增加数据库异常重启处理

ClusterMonitor.py

@@ -9,7 +9,7 @@ import urllib.request
 import pymysql
 from pymysql import OperationalError
 
-formatter = logging.Formatter('%(asctime)s|%(levelname)s|%(message)s')
+formatter = logging.Formatter('%(asctime)s|%(levelname)s|%(lineno)d|%(message)s')
 logger = logging.getLogger(__name__)
 logger.setLevel(logging.DEBUG)
 file_handler = logging.FileHandler("debug.log")
@@ -32,6 +32,10 @@ SLOW_CONFIG = {
 }
 
 
+class DatabaseConnectError(Exception):
+    pass
+
+
 def get_db_conf():
     return {
         'host': "127.0.0.1",  # 只能控制本地数据库
@@ -39,6 +43,7 @@ def get_db_conf():
         'user': "root",
         'password': "Yanei!23",
         'db': "rms_ge_prod",
+        'connect_timeout': 5
     }
 
 
@@ -96,17 +101,19 @@ class Server:
         if self.db_conf is None:
             logger.error(f"db_config error: {self.db_conf}")
             raise ValueError("Database URL is None")
-        self.conn = self.get_connection()
+        self.conn = None
 
     def get_connection(self):
-        while True:
+        for i in range(10):
             try:
                 return pymysql.connect(**self.db_conf)
             except Exception as e:
                 logger.exception(e)
-                time.sleep(5)
+                time.sleep(3)
                 continue
 
+        raise DatabaseConnectError("Connection error")
+
     def get_cluster_status(self):
         while True:
             try:
@@ -117,12 +124,8 @@ class Server:
             except OperationalError as e:
                 logger.error(e)
 
-            try:
+            self.conn.close()
-                self.conn.close()
+            self.conn = self.get_connection()
-                self.conn = self.get_connection()
-            except Exception as e:
-                logger.exception(e)
-                time.sleep(10)
 
     def start_standalone_mode(self):
         logger.info("Start standalone mode...")
@@ -314,6 +317,8 @@ class Server:
                 cur.execute(f"use {get_db_conf()['db']}")
                 cur.execute(f"show tables")
                 logger.info("database is normal")
+            except DatabaseConnectError as e:
+                raise e
             except Exception as e:
                 logger.error(e)
                 time.sleep(1)
@@ -351,15 +356,43 @@ class Server:
                 logger.info("end cluster_run...")
                 return
 
+    def to_standalone_run(self):
+        """切换为单机模式"""
+        # 当前就是单机模式，只能重启尝试
+        if os.path.exists(f"{MYSQL_CONFIG}.cluster_bak"):
+            logger.info("is standalone and try restart")
+            os.system("kill -9 $(ps aux | grep mysqld | awk '{print $2}')")
+            logger.info("kill -9 mysql")
+            time.sleep(3)
+        # 当前是集群模式，切换为单机模式
+        else:
+            logger.info("is cluster and cluster to standalone")
+            self.start_standalone_mode()
+
     def run(self):
         while True:
-            now_status = self.get_cluster_status()
+            try:
-            if int(now_status.get("wsrep_cluster_size") or 0) != 0:
+                if not self.conn:
-                self.cluster_run()
+                    self.conn = self.get_connection()
-            else:
-                self.standalone_run()
 
-            time.sleep(5)
+                now_status = self.get_cluster_status()
+                if int(now_status.get("wsrep_cluster_size") or 0) != 0:
+                    self.cluster_run()
+                else:
+                    self.standalone_run()
+
+                time.sleep(5)
+            except DatabaseConnectError as e:
+                # 数据库异常，切换为单机模式
+                logger.error(e)
+                logger.info("database connect error, to standalone")
+                self.to_standalone_run()
+                time.sleep(5)
+
+                # 重置连接
+                if self.conn:
+                    self.conn.close()
+                self.conn = None
 
 
 if __name__ == '__main__':