qontract-reconcile 0.10.1rc660__py3-none-any.whl → 0.10.1rc662__py3-none-any.whl

{qontract_reconcile-0.10.1rc660.dist-info → qontract_reconcile-0.10.1rc662.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: qontract-reconcile
-Version: 0.10.1rc660
+Version: 0.10.1rc662
 Summary: Collection of tools to reconcile services with their desired state as defined in the app-interface DB.
 Home-page: https://github.com/app-sre/qontract-reconcile
 Author: Red Hat App-SRE Team

{qontract_reconcile-0.10.1rc660.dist-info → qontract_reconcile-0.10.1rc662.dist-info}/RECORD

@@ -616,7 +616,7 @@ reconcile/utils/state.py,sha256=zjsprjbOb0WddzmAvh8ACqAt0fcayrX2YPfz7qceRWw,1609
 reconcile/utils/structs.py,sha256=LcbLEg8WxfRqM6nW7NhcWN0YeqF7SQzxOgntmLs1SgY,352
 reconcile/utils/template.py,sha256=wTvRU4AnAV_o042tD4Mwls2dwWMuk7MKnde3MaCjaYg,331
 reconcile/utils/terraform_client.py,sha256=7L55Rvxfzj3KtJH8AD8D8YRfBnFpHiTFqSa5e2_9jtk,32092
-reconcile/utils/terrascript_aws_client.py,sha256=iM5ouzJh8qE7ixsbpo_JIaeJdLau7yBFu6AFeTqFK3M,271964
+reconcile/utils/terrascript_aws_client.py,sha256=tX-dBE8dmvtvZWDGJ0JjTRNx11MIOIhwZJIueOaEYjA,272396
 reconcile/utils/three_way_diff_strategy.py,sha256=nyqeQsLCoPI6e16k2CF3b9KNgQLU-rPf5RtfdUfVMwE,4468
 reconcile/utils/throughput.py,sha256=iP4UWAe2LVhDo69mPPmgo9nQ7RxHD6_GS8MZe-aSiuM,344
 reconcile/utils/unleash.py,sha256=1D56CsZfE3ShDtN3IErE1T2eeIwNmxhK-yYbCotJ99E,3601
@@ -634,6 +634,9 @@ reconcile/utils/aws_api_typed/organization.py,sha256=dJ7J02BNHu7UDyFa9083b92vSam
 reconcile/utils/aws_api_typed/sts.py,sha256=5Sauncj9Fif3YDLkJYkBZrtOX0v0bGAqOmY0A5Bh9yA,1237
 reconcile/utils/cloud_resource_best_practice/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 reconcile/utils/cloud_resource_best_practice/aws_rds.py,sha256=EvE6XKLsrZ531MJptKqPht2lOETrOjySTHXk6CzMgo0,2279
+reconcile/utils/clusterhealth/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+reconcile/utils/clusterhealth/providerbase.py,sha256=6YmFCw2fbUqgte5wagJsmcM9_gJQabj0dTXKBO0RNJo,1171
+reconcile/utils/clusterhealth/telemeter.py,sha256=a_RDNQMdjELltdeSFAsheK25EEu9ElX48ksLhOOOsvI,1448
 reconcile/utils/glitchtip/__init__.py,sha256=FT6iBhGqoe7KExFdbgL8AYUb64iW_4snF5__Dcl7yt0,258
 reconcile/utils/glitchtip/client.py,sha256=uqRK4XL5V2_bQpRyM3XTX8JKCaMpSOCkcm7jmuqYvwU,10584
 reconcile/utils/glitchtip/models.py,sha256=_oqZXNkyRTsAnx6tF4WUURSBj0cc9UNS4okOQYfAfB4,6435
@@ -720,8 +723,8 @@ tools/test/test_app_interface_metrics_exporter.py,sha256=SX7qL3D1SIRKFo95FoQztvf
 tools/test/test_qontract_cli.py,sha256=OvalpVRfY4pNmpMaWHHYqBjV68b1eGQjX8SCyTAXb1w,3501
 tools/test/test_sd_app_sre_alert_report.py,sha256=v363r9zM7__0kR5K6mvJoGFcM9BvE33fWAayrqkpojA,2116
 tools/test/test_sre_checkpoints.py,sha256=SKqPPTl9ua0RFdSSofnoQX-JZE6dFLO3LRhfQzqtfh8,2607
-qontract_reconcile-0.10.1rc660.dist-info/METADATA,sha256=dtN7pUdCsIWJSJNGAGjZ5gLutIcmOFCi5Y1mhdeJHsY,2382
-qontract_reconcile-0.10.1rc660.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
-qontract_reconcile-0.10.1rc660.dist-info/entry_points.txt,sha256=rIxI5zWtHNlfpDeq1a7pZXAPoqf7HG32KMTN3MeWK_8,429
-qontract_reconcile-0.10.1rc660.dist-info/top_level.txt,sha256=l5ISPoXzt0SdR4jVdkfa7RPSKNc8zAHYWAnR-Dw8Ey8,24
-qontract_reconcile-0.10.1rc660.dist-info/RECORD,,
+qontract_reconcile-0.10.1rc662.dist-info/METADATA,sha256=xAYklbK2n0Hx1Z9xs0KoH9Y1zbE69OON9UxpfPO0_CU,2382
+qontract_reconcile-0.10.1rc662.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+qontract_reconcile-0.10.1rc662.dist-info/entry_points.txt,sha256=rIxI5zWtHNlfpDeq1a7pZXAPoqf7HG32KMTN3MeWK_8,429
+qontract_reconcile-0.10.1rc662.dist-info/top_level.txt,sha256=l5ISPoXzt0SdR4jVdkfa7RPSKNc8zAHYWAnR-Dw8Ey8,24
+qontract_reconcile-0.10.1rc662.dist-info/RECORD,,

reconcile/utils/clusterhealth/providerbase.py

@@ -0,0 +1,40 @@
+from abc import ABC, abstractmethod
+from typing import Optional
+
+from pydantic import BaseModel
+
+
+class ClusterHealth(BaseModel):
+    source: str
+    errors: Optional[set[str]] = None
+
+    def has_health_errors(self) -> bool:
+        return bool(self.errors)
+
+
+def build_assumed_cluster_health() -> ClusterHealth:
+    return ClusterHealth(source="assumption")
+
+
+class ClusterHealthProvider(ABC):
+    """
+    A base class for cluster health providers.
+    """
+
+    @abstractmethod
+    def cluster_health(self, cluster_external_id: str, org_id: str) -> ClusterHealth:
+        """
+        Provides health information for an individual cluster in an organization
+        """
+
+
+class EmptyClusterHealthProvider(ClusterHealthProvider):
+    """
+    A default implementation of a cluster health provider that returns no health
+    information. Not every environment has a healthprovider available and this
+    implementation helps avoiding to check for the existance of a health provider
+    in various code places for such environments.
+    """
+
+    def cluster_health(self, cluster_external_id: str, org_id: str) -> ClusterHealth:
+        return build_assumed_cluster_health()

reconcile/utils/clusterhealth/telemeter.py

@@ -0,0 +1,37 @@
+from functools import lru_cache
+
+from reconcile.utils.clusterhealth.providerbase import (
+    ClusterHealth,
+    ClusterHealthProvider,
+)
+from reconcile.utils.grouping import group_by
+from reconcile.utils.prometheus import PrometheusQuerier
+
+TELEMETER_SOURCE = "telemeter"
+
+
+class TelemeterClusterHealthProvider(ClusterHealthProvider):
+    def __init__(self, querier: PrometheusQuerier):
+        self.querier = querier
+
+    @lru_cache(maxsize=None)
+    def cluster_health_for_org(self, org_id: str) -> dict[str, ClusterHealth]:
+        vectors_by_cluster = group_by(
+            self.querier.instant_vector_query(telemeter_alert_query(org_id)),
+            lambda v: v.mandatory_label("_id"),
+        )
+        return {
+            cluster_id: ClusterHealth(
+                errors={alert.mandatory_label("alertname") for alert in vectors},
+                source=TELEMETER_SOURCE,
+            )
+            for cluster_id, vectors in vectors_by_cluster.items()
+        }
+
+    def cluster_health(self, cluster_external_id: str, org_id: str) -> ClusterHealth:
+        health = self.cluster_health_for_org(org_id=org_id).get(cluster_external_id)
+        return health or ClusterHealth(source=TELEMETER_SOURCE)
+
+
+def telemeter_alert_query(organization_id: str) -> str:
+    return f'alerts{{alertstate="firing", severity="critical"}} * on (_id) group_left(organization) max(ocm_subscription{{organization="{organization_id}"}}) by (_id, organization)'

reconcile/utils/terrascript_aws_client.py

@@ -4071,6 +4071,15 @@ class TerrascriptClient:  # pylint: disable=too-many-public-methods
                     for k, v in cluster_mode.items():
                         values[k] = v
                 values.pop("availability_zones", None)
+            elif spec.provider == "msk":
+                if ebs_volume_size := values.get("broker_node_group_info", {}).pop(
+                    "ebs_volume_size", None
+                ):
+                    values["broker_node_group_info"].setdefault(
+                        "storage_info", {}
+                    ).setdefault("ebs_storage_info", {})[
+                        "volume_size"
+                    ] = ebs_volume_size
 
         return values