def test_05_getCellList(self):
num_partitions = 5
num_replicas = 2
pt = PartitionTable(num_partitions, num_replicas)
# add two kind of node, usable and unsable
uuid1 = self.getStorageUUID()
server1 = ("127.0.0.1", 19001)
sn1 = StorageNode(Mock(), server1, uuid1)
pt.setCell(0, sn1, CellStates.UP_TO_DATE)
uuid2 = self.getStorageUUID()
server2 = ("127.0.0.2", 19001)
sn2 = StorageNode(Mock(), server2, uuid2)
pt.setCell(0, sn2, CellStates.OUT_OF_DATE)
uuid3 = self.getStorageUUID()
server3 = ("127.0.0.3", 19001)
sn3 = StorageNode(Mock(), server3, uuid3)
pt.setCell(0, sn3, CellStates.FEEDING)
uuid4 = self.getStorageUUID()
server4 = ("127.0.0.4", 19001)
sn4 = StorageNode(Mock(), server4, uuid4)
pt.setCell(0, sn4, CellStates.DISCARDED) # won't be added
# now checks result
self.assertEqual(len(pt.partition_list[0]), 3)
for x in xrange(num_partitions):
if x == 0:
# all nodes
all_cell = pt.getCellList(0)
all_nodes = [x.getNode() for x in all_cell]
self.assertEqual(len(all_cell), 3)
self.assertTrue(sn1 in all_nodes)
self.assertTrue(sn2 in all_nodes)
self.assertTrue(sn3 in all_nodes)
self.assertTrue(sn4 not in all_nodes)
# readable nodes
all_cell = pt.getCellList(0, readable=True)
all_nodes = [x.getNode() for x in all_cell]
self.assertEqual(len(all_cell), 2)
self.assertTrue(sn1 in all_nodes)
self.assertTrue(sn2 not in all_nodes)
self.assertTrue(sn3 in all_nodes)
self.assertTrue(sn4 not in all_nodes)
else:
self.assertEqual(len(pt.getCellList(1, False)), 0)
self.assertEqual(len(pt.getCellList(1, True)), 0)
def test_05_getCellList(self):
num_partitions = 5
num_replicas = 2
pt = PartitionTable(num_partitions, num_replicas)
# add two kind of node, usable and unusable
uuid1 = self.getStorageUUID()
server1 = ("127.0.0.1", 19001)
sn1 = self.createStorage(server1, uuid1)
pt._setCell(0, sn1, CellStates.UP_TO_DATE)
uuid2 = self.getStorageUUID()
server2 = ("127.0.0.2", 19001)
sn2 = self.createStorage(server2, uuid2)
pt._setCell(0, sn2, CellStates.OUT_OF_DATE)
uuid3 = self.getStorageUUID()
server3 = ("127.0.0.3", 19001)
sn3 = self.createStorage(server3, uuid3)
pt._setCell(0, sn3, CellStates.FEEDING)
uuid4 = self.getStorageUUID()
server4 = ("127.0.0.4", 19001)
sn4 = self.createStorage(server4, uuid4)
pt._setCell(0, sn4, CellStates.DISCARDED) # won't be added
# now checks result
self.assertEqual(len(pt.partition_list[0]), 3)
for x in xrange(num_partitions):
if x == 0:
# all nodes
all_cell = pt.getCellList(0)
all_nodes = [x.getNode() for x in all_cell]
self.assertEqual(len(all_cell), 3)
self.assertTrue(sn1 in all_nodes)
self.assertTrue(sn2 in all_nodes)
self.assertTrue(sn3 in all_nodes)
self.assertTrue(sn4 not in all_nodes)
# readable nodes
all_cell = pt.getCellList(0, readable=True)
all_nodes = [x.getNode() for x in all_cell]
self.assertEqual(len(all_cell), 2)
self.assertTrue(sn1 in all_nodes)
self.assertTrue(sn2 not in all_nodes)
self.assertTrue(sn3 in all_nodes)
self.assertTrue(sn4 not in all_nodes)
else:
self.assertEqual(len(pt.getCellList(1, False)), 0)
self.assertEqual(len(pt.getCellList(1, True)), 0)
class Application(BaseApplication):
"""The storage node application."""
def __init__(self, config):
super(Application, self).__init__(
config.getSSL(), config.getDynamicMasterList())
for address in config.getMasters():
self.nm.createMaster(address=address)
self.name = config.getCluster()
self.server = config.getBind()
logging.debug('IP address is %s, port is %d', *self.server)
# The partition table is initialized after getting the number of
# partitions.
self.pt = None
self.uuid = config.getUUID()
self.request_handler = MasterRequestEventHandler(self)
self.master_event_handler = MasterEventHandler(self)
self.cluster_state = None
self.reset()
registerLiveDebugger(on_log=self.log)
def close(self):
self.listening_conn = None
super(Application, self).close()
def reset(self):
self.bootstrapped = False
self.master_conn = None
self.master_node = None
def log(self):
self.em.log()
self.nm.log()
if self.pt is not None:
self.pt.log()
def run(self):
try:
self._run()
except Exception:
logging.exception('Pre-mortem data:')
self.log()
logging.flush()
raise
def _run(self):
"""Make sure that the status is sane and start a loop."""
if len(self.name) == 0:
raise RuntimeError, 'cluster name must be non-empty'
# Make a listening port.
handler = AdminEventHandler(self)
self.listening_conn = ListeningConnection(self, handler, self.server)
while self.cluster_state != ClusterStates.STOPPING:
self.connectToPrimary()
try:
while True:
self.em.poll(1)
except PrimaryFailure:
logging.error('primary master is down')
self.listening_conn.close()
while not self.em.isIdle():
self.em.poll(1)
def connectToPrimary(self):
"""Find a primary master node, and connect to it.
If a primary master node is not elected or ready, repeat
the attempt of a connection periodically.
Note that I do not accept any connection from non-master nodes
at this stage.
"""
self.cluster_state = None
# search, find, connect and identify to the primary master
bootstrap = BootstrapManager(self, NodeTypes.ADMIN, self.server)
self.master_node, self.master_conn, num_partitions, num_replicas = \
bootstrap.getPrimaryConnection()
if self.pt is None:
self.pt = PartitionTable(num_partitions, num_replicas)
elif self.pt.getPartitions() != num_partitions:
# XXX: shouldn't we recover instead of raising ?
raise RuntimeError('the number of partitions is inconsistent')
elif self.pt.getReplicas() != num_replicas:
# XXX: shouldn't we recover instead of raising ?
raise RuntimeError('the number of replicas is inconsistent')
# passive handler
self.master_conn.setHandler(self.master_event_handler)
self.master_conn.ask(Packets.AskClusterState())
self.master_conn.ask(Packets.AskPartitionTable())
def sendPartitionTable(self, conn, min_offset, max_offset, uuid):
# we have a pt
self.pt.log()
row_list = []
if max_offset == 0:
max_offset = self.pt.getPartitions()
try:
for offset in xrange(min_offset, max_offset):
row = []
try:
for cell in self.pt.getCellList(offset):
if uuid is None or cell.getUUID() == uuid:
row.append((cell.getUUID(), cell.getState()))
except TypeError:
pass
row_list.append((offset, row))
except IndexError:
conn.send(Errors.ProtocolError('invalid partition table offset'))
else:
conn.answer(Packets.AnswerPartitionList(self.pt.getID(), row_list))
class Application(BaseApplication):
"""The storage node application."""
def __init__(self, config):
super(Application, self).__init__(
config.getSSL(), config.getDynamicMasterList())
for address in config.getMasters():
self.nm.createMaster(address=address)
self.name = config.getCluster()
self.server = config.getBind()
logging.debug('IP address is %s, port is %d', *self.server)
# The partition table is initialized after getting the number of
# partitions.
self.pt = None
self.uuid = config.getUUID()
self.request_handler = MasterRequestEventHandler(self)
self.master_event_handler = MasterEventHandler(self)
self.cluster_state = None
self.reset()
registerLiveDebugger(on_log=self.log)
def close(self):
self.listening_conn = None
super(Application, self).close()
def reset(self):
self.bootstrapped = False
self.master_conn = None
self.master_node = None
def log(self):
self.em.log()
self.nm.log()
if self.pt is not None:
self.pt.log()
def run(self):
try:
self._run()
except Exception:
logging.exception('Pre-mortem data:')
self.log()
logging.flush()
raise
def _run(self):
"""Make sure that the status is sane and start a loop."""
if len(self.name) == 0:
raise RuntimeError, 'cluster name must be non-empty'
# Make a listening port.
handler = AdminEventHandler(self)
self.listening_conn = ListeningConnection(self, handler, self.server)
while self.cluster_state != ClusterStates.STOPPING:
self.connectToPrimary()
try:
while True:
self.em.poll(1)
except PrimaryFailure:
logging.error('primary master is down')
self.listening_conn.close()
while not self.em.isIdle():
self.em.poll(1)
def connectToPrimary(self):
"""Find a primary master node, and connect to it.
If a primary master node is not elected or ready, repeat
the attempt of a connection periodically.
Note that I do not accept any connection from non-master nodes
at this stage.
"""
self.cluster_state = None
# search, find, connect and identify to the primary master
bootstrap = BootstrapManager(self, self.name, NodeTypes.ADMIN,
self.uuid, self.server)
data = bootstrap.getPrimaryConnection()
(node, conn, uuid, num_partitions, num_replicas) = data
self.master_node = node
self.master_conn = conn
self.uuid = uuid
if self.pt is None:
self.pt = PartitionTable(num_partitions, num_replicas)
elif self.pt.getPartitions() != num_partitions:
# XXX: shouldn't we recover instead of raising ?
raise RuntimeError('the number of partitions is inconsistent')
elif self.pt.getReplicas() != num_replicas:
# XXX: shouldn't we recover instead of raising ?
raise RuntimeError('the number of replicas is inconsistent')
# passive handler
self.master_conn.setHandler(self.master_event_handler)
self.master_conn.ask(Packets.AskClusterState())
self.master_conn.ask(Packets.AskNodeInformation())
self.master_conn.ask(Packets.AskPartitionTable())
def sendPartitionTable(self, conn, min_offset, max_offset, uuid):
# we have a pt
self.pt.log()
row_list = []
if max_offset == 0:
max_offset = self.pt.getPartitions()
try:
for offset in xrange(min_offset, max_offset):
row = []
try:
for cell in self.pt.getCellList(offset):
if uuid is None or cell.getUUID() == uuid:
row.append((cell.getUUID(), cell.getState()))
except TypeError:
pass
row_list.append((offset, row))
except IndexError:
conn.notify(Errors.ProtocolError('invalid partition table offset'))
else:
conn.answer(Packets.AnswerPartitionList(self.pt.getID(), row_list))
