class BackupApplication(object):
pt = None
server = None # like in BaseApplication
uuid = None
def __init__(self, app, name, master_addresses):
self.app = weakref.proxy(app)
self.name = name
self.nm = NodeManager()
for master_address in master_addresses:
self.nm.createMaster(address=master_address)
em = property(lambda self: self.app.em)
ssl = property(lambda self: self.app.ssl)
def close(self):
self.nm.close()
del self.__dict__
def setUUID(self, uuid):
if self.uuid != uuid:
self.uuid = uuid
logging.info('Upstream Node ID: %s', uuid_str(uuid))
def log(self):
self.nm.log()
if self.pt is not None:
self.pt.log()
def provideService(self):
logging.info('provide backup')
poll = self.em.poll
app = self.app
pt = app.pt
while True:
app.changeClusterState(ClusterStates.STARTING_BACKUP)
bootstrap = BootstrapManager(self,
NodeTypes.CLIENT,
backup=app.name)
# {offset -> node}
self.primary_partition_dict = {}
# [[tid]]
self.tid_list = tuple([] for _ in xrange(pt.getPartitions()))
try:
while True:
for node in pt.getNodeSet(readable=True):
if not app.isStorageReady(node.getUUID()):
break
else:
break
poll(1)
node, conn = bootstrap.getPrimaryConnection()
try:
app.changeClusterState(ClusterStates.BACKINGUP)
del bootstrap, node
self.ignore_invalidations = True
conn.setHandler(BackupHandler(self))
conn.ask(Packets.AskLastTransaction())
# debug variable to log how big 'tid_list' can be.
self.debug_tid_count = 0
while True:
poll(1)
except PrimaryFailure, msg:
logging.error('upstream master is down: %s', msg)
finally:
app.backup_tid = pt.getBackupTid()
try:
conn.close()
except PrimaryFailure:
pass
try:
del self.pt
except AttributeError:
pass
for node in app.nm.getClientList(True):
node.getConnection().close()
except StateChangedException, e:
if e.args[0] != ClusterStates.STOPPING_BACKUP:
raise
app.changeClusterState(*e.args)
tid = app.backup_tid
# Wait for non-primary partitions to catch up,
# so that all UP_TO_DATE cells are really UP_TO_DATE.
# XXX: Another possibility could be to outdate such cells, and
# they would be quickly updated at the beginning of the
# RUNNING phase. This may simplify code.
# Any unfinished replication from upstream will be truncated.
while pt.getBackupTid(min) < tid:
poll(1)
last_tid = app.getLastTransaction()
handler = EventHandler(app)
if tid < last_tid:
assert tid != ZERO_TID
logging.warning("Truncating at %s (last_tid was %s)",
dump(app.backup_tid), dump(last_tid))
else:
# We will do a dummy truncation, just to leave backup mode,
# so it's fine to start automatically if there's any
# missing storage.
# XXX: Consider using another method to leave backup mode,
# at least when there's nothing to truncate. Because
# in case of StoppedOperation during VERIFYING state,
# this flag will be wrongly set to False.
app._startup_allowed = True
# If any error happened before reaching this line, we'd go back
# to backup mode, which is the right mode to recover.
del app.backup_tid
# Now back to RECOVERY...
return tid
finally:
class BackupApplication(object):
pt = None
def __init__(self, app, name, master_addresses):
self.app = weakref.proxy(app)
self.name = name
self.nm = NodeManager()
for master_address in master_addresses:
self.nm.createMaster(address=master_address)
em = property(lambda self: self.app.em)
ssl = property(lambda self: self.app.ssl)
def close(self):
self.nm.close()
del self.__dict__
def log(self):
self.nm.log()
if self.pt is not None:
self.pt.log()
def provideService(self):
logging.info('provide backup')
poll = self.em.poll
app = self.app
pt = app.pt
while True:
app.changeClusterState(ClusterStates.STARTING_BACKUP)
bootstrap = BootstrapManager(self, self.name, NodeTypes.CLIENT)
# {offset -> node}
self.primary_partition_dict = {}
# [[tid]]
self.tid_list = tuple([] for _ in xrange(pt.getPartitions()))
try:
while True:
for node in pt.getNodeSet(readable=True):
if not app.isStorageReady(node.getUUID()):
break
else:
break
poll(1)
node, conn, uuid, num_partitions, num_replicas = \
bootstrap.getPrimaryConnection()
try:
app.changeClusterState(ClusterStates.BACKINGUP)
del bootstrap, node
if num_partitions != pt.getPartitions():
raise RuntimeError("inconsistent number of partitions")
self.pt = PartitionTable(num_partitions, num_replicas)
conn.setHandler(BackupHandler(self))
conn.ask(Packets.AskNodeInformation())
conn.ask(Packets.AskPartitionTable())
conn.ask(Packets.AskLastTransaction())
# debug variable to log how big 'tid_list' can be.
self.debug_tid_count = 0
while True:
poll(1)
except PrimaryFailure, msg:
logging.error('upstream master is down: %s', msg)
finally:
app.backup_tid = pt.getBackupTid()
try:
conn.close()
except PrimaryFailure:
pass
try:
del self.pt
except AttributeError:
pass
except StateChangedException, e:
if e.args[0] != ClusterStates.STOPPING_BACKUP:
raise
app.changeClusterState(*e.args)
tid = app.backup_tid
# Wait for non-primary partitions to catch up,
# so that all UP_TO_DATE cells are really UP_TO_DATE.
# XXX: Another possibility could be to outdate such cells, and
# they would be quickly updated at the beginning of the
# RUNNING phase. This may simplify code.
# Any unfinished replication from upstream will be truncated.
while pt.getBackupTid(min) < tid:
poll(1)
last_tid = app.getLastTransaction()
handler = EventHandler(app)
if tid < last_tid:
assert tid != ZERO_TID
logging.warning("Truncating at %s (last_tid was %s)",
dump(app.backup_tid), dump(last_tid))
else:
# We will do a dummy truncation, just to leave backup mode,
# so it's fine to start automatically if there's any
# missing storage.
# XXX: Consider using another method to leave backup mode,
# at least when there's nothing to truncate. Because
# in case of StoppedOperation during VERIFYING state,
# this flag will be wrongly set to False.
app._startup_allowed = True
# If any error happened before reaching this line, we'd go back
# to backup mode, which is the right mode to recover.
del app.backup_tid
# Now back to RECOVERY...
return tid
finally:
