Outdate cell of any storage node lost to ensure the replication process to

started when it come back. This cause broadcast partition table changes at
storage connection closure and might be optimized in the future.


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@908 71dcc9de-d417-0410-9af5-da40c76e7ee4

neo/master/app.py

@@ -308,6 +308,12 @@ class Application(object):
                         size -= amt
                         start += amt
 
+    def outdateAndBroadcastPartition(self):
+        " Outdate cell of non-working nodes and broadcast changes """
+        cell_list = self.pt.outdate()
+        if cell_list:
+            self.broadcastPartitionChanges(self.pt.setNextID(), cell_list)
+
     def sendPartitionTable(self, conn):
         """ Send the partition table through the given connection """
         row_list = []

neo/master/handlers/storage.py

@@ -37,10 +37,13 @@ class StorageServiceHandler(BaseServiceHandler):
             conn.notify(protocol.startOperation())
 
     def _nodeLost(self, conn, node):
-        pt = self.app.pt
-        # TODO: check this, is it need ? do we have to broadcast changes ?
-        pt.outdate()
-        if not pt.operational():
+        # XXX: here the cells are outdated to trigger the replication process
+        # when the node will come back. It might be better to reduce network
+        # overload since others nodes known that it's temporarily down and thus,
+        # outdate by themselves its cells.
+        logging.info('storage node lost')
+        self.app.outdateAndBroadcastPartition()
+        if not self.app.pt.operational():
             raise OperationFailure, 'cannot continue operation'
 
     def handleNotifyInformationLocked(self, conn, packet, tid):