Fix issues with raft traffic encryption (#1919)

and run raft behave tests with encryption enabled.

Using the new `pysyncobj` release allowed us to get rid of a lot of hacks with accessing private properties and methods of the parent class and reduce the size of the `raft.py`.

Close https://github.com/zalando/patroni/issues/1746
This commit is contained in:
Alexander Kukushkin
2021-04-30 11:28:41 +02:00
committed by GitHub
parent 294bb43bf1
commit 99626a07f2
7 changed files with 131 additions and 182 deletions
+1 -1
View File
@@ -22,7 +22,7 @@ def install_requirements(what):
for r in read('requirements.txt').split('\n'):
r = r.strip()
if r != '':
extras = {e for e, v in EXTRAS_REQUIRE.items() if v and r.startswith(v[0])}
extras = {e for e, v in EXTRAS_REQUIRE.items() if v and any(r.startswith(x) for x in v)}
if not extras or what == 'all' or what in extras:
requirements.append(r)
+5 -5
View File
@@ -591,10 +591,12 @@ class ExhibitorController(ZooKeeperController):
class RaftController(AbstractDcsController):
CONTROLLER_ADDR = 'localhost:1234'
PASSWORD = '12345'
def __init__(self, context):
super(RaftController, self).__init__(context)
os.environ.update(PATRONI_RAFT_PARTNER_ADDRS="'" + self.CONTROLLER_ADDR + "'", RAFT_PORT='1234')
os.environ.update(PATRONI_RAFT_PARTNER_ADDRS="'" + self.CONTROLLER_ADDR + "'",
PATRONI_RAFT_PASSWORD=self.PASSWORD, RAFT_PORT='1234')
self._raft = None
def _start(self):
@@ -615,18 +617,16 @@ class RaftController(AbstractDcsController):
def cleanup_service_tree(self):
from patroni.dcs.raft import KVStoreTTL
from pysyncobj import SyncObjConf
if self._raft:
self._raft.destroy()
self._raft._SyncObj__thread.join()
self.stop()
os.makedirs(self._work_directory)
self.start()
ready_event = threading.Event()
conf = SyncObjConf(appendEntriesUseBatch=False, dynamicMembershipChange=True, onReady=ready_event.set)
self._raft = KVStoreTTL(None, [self.CONTROLLER_ADDR], conf)
self._raft = KVStoreTTL(ready_event.set, None, None, partner_addrs=[self.CONTROLLER_ADDR], password=self.PASSWORD)
self._raft.startAutoTick()
ready_event.wait()
+86 -126
View File
@@ -5,7 +5,8 @@ import threading
import time
from pysyncobj import SyncObj, SyncObjConf, replicated, FAIL_REASON
from pysyncobj.transport import Node, TCPTransport, CONNECTION_STATE
from pysyncobj.transport import TCPTransport, CONNECTION_STATE
from pysyncobj.utility import TcpUtility, UtilityException
from . import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
from ..utils import validate_directory
@@ -13,122 +14,97 @@ from ..utils import validate_directory
logger = logging.getLogger(__name__)
class MessageNode(Node):
def __init__(self, address):
self.address = address
class UtilityTransport(TCPTransport):
class _TCPTransport(TCPTransport):
def __init__(self, syncObj, selfNode, otherNodes):
super(UtilityTransport, self).__init__(syncObj, selfNode, otherNodes)
self._selfIsReadonlyNode = False
def _connectIfNecessarySingle(self, node):
pass
def connectionState(self, node):
return self._connections[node].state
def isDisconnected(self, node):
return self.connectionState(node) == CONNECTION_STATE.DISCONNECTED
def connectIfRequiredSingle(self, node):
if self.isDisconnected(node):
return self._connections[node].connect(node.ip, node.port)
def disconnectSingle(self, node):
self._connections[node].disconnect()
super(_TCPTransport, self).__init__(syncObj, selfNode, otherNodes)
self.setOnUtilityMessageCallback('members', syncObj.getMembers)
class SyncObjUtility(SyncObj):
class SyncObjUtility(object):
def __init__(self, otherNodes, conf):
autoTick = conf.autoTick
conf.autoTick = False
super(SyncObjUtility, self).__init__(None, otherNodes, conf, transportClass=UtilityTransport)
conf.autoTick = autoTick
self._SyncObj__transport.setOnMessageReceivedCallback(self._onMessageReceived)
self.__result = None
self._nodes = otherNodes
self._utility = TcpUtility(conf.password)
def setPartnerNode(self, partner):
self.__node = partner
def executeCommand(self, command):
try:
return self._utility.executeCommand(self.__node, command)
except UtilityException:
return None
def sendMessage(self, message):
# Abuse the fact that node address is send as a first message
self._SyncObj__transport._selfNode = MessageNode(message)
self._SyncObj__transport.connectIfRequiredSingle(self.__node)
while not self._SyncObj__transport.isDisconnected(self.__node):
self._poller.poll(0.5)
return self.__result
def _onMessageReceived(self, _, message):
self.__result = message
self._SyncObj__transport.disconnectSingle(self.__node)
class MyTCPTransport(TCPTransport):
def _onIncomingMessageReceived(self, conn, message):
if self._syncObj.encryptor and not conn.sendRandKey:
conn.sendRandKey = message
conn.recvRandKey = os.urandom(32)
conn.send(conn.recvRandKey)
return
# Utility messages
if isinstance(message, list) and message[0] == 'members':
conn.send(self._syncObj._get_members())
return True
return super(MyTCPTransport, self)._onIncomingMessageReceived(conn, message)
def getMembers(self):
for self.__node in self._nodes:
response = self.executeCommand(['members'])
if response:
return [member['addr'] for member in response]
class DynMemberSyncObj(SyncObj):
def __init__(self, selfAddress, partnerAddrs, conf):
add_self = False
self.__early_apply_local_log = selfAddress is not None
self.applied_local_log = False
utility = SyncObjUtility(partnerAddrs, conf)
for node in utility._SyncObj__otherNodes:
utility.setPartnerNode(node)
response = utility.sendMessage(['members'])
if response:
partnerAddrs = [member['addr'] for member in response if member['addr'] != selfAddress]
add_self = selfAddress and len(partnerAddrs) == len(response)
break
members = utility.getMembers()
add_self = members and selfAddress not in members
partnerAddrs = [member for member in (members or partnerAddrs) if member != selfAddress]
super(DynMemberSyncObj, self).__init__(selfAddress, partnerAddrs, conf, transportClass=_TCPTransport)
super(DynMemberSyncObj, self).__init__(selfAddress, partnerAddrs, conf, transportClass=MyTCPTransport)
if add_self:
threading.Thread(target=utility.sendMessage, args=(['add', selfAddress],)).start()
thread = threading.Thread(target=utility.executeCommand, args=(['add', selfAddress],))
thread.daemon = True
thread.start()
def _get_members(self):
ret = [{'addr': node.id, 'leader': node == self._getLeader(),
'status': CONNECTION_STATE.CONNECTED if node in self._SyncObj__connectedNodes
else CONNECTION_STATE.DISCONNECTED} for node in self._SyncObj__otherNodes]
ret.append({'addr': self._SyncObj__selfNode.id, 'leader': self._isLeader(),
'status': CONNECTION_STATE.CONNECTED})
return ret
def getMembers(self, args, callback):
callback([{'addr': node.id, 'leader': node == self._getLeader(), 'status': CONNECTION_STATE.CONNECTED
if self.isNodeConnected(node) else CONNECTION_STATE.DISCONNECTED} for node in self.otherNodes] +
[{'addr': self.selfNode.id, 'leader': self._isLeader(), 'status': CONNECTION_STATE.CONNECTED}], None)
def _SyncObj__doChangeCluster(self, request, reverse=False):
ret = False
if not self._SyncObj__selfNode or request[0] != 'add' or reverse or request[1] != self._SyncObj__selfNode.id:
ret = super(DynMemberSyncObj, self)._SyncObj__doChangeCluster(request, reverse)
if ret:
self.forceLogCompaction()
return ret
def _onTick(self, timeToWait=0.0):
super(DynMemberSyncObj, self)._onTick(timeToWait)
# The SyncObj calls onReady callback only when cluster got the leader and is ready for writes.
# In some cases for us it is safe to "signal" the Raft object when the local log is fully applied.
# We are using the `applied_local_log` property for that, but not calling the callback function.
if self.__early_apply_local_log and not self.applied_local_log and self.raftLastApplied == self.raftCommitIndex:
self.applied_local_log = True
class KVStoreTTL(DynMemberSyncObj):
def __init__(self, selfAddress, partnerAddrs, conf, on_set=None, on_delete=None):
def __init__(self, on_ready, on_set, on_delete, **config):
self.__thread = None
self.__on_set = on_set
self.__on_delete = on_delete
self.__limb = {}
self.__retry_timeout = None
self.__early_apply_local_log = selfAddress is not None
self.applied_local_log = False
super(KVStoreTTL, self).__init__(selfAddress, partnerAddrs, conf)
self_addr = config.get('self_addr')
partner_addrs = set(config.get('partner_addrs', []))
if config.get('patronictl'):
if self_addr:
partner_addrs.add(self_addr)
self_addr = None
# Create raft data_dir if necessary
raft_data_dir = config.get('data_dir', '')
if raft_data_dir != '':
validate_directory(raft_data_dir)
file_template = (self_addr or '')
file_template = file_template.replace(':', '_') if os.name == 'nt' else file_template
file_template = os.path.join(raft_data_dir, file_template)
conf = SyncObjConf(password=config.get('password'), autoTick=False, appendEntriesUseBatch=False,
bindAddress=config.get('bind_addr'), commandsWaitLeader=config.get('commandsWaitLeader'),
fullDumpFile=(file_template + '.dump' if self_addr else None),
journalFile=(file_template + '.journal' if self_addr else None),
onReady=on_ready, dynamicMembershipChange=True)
super(KVStoreTTL, self).__init__(self_addr, partner_addrs, conf)
self.__data = {}
@staticmethod
@@ -175,7 +151,7 @@ class KVStoreTTL(DynMemberSyncObj):
if old_value and old_value['created'] != value['created']:
value['created'] = value['updated']
value['index'] = self._SyncObj__raftLastApplied + 1
value['index'] = self.raftLastApplied + 1
self.__data[key] = value
if self.__on_set:
@@ -242,27 +218,29 @@ class KVStoreTTL(DynMemberSyncObj):
return {k: v for k, v in self.__data.items() if k.startswith(key)}
def _onTick(self, timeToWait=0.0):
# The SyncObj starts applying the local log only when there is at least one node connected.
# We want to change this behavior and apply the local log even when there is nobody except us.
# It gives us at least some picture about the last known cluster state.
if self.__early_apply_local_log and not self.applied_local_log and self._SyncObj__needLoadDumpFile:
self._SyncObj__raftCommitIndex = self._SyncObj__getCurrentLogIndex()
self._SyncObj__raftCurrentTerm = self._SyncObj__getCurrentLogTerm()
super(KVStoreTTL, self)._onTick(timeToWait)
# The SyncObj calls onReady callback only when cluster got the leader and is ready for writes.
# In some cases for us it is safe to "signal" the Raft object when the local log is fully applied.
# We are using the `applied_local_log` property for that, but not calling the callback function.
if self.__early_apply_local_log and not self.applied_local_log and self._SyncObj__raftCommitIndex != 1 and \
self._SyncObj__raftLastApplied == self._SyncObj__raftCommitIndex:
self.applied_local_log = True
if self._isLeader():
self.__expire_keys()
else:
self.__limb.clear()
def _autoTickThread(self):
self.__destroying = False
while not self.__destroying:
self.doTick(self.conf.autoTickPeriod)
def startAutoTick(self):
self.__thread = threading.Thread(target=self._autoTickThread)
self.__thread.daemon = True
self.__thread.start()
def destroy(self):
if self.__thread:
self.__destroying = True
self.__thread.join()
super(KVStoreTTL, self).destroy()
class Raft(AbstractDCS):
@@ -270,34 +248,16 @@ class Raft(AbstractDCS):
super(Raft, self).__init__(config)
self._ttl = int(config.get('ttl') or 30)
self_addr = config.get('self_addr')
partner_addrs = config.get('partner_addrs', [])
if self._ctl:
if self_addr:
partner_addrs.append(self_addr)
self_addr = None
# Create raft data_dir if necessary
raft_data_dir = config.get('data_dir', '')
if raft_data_dir != '':
validate_directory(raft_data_dir)
ready_event = threading.Event()
file_template = os.path.join(config.get('data_dir', ''), (self_addr or ''))
conf = SyncObjConf(password=config.get('password'), appendEntriesUseBatch=False,
bindAddress=config.get('bind_addr'), commandsWaitLeader=False,
fullDumpFile=(file_template + '.dump' if self_addr else None),
journalFile=(file_template + '.journal' if self_addr else None),
onReady=ready_event.set, dynamicMembershipChange=True)
self._sync_obj = KVStoreTTL(ready_event.set, self._on_set, self._on_delete, commandsWaitLeader=False, **config)
self._sync_obj.startAutoTick()
self._sync_obj = KVStoreTTL(self_addr, partner_addrs, conf, self._on_set, self._on_delete)
while True:
ready_event.wait(5)
if ready_event.isSet() or self._sync_obj.applied_local_log:
break
else:
logger.info('waiting on raft')
self._sync_obj.forceLogCompaction()
self.set_retry_timeout(int(config.get('retry_timeout') or 10))
def _on_set(self, key, value):
+6 -11
View File
@@ -1,9 +1,7 @@
import logging
import os
from patroni.daemon import AbstractPatroniDaemon, abstract_main
from patroni.dcs.raft import KVStoreTTL
from pysyncobj import SyncObjConf
from .daemon import AbstractPatroniDaemon, abstract_main
from .dcs.raft import KVStoreTTL
logger = logging.getLogger(__name__)
@@ -13,16 +11,13 @@ class RaftController(AbstractPatroniDaemon):
def __init__(self, config):
super(RaftController, self).__init__(config)
raft_config = self.config.get('raft')
self_addr = raft_config['self_addr']
template = os.path.join(raft_config.get('data_dir', ''), self_addr)
self._syncobj_config = SyncObjConf(autoTick=False, appendEntriesUseBatch=False, dynamicMembershipChange=True,
journalFile=template + '.journal', fullDumpFile=template + '.dump')
self._raft = KVStoreTTL(self_addr, raft_config.get('partner_addrs', []), self._syncobj_config)
config = self.config.get('raft')
assert 'self_addr' in config
self._raft = KVStoreTTL(None, None, None, **config)
def _run_cycle(self):
try:
self._raft.doTick(self._syncobj_config.autoTickPeriod)
self._raft.doTick(self._raft.conf.autoTickPeriod)
except Exception:
logger.exception('doTick')
+2 -1
View File
@@ -9,6 +9,7 @@ python-consul>=0.7.1
click>=4.1
prettytable>=0.7
python-dateutil
pysyncobj>=0.3.7
pysyncobj>=0.3.8
cryptography>=1.4
psutil>=2.0.0
ydiff>=1.2.0
+12 -5
View File
@@ -22,8 +22,9 @@ AUTHOR_EMAIL = '[email protected], [email protected], alexk
KEYWORDS = 'etcd governor patroni postgresql postgres ha haproxy confd' +\
' zookeeper exhibitor consul streaming replication kubernetes k8s'
EXTRAS_REQUIRE = {'aws': ['boto'], 'etcd': ['python-etcd'], 'etcd3': ['python-etcd'], 'consul': ['python-consul'],
'exhibitor': ['kazoo'], 'zookeeper': ['kazoo'], 'kubernetes': ['ipaddress'], 'raft': ['pysyncobj']}
EXTRAS_REQUIRE = {'aws': ['boto'], 'etcd': ['python-etcd'], 'etcd3': ['python-etcd'],
'consul': ['python-consul'], 'exhibitor': ['kazoo'], 'zookeeper': ['kazoo'],
'kubernetes': ['ipaddress'], 'raft': ['pysyncobj', 'cryptography']}
COVERAGE_XML = True
COVERAGE_HTML = False
@@ -171,10 +172,16 @@ def setup_package(version):
if r == '':
continue
extra = False
for e, v in EXTRAS_REQUIRE.items():
if v and r.startswith(v[0]):
EXTRAS_REQUIRE[e] = [r] if e != 'kubernetes' or sys.version_info < (3, 0, 0) else []
for e, deps in EXTRAS_REQUIRE.items():
for i, v in enumerate(deps):
if r.startswith(v):
if e != 'kubernetes' or sys.version_info < (3, 0, 0):
deps[i] = r
else:
deps = []
EXTRAS_REQUIRE[e] = deps
extra = True
break
if not extra:
install_requires.append(r)
+18 -32
View File
@@ -3,13 +3,13 @@ import unittest
import tempfile
import time
from mock import Mock, patch
from mock import Mock, PropertyMock, patch
from patroni.dcs.raft import DynMemberSyncObj, KVStoreTTL, Raft, SyncObjUtility
from pysyncobj import SyncObjConf, FAIL_REASON
def remove_files(prefix):
for f in ('journal', 'dump'):
for f in ('journal', 'journal.meta', 'dump'):
f = prefix + f
if os.path.isfile(f):
for i in range(0, 15):
@@ -31,50 +31,38 @@ class TestDynMemberSyncObj(unittest.TestCase):
self.conf = SyncObjConf(appendEntriesUseBatch=False, dynamicMembershipChange=True, autoTick=False)
self.so = DynMemberSyncObj('127.0.0.1:1234', ['127.0.0.1:1235'], self.conf)
@patch.object(SyncObjUtility, 'sendMessage')
def test_add_member(self, mock_send_message):
mock_send_message.return_value = [{'addr': '127.0.0.1:1235'}, {'addr': '127.0.0.1:1236'}]
mock_send_message.ver = 0
@patch.object(SyncObjUtility, 'executeCommand')
def test_add_member(self, mock_execute_command):
mock_execute_command.return_value = [{'addr': '127.0.0.1:1235'}, {'addr': '127.0.0.1:1236'}]
mock_execute_command.ver = 0
DynMemberSyncObj('127.0.0.1:1234', ['127.0.0.1:1235'], self.conf)
self.conf.dynamicMembershipChange = False
DynMemberSyncObj('127.0.0.1:1234', ['127.0.0.1:1235'], self.conf)
def test___onUtilityMessage(self):
self.so._SyncObj__encryptor = Mock()
def test_getMembers(self):
mock_conn = Mock()
mock_conn.sendRandKey = None
self.so._SyncObj__transport._onIncomingMessageReceived(mock_conn, 'randkey')
self.so._SyncObj__transport._onIncomingMessageReceived(mock_conn, ['members'])
self.so._SyncObj__transport._onIncomingMessageReceived(mock_conn, ['status'])
def test__SyncObj__doChangeCluster(self):
self.so._SyncObj__doChangeCluster(['add', '127.0.0.1:1236'])
def test_utility(self):
utility = SyncObjUtility(['127.0.0.1:1235'], self.conf)
utility.setPartnerNode(list(utility._SyncObj__otherNodes)[0])
utility.sendMessage(['members'])
utility._onMessageReceived(0, '')
@patch.object(SyncObjConf, 'fullDumpFile', PropertyMock(return_value=None), create=True)
@patch.object(SyncObjConf, 'journalFile', PropertyMock(return_value=None), create=True)
class TestKVStoreTTL(unittest.TestCase):
@patch.object(SyncObjConf, 'fullDumpFile', PropertyMock(return_value=None), create=True)
@patch.object(SyncObjConf, 'journalFile', PropertyMock(return_value=None), create=True)
def setUp(self):
self.conf = SyncObjConf(appendEntriesUseBatch=False, appendEntriesPeriod=0.001,
raftMinTimeout=0.004, raftMaxTimeout=0.005, autoTickPeriod=0.001)
callback = Mock()
callback.replicated = False
self.so = KVStoreTTL('127.0.0.1:1234', [], self.conf, on_set=callback, on_delete=callback)
self.so = KVStoreTTL(None, callback, callback, self_addr='127.0.0.1:1234')
self.so.startAutoTick()
self.so.set_retry_timeout(10)
@staticmethod
def destroy(so):
so.destroy()
so._SyncObj__thread.join()
def tearDown(self):
if self.so:
self.destroy(self.so)
self.so.destroy()
def test_set(self):
self.assertTrue(self.so.set('foo', 'bar', prevExist=False, ttl=30))
@@ -83,7 +71,7 @@ class TestKVStoreTTL(unittest.TestCase):
self.assertTrue(self.so.retry(self.so._set, 'foo', {'value': 'buz', 'created': 1, 'updated': 1}))
def test_delete(self):
self.conf.autoTickPeriod = 0.1
self.so.autoTickPeriod = 0.2
self.so.set('foo', 'bar')
self.so.set('fooo', 'bar')
self.assertFalse(self.so.delete('foo', prevValue='buz'))
@@ -111,11 +99,10 @@ class TestKVStoreTTL(unittest.TestCase):
def test_on_ready_override(self):
self.assertTrue(self.so.set('foo', 'bar'))
self.destroy(self.so)
self.so.destroy()
self.so = None
self.conf.onReady = Mock()
self.conf.autoTick = False
so = KVStoreTTL('127.0.0.1:1234', ['127.0.0.1:1235'], self.conf)
so = KVStoreTTL(Mock(), None, None, self_addr='127.0.0.1:1234',
partner_addrs=['127.0.0.1:1235'], patronictl=True)
so.doTick(0)
so.destroy()
@@ -150,7 +137,6 @@ class TestRaft(unittest.TestCase):
self.assertTrue(raft.take_leader())
raft.watch(None, 0.001)
raft._sync_obj.destroy()
raft._sync_obj._SyncObj__thread.join()
def tearDown(self):
remove_files(os.path.join(self._TMP, '127.0.0.1:1234.'))