Splity master node handlers:

- All incoming connections start with the identification handler which will
apply the right handler depending on the current cluster state and the node
type. So we are now sure that a node is well identified, got an UUID and is
registered into the node manager when it leaves the identification handler.
- During service, clietnt and storage nodes have their own handler


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@686 71dcc9de-d417-0410-9af5-da40c76e7ee4

neo/master/handler.py

@@ -30,29 +30,6 @@ class MasterEventHandler(EventHandler):
         self.app = app
         EventHandler.__init__(self)
 
-    def acceptNodeIdentification(self, conn, packet, uuid):
-        """ Send a packet to accept the node identification """
-        app = self.app
-        args = (protocol.MASTER_NODE_TYPE, app.uuid, 
-                app.server[0], app.server[1], 
-                app.pt.getPartitions(), app.pt.getReplicas(),
-                uuid)
-        p = protocol.acceptNodeIdentification(*args)
-        conn.answer(p, packet)
-    
-    def registerAdminNode(self, conn, packet, uuid, server):
-        """ Register the connection's peer as an admin node """
-        from neo.master.administration import AdministrationEventHandler
-        from neo.node import AdminNode
-        node = self.app.nm.getNodeByUUID(uuid)
-        if node is None:
-            uuid = self.app.getNewUUID(protocol.ADMIN_NODE_TYPE)
-            self.app.nm.add(AdminNode(uuid=uuid, server=server))
-        conn.setUUID(uuid)
-        conn.setHandler(AdministrationEventHandler(self.app))
-        logging.info('Register admin node %s' % util.dump(uuid))
-        self.acceptNodeIdentification(conn, packet, uuid)
-
     def handleRequestNodeIdentification(self, conn, packet, node_type,
                                         uuid, ip_address, port, name):
         raise NotImplementedError('this method must be overridden')

neo/master/identification.py

+#
+# Copyright (C) 2006-2009  Nexedi SA
+# 
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License
+# as published by the Free Software Foundation; either version 2
+# of the License, or (at your option) any later version.
+# 
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
+
+import logging
+
+from neo import protocol
+from neo.node import AdminNode, MasterNode, ClientNode, StorageNode
+from neo.master.handler import MasterEventHandler
+from neo import decorators
+
+class IdentificationEventHandler(MasterEventHandler):
+    """This class deals with messages from the admin node only"""
+
+    def connectionClosed(self, conn):
+        logging.warning('lost a node in IdentificationEventHandler')
+
+    def timeoutExpired(self, conn):
+        logging.warning('lost a node in IdentificationEventHandler')
+
+    def peerBroken(self, conn):
+        logging.warning('lost a node in IdentificationEventHandler')
+
+    # TODO: move this into a new handler
+    @decorators.identification_required
+    def handleAnnouncePrimaryMaster(self, conn, packet):
+        uuid = conn.getUUID()
+        # I am also the primary... So restart the election.
+        raise ElectionFailure, 'another primary arises'
+
+    def handleReelectPrimaryMaster(self, conn, packet):
+        raise ElectionFailure, 'reelection requested'
+
+    def handleNotifyNodeInformation(self, conn, packet, node_list):
+        # XXX: Secondary master can send this packet 
+        logging.error('ignoring NotifyNodeInformation packet')
+
+    def handleRequestNodeIdentification(self, conn, packet, node_type,
+            uuid, ip_address, port, name):
+
+        # TODO: handle broken nodes
+
+        self.checkClusterName(name)
+        app, nm = self.app, self.app.nm
+        server = (ip_address, port)
+        node_by_uuid = nm.getNodeByUUID(uuid)
+        node_by_addr = nm.getNodeByServer(server)
+
+        if node_by_uuid is not None and node_by_addr is not None and \
+                node_by_uuid is not node_by_addr:
+            # got a conflict, but UUIDs should be more reliable
+            # TODO: delete the old node...
+            raise RuntimeError('node conflict not implemented yet')
+            pass
+        node = node_by_uuid or node_by_addr
+
+        if node is not None and node.getServer() != server:
+            # address changed
+            # TODO: delete or update the old node ?
+            node.setServer(server)
+            raise RuntimeError('node address changement not implemented yet')
+
+        # ask the app the node identification, if refused, an exception is raised
+        result = self.app.identifyNode(node_type, uuid, node) 
+        (uuid, node, state, handler, klass) = result
+        if uuid == protocol.INVALID_UUID:
+            # no valid uuid, give it one
+            uuid = app.getNewUUID(node_type)
+        if node is None:
+            # new node
+            node = klass(uuid=uuid, server=(ip_address, port))
+            app.nm.add(node)
+        handler = handler(self.app)
+        # set up the node
+        node.setUUID(uuid)
+        node.setState(state)
+        # set up the connection
+        conn.setUUID(uuid)
+        conn.setHandler(handler)
+        # XXX: Here we could bin conn and node together
+        # answer
+        args = (protocol.MASTER_NODE_TYPE, app.uuid, app.server[0], app.server[1], 
+                app.pt.getPartitions(), app.pt.getReplicas(), uuid)
+        conn.answer(protocol.acceptNodeIdentification(*args), packet)
+        # trigger the event
+        handler.connectionCompleted(conn)
+        app.broadcastNodeInformation(node)
+

neo/master/recovery.py

@@ -26,161 +26,42 @@ from neo.exception import ElectionFailure
 from neo.protocol import Packet, UnexpectedPacketError, INVALID_UUID, INVALID_PTID
 from neo.node import ClientNode, StorageNode, MasterNode, AdminNode
 from neo.util import dump
-from neo import decorators
 
 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
+        conn.ask(protocol.askLastIDs())
+
     def connectionClosed(self, conn):
+        app = self.app
         uuid = conn.getUUID()
-        if uuid is not None:
-            app = self.app
-            node = app.nm.getNodeByUUID(uuid)
-            if node.getState() == RUNNING_STATE:
-                node.setState(TEMPORARILY_DOWN_STATE)
-                app.broadcastNodeInformation(node)
+        node = app.nm.getNodeByUUID(uuid)
+        if node.getState() == RUNNING_STATE:
+            node.setState(TEMPORARILY_DOWN_STATE)
+            app.broadcastNodeInformation(node)
         MasterEventHandler.connectionClosed(self, conn)
 
     def timeoutExpired(self, conn):
+        app = self.app
         uuid = conn.getUUID()
-        if uuid is not None:
-            app = self.app
-            node = app.nm.getNodeByUUID(uuid)
-            if node.getState() == RUNNING_STATE:
-                node.setState(TEMPORARILY_DOWN_STATE)
-                app.broadcastNodeInformation(node)
+        node = app.nm.getNodeByUUID(uuid)
+        if node.getState() == RUNNING_STATE:
+            node.setState(TEMPORARILY_DOWN_STATE)
+            app.broadcastNodeInformation(node)
         MasterEventHandler.timeoutExpired(self, conn)
 
     def peerBroken(self, conn):
-        uuid = conn.getUUID()
-        if uuid is not None:
-            app = self.app
-            node = app.nm.getNodeByUUID(uuid)
-            if node.getState() != BROKEN_STATE:
-                node.setState(BROKEN_STATE)
-                app.broadcastNodeInformation(node)
-        MasterEventHandler.peerBroken(self, conn)
-
-    def packetReceived(self, conn, packet):
-        MasterEventHandler.packetReceived(self, conn, packet)
-
-    def handleRequestNodeIdentification(self, conn, packet, node_type, uuid, 
-                                        ip_address, port, name):
-        self.checkClusterName(name)
         app = self.app
-        addr = (ip_address, port)
-
-        if node_type == ADMIN_NODE_TYPE:
-            self.registerAdminNode(conn, packet, uuid, addr)
-            return
-
-        if node_type not in (MASTER_NODE_TYPE, STORAGE_NODE_TYPE):
-            logging.info('reject a connection from a client')
-            raise protocol.NotReadyError
-        if node_type is STORAGE_NODE_TYPE and uuid is INVALID_UUID:
-            # refuse an empty storage node (with no UUID) to avoid potential
-            # UUID conflict
-            logging.info('reject empty storage node')
-            raise protocol.NotReadyError
-
-        # Here are many situations. In principle, a node should be identified by
-        # an UUID, since an UUID never change when moving a storage node to a different
-        # server, and an UUID always changes for a master node and a client node whenever
-        # it restarts, so more reliable than a server address.
-        # 
-        # However, master nodes can be known only as the server addresses. And, a node
-        # may claim a server address used by another node.
-
-        node = app.nm.getNodeByUUID(uuid)
-        if not app.isValidUUID(uuid, addr):
-            # Here we have an UUID conflict, assume that's a new node
-            node = None
-        else:
-            # First, get the node by the UUID.
-            node = app.nm.getNodeByUUID(uuid)
-        if node is None:
-            # generate an uuid for this node
-            while not app.isValidUUID(uuid, addr):
-                uuid = app.getNewUUID(node_type)
-            # If nothing is present, try with the server address.
-            node = app.nm.getNodeByServer(addr)
-            if node is None:
-                # Nothing is found. So this must be the first time that this node
-                # connected to me.
-                if node_type == MASTER_NODE_TYPE:
-                    node = MasterNode(server = addr, uuid = uuid)
-                else:
-                    node = StorageNode(server = addr, uuid = uuid)
-                app.nm.add(node)
-                app.broadcastNodeInformation(node)
-            else:
-                # Otherwise, I know it only by the server address or the same server
-                # address but with a different UUID.
-                if node.getUUID() is None:
-                    # This must be a master node.
-                    if node.getNodeType() != MASTER_NODE_TYPE or node_type != MASTER_NODE_TYPE:
-                        # Error. This node uses the same server address as a master
-                        # node.
-                        raise protocol.ProtocolError('invalid server address') 
-                    node.setUUID(uuid)
-                    if node.getState() != RUNNING_STATE:
-                        node.setState(RUNNING_STATE)
-                    app.broadcastNodeInformation(node)
-                else:
-                    # This node has a different UUID.
-                    if node.getState() == RUNNING_STATE:
-                        # If it is still running, reject this node.
-                        raise protocol.ProtocolError('invalid server address')
-                    # Otherwise, forget the old one.
-                    node.setState(BROKEN_STATE)
-                    app.broadcastNodeInformation(node)
-                    # And insert a new one.
-                    node.setUUID(uuid)
-                    node.setState(RUNNING_STATE)
-                    app.broadcastNodeInformation(node)
-        else:
-            # I know this node by the UUID.
-            if node.getServer() != addr:
-                # This node has a different server address.
-                if node.getState() == RUNNING_STATE:
-                    # If it is still running, reject this node.
-                    raise protocol.ProtocolError('invalid server address')
-                # Otherwise, forget the old one.
-                node.setState(BROKEN_STATE)
-                app.broadcastNodeInformation(node)
-                # And insert a new one.
-                node.setServer(addr)
-                node.setState(RUNNING_STATE)
-                app.broadcastNodeInformation(node)
-            else:
-                # If this node is broken, reject it. Otherwise, assume that it is
-                # working again.
-                if node.getState() == BROKEN_STATE:
-                    raise protocol.BrokenNodeDisallowedError
-                node.setUUID(uuid)
-                node.setState(RUNNING_STATE)
-                app.broadcastNodeInformation(node)
-
-        conn.setUUID(uuid)
-
-        self.acceptNodeIdentification(conn, packet, uuid)
-
-        if node_type is STORAGE_NODE_TYPE:
-            # ask the last IDs.
-            conn.ask(protocol.askLastIDs())
-
-    @decorators.identification_required
-    def handleAnnouncePrimaryMaster(self, conn, packet):
         uuid = conn.getUUID()
-        # I am also the primary... So restart the election.
-        raise ElectionFailure, 'another primary arises'
-
-    def handleReelectPrimaryMaster(self, conn, packet):
-        raise ElectionFailure, 'reelection requested'
+        node = app.nm.getNodeByUUID(uuid)
+        if node.getState() != BROKEN_STATE:
+            node.setState(BROKEN_STATE)
+            app.broadcastNodeInformation(node)
+        MasterEventHandler.peerBroken(self, conn)
 
-    @decorators.identification_required
     def handleNotifyNodeInformation(self, conn, packet, node_list):
-        uuid = conn.getUUID()
         app = self.app
         for node_type, ip_address, port, uuid, state in node_list:
             if node_type in (CLIENT_NODE_TYPE, ADMIN_NODE_TYPE):
@@ -223,14 +104,12 @@ class RecoveryEventHandler(MasterEventHandler):
             # Something wrong happened possibly. Cut the connection to this node,
             # if any, and notify the information to others.
             # XXX this can be very slow.
-            for c in app.em.getConnectionList():
-                if c.getUUID() == uuid:
-                    c.close()
+            c = app.em.getConnectionByUUID(uuid)
+            if c is not None:
+                c.close()
             node.setState(state)
             app.broadcastNodeInformation(node)
 
-    @decorators.identification_required
-    @decorators.restrict_node_types(STORAGE_NODE_TYPE)
     def handleAnswerLastIDs(self, conn, packet, loid, ltid, lptid):
         uuid = conn.getUUID()
         app = self.app
@@ -251,8 +130,6 @@ class RecoveryEventHandler(MasterEventHandler):
         elif app.pt.getID() == lptid and app.target_uuid is None:
             app.target_uuid = uuid
 
-    @decorators.identification_required
-    @decorators.restrict_node_types(STORAGE_NODE_TYPE)
     def handleAnswerPartitionTable(self, conn, packet, ptid, row_list):
         uuid = conn.getUUID()
         app = self.app

neo/protocol.py

@@ -314,8 +314,10 @@ INTERNAL_ERROR_CODE = 8
 # Cluster states
 cluster_states = Enum({
     'BOOTING': 1,
-    'RUNNING': 2,
-    'STOPPING': 3,
+    'RECOVERING': 2,
+    'VERIFYING': 3,
+    'RUNNING': 4,
+    'STOPPING': 5,
 })
 VALID_CLUSTER_STATE_LIST = (BOOTING, RUNNING, STOPPING)