Bug fix: Call abort() instead of close().

The NotifyNodeInformation packet was not always sent before connection
closure, call abort() to ensure the write buffer is flushed before closure
so the peer can understand that it is not a primary failure.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1675 71dcc9de-d417-0410-9af5-da40c76e7ee4

neo/master/handlers/administration.py

@@ -87,7 +87,7 @@ class AdministrationHandler(MasterHandler):
                 node_list = [node.asTuple()]
                 storage_conn.notify(Packets.NotifyNodeInformation(node_list))
                 # close to avoid handle the closure as a connection lost
-                storage_conn.close()
+                storage_conn.abort()
             # modify the partition table if required
             cell_list = []
             if modify_partition_table: