class Application(BaseApplication):
"""The master node application."""
packing = None
# Latest completely commited TID
last_transaction = ZERO_TID
backup_tid = None
backup_app = None
uuid = None
truncate_tid = None
def __init__(self, config):
super(Application, self).__init__(
config.getSSL(), config.getDynamicMasterList())
self.tm = TransactionManager(self.onTransactionCommitted)
self.name = config.getCluster()
self.server = config.getBind()
self.autostart = config.getAutostart()
self.storage_readiness = set()
for master_address in config.getMasters():
self.nm.createMaster(address=master_address)
logging.debug('IP address is %s, port is %d', *self.server)
# Partition table
replicas, partitions = config.getReplicas(), config.getPartitions()
if replicas < 0:
raise RuntimeError, 'replicas must be a positive integer'
if partitions <= 0:
raise RuntimeError, 'partitions must be more than zero'
self.pt = PartitionTable(partitions, replicas)
logging.info('Configuration:')
logging.info('Partitions: %d', partitions)
logging.info('Replicas : %d', replicas)
logging.info('Name : %s', self.name)
self.listening_conn = None
self.primary = None
self.primary_master_node = None
self.cluster_state = None
uuid = config.getUUID()
if uuid:
self.uuid = uuid
# election related data
self.unconnected_master_node_set = set()
self.negotiating_master_node_set = set()
self.master_address_dict = weakref.WeakKeyDictionary()
self._current_manager = None
# backup
upstream_cluster = config.getUpstreamCluster()
if upstream_cluster:
if upstream_cluster == self.name:
raise ValueError("upstream cluster name must be"
" different from cluster name")
self.backup_app = BackupApplication(self, upstream_cluster,
config.getUpstreamMasters())
self.administration_handler = administration.AdministrationHandler(
self)
self.secondary_master_handler = secondary.SecondaryMasterHandler(self)
self.client_service_handler = client.ClientServiceHandler(self)
self.storage_service_handler = storage.StorageServiceHandler(self)
registerLiveDebugger(on_log=self.log)
def close(self):
self.listening_conn = None
if self.backup_app is not None:
self.backup_app.close()
super(Application, self).close()
def log(self):
self.em.log()
if self.backup_app is not None:
self.backup_app.log()
self.nm.log()
self.tm.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."""
# Make a listening port.
self.listening_conn = ListeningConnection(self, None, self.server)
# Start a normal operation.
while self.cluster_state != ClusterStates.STOPPING:
# (Re)elect a new primary master.
self.primary = not self.nm.getMasterList()
if not self.primary:
self.electPrimary()
try:
if self.primary:
self.playPrimaryRole()
else:
self.playSecondaryRole()
raise RuntimeError, 'should not reach here'
except (ElectionFailure, PrimaryFailure):
# Forget all connections.
for conn in self.em.getClientList():
conn.close()
def electPrimary(self):
"""Elect a primary master node.
The difficulty is that a master node must accept connections from
others while attempting to connect to other master nodes at the
same time. Note that storage nodes and client nodes may connect
to self as well as master nodes."""
logging.info('begin the election of a primary master')
client_handler = election.ClientElectionHandler(self)
self.unconnected_master_node_set.clear()
self.negotiating_master_node_set.clear()
self.master_address_dict.clear()
self.listening_conn.setHandler(election.ServerElectionHandler(self))
getByAddress = self.nm.getByAddress
while True:
# handle new connected masters
for node in self.nm.getMasterList():
node.setUnknown()
self.unconnected_master_node_set.add(node.getAddress())
# start the election process
self.primary = None
self.primary_master_node = None
try:
while (self.unconnected_master_node_set or
self.negotiating_master_node_set):
for addr in self.unconnected_master_node_set:
self.negotiating_master_node_set.add(addr)
ClientConnection(self, client_handler,
# XXX: Ugly, but the whole election code will be
# replaced soon
getByAddress(addr))
self.unconnected_master_node_set.clear()
self.em.poll(1)
except ElectionFailure, m:
# something goes wrong, clean then restart
logging.error('election failed: %s', m)
# Ask all connected nodes to reelect a single primary master.
for conn in self.em.getClientList():
conn.notify(Packets.ReelectPrimary())
conn.abort()
# Wait until the connections are closed.
self.primary = None
self.primary_master_node = None
# XXX: Since poll does not wake up anymore every second,
# the following time condition should be reviewed.
# See also playSecondaryRole.
t = time() + 10
while self.em.getClientList() and time() < t:
try:
self.em.poll(1)
except ElectionFailure:
pass
# Close all connections.
for conn in self.em.getClientList() + self.em.getServerList():
conn.close()
else:
# election succeed, stop the process
self.primary = self.primary is None
break
class Application(BaseApplication):
"""The master node application."""
packing = None
# Latest completely commited TID
last_transaction = ZERO_TID
backup_tid = None
backup_app = None
uuid = None
truncate_tid = None
def __init__(self, config):
super(Application, self).__init__(config.getSSL(),
config.getDynamicMasterList())
self.tm = TransactionManager(self.onTransactionCommitted)
self.name = config.getCluster()
self.server = config.getBind()
self.autostart = config.getAutostart()
self.storage_readiness = set()
for master_address in config.getMasters():
self.nm.createMaster(address=master_address)
logging.debug('IP address is %s, port is %d', *self.server)
# Partition table
replicas, partitions = config.getReplicas(), config.getPartitions()
if replicas < 0:
raise RuntimeError, 'replicas must be a positive integer'
if partitions <= 0:
raise RuntimeError, 'partitions must be more than zero'
self.pt = PartitionTable(partitions, replicas)
logging.info('Configuration:')
logging.info('Partitions: %d', partitions)
logging.info('Replicas : %d', replicas)
logging.info('Name : %s', self.name)
self.listening_conn = None
self.primary = None
self.primary_master_node = None
self.cluster_state = None
uuid = config.getUUID()
if uuid:
self.uuid = uuid
# election related data
self.unconnected_master_node_set = set()
self.negotiating_master_node_set = set()
self.master_address_dict = weakref.WeakKeyDictionary()
self._current_manager = None
# backup
upstream_cluster = config.getUpstreamCluster()
if upstream_cluster:
if upstream_cluster == self.name:
raise ValueError("upstream cluster name must be"
" different from cluster name")
self.backup_app = BackupApplication(self, upstream_cluster,
config.getUpstreamMasters())
self.administration_handler = administration.AdministrationHandler(
self)
self.secondary_master_handler = secondary.SecondaryMasterHandler(self)
self.client_service_handler = client.ClientServiceHandler(self)
self.storage_service_handler = storage.StorageServiceHandler(self)
registerLiveDebugger(on_log=self.log)
def close(self):
self.listening_conn = None
if self.backup_app is not None:
self.backup_app.close()
super(Application, self).close()
def log(self):
self.em.log()
if self.backup_app is not None:
self.backup_app.log()
self.nm.log()
self.tm.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."""
# Make a listening port.
self.listening_conn = ListeningConnection(self, None, self.server)
# Start a normal operation.
while self.cluster_state != ClusterStates.STOPPING:
# (Re)elect a new primary master.
self.primary = not self.nm.getMasterList()
if not self.primary:
self.electPrimary()
try:
if self.primary:
self.playPrimaryRole()
else:
self.playSecondaryRole()
raise RuntimeError, 'should not reach here'
except (ElectionFailure, PrimaryFailure):
# Forget all connections.
for conn in self.em.getClientList():
conn.close()
def electPrimary(self):
"""Elect a primary master node.
The difficulty is that a master node must accept connections from
others while attempting to connect to other master nodes at the
same time. Note that storage nodes and client nodes may connect
to self as well as master nodes."""
logging.info('begin the election of a primary master')
client_handler = election.ClientElectionHandler(self)
self.unconnected_master_node_set.clear()
self.negotiating_master_node_set.clear()
self.master_address_dict.clear()
self.listening_conn.setHandler(election.ServerElectionHandler(self))
getByAddress = self.nm.getByAddress
while True:
# handle new connected masters
for node in self.nm.getMasterList():
node.setUnknown()
self.unconnected_master_node_set.add(node.getAddress())
# start the election process
self.primary = None
self.primary_master_node = None
try:
while (self.unconnected_master_node_set
or self.negotiating_master_node_set):
for addr in self.unconnected_master_node_set:
self.negotiating_master_node_set.add(addr)
ClientConnection(
self,
client_handler,
# XXX: Ugly, but the whole election code will be
# replaced soon
getByAddress(addr))
self.unconnected_master_node_set.clear()
self.em.poll(1)
except ElectionFailure, m:
# something goes wrong, clean then restart
logging.error('election failed: %s', m)
# Ask all connected nodes to reelect a single primary master.
for conn in self.em.getClientList():
conn.notify(Packets.ReelectPrimary())
conn.abort()
# Wait until the connections are closed.
self.primary = None
self.primary_master_node = None
# XXX: Since poll does not wake up anymore every second,
# the following time condition should be reviewed.
# See also playSecondaryRole.
t = time() + 10
while self.em.getClientList() and time() < t:
try:
self.em.poll(1)
except ElectionFailure:
pass
# Close all connections.
for conn in self.em.getClientList() + self.em.getServerList():
conn.close()
else:
# election succeed, stop the process
self.primary = self.primary is None
break
