Fix implementation of setNodeState in master administration handler, it's now

allowed to set a disconnected node to DOWN state and drop it from the partition
table. A global review of node state alteration by the admin node is required to
ensure all and no less-or-more is allowed here. Add a TODO entry about that.


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@1240 71dcc9de-d417-0410-9af5-da40c76e7ee4

TODO

@@ -79,6 +79,9 @@ RC  - Review output of pylint (CODE)
     - Fix packet types (CODE)
       Some packets are of response type but have a name starting with 'notify'.
       Decide wether they are notifications or responses, and fix the right one.
+    - Consider replace setNodeState admin packet by one per action, like
+      dropNode to reduce packet processing complexity and reduce bad actions
+      like set a node in TEMPORARILY_DOWN state.
 
     Storage
     - Implement incremental storage verification (BANDWITH)

neo/master/app.py

@@ -311,6 +311,7 @@ class Application(object):
 
     def broadcastPartitionChanges(self, ptid, cell_list):
         """Broadcast a Notify Partition Changes packet."""
+        # XXX: don't send if cell_list is empty, to have an unique check
         logging.debug('broadcastPartitionChanges')
         self.pt.log()
         for c in self.em.getConnectionList():

neo/master/handlers/administration.py

@@ -48,20 +48,12 @@ class AdministrationHandler(MasterHandler):
         app = self.app
         node = app.nm.getNodeByUUID(uuid)
         if node is None:
-            p = protocol.protocolError('invalid uuid')
-            conn.answer(p, packet.getId())
-            return
+            raise protocol.ProtocolError('unknown node')
 
         if uuid == app.uuid:
-            # get message for self
-            if state == RUNNING_STATE:
-                # yes I know
-                p = protocol.noError('node state changed')
-                conn.answer(p, packet.getId())
-                return
-            else:
-                # I was asked to shutdown
             node.setState(state)
+            # get message for self
+            if state != RUNNING_STATE:
                 p = protocol.noError('node state changed')
                 conn.answer(p, packet.getId())
                 app.shutdown()
@@ -70,34 +62,24 @@ class AdministrationHandler(MasterHandler):
             # no change, just notify admin node
             p = protocol.noError('node state changed')
             conn.answer(p, packet.getId())
-        else:
-            # FIXME: this is wrong, what about a disconnected node to set in 
-            # down state ? we can't get a connection to it but still set the 
-            # state.
+            return
+
+        if state == protocol.RUNNING_STATE:
             # first make sure to have a connection to the node
             node_conn = None
-            conn_found = False
             for node_conn in app.em.getConnectionList():
                 if node_conn.getUUID() == node.getUUID():
-                    conn_found = True
                     break
-            if conn_found is False:
+            else:
                 # no connection to the node
-                p = protocol.protocolError('no connection to the node')
-                conn.notify(p)
-                return
+                raise protocol.ProtocolError('no connection to the node')
 
+        elif state == protocol.DOWN_STATE and node.isStorage():
             # modify the partition table if required
-        if modify_partition_table and node.isStorage():
-            if state in (DOWN_STATE, TEMPORARILY_DOWN_STATE, HIDDEN_STATE):
+            cell_list = []
+            if modify_partition_table:
                 # remove from pt
                 cell_list = app.pt.dropNode(node)
-            else:
-                # add to pt
-                cell_list = app.pt.addNode(node)
-            if len(cell_list) != 0:
-                ptid = app.pt.setNextID()
-                app.broadcastPartitionChanges(ptid, cell_list)
             else:
                 # outdate node in partition table
                 cell_list = app.pt.outdate()
@@ -105,16 +87,11 @@ class AdministrationHandler(MasterHandler):
                 ptid = app.pt.setNextID()
                 app.broadcastPartitionChanges(ptid, cell_list)
 
-        if node.getState() != state:
+        # /!\ send the node information *after* the partition table change
         node.setState(state)
         p = protocol.noError('state changed')
         conn.answer(p, packet.getId())
         app.broadcastNodeInformation(node)
-            # If this is a storage node, ask it to start.
-            if node.isStorage() and state == RUNNING_STATE  \
-                   and self.app.cluster_state == RUNNING:
-                logging.info("asking sn to start operation")
-                node_conn.notify(protocol.startOperation())
 
     def handleAddPendingNodes(self, conn, packet, uuid_list):
         uuids = ', '.join([dump(uuid) for uuid in uuid_list])