qontract-reconcile 0.10.2.dev105__tar.gz → 0.10.2.dev107__tar.gz

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: qontract-reconcile
-Version: 0.10.2.dev105
+Version: 0.10.2.dev107
 Summary: Collection of tools to reconcile services with their desired state as defined in the app-interface DB.
 Project-URL: homepage, https://github.com/app-sre/qontract-reconcile
 Project-URL: repository, https://github.com/app-sre/qontract-reconcile

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/reconcile/fleet_labeler/integration.py

@@ -66,11 +66,11 @@ class FleetLabelerIntegration(QontractReconcileIntegration[NoParams]):
             secret_reader=self.secret_reader,
             dry_run=dry_run,
         )
-        self._metrics = dependencies.metrics
         self.reconcile(dependencies=dependencies)
 
     def reconcile(self, dependencies: Dependencies) -> None:
         validate_label_specs(specs=dependencies.label_specs_by_name)
+        all_cluster_ids: set[str] = set()
         for spec_name, ocm in dependencies.ocm_clients_by_label_spec_name.items():
             spec = dependencies.label_specs_by_name[spec_name]
             discovered_clusters = self._discover_desired_clusters(
@@ -99,6 +99,16 @@ class FleetLabelerIntegration(QontractReconcileIntegration[NoParams]):
                 ocm=ocm,
                 dry_run=synch_labels,
             )
+            num_labels = 0
+            for cluster in spec.clusters:
+                all_cluster_ids.add(cluster.cluster_id)
+                num_labels += len(cluster.subscription_labels)
+            dependencies.metrics.set_managed_labels_gauge(
+                ocm_name=spec.ocm_env.name,
+                spec_name=spec_name,
+                value=num_labels,
+            )
+        dependencies.metrics.set_managed_clusters_gauge(value=len(all_cluster_ids))
 
     def _discover_desired_clusters(
         self, spec: FleetLabelsSpecV1, ocm: OCMClient

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/reconcile/fleet_labeler/metrics.py

@@ -36,14 +36,34 @@ class FleetLabelerMetrics:
             value,
         )
 
+    def set_managed_labels_gauge(
+        self, ocm_name: str, spec_name: str, value: int
+    ) -> None:
+        metrics.set_gauge(
+            FleetLabelerManagedLabelsGauge(
+                ocm_name=ocm_name,
+                spec=spec_name,
+            ),
+            value,
+        )
+
+    def set_managed_clusters_gauge(self, value: int) -> None:
+        metrics.set_gauge(
+            FleetLabelerManagedClustersGauge(),
+            value,
+        )
+
 
 class FleetLabelerBaseMetric(BaseModel):
     integration: str = QONTRACT_INTEGRATION
+
+
+class FleetLabelerSpecBaseMetric(FleetLabelerBaseMetric):
     ocm_name: str
     spec: str
 
 
-class FleetLabelerDuplicateClusterMatchesGauge(FleetLabelerBaseMetric, GaugeMetric):
+class FleetLabelerDuplicateClusterMatchesGauge(FleetLabelerSpecBaseMetric, GaugeMetric):
     """
     Gauge for the number of clusters that have duplicate matches. Clusters with
     duplicate matches are being ignored by fleet labeler, as it cannot clearly
@@ -56,7 +76,7 @@ class FleetLabelerDuplicateClusterMatchesGauge(FleetLabelerBaseMetric, GaugeMetr
         return "fleet_labeler_duplicate_cluster_matches"
 
 
-class FleetLabelerLabelRenderingErrorGauge(FleetLabelerBaseMetric, GaugeMetric):
+class FleetLabelerLabelRenderingErrorGauge(FleetLabelerSpecBaseMetric, GaugeMetric):
     """
     Gauge for the number of clusters that have label render errors. This
     happens when the fleet labeler is unable to render the subscription labels
@@ -66,3 +86,28 @@ class FleetLabelerLabelRenderingErrorGauge(FleetLabelerBaseMetric, GaugeMetric):
     @classmethod
     def name(cls) -> str:
         return "fleet_labeler_label_rendering_error"
+
+
+class FleetLabelerManagedLabelsGauge(FleetLabelerSpecBaseMetric, GaugeMetric):
+    """
+    Gauge for the current number of labels under management.
+    Note, that label<->cluster combination must be unique across all
+    specs, hence we can give the spec as a label here.
+    """
+
+    @classmethod
+    def name(cls) -> str:
+        return "fleet_labeler_managed_labels"
+
+
+class FleetLabelerManagedClustersGauge(FleetLabelerBaseMetric, GaugeMetric):
+    """
+    Gauge for the current number of clusters under management.
+    Note, that a cluster can be part of multiple spec inventories.
+    This metric only cares about the total number of unique clusters,
+    thus the spec is not a label here.
+    """
+
+    @classmethod
+    def name(cls) -> str:
+        return "fleet_labeler_managed_clusters"

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/reconcile/gql_definitions/introspection.json

@@ -46575,6 +46575,77 @@
                             },
                             "isDeprecated": false,
                             "deprecationReason": null
+                        },
+                        {
+                            "name": "health_check",
+                            "description": null,
+                            "args": [],
+                            "type": {
+                                "kind": "OBJECT",
+                                "name": "NamespaceTerraformResourceALBTargetHealthcheck_v1",
+                                "ofType": null
+                            },
+                            "isDeprecated": false,
+                            "deprecationReason": null
+                        }
+                    ],
+                    "inputFields": null,
+                    "interfaces": [],
+                    "enumValues": null,
+                    "possibleTypes": null
+                },
+                {
+                    "kind": "OBJECT",
+                    "name": "NamespaceTerraformResourceALBTargetHealthcheck_v1",
+                    "description": null,
+                    "fields": [
+                        {
+                            "name": "unhealthy_threshold",
+                            "description": null,
+                            "args": [],
+                            "type": {
+                                "kind": "SCALAR",
+                                "name": "Int",
+                                "ofType": null
+                            },
+                            "isDeprecated": false,
+                            "deprecationReason": null
+                        },
+                        {
+                            "name": "timeout",
+                            "description": null,
+                            "args": [],
+                            "type": {
+                                "kind": "SCALAR",
+                                "name": "Int",
+                                "ofType": null
+                            },
+                            "isDeprecated": false,
+                            "deprecationReason": null
+                        },
+                        {
+                            "name": "interval",
+                            "description": null,
+                            "args": [],
+                            "type": {
+                                "kind": "SCALAR",
+                                "name": "Int",
+                                "ofType": null
+                            },
+                            "isDeprecated": false,
+                            "deprecationReason": null
+                        },
+                        {
+                            "name": "healthy_threshold",
+                            "description": null,
+                            "args": [],
+                            "type": {
+                                "kind": "SCALAR",
+                                "name": "Int",
+                                "ofType": null
+                            },
+                            "isDeprecated": false,
+                            "deprecationReason": null
                         }
                     ],
                     "inputFields": null,

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/reconcile/gql_definitions/terraform_resources/terraform_resources_namespaces.py

@@ -319,6 +319,12 @@ query TerraformResourcesNamespaces {
                     openshift_service
                     protocol
                     protocol_version
+                    health_check {
+                        unhealthy_threshold
+                        timeout
+                        interval
+                        healthy_threshold
+                    }
                 }
                 rules {
                     condition {
@@ -799,6 +805,13 @@ class NamespaceTerraformResourceS3CloudFrontPublicKeyV1(NamespaceTerraformResour
     annotations: Optional[str] = Field(..., alias="annotations")
 
 
+class NamespaceTerraformResourceALBTargetHealthcheckV1(ConfiguredBaseModel):
+    unhealthy_threshold: Optional[int] = Field(..., alias="unhealthy_threshold")
+    timeout: Optional[int] = Field(..., alias="timeout")
+    interval: Optional[int] = Field(..., alias="interval")
+    healthy_threshold: Optional[int] = Field(..., alias="healthy_threshold")
+
+
 class NamespaceTerraformResourceALBTargetsV1(ConfiguredBaseModel):
     name: str = Field(..., alias="name")
     default: bool = Field(..., alias="default")
@@ -806,6 +819,7 @@ class NamespaceTerraformResourceALBTargetsV1(ConfiguredBaseModel):
     openshift_service: Optional[str] = Field(..., alias="openshift_service")
     protocol: Optional[str] = Field(..., alias="protocol")
     protocol_version: Optional[str] = Field(..., alias="protocol_version")
+    health_check: Optional[NamespaceTerraformResourceALBTargetHealthcheckV1] = Field(..., alias="health_check")
 
 
 class NamespaceTerraformResourceALBConditionV1(ConfiguredBaseModel):

{qontract_reconcile-0.10.2.dev105 → qontract_reconcile-0.10.2.dev107}/reconcile/utils/terrascript_aws_client.py

@@ -5376,6 +5376,25 @@ class TerrascriptClient:  # pylint: disable=too-many-public-methods
                 f"{identifier}-{target_name}", **lbt_random_id_values
             )
             tf_resources.append(lbt_random_id)
+            health_check_interval = 10
+            health_check_timeout = 5
+            health_check_unhealthy_threshold = 3
+            health_check_healthy_threshold = 3
+            if health_check := t.get("health_check"):
+                health_check_interval = (
+                    health_check.get("interval") or health_check_interval
+                )
+                health_check_timeout = (
+                    health_check.get("timeout") or health_check_timeout
+                )
+                health_check_unhealthy_threshold = (
+                    health_check.get("unhealthy_threshold")
+                    or health_check_unhealthy_threshold
+                )
+                health_check_healthy_threshold = (
+                    health_check.get("healthy_threshold")
+                    or health_check_healthy_threshold
+                )
 
             # https://www.terraform.io/docs/providers/aws/r/
             # lb_target_group.html
@@ -5388,7 +5407,10 @@ class TerrascriptClient:  # pylint: disable=too-many-public-methods
                 "target_type": "ip",
                 "vpc_id": vpc_id,
                 "health_check": {
-                    "interval": 10,
+                    "interval": health_check_interval,
+                    "timeout": health_check_timeout,
+                    "unhealthy_threshold": health_check_unhealthy_threshold,
+                    "healthy_threshold": health_check_healthy_threshold,
                     "path": "/",
                     "protocol": "HTTPS",
                     "port": 443,