cluster: summarize node load by hardware

ariadne/services/cluster_state.py

@@ -66,6 +66,53 @@ _LOAD_TOP_COUNT = 5
 _NAMESPACE_TOP_COUNT = 5
 
 
+def _node_usage_by_hardware(node_load: list[dict[str, Any]], node_details: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    if not node_load or not node_details:
+        return []
+    hardware_by_node = _hardware_map(node_details)
+    buckets: dict[str, dict[str, list[float]]] = {}
+    for entry in node_load:
+        if not isinstance(entry, dict):
+            continue
+        node = entry.get("node")
+        if not isinstance(node, str) or not node:
+            continue
+        hardware = hardware_by_node.get(node, "unknown")
+        _append_hardware_usage(buckets, str(hardware), entry)
+    return _finalize_hardware_usage(buckets)
+
+
+def _hardware_map(node_details: list[dict[str, Any]]) -> dict[str, str]:
+    mapping: dict[str, str] = {}
+    for node in node_details:
+        if not isinstance(node, dict):
+            continue
+        name = node.get("name")
+        if isinstance(name, str) and name:
+            mapping[name] = str(node.get("hardware") or "unknown")
+    return mapping
+
+
+def _append_hardware_usage(buckets: dict[str, dict[str, list[float]]], hardware: str, entry: dict[str, Any]) -> None:
+    bucket = buckets.setdefault(hardware, {"load_index": [], "cpu": [], "ram": [], "net": [], "io": []})
+    for key in ("load_index", "cpu", "ram", "net", "io"):
+        value = entry.get(key)
+        if isinstance(value, (int, float)):
+            bucket[key].append(float(value))
+
+
+def _finalize_hardware_usage(buckets: dict[str, dict[str, list[float]]]) -> list[dict[str, Any]]:
+    output: list[dict[str, Any]] = []
+    for hardware, metrics in buckets.items():
+        row: dict[str, Any] = {"hardware": hardware}
+        for key, values in metrics.items():
+            if values:
+                row[key] = sum(values) / len(values)
+        output.append(row)
+    output.sort(key=lambda item: (-(item.get("load_index") or 0), item.get("hardware") or ""))
+    return output
+
+
 @dataclass(frozen=True)
 class ClusterStateSummary:
     nodes_total: int | None
@@ -1771,6 +1818,7 @@ def collect_cluster_state() -> tuple[dict[str, Any], ClusterStateSummary]:
         node_pods,
     )
     metrics["node_load_summary"] = _node_load_summary(metrics.get("node_load", []))
+    metrics["node_load_by_hardware"] = _node_usage_by_hardware(metrics.get("node_load", []), node_details)
 
     snapshot = {
         "collected_at": collected_at.isoformat(),