加入告警去重與洞察向量回補

services/openclaw_learning_service.py

@@ -62,6 +62,41 @@ def enqueue_insight_embedding(insight_id: int, insight_type: str, content: str,
     return _enqueue_embedding("ai_insights", int(insight_id), embed_target_text)
 
 
+def enqueue_missing_insight_embeddings(limit: int = 200) -> dict:
+    """Backfill existing ai_insights that have not yet entered the embedding queue."""
+    session = get_session()
+    try:
+        rows = session.execute(
+            text("""
+                SELECT id, insight_type, period, content
+                  FROM ai_insights i
+                 WHERE i.embedding IS NULL
+                   AND COALESCE(i.status, 'approved') NOT IN ('archived', 'rejected')
+                   AND NOT EXISTS (
+                       SELECT 1
+                         FROM embedding_retry_queue q
+                        WHERE q.target_table = 'ai_insights'
+                          AND q.target_id = i.id
+                          AND q.status IN ('pending', 'processing', 'done')
+                   )
+                 ORDER BY i.created_at DESC
+                 LIMIT :lim
+            """),
+            {"lim": limit},
+        ).fetchall()
+    except Exception as exc:
+        sys_log.warning(f"[OCLearn] embedding backfill 略過 (schema/pgvector 未就緒?): {exc}")
+        return {"scanned": 0, "enqueued": 0, "status": "skipped", "error": str(exc)[:200]}
+    finally:
+        session.close()
+
+    enqueued = 0
+    for row in rows:
+        if enqueue_insight_embedding(row.id, row.insight_type, row.content, row.period):
+            enqueued += 1
+    return {"scanned": len(rows), "enqueued": enqueued, "status": "ok"}
+
+
 def _process_one_embedding(row_id: int, target_table: str, target_id: int,
                            text_content: str, model: str) -> bool:
     """處理單筆 embedding，成功寫回目標表，失敗累加 attempts"""