cluster_state: add delta and pod issue summaries

ariadne/settings.py

@@ -189,6 +189,7 @@ class Settings:
     k8s_api_timeout_sec: float
     vm_url: str
     cluster_state_vm_timeout_sec: float
+    alertmanager_url: str
 
     mailu_sync_cron: str
     nextcloud_sync_cron: str
@@ -469,6 +470,7 @@ class Settings:
                 "http://victoria-metrics-single-server.monitoring.svc.cluster.local:8428",
             ).rstrip("/"),
             "cluster_state_vm_timeout_sec": _env_float("ARIADNE_CLUSTER_STATE_VM_TIMEOUT_SEC", 5.0),
+            "alertmanager_url": _env("ARIADNE_ALERTMANAGER_URL", "").rstrip("/"),
             "cluster_state_cron": _env("ARIADNE_SCHEDULE_CLUSTER_STATE", "*/15 * * * *"),
             "cluster_state_keep": _env_int("ARIADNE_CLUSTER_STATE_KEEP", 168),
         }

tests/test_cluster_state.py

@@ -143,6 +143,19 @@ def test_collect_cluster_state(monkeypatch) -> None:
     assert snapshot["metrics"]["pod_mem_top"] == []
     assert snapshot["metrics"]["job_failures_24h"] == []
     assert snapshot["metrics"]["pvc_usage_top"] == []
+    assert snapshot["summary"]["counts"]["nodes_total"] == 5.0
+    assert snapshot["summary"]["counts"]["nodes_ready"] == 5.0
+    assert snapshot["summary"]["counts"]["pods_running"] == 5.0
+    assert snapshot["summary"]["top"]["namespace_pods"][0]["namespace"] == "media"
+    assert snapshot["summary"]["baseline_window"]
+    assert "workload_not_ready" in snapshot["summary"]["top"]
+    assert "pod_restarts" in snapshot["summary"]["top"]
+    assert "attention_ranked" in snapshot["summary"]
+    assert snapshot["summary"]["health_bullets"]
+    assert snapshot["summary"]["unknowns"] == []
+    assert snapshot["context"]["nodes"]
+    assert snapshot["context"]["namespaces"]
+    assert "baseline" in snapshot["context"]["nodes"][0]
     assert summary.nodes_total == 2
     assert summary.nodes_ready == 1
     assert summary.pods_running == 5.0