cluster_state: add delta and pod issue summaries

ariadne/services/cluster_state.py

@@ -103,6 +103,8 @@ _POD_WAITING_REASONS = {
     "err_image_pull": "ErrImagePull",
     "create_config_error": "CreateContainerConfigError",
 }
+_DELTA_TOP_LIMIT = 6
+_REASON_TOP_LIMIT = 5
 
 
 def _node_usage_by_hardware(node_load: list[dict[str, Any]], node_details: list[dict[str, Any]]) -> list[dict[str, Any]]:
@@ -2860,6 +2862,53 @@ def _delta_severity(delta: float) -> str:
     return "info"
 
 
+def _delta_entry_label(entry: dict[str, Any]) -> tuple[str, str]:
+    if "node" in entry:
+        return ("node", str(entry.get("node") or ""))
+    return ("namespace", str(entry.get("namespace") or ""))
+
+
+def _delta_top(entries: list[dict[str, Any]], key: str, limit: int = _DELTA_TOP_LIMIT) -> list[dict[str, Any]]:
+    output: list[dict[str, Any]] = []
+    for entry in entries:
+        if not isinstance(entry, dict):
+            continue
+        deltas = entry.get("baseline_delta") if isinstance(entry.get("baseline_delta"), dict) else {}
+        delta = deltas.get(key)
+        if not isinstance(delta, (int, float)):
+            continue
+        label_key, label_value = _delta_entry_label(entry)
+        output.append(
+            {
+                label_key: label_value,
+                "metric": key,
+                "delta": delta,
+                "severity": _delta_severity(float(delta)),
+            }
+        )
+    output.sort(key=lambda item: (-(abs(item.get("delta") or 0)), item.get("metric") or ""))
+    return output[:limit]
+
+
+def _reason_top(counts: dict[str, Any], limit: int = _REASON_TOP_LIMIT) -> list[dict[str, Any]]:
+    output: list[dict[str, Any]] = []
+    for reason, value in counts.items() if isinstance(counts, dict) else []:
+        if isinstance(reason, str) and reason and isinstance(value, (int, float)):
+            output.append({"reason": reason, "count": int(value)})
+    output.sort(key=lambda item: (-item.get("count", 0), item.get("reason") or ""))
+    return output[:limit]
+
+
+def _pod_issue_summary(pod_issues: dict[str, Any], metrics: dict[str, Any]) -> dict[str, Any]:
+    waiting = pod_issues.get("waiting_reasons") if isinstance(pod_issues, dict) else {}
+    phase = pod_issues.get("phase_reasons") if isinstance(pod_issues, dict) else {}
+    return {
+        "waiting_reasons_top": _reason_top(waiting),
+        "phase_reasons_top": _reason_top(phase),
+        "namespace_issue_top": metrics.get("namespace_issue_top") or {},
+    }
+
+
 def _delta_hit(delta: Any) -> bool:
     if not isinstance(delta, (int, float)):
         return False
@@ -3546,9 +3595,13 @@ def collect_cluster_state() -> tuple[dict[str, Any], ClusterStateSummary]:
         },
         "pressure_summary": pressure_summary,
         "trend_summary": metrics.get("trend_summary"),
+        "trend_requests": metrics.get("namespace_request_trends"),
         "time_series": metrics.get("cluster_trends"),
         "node_condition_trends": metrics.get("node_condition_trends"),
+        "pod_waiting_trends": metrics.get("pod_waiting_trends"),
+        "pod_terminated_trends": metrics.get("pod_terminated_trends"),
         "pod_reason_totals": metrics.get("pod_reason_totals"),
+        "pod_issue_summary": _pod_issue_summary(pod_issues, metrics),
         "offenders": _build_offenders(metrics),
         "alerts": metrics.get("alerts", {}),
         "top": {
@@ -3573,6 +3626,19 @@ def collect_cluster_state() -> tuple[dict[str, Any], ClusterStateSummary]:
             "workloads": _workload_index(workloads),
             "namespaces": _namespace_nodes_top(namespace_context, 12),
         },
+        "baseline_deltas": {
+            "nodes": {
+                "cpu": _delta_top(node_context, "cpu"),
+                "ram": _delta_top(node_context, "ram"),
+                "net": _delta_top(node_context, "net"),
+                "io": _delta_top(node_context, "io"),
+                "disk": _delta_top(node_context, "disk"),
+            },
+            "namespaces": {
+                "cpu": _delta_top(namespace_context, "cpu"),
+                "mem": _delta_top(namespace_context, "mem"),
+            },
+        },
         "attention_ranked": _build_attention_ranked(metrics, node_context, pod_issues, workload_health),
         "signals": signals,
         "profiles": profiles,