def test_initial_consensus_state(some_item, other_item, validators):
name = some_item(validators)
primary = other_item(validators)
data = ConsensusSharedData(name, validators, 0)
data.primary_name = primary
data.set_validators(validators)
# General info
assert data.name == name
# Validators
assert data.validators == validators
assert data.quorums.n == len(validators)
assert data.primary_name == primary
# View
assert data.view_no == 0
assert not data.waiting_for_new_view
# 3PC votes
assert data.preprepared == []
assert data.prepared == []
# Checkpoints
assert data.stable_checkpoint == 0
assert list(data.checkpoints) == [data.initial_checkpoint]
def validator(view_no):
validators = genNodeNames(4)
inst_id = 0
cd = ConsensusSharedData(generateName(validators[0], inst_id), validators,
inst_id, True)
cd.pp_seq_no = 1
cd.view_no = view_no
cd.node_mode = Mode.participating
return OrderingServiceMsgValidator(data=cd)
def view_change_trigger_service(internal_bus, external_bus, timer, stasher, validators):
# TODO: Use validators fixture
data = ConsensusSharedData("some_name", genNodeNames(4), 0)
data.node_mode = Mode.participating
data.node_status = Status.started
return ViewChangeTriggerService(data=data,
timer=timer,
bus=internal_bus,
network=external_bus,
db_manager=DatabaseManager(),
stasher=stasher,
is_master_degraded=lambda: False)
def test_set_validators_change_vc_votes_quorums(validators):
name = "SomeData"
validators_changes_f = validators + ["A", "B", "C"]
data = ConsensusSharedData(name, validators, 0)
qb = data.view_change_votes._quorums
na = len(validators_changes_f)
data.set_validators(validators_changes_f)
qa = data.view_change_votes._quorums
assert qb.n != qa.n
assert qa.n == na
def primary_connection_monitor_service(internal_bus, external_bus, timer):
# TODO: Use validators fixture
nodes = genNodeNames(4)
data = ConsensusSharedData("some_name", nodes, 0)
data.node_mode = Mode.participating
data.node_status = Status.started
data.primary_name = nodes[0]
service = PrimaryConnectionMonitorService(data=data,
timer=timer,
bus=internal_bus,
network=external_bus)
internal_bus.send(PrimarySelected())
return service
def test_process_message_req_new_view_by_non_primary(
message_req_service: MessageReqService, external_bus,
data: ConsensusSharedData, new_view_message: NewView):
frm = "frm"
data.primary_name = "a" + data.name
data.new_view = new_view_message
message_req = MessageReq(
**{
f.MSG_TYPE.nm: NEW_VIEW,
f.PARAMS.nm: {
f.INST_ID.nm: data.inst_id,
f.VIEW_NO.nm: data.view_no
},
})
external_bus.process_incoming(message_req, frm)
assert len(external_bus.sent_messages) == 0
def __init__(self,
name: str,
validators: List[str],
primary_name: str,
timer: TimerService,
bus: InternalBus,
network: ExternalBus,
write_manager: WriteRequestManager = None,
bls_bft_replica: BlsBftReplica = None):
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = primary_name
self._orderer = OrderingService(data=self._data,
timer=timer,
bus=bus,
network=network,
write_manager=write_manager,
bls_bft_replica=bls_bft_replica)
self._checkpointer = CheckpointService(self._data, bus, network)
self._view_changer = ViewChangeService(self._data, timer, bus, network)
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0,
viewNo=0,
seqNoStart=0,
seqNoEnd=0,
digest='empty'))
def view_change_service(internal_bus, external_bus, timer, stasher,
validators):
# TODO: Use validators fixture
data = ConsensusSharedData("some_name", genNodeNames(4), 0)
primaries_selector = RoundRobinConstantNodesPrimariesSelector(validators)
return ViewChangeService(data, timer, internal_bus, external_bus, stasher,
primaries_selector)
def __init__(self, name: str, validators: List[str], primary_name: str,
timer: TimerService, bus: InternalBus, network: ExternalBus,
write_manager: WriteRequestManager,
bls_bft_replica: BlsBftReplica=None):
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = primary_name
config = getConfig()
stasher = StashingRouter(config.REPLICA_STASH_LIMIT, buses=[bus, network])
self._orderer = OrderingService(data=self._data,
timer=timer,
bus=bus,
network=network,
write_manager=write_manager,
bls_bft_replica=bls_bft_replica,
freshness_checker=FreshnessChecker(
freshness_timeout=config.STATE_FRESHNESS_UPDATE_INTERVAL),
stasher=stasher)
self._checkpointer = CheckpointService(self._data, bus, network, stasher,
write_manager.database_manager)
self._view_changer = ViewChangeService(self._data, timer, bus, network, stasher)
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0, viewNo=0, seqNoStart=0, seqNoEnd=0,
digest='4F7BsTMVPKFshM1MwLf6y23cid6fL3xMpazVoF9krzUw'))
def __init__(self,
name: str,
validators: List[str],
primary_name: str,
timer: TimerService,
bus: InternalBus,
network: ExternalBus,
write_manager: WriteRequestManager,
bls_bft_replica: BlsBftReplica = None):
# ToDo: Maybe ConsensusSharedData should be initiated before and passed already prepared?
self._internal_bus = bus
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = generateName(primary_name,
self._data.inst_id)
self.config = getConfig()
self.stasher = StashingRouter(self.config.REPLICA_STASH_LIMIT,
buses=[bus, network])
self._write_manager = write_manager
self._primaries_selector = RoundRobinNodeRegPrimariesSelector(
self._write_manager.node_reg_handler)
self._orderer = OrderingService(
data=self._data,
timer=timer,
bus=bus,
network=network,
write_manager=self._write_manager,
bls_bft_replica=bls_bft_replica,
freshness_checker=FreshnessChecker(
freshness_timeout=self.config.STATE_FRESHNESS_UPDATE_INTERVAL),
primaries_selector=self._primaries_selector,
stasher=self.stasher)
self._checkpointer = CheckpointService(self._data, bus, network,
self.stasher,
write_manager.database_manager)
self._view_changer = ViewChangeService(self._data, timer, bus, network,
self.stasher,
self._primaries_selector)
self._message_requestor = MessageReqService(self._data, bus, network)
self._add_ledgers()
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0,
viewNo=0,
seqNoStart=0,
seqNoEnd=0,
digest='4F7BsTMVPKFshM1MwLf6y23cid6fL3xMpazVoF9krzUw'))
# ToDo: it should be done in Zero-view stage.
write_manager.on_catchup_finished()
self._data.primaries = self._view_changer._primaries_selector.select_primaries(
self._data.view_no)
# ToDo: ugly way to understand node_reg changing
self._previous_node_reg = self._write_manager.node_reg_handler.committed_node_reg
bus.subscribe(Ordered, self.emulate_ordered_processing)
def test_process_message_req_new_view(message_req_service: MessageReqService,
external_bus, data: ConsensusSharedData,
new_view_message: NewView):
frm = "frm"
data.primary_name = data.name
data.new_view = new_view_message
message_req = MessageReq(
**{
f.MSG_TYPE.nm: NEW_VIEW,
f.PARAMS.nm: {
f.INST_ID.nm: data.inst_id,
f.VIEW_NO.nm: data.view_no
},
})
external_bus.process_incoming(message_req, frm)
assert len(external_bus.sent_messages) == 1
assert external_bus.sent_messages[0] == (MessageRep(
message_req.msg_type, message_req.params,
new_view_message._asdict()), [frm])
def validator(view_no):
validators = genNodeNames(4)
inst_id = 0
cd = ConsensusSharedData(generateName(validators[0], inst_id), validators, inst_id, True)
cd.pp_seq_no = 1
cd.view_no = view_no
cd.node_mode = Mode.participating
cd.node_status = Status.started
cd.prev_view_prepare_cert = cd.last_ordered_3pc[1]
return OrderingServiceMsgValidator(data=cd)
def __init__(self, name: str, validators: List[str], primary_name: str,
timer: TimerService, bus: InternalBus, network: ExternalBus):
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = primary_name
self._orderer = OrderingService(self._data, bus, network)
self._checkpointer = CheckpointService(self._data, bus, network)
self._view_changer = ViewChangeService(self._data, timer, bus, network)
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0,
viewNo=0,
seqNoStart=0,
seqNoEnd=0,
digest='empty'))
def test_process_message_req_new_view(message_req_service: MessageReqService,
external_bus, data: ConsensusSharedData,
new_view_message: NewView):
frm = "frm"
data.primary_name = data.name
data.new_view_votes.add_new_view(new_view_message, data.primary_name)
message_req = MessageReq(
**{
f.MSG_TYPE.nm: NEW_VIEW,
f.PARAMS.nm: {
f.INST_ID.nm: data.inst_id,
f.VIEW_NO.nm: data.view_no
},
})
external_bus.process_incoming(message_req, frm)
assert len(external_bus.sent_messages) == 1
expected_new_view = new_view_message._asdict()
expected_new_view.update(
{f.PRIMARY.nm: replica_name_to_node_name(data.primary_name)})
assert external_bus.sent_messages[0] == (MessageRep(
message_req.msg_type, message_req.params, expected_new_view), [frm])
def __init__(self,
name: str,
validators: List[str],
primary_name: str,
timer: TimerService,
bus: InternalBus,
network: ExternalBus,
write_manager: WriteRequestManager,
bls_bft_replica: BlsBftReplica = None):
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = primary_name
config = getConfig()
stasher = StashingRouter(config.REPLICA_STASH_LIMIT)
self._orderer = OrderingService(data=self._data,
timer=timer,
bus=bus,
network=network,
write_manager=write_manager,
bls_bft_replica=bls_bft_replica,
stasher=stasher)
self._checkpointer = CheckpointService(
self._data,
bus,
network,
stasher,
write_manager.database_manager,
old_stasher=FakeSomething(unstash_watermarks=lambda: None))
self._view_changer = ViewChangeService(self._data, timer, bus, network)
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0,
viewNo=0,
seqNoStart=0,
seqNoEnd=0,
digest='empty'))
class Replica(HasActionQueue, MessageProcessor):
STASHED_CHECKPOINTS_BEFORE_CATCHUP = 1
HAS_NO_PRIMARY_WARN_THRESCHOLD = 10
def __init__(self,
node: 'plenum.server.node.Node',
instId: int,
config=None,
isMaster: bool = False,
bls_bft_replica: BlsBftReplica = None,
metrics: MetricsCollector = NullMetricsCollector(),
get_current_time=None,
get_time_for_3pc_batch=None):
"""
Create a new replica.
:param node: Node on which this replica is located
:param instId: the id of the protocol instance the replica belongs to
:param isMaster: is this a replica of the master protocol instance
"""
HasActionQueue.__init__(self)
self.get_current_time = get_current_time or time.perf_counter
self.get_time_for_3pc_batch = get_time_for_3pc_batch or node.utc_epoch
# self.stats = Stats(TPCStat)
self.config = config or getConfig()
self.metrics = metrics
self.node = node
self.instId = instId
self.name = self.generateName(node.name, self.instId)
self.logger = getlogger(self.name)
self.validator = ReplicaValidator(self)
self.outBox = deque()
"""
This queue is used by the replica to send messages to its node. Replica
puts messages that are consumed by its node
"""
self.inBox = deque()
"""
This queue is used by the replica to receive messages from its node.
Node puts messages that are consumed by the replica
"""
self.inBoxStash = deque()
"""
If messages need to go back on the queue, they go here temporarily and
are put back on the queue on a state change
"""
self._is_master = isMaster
# Dictionary to keep track of the which replica was primary during each
# view. Key is the view no and value is the name of the primary
# replica during that view
self.primaryNames = OrderedDict() # type: OrderedDict[int, str]
# Flag being used for preterm exit from the loop in the method
# `processStashedMsgsForNewWaterMarks`. See that method for details.
self.consumedAllStashedMsgs = True
self._freshness_checker = FreshnessChecker(
freshness_timeout=self.config.STATE_FRESHNESS_UPDATE_INTERVAL)
self._bls_bft_replica = bls_bft_replica
self._state_root_serializer = state_roots_serializer
# Did we log a message about getting request while absence of primary
self.warned_no_primary = False
self._consensus_data = ConsensusSharedData(
self.name, self.node.poolManager.node_names_ordered_by_rank(),
self.instId, self.isMaster)
self._internal_bus = InternalBus()
self._external_bus = ExternalBus(send_handler=self.send)
self.stasher = self._init_replica_stasher()
self._subscription = Subscription()
self._bootstrap_consensus_data()
self._subscribe_to_external_msgs()
self._subscribe_to_internal_msgs()
self._checkpointer = self._init_checkpoint_service()
self._ordering_service = self._init_ordering_service()
self._message_req_service = self._init_message_req_service()
self._view_change_service = self._init_view_change_service()
for ledger_id in self.ledger_ids:
self.register_ledger(ledger_id)
@property
def internal_bus(self):
return self._internal_bus
def cleanup(self):
# Aggregate all the currently forwarded requests
req_keys = set()
for msg in self.inBox:
if isinstance(msg, ReqKey):
req_keys.add(msg.digest)
for req_queue in self._ordering_service.requestQueues.values():
for req_key in req_queue:
req_keys.add(req_key)
for pp in self._ordering_service.sent_preprepares.values():
for req_key in pp.reqIdr:
req_keys.add(req_key)
for pp in self._ordering_service.prePrepares.values():
for req_key in pp.reqIdr:
req_keys.add(req_key)
for req_key in req_keys:
if req_key in self.requests:
self.requests.ordered_by_replica(req_key)
self.requests.free(req_key)
self._ordering_service.cleanup()
self._checkpointer.cleanup()
self._subscription.unsubscribe_all()
self.stasher.unsubscribe_from_all()
@property
def external_bus(self):
return self._external_bus
@property
def isMaster(self):
return self._is_master
@isMaster.setter
def isMaster(self, value):
self._is_master = value
self._consensus_data.is_master = value
def _bootstrap_consensus_data(self):
self._consensus_data.requests = self.requests
self._consensus_data.node_mode = self.node.mode
self._consensus_data.quorums = self.quorums
def set_primaries(self, primaries):
self._consensus_data.primaries = primaries
def _subscribe_to_external_msgs(self):
# self._subscription.subscribe(self._external_bus, ReqKey, self.readyFor3PC)
pass
def _process_new_view_accepted(self, msg: NewViewAccepted):
self.clear_requests_and_fix_last_ordered()
def _subscribe_to_internal_msgs(self):
self._subscription.subscribe(self.internal_bus, Ordered,
self._send_ordered)
self._subscription.subscribe(self.internal_bus, NeedBackupCatchup,
self._caught_up_backup)
self._subscription.subscribe(self.internal_bus, ReqKey,
self.readyFor3PC)
self._subscription.subscribe(self.internal_bus, RaisedSuspicion,
self._process_suspicious_node)
self._subscription.subscribe(self.internal_bus, NewViewAccepted,
self._process_new_view_accepted)
def register_ledger(self, ledger_id):
# Using ordered set since after ordering each PRE-PREPARE,
# the request key is removed, so fast lookup and removal of
# request key is needed. Need the collection to be ordered since
# the request key needs to be removed once its ordered
if ledger_id not in self._ordering_service.requestQueues:
self._ordering_service.requestQueues[ledger_id] = OrderedSet()
if ledger_id != AUDIT_LEDGER_ID:
self._freshness_checker.register_ledger(
ledger_id=ledger_id,
initial_time=self.get_time_for_3pc_batch())
@property
def last_prepared_before_view_change(self):
return self._consensus_data.legacy_last_prepared_before_view_change
@last_prepared_before_view_change.setter
def last_prepared_before_view_change(self, lst):
self._consensus_data.legacy_last_prepared_before_view_change = lst
@property
def h(self) -> int:
return self._consensus_data.low_watermark
@property
def H(self) -> int:
return self._consensus_data.high_watermark
@property
def last_ordered_3pc(self) -> tuple:
return self._ordering_service.last_ordered_3pc
@last_ordered_3pc.setter
def last_ordered_3pc(self, key3PC):
self._ordering_service.last_ordered_3pc = key3PC
@property
def lastPrePrepareSeqNo(self):
return self._ordering_service.lastPrePrepareSeqNo
@property
def requests(self):
return self.node.requests
@property
def ledger_ids(self):
return self.node.ledger_ids
@property
def quorums(self):
return self.node.quorums
@property
def utc_epoch(self):
return self.node.utc_epoch()
@staticmethod
def generateName(nodeName: str, instId: int):
"""
Create and return the name for a replica using its nodeName and
instanceId.
Ex: Alpha:1
"""
if isinstance(nodeName, str):
# Because sometimes it is bytes (why?)
if ":" in nodeName:
# Because in some cases (for requested messages) it
# already has ':'. This should be fixed.
return nodeName
return "{}:{}".format(nodeName, instId)
@staticmethod
def getNodeName(replicaName: str):
return replicaName.split(":")[0]
@property
def isPrimary(self):
"""
Is this node primary?
:return: True if this node is primary, False if not, None if primary status not known
"""
return self._consensus_data.is_primary
@property
def hasPrimary(self):
return self.primaryName is not None
@property
def primaryName(self):
"""
Name of the primary replica of this replica's instance
:return: Returns name if primary is known, None otherwise
"""
return self._consensus_data.primary_name
@primaryName.setter
def primaryName(self, value: Optional[str]) -> None:
"""
Set the value of isPrimary.
:param value: the value to set isPrimary to
"""
if value is not None:
self.warned_no_primary = False
self.primaryNames[self.viewNo] = value
self.compact_primary_names()
if value != self._consensus_data.primary_name:
self._consensus_data.primary_name = value
self.logger.info(
"{} setting primaryName for view no {} to: {}".format(
self, self.viewNo, value))
if value is None:
# Since the GC needs to happen after a primary has been
# decided.
return
self._gc_before_new_view()
def compact_primary_names(self):
min_allowed_view_no = self.viewNo - 1
views_to_remove = []
for view_no in self.primaryNames:
if view_no >= min_allowed_view_no:
break
views_to_remove.append(view_no)
for view_no in views_to_remove:
self.primaryNames.pop(view_no)
def primaryChanged(self, primaryName):
self._ordering_service.batches.clear()
if self.isMaster:
# Since there is no temporary state data structure and state root
# is explicitly set to correct value
for lid in self.ledger_ids:
try:
ledger = self.node.getLedger(lid)
except KeyError:
continue
ledger.reset_uncommitted()
self.primaryName = primaryName
self._setup_for_non_master_after_view_change(self.viewNo)
def on_view_change_start(self):
if self.isMaster:
lst = self._ordering_service.l_last_prepared_certificate_in_view()
self._consensus_data.legacy_last_prepared_before_view_change = lst
self.logger.info(
'{} setting last prepared for master to {}'.format(self, lst))
def on_view_change_done(self):
if self.isMaster:
self.last_prepared_before_view_change = None
self.stasher.process_all_stashed(STASH_VIEW_3PC)
def _clear_all_3pc_msgs(self):
self._ordering_service._clear_all_3pc_msgs()
def clear_requests_and_fix_last_ordered(self):
if self.isMaster:
return
self._internal_bus.send(CheckpointStabilized(self.last_ordered_3pc))
self._ordering_service._lastPrePrepareSeqNo = 0
self._ordering_service.last_ordered_3pc = (self.viewNo, 0)
self._clear_all_3pc_msgs()
def on_propagate_primary_done(self):
if self.isMaster:
# if this is a Primary that is re-connected (that is view change is not actually changed,
# we just propagate it, then make sure that we did't break the sequence
# of ppSeqNo
self._checkpointer.update_watermark_from_3pc()
elif not self.isPrimary:
self._checkpointer.set_watermarks(low_watermark=0,
high_watermark=sys.maxsize)
def get_lowest_probable_prepared_certificate_in_view(
self, view_no) -> Optional[int]:
"""
Return lowest pp_seq_no of the view for which can be prepared but
choose from unprocessed PRE-PREPAREs and PREPAREs.
"""
# TODO: Naive implementation, dont need to iterate over the complete
# data structures, fix this later
seq_no_pp = SortedList() # pp_seq_no of PRE-PREPAREs
# pp_seq_no of PREPAREs with count of PREPAREs for each
seq_no_p = set()
for (v, p) in self._ordering_service.prePreparesPendingPrevPP:
if v == view_no:
seq_no_pp.add(p)
if v > view_no:
break
for (
v, p
), pr in self._ordering_service.preparesWaitingForPrePrepare.items():
if v == view_no and len(pr) >= self.quorums.prepare.value:
seq_no_p.add(p)
for n in seq_no_pp:
if n in seq_no_p:
return n
return None
def _setup_for_non_master_after_view_change(self, current_view):
if not self.isMaster:
for v in list(self.stashed_out_of_order_commits.keys()):
if v < current_view:
self.stashed_out_of_order_commits.pop(v)
def is_primary_in_view(self, viewNo: int) -> Optional[bool]:
"""
Return whether this replica was primary in the given view
"""
if viewNo not in self.primaryNames:
return False
return self.primaryNames[viewNo] == self.name
def isMsgForCurrentView(self, msg):
"""
Return whether this request's view number is equal to the current view
number of this replica.
"""
viewNo = getattr(msg, "viewNo", None)
return viewNo == self.viewNo
def isPrimaryForMsg(self, msg) -> Optional[bool]:
"""
Return whether this replica is primary if the request's view number is
equal this replica's view number and primary has been selected for
the current view.
Return None otherwise.
:param msg: message
"""
return self.isPrimary if self.isMsgForCurrentView(msg) \
else self.is_primary_in_view(msg.viewNo)
def isMsgFromPrimary(self, msg, sender: str) -> bool:
"""
Return whether this message was from primary replica
:param msg:
:param sender:
:return:
"""
if self.isMsgForCurrentView(msg):
return self.primaryName == sender
try:
return self.primaryNames[msg.viewNo] == sender
except KeyError:
return False
def __repr__(self):
return self.name
@property
def f(self) -> int:
"""
Return the number of Byzantine Failures that can be tolerated by this
system. Equal to (N - 1)/3, where N is the number of nodes in the
system.
"""
return self.node.f
@property
def viewNo(self):
"""
Return the current view number of this replica.
"""
return self._consensus_data.view_no
@property
def stashed_out_of_order_commits(self):
# Commits which are not being ordered since commits with lower
# sequence numbers have not been ordered yet. Key is the
# viewNo and value a map of pre-prepare sequence number to commit
# type: Dict[int,Dict[int,Commit]]
return self._ordering_service.stashed_out_of_order_commits
def send_3pc_batch(self):
return self._ordering_service.send_3pc_batch()
@staticmethod
def batchDigest(reqs):
return replica_batch_digest(reqs)
def readyFor3PC(self, key: ReqKey):
try:
fin_req = self.requests[key.digest].finalised
except KeyError:
# Means that this request is outdated and is dropped from the main requests queue
self.logger.debug(
'{} reports request {} is ready for 3PC but it has been dropped '
'from requests queue, ignore this request'.format(self, key))
return
queue = self._ordering_service.requestQueues[
self.node.ledger_id_for_request(fin_req)]
queue.add(key.digest)
if not self.hasPrimary and len(
queue
) >= self.HAS_NO_PRIMARY_WARN_THRESCHOLD and not self.warned_no_primary:
self.logger.warning(
'{} is getting requests but still does not have '
'a primary so the replica will not process the request '
'until a primary is chosen'.format(self))
self.warned_no_primary = True
@measure_replica_time(MetricsName.SERVICE_REPLICA_QUEUES_TIME,
MetricsName.SERVICE_BACKUP_REPLICAS_QUEUES_TIME)
def serviceQueues(self, limit=None):
"""
Process `limit` number of messages in the inBox.
:param limit: the maximum number of messages to process
:return: the number of messages successfully processed
"""
# TODO should handle SuspiciousNode here
r = self.dequeue_pre_prepares()
# r += self.inBoxRouter.handleAllSync(self.inBox, limit)
r += self._handle_external_messages(self.inBox, limit)
r += self.send_3pc_batch()
r += self._serviceActions()
return r
# Messages that can be processed right now needs to be added back to the
# queue. They might be able to be processed later
def _handle_external_messages(self, deq: deque, limit=None) -> int:
"""
Synchronously handle all items in a deque.
:param deq: a deque of items to be handled by this router
:param limit: the number of items in the deque to the handled
:return: the number of items handled successfully
"""
count = 0
while deq and (not limit or count < limit):
count += 1
msg = deq.popleft()
external_msg, sender = msg if len(msg) == 2 else (msg, None)
# TODO: get rid of appending instId to sender
sender = self.generateName(sender, self.instId)
self._external_bus.process_incoming(external_msg, sender)
return count
def _gc_before_new_view(self):
# Trigger GC for all batches of old view
# Clear any checkpoints, since they are valid only in a view
# ToDo: Need to send a cmd like ViewChangeStart into internal bus
# self._gc(self.last_ordered_3pc)
self._ordering_service.gc(self.last_ordered_3pc)
# ToDo: get rid of directly calling
self._ordering_service._clear_prev_view_pre_prepares()
# self._clear_prev_view_pre_prepares()
def has_already_ordered(self, view_no, pp_seq_no):
return compare_3PC_keys(
(view_no, pp_seq_no), self.last_ordered_3pc) >= 0
def dequeue_pre_prepares(self):
return self._ordering_service.dequeue_pre_prepares()
def send(self, msg, to_nodes=None) -> None:
"""
Send a message to the node on which this replica resides.
:param stat:
:param rid: remote id of one recipient (sends to all recipients if None)
:param msg: the message to send
"""
self.logger.trace("{} sending {}".format(self, msg.__class__.__name__),
extra={
"cli": True,
"tags": ['sending']
})
self.logger.trace("{} sending {}".format(self, msg))
if to_nodes:
self.node.sendToNodes(msg, names=to_nodes)
return
self.outBox.append(msg)
def revert_unordered_batches(self):
"""
Revert changes to ledger (uncommitted) and state made by any requests
that have not been ordered.
"""
return self._ordering_service.revert_unordered_batches()
def on_catch_up_finished(self,
last_caught_up_3PC=None,
master_last_ordered_3PC=None):
if master_last_ordered_3PC and last_caught_up_3PC and \
compare_3PC_keys(master_last_ordered_3PC,
last_caught_up_3PC) > 0:
if self.isMaster:
self._caught_up_till_3pc(last_caught_up_3PC)
else:
self._ordering_service.first_batch_after_catchup = True
self._catchup_clear_for_backup()
self.stasher.process_all_stashed(STASH_CATCH_UP)
def discard_req_key(self, ledger_id, req_key):
return self._ordering_service.discard_req_key(ledger_id, req_key)
def _caught_up_backup(self, msg: NeedBackupCatchup):
self._caught_up_till_3pc(msg.caught_up_till_3pc)
def _caught_up_till_3pc(self, last_caught_up_3PC):
self._ordering_service._caught_up_till_3pc(last_caught_up_3PC)
self._checkpointer.caught_up_till_3pc(last_caught_up_3PC)
def _catchup_clear_for_backup(self):
if not self.isPrimary:
self.outBox.clear()
self._checkpointer.catchup_clear_for_backup()
self._ordering_service.catchup_clear_for_backup()
def _remove_ordered_from_queue(self, last_caught_up_3PC=None):
"""
Remove any Ordered that the replica might be sending to node which is
less than or equal to `last_caught_up_3PC` if `last_caught_up_3PC` is
passed else remove all ordered, needed in catchup
"""
to_remove = []
for i, msg in enumerate(self.outBox):
if isinstance(msg, Ordered) and \
(not last_caught_up_3PC or
compare_3PC_keys((msg.viewNo, msg.ppSeqNo), last_caught_up_3PC) >= 0):
to_remove.append(i)
self.logger.trace(
'{} going to remove {} Ordered messages from outbox'.format(
self, len(to_remove)))
# Removing Ordered from queue but returning `Ordered` in order that
# they should be processed.
removed = []
for i in reversed(to_remove):
removed.insert(0, self.outBox[i])
del self.outBox[i]
return removed
def _get_last_timestamp_from_state(self, ledger_id):
if ledger_id == DOMAIN_LEDGER_ID:
ts_store = self.node.db_manager.get_store(TS_LABEL)
if ts_store:
last_timestamp = ts_store.get_last_key()
if last_timestamp:
last_timestamp = int(last_timestamp.decode())
self.logger.debug(
"Last ordered timestamp from store is : {}"
"".format(last_timestamp))
return last_timestamp
return None
def get_ledgers_last_update_time(self) -> dict:
if self._freshness_checker:
return self._freshness_checker.get_last_update_time()
def get_valid_req_ids_from_all_requests(self, reqs, invalid_indices):
return [
req.key for idx, req in enumerate(reqs)
if idx not in invalid_indices
]
def report_suspicious_node(self, ex):
if self.isMaster:
self.node.reportSuspiciousNodeEx(ex)
else:
self.warn_suspicious_backup(ex.node, ex.reason, ex.code)
def warn_suspicious_backup(self, nodeName, reason, code):
self.logger.warning(
"backup replica {} raised suspicion on node {} for {}; suspicion code "
"is {}".format(self, nodeName, reason, code))
def set_validators(self, validators):
self._consensus_data.set_validators(validators)
def set_view_no(self, view_no):
self._consensus_data.view_no = view_no
def set_view_change_status(self, legacy_vc_in_progress):
self._consensus_data.legacy_vc_in_progress = legacy_vc_in_progress
def set_mode(self, mode):
self._consensus_data.node_mode = mode
def set_primaries_batch_needed(self, value):
self._ordering_service.primaries_batch_needed = value
def update_connecteds(self, connecteds: dict):
self._external_bus.update_connecteds(connecteds)
def _init_replica_stasher(self):
return StashingRouter(self.config.REPLICA_STASH_LIMIT,
buses=[self.internal_bus, self._external_bus],
unstash_handler=self._add_to_inbox)
def _process_suspicious_node(self, msg: RaisedSuspicion):
if msg.inst_id != self.instId:
return
self.report_suspicious_node(msg.ex)
def _send_ordered(self, msg: Ordered):
self.send(msg)
def _init_checkpoint_service(self) -> CheckpointService:
return CheckpointService(data=self._consensus_data,
bus=self.internal_bus,
network=self._external_bus,
stasher=self.stasher,
db_manager=self.node.db_manager,
metrics=self.metrics)
def _init_ordering_service(self) -> OrderingService:
return OrderingService(
data=self._consensus_data,
timer=self.node.timer,
bus=self.internal_bus,
network=self._external_bus,
write_manager=self.node.write_manager,
bls_bft_replica=self._bls_bft_replica,
freshness_checker=self._freshness_checker,
get_current_time=self.get_current_time,
get_time_for_3pc_batch=self.get_time_for_3pc_batch,
stasher=self.stasher,
metrics=self.metrics)
def _init_message_req_service(self) -> MessageReqService:
return MessageReqService(data=self._consensus_data,
bus=self.internal_bus,
network=self._external_bus,
metrics=self.metrics)
def _init_view_change_service(self) -> ViewChangeService:
return ViewChangeService(data=self._consensus_data,
timer=self.node.timer,
bus=self.internal_bus,
network=self._external_bus,
stasher=self.stasher)
def _add_to_inbox(self, message):
self.inBox.append(message)
def _data(name):
data = ConsensusSharedData(name, validators, 0)
data.view_no = initial_view_no
data.checkpoints = initial_checkpoints
return data
def consensus_data_provider():
validators = genNodeNames(N)
return ConsensusSharedData("nodeA", validators, 0)
def view_change_service():
data = ConsensusSharedData("some_name", genNodeNames(4), 0)
return ViewChangeService(data, MockTimer(0), InternalBus(), MockNetwork())
def _data(name):
data = ConsensusSharedData(name, validators, 0, is_master)
data.view_no = initial_view_no
data.checkpoints.update(initial_checkpoints)
return data
def __init__(self,
node: 'plenum.server.node.Node',
instId: int,
config=None,
isMaster: bool = False,
bls_bft_replica: BlsBftReplica = None,
metrics: MetricsCollector = NullMetricsCollector(),
get_current_time=None,
get_time_for_3pc_batch=None):
"""
Create a new replica.
:param node: Node on which this replica is located
:param instId: the id of the protocol instance the replica belongs to
:param isMaster: is this a replica of the master protocol instance
"""
HasActionQueue.__init__(self)
self.get_current_time = get_current_time or time.perf_counter
self.get_time_for_3pc_batch = get_time_for_3pc_batch or node.utc_epoch
# self.stats = Stats(TPCStat)
self.config = config or getConfig()
self.metrics = metrics
self.node = node
self.instId = instId
self.name = self.generateName(node.name, self.instId)
self.logger = getlogger(self.name)
self.validator = ReplicaValidator(self)
self.outBox = deque()
"""
This queue is used by the replica to send messages to its node. Replica
puts messages that are consumed by its node
"""
self.inBox = deque()
"""
This queue is used by the replica to receive messages from its node.
Node puts messages that are consumed by the replica
"""
self.inBoxStash = deque()
"""
If messages need to go back on the queue, they go here temporarily and
are put back on the queue on a state change
"""
self._is_master = isMaster
# Dictionary to keep track of the which replica was primary during each
# view. Key is the view no and value is the name of the primary
# replica during that view
self.primaryNames = OrderedDict() # type: OrderedDict[int, str]
# Flag being used for preterm exit from the loop in the method
# `processStashedMsgsForNewWaterMarks`. See that method for details.
self.consumedAllStashedMsgs = True
self._freshness_checker = FreshnessChecker(
freshness_timeout=self.config.STATE_FRESHNESS_UPDATE_INTERVAL)
self._bls_bft_replica = bls_bft_replica
self._state_root_serializer = state_roots_serializer
# Did we log a message about getting request while absence of primary
self.warned_no_primary = False
self._consensus_data = ConsensusSharedData(
self.name, self.node.poolManager.node_names_ordered_by_rank(),
self.instId, self.isMaster)
self._internal_bus = InternalBus()
self._external_bus = ExternalBus(send_handler=self.send)
self.stasher = self._init_replica_stasher()
self._subscription = Subscription()
self._bootstrap_consensus_data()
self._subscribe_to_external_msgs()
self._subscribe_to_internal_msgs()
self._checkpointer = self._init_checkpoint_service()
self._ordering_service = self._init_ordering_service()
self._message_req_service = self._init_message_req_service()
self._view_change_service = self._init_view_change_service()
for ledger_id in self.ledger_ids:
self.register_ledger(ledger_id)
def _data(name):
data = ConsensusSharedData(generateName(name, 0), validators, 0, is_master)
data.view_no = initial_view_no
return data
def view_change_service(internal_bus, external_bus, timer, stasher):
data = ConsensusSharedData("some_name", genNodeNames(4), 0)
return ViewChangeService(data, timer, internal_bus, external_bus, stasher)
def __init__(self, name: str, validators: List[str], primary_name: str,
timer: TimerService, bus: InternalBus, network: ExternalBus,
write_manager: WriteRequestManager,
bls_bft_replica: BlsBftReplica = None):
# ToDo: Maybe ConsensusSharedData should be initiated before and passed already prepared?
self._network = network
self._data = ConsensusSharedData(name, validators, 0)
self._data.primary_name = generateName(primary_name, self._data.inst_id)
self._timer = timer
self.config = getConfig()
self.stasher = StashingRouter(self.config.REPLICA_STASH_LIMIT, buses=[bus, network])
self._write_manager = write_manager
self._primaries_selector = RoundRobinNodeRegPrimariesSelector(self._write_manager.node_reg_handler)
self._freshness_checker = FreshnessChecker(freshness_timeout=self.config.STATE_FRESHNESS_UPDATE_INTERVAL)
for ledger_id in [POOL_LEDGER_ID, DOMAIN_LEDGER_ID, CONFIG_LEDGER_ID]:
self._freshness_checker.register_ledger(ledger_id=ledger_id,
initial_time=self.get_time_for_3pc_batch())
self._orderer = OrderingService(data=self._data,
timer=self._timer,
bus=bus,
network=network,
write_manager=self._write_manager,
bls_bft_replica=bls_bft_replica,
freshness_checker=self._freshness_checker,
get_time_for_3pc_batch=self.get_time_for_3pc_batch,
stasher=self.stasher)
self._checkpointer = CheckpointService(self._data, bus, network, self.stasher,
write_manager.database_manager)
self._view_changer = ViewChangeService(self._data, self._timer, bus, network, self.stasher, self._primaries_selector)
self._view_change_trigger = ViewChangeTriggerService(data=self._data,
timer=self._timer,
bus=bus,
network=network,
db_manager=write_manager.database_manager,
is_master_degraded=lambda: False,
stasher=self.stasher)
self._primary_connection_monitor = PrimaryConnectionMonitorService(data=self._data,
timer=self._timer,
bus=bus,
network=network)
self._freshness_monitor = FreshnessMonitorService(data=self._data,
timer=self._timer,
bus=bus,
network=network,
freshness_checker=self._freshness_checker,
get_time_for_3pc_batch=self.get_time_for_3pc_batch)
self._message_requestor = MessageReqService(self._data, bus, network)
self._add_ledgers()
# TODO: This is just for testing purposes only
self._data.checkpoints.append(
Checkpoint(instId=0, viewNo=0, seqNoStart=0, seqNoEnd=0,
digest='4F7BsTMVPKFshM1MwLf6y23cid6fL3xMpazVoF9krzUw'))
# ToDo: it should be done in Zero-view stage.
write_manager.on_catchup_finished()
# Simulate node behavior
self._internal_bus = bus
self._internal_bus.subscribe(NodeNeedViewChange, self.process_node_need_view_change)
self._internal_bus.subscribe(Ordered, self.emulate_ordered_processing)
# ToDo: ugly way to understand node_reg changing
self._previous_node_reg = self._write_manager.node_reg_handler.committed_node_reg
def patched_set_validators(self, validators):
ConsensusSharedData.set_validators(self, validators)
self.quorums.commit = Quorum(nodeCount)
def consensus_data_helper(validators):
return ConsensusDataHelper(ConsensusSharedData('sample', validators, 0))