c-foreach-restart: add arguments for retries and delay

eevans · eevans · commit 371988827303 · 2016-09-23T13:19:23.000-05:00
diff --git a/c-foreach-restart b/c-foreach-restart
@@ -1,18 +1,29 @@
 #!/usr/bin/python
 
 
+import argparse
 import logging
+import sys
 
 from cassandra.tools import get_instances
 from cassandra.tools.config import LOG_LEVEL
 
 logging.basicConfig(level=LOG_LEVEL)
 
 
-def main():
+def main(retries, delay):
     for instance in get_instances():
-        instance.restart()
+        instance.restart(retries=retries, delay=delay)
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Cassandra instance restarter")
+    parser.add_argument("-r", "--retries", metavar="RETRIES", default=10, type=int,
+                        help="Maximum number of times to check if service is up.")
+    parser.add_argument("-d", "--delay", metavar="DELAY", default=6, type=int,
+                        help="Number seconds between connection attempts, in seconds.")
+    return parser.parse_args(sys.argv[1:])
 
 
 if __name__ == "__main__":
-    main()
+    args = parse_args()
+    main(args.retries, args.delay)
diff --git a/cassandra/tools/instances.py b/cassandra/tools/instances.py
@@ -45,7 +45,8 @@ def __get(key):
     """
     Restart this Cassandra instance.
     """
-    def restart(self):
+    def restart(self, retries=10, delay=6):
+        self.__log_debug("Restarting instance... (retries=%s, delay=%s)", retries, delay)
         self.__log_info("Disabling client ports...")
         self.nodetool.run("disablebinary")
         self.nodetool.run("disablethrift")
@@ -54,8 +55,7 @@ def restart(self):
         self.__log_info("Restarting service %s", self.service_name)
         call("systemctl", "restart", self.service_name)
         listening = False
-        # TODO: The number of retries and timeout used should be configurable
-        for i in range(0, 10):
+        for i in range(0, retries):
             logging.debug("Testing CQL port (attempt #%s)", (i + 1))
             if self.listening(self.rpc_address, self.native_transport_port):
                 self.__log_info("CQL (%s:%s) is UP", self.rpc_address, self.native_transport_port)
@@ -67,11 +67,14 @@ def restart(self):
                     self.rpc_address,
                     self.native_transport_port
                 )
-            sleep(6)
+            sleep(delay)
         if not listening:
             self.__log_error("CQL (%s:%s) DOWN", self.rpc_address, self.native_transport_port)
             raise Exception("{} restart FAILED".format(self.service_name))
 
+    def __log_debug(self, msg, *args, **kwargs):
+        self.__log(logging.DEBUG, msg, *args, **kwargs)
+
     def __log_info(self, msg, *args, **kwargs):
         self.__log(logging.INFO, msg, *args, **kwargs)
 
