Skip to content

N
neo
Commit ecd05c50 authored by Grégory Wisniewski's avatar Grégory Wisniewski
Browse files
Options

Split storage handlers. As for the primary master, all incoming connections are 

handled by the identification handler which apply the right handler depending on
the peer node type. Client and storage incoming connections are refused until
the storage reach the operation state, but the listening port is open to ensure
it's available, tryAgain errors are sent to wrong nodes.
Remove decorators because connection type is implied by handler type.


git-svn-id: https://svn.erp5.org/repos/neo/branches/prototype3@752 71dcc9de-d417-0410-9af5-da40c76e7ee4
parent 733e7e90
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 242 additions and 279 deletions
neo/storage/app.py
View file @ ecd05c50
......@@ -32,9 +32,10 @@ from neo.connection import ListeningConnection, ClientConnection
from neo.exception import OperationFailure, PrimaryFailure
from neo.storage.bootstrap import BootstrapEventHandler
from neo.storage.verification import VerificationEventHandler
from neo.storage.operation import OperationEventHandler
from neo.storage.operation import MasterOperationEventHandler
from neo.storage.hidden import HiddenEventHandler
from neo.storage.replicator import Replicator
from neo.storage.identification import IdentificationEventHandler
from neo.connector import getConnectorHandler
from neo.pt import PartitionTable
from neo.util import dump
......@@ -72,6 +73,8 @@ class Application(object):
self.listening_conn = None
self.master_conn = None
# ready is True when operational and got all informations
self.ready = False
self.has_node_information = False
self.has_partition_table = False
......@@ -134,23 +137,21 @@ class Application(object):
for server in self.master_node_list:
self.nm.add(MasterNode(server = server))
# Make a listening port
handler = IdentificationEventHandler(self)
self.listening_conn = ListeningConnection(self.em, handler,
addr=self.server, connector_handler=self.connector_handler)
# Connect to a primary master node, verify data, and
# start the operation. This cycle will be executed permentnly,
# until the user explicitly requests a shutdown.
while 1:
self.operational = False
# refuse any incoming connections for now
if self.listening_conn is not None:
self.listening_conn.close()
self.listening_conn = None
# look for the primary master
self.connectToPrimaryMaster()
assert self.master_conn is not None
if self.uuid == INVALID_UUID:
raise RuntimeError, 'No UUID supplied from the primary master'
# Make a listening port when connected to the primary
self.listening_conn = ListeningConnection(self.em, None,
addr=self.server, connector_handler=self.connector_handler)
try:
while 1:
try:
......@@ -241,32 +242,28 @@ class Application(object):
logging.info('verifying data')
handler = VerificationEventHandler(self)
self.master_conn.setHandler(handler)
em = self.em
# Make sure that every connection has the verfication event handler.
for conn in em.getConnectionList():
conn.setHandler(handler)
while not self.operational:
em.poll(1)
# ask node list
# ask node list and partition table
self.master_conn.ask(protocol.askNodeInformation())
self.master_conn.ask(protocol.askPartitionTable(()))
while not self.has_node_information or not self.has_partition_table:
em.poll(1)
self.ready = True
def doOperation(self):
"""Handle everything, including replications and transactions."""
logging.info('doing operation')
handler = OperationEventHandler(self)
em = self.em
nm = self.nm
# Make sure that every connection has the verfication event handler.
for conn in em.getConnectionList():
conn.setHandler(handler)
handler = MasterOperationEventHandler(self)
self.master_conn.setHandler(handler)
# Forget all unfinished data.
self.dm.dropUnfinishedData()
......@@ -321,7 +318,6 @@ class Application(object):
def getPartition(self, oid_or_tid):
return unpack('!Q', oid_or_tid)[0] % self.num_partitions
def shutdown(self):
"""Close all connections and exit"""
for c in self.em.getConnectionList():
......
neo/storage/identification.py 0 → 100644
View file @ ecd05c50
#
# 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.storage.handler import StorageEventHandler
from neo.protocol import INVALID_SERIAL, INVALID_TID, \
INVALID_PARTITION, BROKEN_STATE, TEMPORARILY_DOWN_STATE, \
MASTER_NODE_TYPE, STORAGE_NODE_TYPE, CLIENT_NODE_TYPE, \
DISCARDED_STATE, OUT_OF_DATE_STATE
from neo.util import dump
class IdentificationEventHandler(StorageEventHandler):
""" Handler used for incoming connections during operation state """
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')
def handleRequestNodeIdentification(self, conn, packet, node_type,
uuid, ip_address, port, name):
self.checkClusterName(name)
# reject any incoming connections if not ready
if not self.app.ready:
raise protocol.notReady('try again')
app, nm = self.app, self.app.nm
server = (ip_address, port)
node = app.nm.getNodeByUUID(uuid)
if node is None:
logging.error('reject an unknown node %s', dump(uuid))
raise protocol.NotReadyError
# If this node is broken, reject it.
if node.getState() == BROKEN_STATE:
raise protocol.BrokenNodeDisallowedError
# choose the handler according to the node type
if node_type == protocol.CLIENT_NODE_TYPE:
from neo.storage.operation import ClientOperationEventHandler
handler = ClientOperationEventHandler
elif node_type == protocol.STORAGE_NODE_TYPE:
from neo.storage.operation import StorageOperationEventHandler
handler = StorageOperationEventHandler
else:
raise protocol.protocolError('reject non-client-or-storage node')
# apply the handler and set up the connection
handler = handler(self.app)
conn.setHandler(handler)
conn.setUUID(uuid)
node.setUUID(uuid)
# FIXME: here we should use pt.getPartitions() and pt.getReplicas()
args = (STORAGE_NODE_TYPE, app.uuid, app.server[0], app.server[1],
app.num_partitions, app.num_replicas, uuid)
# accept the identification and trigger an event
conn.answer(protocol.acceptNodeIdentification(*args), packet)
handler.connectionCompleted(conn)
neo/storage/operation.py
View file @ ecd05c50
This diff is collapsed.
neo/storage/verification.py
View file @ ecd05c50
......@@ -24,10 +24,8 @@ from neo.protocol import INVALID_OID, INVALID_TID, \
MASTER_NODE_TYPE, STORAGE_NODE_TYPE, \
Packet, UnexpectedPacketError
from neo.util import dump
from neo.node import MasterNode, StorageNode, ClientNode
from neo.connection import ClientConnection
from neo.node import StorageNode
from neo.exception import PrimaryFailure, OperationFailure
from neo import decorators
class VerificationEventHandler(StorageEventHandler):
"""This class deals with events for a verification phase."""
......@@ -39,67 +37,23 @@ class VerificationEventHandler(StorageEventHandler):
StorageEventHandler.connectionAccepted(self, conn, s, addr)
def timeoutExpired(self, conn):
if not conn.isServerConnection():
# If a primary master node timeouts, I cannot continue.
logging.critical('the primary master node times out')
raise PrimaryFailure('the primary master node times out')
# If a primary master node timeouts, I cannot continue.
logging.critical('the primary master node times out')
raise PrimaryFailure('the primary master node times out')
StorageEventHandler.timeoutExpired(self, conn)
def connectionClosed(self, conn):
if not conn.isServerConnection():
# If a primary master node closes, I cannot continue.
logging.critical('the primary master node is dead')
raise PrimaryFailure('the primary master node is dead')
# If a primary master node closes, I cannot continue.
logging.critical('the primary master node is dead')
raise PrimaryFailure('the primary master node is dead')
StorageEventHandler.connectionClosed(self, conn)
def peerBroken(self, conn):
if not conn.isServerConnection():
# If a primary master node gets broken, I cannot continue.
logging.critical('the primary master node is broken')
raise PrimaryFailure('the primary master node is broken')
# If a primary master node gets broken, I cannot continue.
logging.critical('the primary master node is broken')
raise PrimaryFailure('the primary master node is broken')
StorageEventHandler.peerBroken(self, conn)
@decorators.server_connection_required
def handleRequestNodeIdentification(self, conn, packet, node_type,
uuid, ip_address, port, name):
self.checkClusterName(name)
app = self.app
if node_type != MASTER_NODE_TYPE:
logging.info('reject a connection from a non-master')
raise protocol.NotReadyError
addr = (ip_address, port)
node = app.nm.getNodeByServer(addr)
if node is None:
node = MasterNode(server = addr, uuid = uuid)
app.nm.add(node)
else:
# If this node is broken, reject it.
if node.getUUID() == uuid:
if node.getState() == BROKEN_STATE:
raise protocol.BrokenNodeDisallowedError
# Trust the UUID sent by the peer.
node.setUUID(uuid)
conn.setUUID(uuid)
p = protocol.acceptNodeIdentification(STORAGE_NODE_TYPE, app.uuid,
app.server[0], app.server[1], app.num_partitions,
app.num_replicas, uuid)
conn.answer(p, packet)
# Now the master node should know that I am not the right one.
conn.abort()
def handleAcceptNodeIdentification(self, conn, packet, node_type,
uuid, ip_address, port,
num_partitions, num_replicas, your_uuid):
raise UnexpectedPacketError
@decorators.client_connection_required
def handleAnswerPrimaryMaster(self, conn, packet, primary_uuid,
known_master_list):
app = self.app
......@@ -109,7 +63,6 @@ class VerificationEventHandler(StorageEventHandler):
# But a primary master node is supposed not to send any info
# with this packet, so it would be useless.
@decorators.client_connection_required
def handleAskLastIDs(self, conn, packet):
app = self.app
oid = app.dm.getLastOID() or INVALID_OID
......@@ -124,8 +77,9 @@ class VerificationEventHandler(StorageEventHandler):
def handleAnswerPartitionTable(self, conn, packet, ptid, row_list):
assert not row_list
self.app.has_partition_table = True
logging.info('Got the partition table :')
self.app.pt.log()
@decorators.client_connection_required
def handleAskPartitionTable(self, conn, packet, offset_list):
app = self.app
row_list = []
......@@ -144,7 +98,6 @@ class VerificationEventHandler(StorageEventHandler):
p = protocol.answerPartitionTable(app.ptid, row_list)
conn.answer(p, packet)
@decorators.client_connection_required
def handleSendPartitionTable(self, conn, packet, ptid, row_list):
"""A primary master node sends this packet to synchronize a partition
table. Note that the message can be split into multiple packets."""
......@@ -176,7 +129,6 @@ class VerificationEventHandler(StorageEventHandler):
cell.getState()))
app.dm.setPartitionTable(ptid, cell_list)
@decorators.client_connection_required
def handleNotifyPartitionChanges(self, conn, packet, ptid, cell_list):
"""This is very similar to Send Partition Table, except that
the information is only about changes from the previous."""
......@@ -203,15 +155,12 @@ class VerificationEventHandler(StorageEventHandler):
# Then, the database.
app.dm.changePartitionTable(ptid, cell_list)
@decorators.client_connection_required
def handleStartOperation(self, conn, packet):
self.app.operational = True
@decorators.client_connection_required
def handleStopOperation(self, conn, packet):
raise OperationFailure('operation stopped')
@decorators.client_connection_required
def handleAskUnfinishedTransactions(self, conn, packet):
tid_list = self.app.dm.getUnfinishedTIDList()
p = protocol.answerUnfinishedTransactions(tid_list)
......@@ -225,6 +174,8 @@ class VerificationEventHandler(StorageEventHandler):
# not been finished.
t = app.dm.getTransaction(tid, all = True)
else:
# XXX: this should never be used since we don't accept incoming
# connections out of the operation state.
t = app.dm.getTransaction(tid)
if t is None:
......@@ -233,7 +184,6 @@ class VerificationEventHandler(StorageEventHandler):
p = protocol.answerTransactionInformation(tid, t[1], t[2], t[3], t[0])
conn.answer(p, packet)
@decorators.client_connection_required
def handleAskObjectPresent(self, conn, packet, oid, tid):
if self.app.dm.objectPresent(oid, tid):
p = protocol.answerObjectPresent(oid, tid)
......@@ -242,16 +192,9 @@ class VerificationEventHandler(StorageEventHandler):
'%s:%s do not exist' % (dump(oid), dump(tid)))
conn.answer(p, packet)
@decorators.client_connection_required
def handleDeleteTransaction(self, conn, packet, tid):
self.app.dm.deleteTransaction(tid, all = True)
@decorators.client_connection_required
def handleCommitTransaction(self, conn, packet, tid):
self.app.dm.finishTransaction(tid)
def handleLockInformation(self, conn, packet, tid):
pass
def handleUnlockInformation(self, conn, packet, tid):
pass
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7