fix(monitoring): classify host memory pressure accurately

scripts/cold_start_playbooks.py

@@ -104,18 +104,18 @@ PLAYBOOK_TEMPLATES = [
         "severity_range": ["P2", "P3"],
     },
     {
-        "name": "節點記憶體不足 — 清理 Cache",
-        "alert_type": "HostOutOfMemory",
-        "description": "主機記憶體不足，清理 Page Cache 釋放空間",
+        "name": "節點記憶體壓力 — 唯讀診斷",
+        "alert_type": "HostMemoryUsageHigh",
+        "description": "主機 MemAvailable 偏低；先驗證 OOM、swap、程序與服務狀態，不做破壞性快取清理",
         "category": "infrastructure",
         "repair_steps": [
-            {"command": "ssh {target} 'free -h'", "purpose": "確認記憶體狀況"},
-            {"command": "ssh {target} 'sudo sync && sudo sysctl vm.drop_caches=3'", "purpose": "清理 Page Cache"},
-            {"command": "ssh {target} 'free -h'", "purpose": "確認記憶體已釋放"},
+            {"command": "ssh {target} 'free -h'", "purpose": "讀取 MemAvailable 與 swap 狀況"},
+            {"command": "ssh {target} \"awk '$1 == \\\"oom_kill\\\" { print }' /proc/vmstat\"", "purpose": "確認 kernel OOM kill 累計值"},
+            {"command": "ssh {target} 'ps aux --sort=-%mem | head -20'", "purpose": "讀取主要記憶體程序"},
         ],
         "estimated_minutes": 5,
-        "symptom_alertnames": ["HostOutOfMemory", "NodeMemoryPressure"],
-        "severity_range": ["P1", "P2"],
+        "symptom_alertnames": ["HostMemoryUsageHigh", "HostOutOfMemory", "NodeMemoryPressure"],
+        "severity_range": ["P2", "P3"],
     },
     {
         "name": "磁碟使用率過高 — 清理舊日誌",

scripts/ops/tests/test_host_pressure_alert_contract.py

@@ -7,6 +7,8 @@ import yaml
 
 ROOT = Path(__file__).resolve().parents[3]
 ALERTS = ROOT / "ops" / "monitoring" / "alerts-unified.yml"
+LEGACY_ALERTS = ROOT / "ops" / "monitoring" / "alerts.yml"
+COLD_START_PLAYBOOKS = ROOT / "scripts" / "cold_start_playbooks.py"
 
 
 def load_alerts() -> dict[str, dict]:
@@ -54,6 +56,40 @@ def test_critical_sustained_load_alert_uses_deployed_controller_path() -> None:
     assert "scripts/ops/host-sustained-load-controller.py" not in action
 
 
+def test_host_memory_pressure_is_not_mislabeled_as_oom_or_host_188() -> None:
+    alerts = load_alerts()
+    rule = alerts["HostMemoryUsageHigh"]
+    annotations = rule["annotations"]
+
+    assert "HostOutOfMemory" not in alerts
+    assert rule["labels"]["layer"] == "systemd-{{ $labels.host }}"
+    assert rule["labels"]["auto_repair"] == "true"
+    assert "不代表已發生 OOM" in annotations["description"]
+    assert "oom_kill" in annotations["auto_repair_action"]
+    assert "drop_caches" not in annotations["auto_repair_action"]
+    assert "restart" not in annotations["auto_repair_action"]
+    assert "reboot" not in annotations["auto_repair_action"]
+    assert "禁止 drop_caches、restart、reboot" in annotations["runbook"]
+
+
+def test_legacy_rule_and_seed_playbook_use_same_read_only_memory_contract() -> None:
+    legacy_payload = yaml.safe_load(LEGACY_ALERTS.read_text(encoding="utf-8"))
+    legacy_alerts = {
+        rule["alert"]: rule
+        for group in legacy_payload["groups"]
+        for rule in group.get("rules", [])
+        if "alert" in rule
+    }
+    playbook_source = COLD_START_PLAYBOOKS.read_text(encoding="utf-8")
+
+    assert "HostMemoryUsageHigh" in legacy_alerts
+    assert "HostOutOfMemory" not in legacy_alerts
+    assert legacy_alerts["HostMemoryUsageHigh"]["labels"]["layer"] == "systemd-{{ $labels.host }}"
+    assert "sudo sysctl vm.drop_caches" not in playbook_source
+    assert '"alert_type": "HostMemoryUsageHigh"' in playbook_source
+    assert '"HostOutOfMemory"' in playbook_source  # historical persisted-event alias only
+
+
 def test_backup_aggregate_alert_excludes_old_wrapper_noise() -> None:
     alerts = load_alerts()
     expr = str(alerts["BackupAggregateRunFailed"]["expr"])