Raise error if any on-watch service is not running

slapos/grid/SlapObject.py

@@ -892,50 +892,21 @@ class Partition(object):
     service_list = []
     if os.path.exists(self.service_path):
       if os.path.isdir(self.service_path):
-        service_list = sorted(os.listdir(self.service_path))
-    """Asks supervisord to check if one of the processes are in the state_list."""
-    service_list = []
-    print(os.listdir(self.service_path))
-    state_list = ['RUNNING',]
-    try: 
-      output = subprocess.check_output(['slapos', 'node', 'status'])
-    except Exception as e:
-      output = str(e.output)
-    content = output.split('\\n')
-    for line in content:
-      # print(line)
-      tokens = line.split(' ')
-      # print(service_name)
-      if 'on-watch' in tokens[0]:
-        service_name = tokens[0].split(':')[1]
-        # print(service_name)
-        service_list.append(service_name)
-    bad_service_list = []
+        for service in os.listdir(self.service_path):
+          service_list.append(service + '-on-watch')
+
     print(service_list)
-    for process in service_list:
-      with self.getSupervisorRPC() as supervisor:
-        print("Fuck!!!")
-        # info = supervisor.getProcessInfo(process_name)
-        print(info)
-      if info['statename'] in state_list:
-        continue
-      # try:
-      #   with self.getSupervisorRPC() as supervisor:
-      #     print("Fuck!!!")
-      #     info = supervisor.getProcessInfo(process_name)
-      #     print(info)
-      #   if info['statename'] in state_list:
-      #     continue
-      # except xmlrpclib.Fault as exc:
-      #   print(exc)
-    #     if exc.faultString.startswith('BAD_NAME:'):
-    #       self.logger.debug("BAD process name: %r" % process_name)
-    #       print("Service name is incorrect: %r" % process_name)
-    #       continue
-    #     else:
-    #       print("Service %r is not running." % process_name)
-    # else:
-    #   print("All on-watch services are running.")
+
+    bad_service_list = []
+    with self.getSupervisorRPC() as supervisor:
+      all_process = supervisor.getAllProcessInfo()
+      print(all_process)
+      for process in all_process:
+        if 'on-watch' in process['name'] and process['statename'] != "RUNNING":
+          self.logger.info('On watch service %r is not running' % process['name'])
+          bad_service_list.append(process['name'])
+    if len(bad_service_list) > 0:
+      raise ChildProcessError("Some on-watch process is not running!")
     
   def cleanupFolder(self, folder_path):
     """Delete all files and folders in a specified directory

slapos/grid/slapgrid.py

@@ -1292,9 +1292,6 @@ stderr_logfile_backups=1
         if not self.force_stop:
           self._checkPromiseList(local_partition)
           computer_partition.started()
-          print("******************************")
-          print("******************************")
-          print("******************************")
           local_partition.checkOnWatchServiceStatus()
         self._endInstallationTransaction(computer_partition)
       elif computer_partition_state == COMPUTER_PARTITION_STOPPED_STATE: