From 5ed396e390130faf60fb4bdbf8114365508eb558 Mon Sep 17 00:00:00 2001
From: Your Name <you@example.com>
Date: Wed, 6 May 2026 00:19:35 +0800
Subject: [PATCH] fix(decision): derive telegram dedup from incident signals

---
 apps/api/src/services/decision_manager.py     | 18 ++++++++--
 .../test_decision_manager_telegram_dedup.py   | 35 +++++++++++++++++++
 docs/LOGBOOK.md                               |  9 +++++
 3 files changed, 60 insertions(+), 2 deletions(-)
 create mode 100644 apps/api/tests/test_decision_manager_telegram_dedup.py

diff --git a/apps/api/src/services/decision_manager.py b/apps/api/src/services/decision_manager.py
index 68e9b48b..c3a83588 100644
--- a/apps/api/src/services/decision_manager.py
+++ b/apps/api/src/services/decision_manager.py
@@ -55,6 +55,20 @@ def _fire_and_forget(coro) -> asyncio.Task:
     return task
 
 
+def _incident_alertname_for_dedup(incident: Incident) -> str:
+    """Return a stable alert name for Telegram fingerprint dedup."""
+    if incident.signals:
+        signal = incident.signals[0]
+        return (
+            signal.labels.get("alertname")
+            or signal.alert_name
+            or signal.annotations.get("summary")
+            or signal.annotations.get("description")
+            or incident.incident_id
+        )
+    return incident.incident_id
+
+
 def _phase2_fallback_reason(package: Any) -> str | None:
     """Return why a Phase 2 package should continue to Playbook/LLM fallback.
 
@@ -212,7 +226,7 @@ async def _push_decision_to_telegram(
         # 改成 alertname+target 構造的 fingerprint key + TTL 86400s，同症狀共用 dedup。
         # Incident 真正 RESOLVED/CLOSED 時走 line 220-226 的 status check 提早 return，不影響復發偵測。
         redis = get_redis()
-        _alertname_fp = (incident.title or "unknown").strip().lower().replace(" ", "_")[:60]
+        _alertname_fp = _incident_alertname_for_dedup(incident).strip().lower().replace(" ", "_")[:60]
         _target_fp = (
             incident.affected_services[0] if incident.affected_services else "unknown"
         ).lower()[:40]
@@ -3236,7 +3250,7 @@ class DecisionManager:
                         # 與 line 217-218 同邏輯，避免 pod restart resend 路徑繞過 fingerprint dedup。
                         # 原本 telegram_sent:{incident_id} TTL 600s 早就過期 → 重啟必重發；
                         # 改 fingerprint + 24h TTL → 同症狀 24h 內任何 INC ID 都不會重推。
-                        _alertname_fp = (getattr(incident, "title", None) or "unknown").strip().lower().replace(" ", "_")[:60]
+                        _alertname_fp = _incident_alertname_for_dedup(incident).strip().lower().replace(" ", "_")[:60]
                         _affected = getattr(incident, "affected_services", None) or []
                         _target_fp = (_affected[0] if _affected else "unknown").lower()[:40]
                         dedup_key = f"telegram_sent:fp:{_alertname_fp}:{_target_fp}"
diff --git a/apps/api/tests/test_decision_manager_telegram_dedup.py b/apps/api/tests/test_decision_manager_telegram_dedup.py
new file mode 100644
index 00000000..84557021
--- /dev/null
+++ b/apps/api/tests/test_decision_manager_telegram_dedup.py
@@ -0,0 +1,35 @@
+from datetime import UTC, datetime
+
+from src.models.incident import Incident, Severity, Signal
+from src.services.decision_manager import _incident_alertname_for_dedup
+
+
+def test_telegram_dedup_alertname_uses_signal_not_missing_title() -> None:
+    incident = Incident(
+        incident_id="INC-TEST",
+        severity=Severity.P3,
+        signals=[
+            Signal(
+                alert_name="FallbackAlertName",
+                severity=Severity.P3,
+                source="alertmanager",
+                fired_at=datetime.now(UTC),
+                labels={"alertname": "DockerContainerMemoryLimitPressure"},
+            )
+        ],
+        affected_services=["node-exporter-110"],
+    )
+
+    assert not hasattr(incident, "title")
+    assert _incident_alertname_for_dedup(incident) == "DockerContainerMemoryLimitPressure"
+
+
+def test_telegram_dedup_alertname_falls_back_to_incident_id() -> None:
+    incident = Incident(
+        incident_id="INC-NO-SIGNAL",
+        severity=Severity.P3,
+        signals=[],
+        affected_services=[],
+    )
+
+    assert _incident_alertname_for_dedup(incident) == "INC-NO-SIGNAL"
diff --git a/docs/LOGBOOK.md b/docs/LOGBOOK.md
index 0cdbc4c7..663d27b1 100644
--- a/docs/LOGBOOK.md
+++ b/docs/LOGBOOK.md
@@ -6,6 +6,15 @@
 
 ---
 
+## 2026-05-06 | Decision Telegram dedup no longer reads missing Incident.title
+
+**背景**：新 Ollama-first 部署後，production log 顯示 alert diagnosis 已走 `ollama_gcp_a -> ollama_gcp_b -> ollama_local -> gemini` 且 `phase24_ai_router_used` provider=`ollama`，但 DecisionManager 推送 Telegram decision card 時出現 `telegram_decision_push_failed: 'Incident' object has no attribute 'title'`。
+
+**本次修補**：
+- 新增 `_incident_alertname_for_dedup()`，Telegram fingerprint dedup 改從 `signal.labels.alertname -> signal.alert_name -> signal.annotations -> incident_id` 取值。
+- `_push_decision_to_telegram()` 與 stale READY token resend 共用同一個 dedup helper，避免兩條路徑再次漂移。
+- 補 `test_decision_manager_telegram_dedup.py`，鎖住 `Incident` 無 `title` 欄位時仍能產出 alertname fingerprint。
+
 ## 2026-05-05 | Alert diagnosis prioritizes resolution over speed
 
 **背景**：統帥明確修正策略：告警不是為了快速發卡片，而是為了把問題想清楚並完成 AI 自動化解決；GCP-A/GCP-B 有 SSD，可承擔深度診斷等待時間，Gemini 只能作 GCP-A → GCP-B → 111 全失敗後的備援。