotations
+
+    # Populate text blocks.
+    alert_name = alert_instance.labels.get("alertname") or "Alertmanager Alert"
+    description = annotations.get("description") or ""
+    reason_data["text_blocks"].append(
+        {
+            "title": alert_name,
+            "text": description,
+        }
+    )
+    if summary := annotations.get("summary"):
+        reason_data["text_blocks"].append(
+            {
+                "title": "Summary",
+                "text": summary,
+            }
+        )
+
+    # Populate URL blocks.
+    reason_data["url_blocks"].append(
+        {"title": "Alert", "url": alert_instance.generatorURL}
+    )
+    if dashboard_url := alert_instance.dashboardURL:
+        reason_data["url_blocks"].append({"title": "Dashboard", "url": dashboard_url})
+    if panel_url := alert_instance.panelURL:
+        reason_data["url_blocks"].append({"title": "Panel", "url": panel_url})
+    if runbook_url := annotations.get("runbook_url"):
+        reason_data["url_blocks"].append({"title": "Runbook", "url": runbook_url})
+
+    return reason_data
+
+
+def map_alert_instance_to_feature_health_event_data(
+    alert_instance: GrafanaAlertInstance,
+) -> FeatureHealthEventData | None:
+    labels = alert_instance.labels
+    if feature_name := labels.get(GRAFANA_FEATURE_LABEL_NAME):
+        if alert_instance.status == "firing":
+            created_at = alert_instance.startsAt
+            event_type = FeatureHealthEventType.UNHEALTHY
+        else:
+            created_at = alert_instance.endsAt
+            event_type = FeatureHealthEventType.HEALTHY
+        return FeatureHealthEventData(
+            created_at=created_at,
+            feature_name=feature_name,
+            environment_name=labels.get(GRAFANA_ENVIRONMENT_LABEL_NAME),
+            type=event_type,
+            external_id=alert_instance.fingerprint,
+            reason=map_alert_instance_to_feature_health_event_reason(alert_instance),
+            provider_name=FeatureHealthProviderName.GRAFANA.value,
+        )
+    return None

api/features/feature_health/providers/grafana/services.py

@@ -0,0 +1,20 @@
+from features.feature_health.providers.grafana.mappers import (
+    map_alert_instance_to_feature_health_event_data,
+    map_payload_to_alert_instances,
+)
+from features.feature_health.types import (
+    FeatureHealthEventData,
+    FeatureHealthProviderResponse,
+)
+
+
+def get_provider_response(payload: str) -> FeatureHealthProviderResponse:
+    events: list[FeatureHealthEventData] = []
+
+    for alert_instance in map_payload_to_alert_instances(payload):
+        if event_data := map_alert_instance_to_feature_health_event_data(
+            alert_instance
+        ):
+            events.append(event_data)
+
+    return FeatureHealthProviderResponse(events=events)

api/features/feature_health/providers/grafana/types.py

@@ -0,0 +1,23 @@
+from datetime import datetime
+
+from pydantic import BaseModel
+
+
+class AlertmanagerAlertInstance(BaseModel):
+    annotations: dict[str, str]
+    generatorURL: str
+    endsAt: datetime
+    fingerprint: str
+    labels: dict[str, str]
+    startsAt: datetime
+    status: str
+
+
+class GrafanaAlertInstance(AlertmanagerAlertInstance):
+    dashboardURL: str = ""
+    panelURL: str = ""
+    silenceURL: str = ""
+
+
+class GrafanaWebhookData(BaseModel):
+    alerts: list[GrafanaAlertInstance]

api/features/feature_health/providers/sample/__init__.py

@@ -1,5 +1,5 @@
-from features.feature_health.providers.sample.mappers import (
-    map_payload_to_provider_response,
+from features.feature_health.providers.sample.services import (
+    get_provider_response,
 )
 
-__all__ = ("map_payload_to_provider_response",)
+__all__ = ("get_provider_response",)

api/features/feature_health/providers/sample/mappers.py

@@ -1,11 +1,19 @@
-import json
+from pydantic.type_adapter import TypeAdapter
 
-from features.feature_health.models import FeatureHealthEventType
+from features.feature_health.models import (
+    FeatureHealthEventType,
+    FeatureHealthProviderName,
+)
 from features.feature_health.providers.sample.types import (
     SampleEvent,
     SampleEventStatus,
 )
-from features.feature_health.types import FeatureHealthProviderResponse
+from features.feature_health.types import (
+    FeatureHealthEventData,
+    FeatureHealthProviderResponse,
+)
+
+_sample_event_type_adapter = TypeAdapter(SampleEvent)
 
 
 def map_sample_event_status_to_feature_health_event_type(
@@ -21,13 +29,18 @@ def map_sample_event_status_to_feature_health_event_type(
 def map_payload_to_provider_response(
     payload: str,
 ) -> FeatureHealthProviderResponse:
-    event_data: SampleEvent = json.loads(payload)
+    event_data: SampleEvent = _sample_event_type_adapter.validate_json(payload)
 
     return FeatureHealthProviderResponse(
-        feature_name=event_data["feature"],
-        environment_name=event_data.get("environment"),
-        event_type=map_sample_event_status_to_feature_health_event_type(
-            event_data["status"]
-        ),
-        reason=event_data.get("reason", ""),
+        events=[
+            FeatureHealthEventData(
+                feature_name=event_data["feature"],
+                environment_name=event_data.get("environment"),
+                type=map_sample_event_status_to_feature_health_event_type(
+                    event_data["status"]
+                ),
+                reason=event_data.get("reason"),
+                provider_name=FeatureHealthProviderName.SAMPLE.value,
+            ),
+        ],
     )