Add disconnect on Nth retry to make it reresolving IP

salt/transport/zeromq.py

@@ -179,7 +179,7 @@ def __init__(self, opts, io_loop, **kwargs):
             self._socket.setsockopt(zmq.IPV4ONLY, 0)
 
         if HAS_ZMQ_MONITOR and self.opts["zmq_monitor"]:
-            self._monitor = ZeroMQSocketMonitor(self._socket)
+            self._monitor = ZeroMQSocketMonitor(self._socket, opts)
             self._monitor.start_io_loop(self.io_loop)
 
     def close(self):
@@ -641,7 +641,7 @@ def mark_future(msg):
 class ZeroMQSocketMonitor:
     __EVENT_MAP = None
 
-    def __init__(self, socket):
+    def __init__(self, socket, opts=None):
         """
         Create ZMQ monitor sockets
 
@@ -651,6 +651,11 @@ def __init__(self, socket):
         self._socket = socket
         self._monitor_socket = self._socket.get_monitor_socket()
         self._monitor_stream = None
+        self.disconnect_callback = None
+        self.disconnect_on_retry = None
+        self._connect_retry = None
+        if opts is not None:
+            self.disconnect_on_retry = opts.get("zmq_disconnect_on_retry", 10)
 
     def start_io_loop(self, io_loop):
         log.trace("Event monitor start!")
@@ -688,11 +693,20 @@ def monitor_callback(self, msg):
         if evt["event"] == zmq.EVENT_MONITOR_STOPPED:
             self.stop()
         elif evt["event"] == zmq.EVENT_DISCONNECTED:
+            if self.disconnect_callback is not None:
+                self.disconnect_callback()
+        elif evt["event"] == zmq.EVENT_CONNECT_RETRIED:
             if (
-                hasattr(self, "disconnect_callback")
+                self.disconnect_on_retry is not None
                 and self.disconnect_callback is not None
             ):
-                self.disconnect_callback()
+                if self._connect_retry is None:
+                    self._connect_retry = self.disconnect_on_retry
+                self._connect_retry -= 1
+                if self._connect_retry <= 0:
+                    log.debug("Calling disconnect callback as number of retries reached.")
+                    self.disconnect_callback()
+                    self._connect_retry = self.disconnect_on_retry
 
     def stop(self):
         if self._socket is None: