Remove dropNode() calls where it's not safe to change the partition table

without give the choice to an admin to discard the node's content. Now a node
can be dropped only when requested by the admin node or when a storage has
changed its address (this case should be checked and discussed).


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@732 71dcc9de-d417-0410-9af5-da40c76e7ee4

neo/master/app.py

@@ -646,11 +646,15 @@ class Application(object):
                     for node in nm.getStorageNodeList():
                         if node.getState() == TEMPORARILY_DOWN_STATE \
                                and node.getLastStateChange() + expiration < current_time:
-                            logging.info('%s is down' % (node, ))
-                            node.setState(DOWN_STATE)
-                            self.broadcastNodeInformation(node)
-                            cell_list = self.pt.dropNode(node)
-                            self.broadcastPartitionChanges(self.pt.setNextID(), cell_list)
+                            logging.warning('%s is down, have to notify the admin' % (node, ))
+                            # XXX: here we should notify the administrator that
+                            # a node seems dead and should be dropped frop the
+                            # partition table. This should not be done
+                            # automaticaly to avoid data lost.
+                            #node.setState(DOWN_STATE)
+                            #self.broadcastNodeInformation(node)
+                            #cell_list = self.pt.dropNode(node)
+                            #self.broadcastPartitionChanges(self.pt.setNextID(), cell_list)
                             if not self.pt.operational():
                                 # Catastrophic.
                                 raise OperationFailure, 'cannot continue operation'

neo/master/identification.py

@@ -42,10 +42,10 @@ class IdentificationEventHandler(MasterEventHandler):
         node_by_addr = nm.getNodeByServer(server)
 
         def changeNodeAddress(node, server):
-            from copy import copy
             if node_type == protocol.STORAGE_NODE_TYPE:
                 args = (node.getServer(), server)
                 # remove storage from partition table
+                # XXX: this should be safe but need to be checked
                 cell_list = app.pt.dropNode(node)
                 if cell_list:
                     ptid = app.pt.setNextID()
@@ -70,11 +70,15 @@ class IdentificationEventHandler(MasterEventHandler):
                 if node.getState() == protocol.RUNNING_STATE:
                     # still running, reject this new node
                     raise protocol.ProtocolError('invalid server address')
+                # this node has changed its address
                 node = changeNodeAddress(node, server)
         if node_by_uuid is None and node_by_addr is not None:
             if node.getState() == protocol.RUNNING_STATE:
                 # still running, reject this new node
                 raise protocol.ProtocolError('invalid server address')
+            # FIXME: here the node was known with a different uuid but with the
+            # same address, is it safe to forgot the old, even if he's not
+            # running ?
             node = changeNodeAddress(node, server)
 
         # ask the app the node identification, if refused, an exception is raised

neo/master/recovery.py

@@ -31,7 +31,7 @@ class RecoveryEventHandler(MasterEventHandler):
     """This class deals with events for a recovery phase."""
 
     def connectionCompleted(self, conn):
-        # ask the last IDs to perform the the recovery
+        # ask the last IDs to perform the recovery
         conn.ask(protocol.askLastIDs())
 
     def connectionClosed(self, conn):

neo/master/service.py

@@ -99,14 +99,7 @@ class ServiceEventHandler(MasterEventHandler):
 
             app.broadcastNodeInformation(node)
             if node.getNodeType() == STORAGE_NODE_TYPE:
-                if state in (DOWN_STATE, BROKEN_STATE):
-                    # XXX still required to change here ??? who can send
-                    # this kind of message with these status except admin node
-                    cell_list = app.pt.dropNode(node)
-                    if len(cell_list) != 0:
-                        ptid = app.pt.setNextID()
-                        app.broadcastPartitionChanges(ptid, cell_list)
-                elif state == TEMPORARILY_DOWN_STATE:
+                if state == TEMPORARILY_DOWN_STATE:
                     cell_list = app.pt.outdate()
                     if len(cell_list) != 0:
                         ptid = app.pt.setNextID()

neo/master/shutdown.py

@@ -103,14 +103,7 @@ class ShutdownEventHandler(ServiceEventHandler):
             logging.debug('broadcasting node information')
             app.broadcastNodeInformation(node)
             if node.getNodeType() == STORAGE_NODE_TYPE:
-                if state in (DOWN_STATE, BROKEN_STATE):
-                    # XXX still required to change here ??? who can send
-                    # this kind of message with these status except admin node
-                    cell_list = app.pt.dropNode(node)
-                    if len(cell_list) != 0:
-                        ptid = app.pt.setNextID()
-                        app.broadcastPartitionChanges(ptid, cell_list)
-                elif state == TEMPORARILY_DOWN_STATE:
+                if state == TEMPORARILY_DOWN_STATE:
                     cell_list = app.pt.outdate()
                     if len(cell_list) != 0:
                         ptid = app.pt.setNextID()

neo/master/verification.py

@@ -33,43 +33,33 @@ class VerificationEventHandler(MasterEventHandler):
     def connectionCompleted(self, conn):
         pass
 
-    def connectionClosed(self, conn):
+    def _dropIt(self, conn, node, new_state):
         app = self.app
-        uuid = conn.getUUID()
-        node = app.nm.getNodeByUUID(uuid)
+        node.setState(new_state)
+        app.broadcastNodeInformation(node)
+        if not app.pt.operational():
+            raise VerificationFailure, 'cannot continue verification'
+
+    def connectionClosed(self, conn):
+        node = self.app.nm.getNodeByUUID(conn.getUUID())
         if node.getState() == RUNNING_STATE:
-            node.setState(TEMPORARILY_DOWN_STATE)
-            app.broadcastNodeInformation(node)
-            if not app.pt.operational():
-                # Catastrophic.
-                raise VerificationFailure, 'cannot continue verification'
+            self._dropIt(conn, node, TEMPORARILY_DOWN_STATE)
         MasterEventHandler.connectionClosed(self, conn)
 
     def timeoutExpired(self, conn):
-        app = self.app
-        uuid = conn.getUUID()
-        node = app.nm.getNodeByUUID(uuid)
+        node = self.app.nm.getNodeByUUID(conn.getUUID())
         if node.getState() == RUNNING_STATE:
-            node.setState(TEMPORARILY_DOWN_STATE)
-            app.broadcastNodeInformation(node)
-            if not app.pt.operational():
-                # Catastrophic.
-                raise VerificationFailure, 'cannot continue verification'
+            self._dropIt(conn, node, TEMPORARILY_DOWN_STATE)
         MasterEventHandler.timeoutExpired(self, conn)
 
     def peerBroken(self, conn):
-        app = self.app
-        uuid = conn.getUUID()
-        node = app.nm.getNodeByUUID(uuid)
+        node = self.app.nm.getNodeByUUID(conn.getUUID())
         if node.getState() != BROKEN_STATE:
-            node.setState(BROKEN_STATE)
-            app.broadcastNodeInformation(node)
-            cell_list = app.pt.dropNode(node)
-            ptid = app.pt.setNextID()
-            app.broadcastPartitionChanges(ptid, cell_list)
-            if not app.pt.operational():
-                # Catastrophic.
-                raise VerificationFailure, 'cannot continue verification'
+            self._dropIt(conn, node, BROKEN_STATE)
+            # here the node is no more dropped from the partition table anymore
+            # because it's under the responsability of an administrator to
+            # restore the node, backup the node content or drop it definitely
+            # and loose all it's content.
         MasterEventHandler.peerBroken(self, conn)
 
     def handleNotifyNodeInformation(self, conn, packet, node_list):