ariadne/tests/test_cluster_state.py

from __future__ import annotations

from datetime import datetime, timezone

import ariadne.services.cluster_state as cluster_state


class DummyStorage:
    def __init__(self) -> None:
        self.snapshot = None
        self.keep = None

    def record_cluster_state(self, snapshot):  # type: ignore[no-untyped-def]
        self.snapshot = snapshot

    def prune_cluster_state(self, keep: int) -> None:
        self.keep = keep


def test_collect_cluster_state(monkeypatch) -> None:
    def fake_get_json(path: str):
        if path.endswith("/nodes"):
            return {
                "items": [
                    {
                        "metadata": {"name": "node-a", "labels": {"kubernetes.io/arch": "arm64"}},
                        "status": {
                            "conditions": [{"type": "Ready", "status": "True"}],
                            "nodeInfo": {"architecture": "arm64"},
                            "addresses": [{"type": "InternalIP", "address": "10.0.0.1"}],
                        },
                    },
                    {
                        "metadata": {
                            "name": "node-b",
                            "labels": {"kubernetes.io/arch": "amd64"},
                            "creationTimestamp": "2026-01-01T00:00:00Z",
                        },
                        "spec": {
                            "taints": [
                                {"key": "node-role.kubernetes.io/control-plane", "effect": "NoSchedule"}
                            ]
                        },
                        "status": {
                            "conditions": [{"type": "Ready", "status": "False"}],
                            "nodeInfo": {"architecture": "amd64"},
                        },
                    },
                ]
            }
        if path.startswith("/api/v1/pods"):
            return {
                "items": [
                    {
                        "metadata": {
                            "name": "jellyfin-0",
                            "namespace": "media",
                            "labels": {"app": "jellyfin"},
                        },
                        "spec": {"nodeName": "node-a"},
                        "status": {"phase": "Running"},
                    }
                ]
            }
        if path.startswith("/api/v1/events"):
            return {"items": []}
        if path.startswith("/apis/apps/v1/deployments"):
            return {
                "items": [
                    {
                        "metadata": {"name": "api", "namespace": "apps"},
                        "spec": {"replicas": 2},
                        "status": {"readyReplicas": 1, "availableReplicas": 1, "updatedReplicas": 1},
                    }
                ]
            }
        if path.startswith("/apis/apps/v1/statefulsets"):
            return {
                "items": [
                    {
                        "metadata": {"name": "db", "namespace": "apps"},
                        "spec": {"replicas": 1},
                        "status": {"readyReplicas": 1, "currentReplicas": 1, "updatedReplicas": 1},
                    }
                ]
            }
        if path.startswith("/apis/apps/v1/daemonsets"):
            return {
                "items": [
                    {
                        "metadata": {"name": "agent", "namespace": "apps"},
                        "status": {"desiredNumberScheduled": 3, "numberReady": 3, "updatedNumberScheduled": 3},
                    }
                ]
            }
        return {
            "items": [
                {
                    "metadata": {"name": "apps", "namespace": "flux-system"},
                    "spec": {"suspend": False},
                    "status": {"conditions": [{"type": "Ready", "status": "True"}]},
                },
                {
                    "metadata": {"name": "broken", "namespace": "flux-system"},
                    "spec": {"suspend": False},
                    "status": {"conditions": [{"type": "Ready", "status": "False", "reason": "Fail"}]},
                },
            ]
        }

    monkeypatch.setattr(cluster_state, "get_json", fake_get_json)
    monkeypatch.setattr(cluster_state, "_vm_scalar", lambda _expr: 5.0)
    monkeypatch.setattr(cluster_state, "_vm_vector", lambda _expr: [])

    snapshot, summary = cluster_state.collect_cluster_state()
    assert snapshot["nodes"]["total"] == 2
    assert snapshot["nodes"]["ready"] == 1
    assert snapshot["flux"]["not_ready"] == 1
    assert snapshot["nodes_summary"]["total"] == 2
    assert snapshot["nodes_summary"]["ready"] == 1
    assert "pressure_nodes" in snapshot["nodes_summary"]
    assert snapshot["nodes_detail"]
    assert snapshot["nodes_detail"][1]["age_hours"] is not None
    assert snapshot["nodes_detail"][1]["taints"]
    assert snapshot["workloads"]
    assert snapshot["namespace_pods"]
    assert snapshot["namespace_pods"][0]["namespace"] == "media"
    assert snapshot["namespace_nodes"]
    assert snapshot["node_pods"]
    assert "pod_issues" in snapshot
    assert "workloads_health" in snapshot
    assert snapshot["workloads_health"]["deployments"]["total"] == 1
    assert snapshot["workloads_health"]["deployments"]["not_ready"] == 1
    assert snapshot["events"]["warnings_total"] == 0
    assert "node_usage_stats" in snapshot["metrics"]
    assert snapshot["metrics"]["namespace_cpu_top"] == []
    assert snapshot["metrics"]["namespace_mem_top"] == []
    assert snapshot["metrics"]["namespace_cpu_requests_top"] == []
    assert snapshot["metrics"]["namespace_mem_requests_top"] == []
    assert snapshot["metrics"]["namespace_net_top"] == []
    assert snapshot["metrics"]["namespace_io_top"] == []
    assert snapshot["metrics"]["pod_cpu_top"] == []
    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


def test_run_cluster_state_records(monkeypatch) -> None:
    dummy = DummyStorage()
    snapshot = {"collected_at": datetime.now(timezone.utc).isoformat()}
    summary = cluster_state.ClusterStateSummary(1, 1, 1.0, 0, 0)

    monkeypatch.setattr(cluster_state, "collect_cluster_state", lambda: (snapshot, summary))

    result = cluster_state.run_cluster_state(dummy)
    assert result == summary
    assert dummy.snapshot == snapshot
    assert dummy.keep == cluster_state.settings.cluster_state_keep