grid/slapgrid: reduce cases when slapgrid switch to offline mode

slapos/grid/slapgrid.py

@@ -335,6 +335,12 @@ def check_required_only_partitions(existing, required):
   return required
 
 
+class SlapgridException(Exception):
+  pass
+
+class NotConnectedToSlapOSMaster(SlapgridException):
+  pass
+
 class Slapgrid(object):
   """ Main class for SlapGrid. Fetches and processes informations from master
   server and pushes usage information to master server.
@@ -1422,12 +1428,10 @@ stderr_logfile_backups=1
   def processComputerPartitionList(self):
     try:
       return self.processComputerPartitionListOnline()
-    except (RequestException, ConnectionError):
-      # XXX Romain Why catching all errors?
+    except NotConnectedToSlapOSMaster:
       # We should only check getting global state of comp
       # and going offline in this case
-      raise
-      # return self.processComputerPartitionListOffline()
+      return self.processComputerPartitionListOffline()
 
   def processComputerPartitionListOnline(self):
     """
@@ -1443,7 +1447,15 @@ stderr_logfile_backups=1
     # Boolean to know if every promises correctly passed
     clean_run_promise = True
 
-    computer_partition_list = self.getRequiredComputerPartitionList()
+    try:
+      computer_partition_list = self.getRequiredComputerPartitionList()
+    except (RequestException, ConnectionError) as exc:
+      # If connection to the slapos master fails
+      # while retrieving compute node state,
+      # switch to the offline mode
+      # Log the exception to understand why it switched to offline mode
+      self.logger.error(exc)
+      raise NotConnectedToSlapOSMaster
 
     process_error_partition_list = []
     promise_error_partition_list = []