snapshot: add namespace pod rollup

ariadne/services/cluster_state.py

@@ -363,6 +363,40 @@ def _summarize_workloads(payload: dict[str, Any]) -> list[dict[str, Any]]:
     return output
 
 
+def _summarize_namespace_pods(payload: dict[str, Any]) -> list[dict[str, Any]]:
+    namespaces: dict[str, dict[str, Any]] = {}
+    for pod in _items(payload):
+        metadata = pod.get("metadata") if isinstance(pod.get("metadata"), dict) else {}
+        status = pod.get("status") if isinstance(pod.get("status"), dict) else {}
+        namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
+        if not _namespace_allowed(namespace):
+            continue
+        phase = status.get("phase") if isinstance(status.get("phase"), str) else ""
+        entry = namespaces.setdefault(
+            namespace,
+            {
+                "namespace": namespace,
+                "pods_total": 0,
+                "pods_running": 0,
+                "pods_pending": 0,
+                "pods_failed": 0,
+                "pods_succeeded": 0,
+            },
+        )
+        entry["pods_total"] += 1
+        if phase == "Running":
+            entry["pods_running"] += 1
+        elif phase == "Pending":
+            entry["pods_pending"] += 1
+        elif phase == "Failed":
+            entry["pods_failed"] += 1
+        elif phase == "Succeeded":
+            entry["pods_succeeded"] += 1
+    output = list(namespaces.values())
+    output.sort(key=lambda item: (-item.get("pods_total", 0), item.get("namespace") or ""))
+    return output
+
+
 def _vm_query(expr: str) -> list[dict[str, Any]] | None:
     base = settings.vm_url
     if not base:
@@ -562,9 +596,11 @@ def collect_cluster_state() -> tuple[dict[str, Any], ClusterStateSummary]:
         errors.append(f"flux: {exc}")
 
     workloads: list[dict[str, Any]] = []
+    namespace_pods: list[dict[str, Any]] = []
     try:
         pods_payload = get_json("/api/v1/pods?limit=5000")
         workloads = _summarize_workloads(pods_payload)
+        namespace_pods = _summarize_namespace_pods(pods_payload)
     except Exception as exc:
         errors.append(f"pods: {exc}")
 
@@ -577,6 +613,7 @@ def collect_cluster_state() -> tuple[dict[str, Any], ClusterStateSummary]:
         "nodes_detail": node_details,
         "flux": kustomizations or {},
         "workloads": workloads,
+        "namespace_pods": namespace_pods,
         "metrics": metrics,
         "errors": errors,
     }

tests/test_cluster_state.py

@@ -80,6 +80,8 @@ def test_collect_cluster_state(monkeypatch) -> None:
     assert snapshot["nodes_summary"]["ready"] == 1
     assert snapshot["nodes_detail"]
     assert snapshot["workloads"]
+    assert snapshot["namespace_pods"]
+    assert snapshot["namespace_pods"][0]["namespace"] == "media"
     assert summary.nodes_total == 2
     assert summary.nodes_ready == 1
     assert summary.pods_running == 5.0