add ctrl+c as first try to kill daemons/TRex.

master daemon change directory to "/" to deal with case it's directory was deleted and update_trex() was used.
remove check connection to daemons as part of init of client (can be done manually per needed daemon)

4 files changed, 52 insertions, 76 deletions

diff --git a/scripts/automation/trex_control_plane/server/singleton_daemon.py b/scripts/automation/trex_control_plane/server/singleton_daemon.py
index 8fdedc6e..79deace1 100755
--- a/scripts/automation/trex_control_plane/server/singleton_daemon.py
+++ b/scripts/automation/trex_control_plane/server/singleton_daemon.py
@@ -2,6 +2,7 @@ import errno
 import os
 import shlex
 import socket
+import signal
 import tempfile
 import types
 from subprocess import Popen
@@ -73,31 +74,27 @@ class SingletonDaemon(object):
         if pid:
             return pid
 
-
-    # kill daemon
-    def kill(self, timeout = 10):
-        pid = self.get_pid()
-        if not pid:
-            return False
-        ret_code, stdout, stderr = run_command('kill %s' % pid) # usual kill
-        if ret_code:
-            raise Exception('Failed to run kill command for %s: %s' % (self.name, [ret_code, stdout, stderr]))
+    def kill_by_signal(self, pid, signal_name, timeout):
+        os.kill(pid, signal_name)
         poll_rate = 0.1
         for i in range(int(timeout / poll_rate)):
             if not self.is_running():
                 return True
             sleep(poll_rate)
-        ret_code, stdout, stderr = run_command('kill -9 %s' % pid) # unconditional kill
-        if ret_code:
-            raise Exception('Failed to run kill -9 command for %s: %s' % (self.name, [ret_code, stdout, stderr]))
-        for i in range(int(timeout / poll_rate)):
-            if not self.is_running():
+
+    # kill daemon, with verification
+    def kill(self, timeout = 15):
+        pid = self.get_pid()
+        if not pid:
+            return False
+        # try Ctrl+C, usual kill, kill -9
+        for signal_name in [signal.SIGINT, signal.SIGTERM, signal.SIGKILL]:
+            if self.kill_by_signal(pid, signal_name, timeout):
                 return True
-            sleep(poll_rate)
         raise Exception('Could not kill %s, even with -9' % self.name)
 
     # try connection as RPC client, return True upon success, False if fail
-    def check_connectivity(self, timeout = 5):
+    def check_connectivity(self, timeout = 15):
         daemon = jsonrpclib.Server('http://127.0.0.1:%s/' % self.port)
         poll_rate = 0.1
         for i in range(int(timeout/poll_rate)):
@@ -140,7 +137,7 @@ class SingletonDaemon(object):
             raise Exception('%s failed to run.' % self.name)
 
     # restart the daemon
-    def restart(self, timeout = 5):
+    def restart(self, timeout = 15):
         if self.is_running():
             self.kill(timeout)
         return self.start(timeout)
diff --git a/scripts/automation/trex_control_plane/server/trex_server.py b/scripts/automation/trex_control_plane/server/trex_server.py
index 5e278a74..3da629ec 100755
--- a/scripts/automation/trex_control_plane/server/trex_server.py
+++ b/scripts/automation/trex_control_plane/server/trex_server.py
@@ -387,23 +387,14 @@ class CTRexServer(object):
         return trex_cmds_list
 
 
-    def kill_all_trexes(self):
+    # Silently tries to kill TRexes with given signal.
+    # Responsibility of client to verify with get_trex_cmds.
+    def kill_all_trexes(self, signal_name):
         logger.info('Processing kill_all_trexes() command.')
         trex_cmds_list = self.get_trex_cmds()
-        if not trex_cmds_list:
-            return False
         for pid, cmd in trex_cmds_list:
-            logger.info('Killing process %s %s' % (pid, cmd))
-            run_command('kill %s' % pid)
-            ret_code_ps, _, _ = run_command('ps -p %s' % pid)
-            if not ret_code_ps:
-                logger.info('Killing with -9.')
-                run_command('kill -9 %s' % pid)
-                ret_code_ps, _, _ = run_command('ps -p %s' % pid)
-                if not ret_code_ps:
-                    logger.info('Could not kill process.')
-                    raise Exception('Could not kill process %s %s' % (pid, cmd))
-        return True
+            logger.info('Killing with signal %s process %s %s' % (signal_name, pid, cmd))
+            os.kill(int(pid), signal_name)
 
 
     def wait_until_kickoff_finish (self, timeout = 40):
diff --git a/scripts/automation/trex_control_plane/stf/trex_stf_lib/trex_client.py b/scripts/automation/trex_control_plane/stf/trex_stf_lib/trex_client.py
index 3b01560a..e1b298db 100755
--- a/scripts/automation/trex_control_plane/stf/trex_stf_lib/trex_client.py
+++ b/scripts/automation/trex_control_plane/stf/trex_stf_lib/trex_client.py
@@ -13,6 +13,7 @@ from distutils.util import strtobool
 from collections import deque, OrderedDict
 from json import JSONDecoder
 import traceback
+import signal
 
 try:
     from . import outer_packages
@@ -92,42 +93,9 @@ class CTRexClient(object):
         self.decoder            = JSONDecoder()
         self.history            = jsonrpclib.history.History()
         self.master_daemon_path = "http://{hostname}:{port}/".format( hostname = self.trex_host, port = master_daemon_port )
+        self.master_daemon      = jsonrpclib.Server(self.master_daemon_path, history = self.history)
         self.trex_server_path   = "http://{hostname}:{port}/".format( hostname = self.trex_host, port = trex_daemon_port )
-        self.connect_master()
-        self.connect_server()
-
-
-    def connect_master(self):
-        '''
-        Connects to Master daemon via JsonRPC.
-        This daemon controls TRex daemon server.
-        Return true if success, false if fail
-        '''
-        try:
-            print('Connecting to Master daemon @ %s ...' % self.master_daemon_path)
-            self.master_daemon      = jsonrpclib.Server(self.master_daemon_path, history = self.history)
-            self.check_master_connectivity()
-            print('Connected to Master daemon.')
-            return True
-        except Exception as e:
-            print(e)
-            return False
-
-    def connect_server(self):
-        '''
-        Connects to TRex daemon server via JsonRPC.
-        This daemon controls TRex. (start/stop)
-        Return true if success, false if fail
-        '''
-        try:
-            print('Connecting to TRex daemon server @ %s ...' % self.trex_server_path)
-            self.server             = jsonrpclib.Server(self.trex_server_path, history = self.history)
-            self.check_server_connectivity()
-            print('Connected TRex server daemon.')
-            return True
-        except Exception as e:
-            print(e)
-            return False
+        self.server             = jsonrpclib.Server(self.trex_server_path, history = self.history)
 
 
     def add (self, x, y):
@@ -322,18 +290,28 @@ class CTRexClient(object):
         finally:
             self.prompt_verbose_data()
 
-    def kill_all_trexes(self):
+    def kill_all_trexes(self, timeout = 15):
         """
         Kills running TRex processes (if exists) on the server, not only owned by current daemon.
         Raises exception upon error killing.
 
         :return: 
-            + **True** if any process killed 
+            + **True** if processes killed/not running
             + **False** otherwise.
 
         """
         try:
-            return self.server.kill_all_trexes()
+            poll_rate = 0.1
+            # try Ctrl+C, usual kill, -9
+            for signal_name in [signal.SIGINT, signal.SIGTERM, signal.SIGKILL]:
+                self.server.kill_all_trexes(signal_name)
+                for i in range(int(timeout / poll_rate)):
+                    if not self.get_trex_cmds():
+                        return True
+                    sleep(poll_rate)
+            if self.get_trex_cmds():
+                return False
+            return True
         except AppError as err:
             self._handle_AppError_exception(err.args[0])
         finally:
@@ -1518,8 +1496,15 @@ class CTRexResult(object):
         return result
 
 
-
-
 if __name__ == "__main__":
-    pass
+    c = CTRexClient('127.0.0.1')
+    print('restarting daemon')
+    c.restart_trex_daemon()
+    print('kill any running')
+    c.kill_all_trexes()
+    print('start')
+    c.start_stateless()
+    print('sleep')
+    time.sleep(5)
+    print('done')
 
diff --git a/scripts/master_daemon.py b/scripts/master_daemon.py
index 3f130bdb..e50d49ee 100755
--- a/scripts/master_daemon.py
+++ b/scripts/master_daemon.py
@@ -119,14 +119,16 @@ def start_master_daemon_func():
         server.register_function(stl_rpc_proxy.start, 'start_stl_rpc_proxy')
         server.register_function(stl_rpc_proxy.stop, 'stop_stl_rpc_proxy')
         server.register_function(server.funcs.keys, 'get_methods') # should be last
-        signal.signal(signal.SIGTSTP, stop_handler)
-        signal.signal(signal.SIGTERM, stop_handler)
+        signal.signal(signal.SIGTSTP, stop_handler) # ctrl+z
+        signal.signal(signal.SIGTERM, stop_handler) # kill
         server.serve_forever()
+    except KeyboardInterrupt:
+        logging.info('Ctrl+C')
     except Exception as e:
         logging.error('Closing due to error: %s' % e)
 
-def stop_handler(*args, **kwargs):
-    logging.info('Got killed explicitly.')
+def stop_handler(signalnum, *args, **kwargs):
+    logging.info('Got signal %s, exiting.' % signalnum)
     sys.exit(0)
 
 # returns True if given path is under current dir or /tmp
@@ -191,6 +193,7 @@ master_daemon      = SingletonDaemon('Master daemon', 'trex_master_daemon', args
 tmp_dir = '/tmp/trex-tmp'
 logging_file = '/var/log/trex/master_daemon.log'
 logging_file_bu = '/var/log/trex/master_daemon.log_bu'
+os.chdir('/')
 
 if not _check_path_under_current_or_temp(args.trex_dir):
     raise Exception('Only allowed to use path under /tmp or current directory')