开发机迁移

apps/backend/app/database.py

@@ -31,6 +31,31 @@ from app.config import (
     ETL_DB_USER,
 )
 
+# TCP keepalive 参数：防止长期运行的后台服务连接被网络设备/PostgreSQL 回收
+_KEEPALIVE_KWARGS = {
+    "keepalives": 1,
+    "keepalives_idle": 60,       # 空闲 60 秒后开始探测
+    "keepalives_interval": 10,   # 每 10 秒探测一次
+    "keepalives_count": 3,       # 连续 3 次失败判定断开
+}
+
+# 连接重试参数：应对 PostgreSQL 瞬时不可用（Tailscale 网络抖动等）
+_CONNECT_MAX_RETRIES = 3
+_CONNECT_RETRY_DELAY = 1.0  # 秒
+
+
+def _connect_with_retry(connect_fn, max_retries=_CONNECT_MAX_RETRIES):
+    """带重试的数据库连接，应对服务端瞬时拒绝连接。"""
+    last_exc = None
+    for attempt in range(max_retries):
+        try:
+            return connect_fn()
+        except psycopg2.OperationalError as e:
+            last_exc = e
+            if attempt < max_retries - 1:
+                time.sleep(_CONNECT_RETRY_DELAY * (attempt + 1))
+    raise last_exc
+
 
 def get_connection() -> PgConnection:
     """
@@ -49,13 +74,14 @@ def get_connection() -> PgConnection:
 
     start = time.perf_counter() if should_trace else 0.0
 
-    conn = psycopg2.connect(
+    conn = _connect_with_retry(lambda: psycopg2.connect(
         host=DB_HOST,
         port=DB_PORT,
         user=DB_USER,
         password=DB_PASSWORD,
         dbname=APP_DB_NAME,
-    )
+        **_KEEPALIVE_KWARGS,
+    ))
 
     if should_trace:
         from datetime import datetime
@@ -86,13 +112,14 @@ def get_etl_global_readonly_connection() -> PgConnection:
 
     用于系统管理后台等不需要门店隔离的场景（如 ETL 状态监控）。
     """
-    conn = psycopg2.connect(
+    conn = _connect_with_retry(lambda: psycopg2.connect(
         host=ETL_DB_HOST,
         port=ETL_DB_PORT,
         user=ETL_DB_USER,
         password=ETL_DB_PASSWORD,
         dbname=ETL_DB_NAME,
-    )
+        **_KEEPALIVE_KWARGS,
+    ))
     try:
         conn.autocommit = False
         with conn.cursor() as cur:
@@ -121,13 +148,14 @@ def get_etl_readonly_connection(site_id: int | str) -> PgConnection:
         finally:
             conn.close()
     """
-    conn = psycopg2.connect(
+    conn = _connect_with_retry(lambda: psycopg2.connect(
         host=ETL_DB_HOST,
         port=ETL_DB_PORT,
         user=ETL_DB_USER,
         password=ETL_DB_PASSWORD,
         dbname=ETL_DB_NAME,
-    )
+        **_KEEPALIVE_KWARGS,
+    ))
     try:
         conn.autocommit = False
         with conn.cursor() as cur:
@@ -142,3 +170,23 @@ def get_etl_readonly_connection(site_id: int | str) -> PgConnection:
         conn.close()
         raise
     return conn
+
+
+def get_etl_write_connection() -> PgConnection:
+    """
+    获取 ETL 数据库（etl_feiqiu）的可写连接。
+
+    仅用于后端需要回写 ETL 汇总表的场景（如 task_generator 回写关系指数统计）。
+    不设置 RLS 隔离，调用方需在 SQL 中显式指定 site_id。
+    调用方负责关闭连接。
+    """
+    conn = _connect_with_retry(lambda: psycopg2.connect(
+        host=ETL_DB_HOST,
+        port=ETL_DB_PORT,
+        user=ETL_DB_USER,
+        password=ETL_DB_PASSWORD,
+        dbname=ETL_DB_NAME,
+        **_KEEPALIVE_KWARGS,
+    ))
+    conn.autocommit = False
+    return conn