def setUp(self):
self.tearDown()
self._connection = get_central_connection()
self._connection.begin_transaction()
purge_customer(self._connection, _test_username)
create_customer(self._connection, _test_username)
add_key_to_customer(self._connection, _test_username)
self._connection.commit()
def setUp(self):
self.tearDown()
self._connection = get_central_connection()
self._connection.begin_transaction()
purge_customer(self._connection, _test_username)
create_customer(self._connection, _test_username)
add_key_to_customer(self._connection, _test_username)
self._connection.commit()
def _load_collection_id_queue(collection_id_queue):
"""
load every known collection into the queue
"""
connection = get_central_connection()
rows = connection.fetch_all_rows(
"select id from nimbusio_central.collection order by id")
connection.close()
for (collection_id, ) in rows:
collection_id_queue.append(collection_id)
def _load_collection_id_queue(collection_id_queue):
"""
load every known collection into the queue
"""
connection = get_central_connection()
rows = connection.fetch_all_rows(
"select id from nimbusio_central.collection order by id"
)
connection.close()
for (collection_id, ) in rows:
collection_id_queue.append(collection_id)
def delete_all_motoboto_test_segments():
central_conn = get_central_connection()
local_conn = get_node_local_connection()
collection_id_rows = central_conn.fetch_all_rows(_test_collections_query, [])
central_conn.close()
local_conn.begin_transaction()
local_conn.execute("create temp table tmp_motoboto_collection_ids (id int4 not null)", [])
for row in collection_id_rows:
local_conn.execute("insert into tmp_motoboto_collection_ids values (%s)", row)
for query in _delete_test_collections_data:
rowcount = local_conn.execute(query, [])
if rowcount:
print "Deleted %s via %s" % (rowcount, query.split("\n", 1)[0])
local_conn.commit()
def main():
"""
main entry point
"""
options = _parse_command_line()
connection = get_central_connection()
try:
_dispatch_table[options.command](connection, options)
except Exception:
connection.rollback()
raise
else:
connection.commit()
finally:
connection.close()
return 0
def main():
"""
main entry point
"""
options = _parse_command_line()
connection = get_central_connection()
connection.begin_transaction()
try:
_dispatch_table[options.command](connection, options)
except Exception:
connection.rollback()
raise
else:
connection.commit()
finally:
connection.close()
return 0
def get_versioned_collections():
"""
return the set of all collection ids for which versioning is true
"""
log = logging.getLogger("get_versioned_collections")
versioned_collections = set()
connection = get_central_connection()
try:
for (collection_id, ) in connection.fetch_all_rows(_query, []):
versioned_collections.add(collection_id)
finally:
connection.close()
log.info("found {0} versioned collectons".format(
len(versioned_collections)))
return versioned_collections
def get_versioned_collections():
"""
return the set of all collection ids for which versioning is true
"""
log = logging.getLogger("get_versioned_collections")
versioned_collections = set()
connection = get_central_connection()
try:
for (collection_id, ) in connection.fetch_all_rows(_query, []):
versioned_collections.add(collection_id)
finally:
connection.close()
log.info("found {0} versioned collectons".format(
len(versioned_collections)
))
return versioned_collections
def delete_all_motoboto_test_segments():
central_conn = get_central_connection()
local_conn = get_node_local_connection()
collection_id_rows = central_conn.fetch_all_rows(_test_collections_query,
[])
central_conn.close()
local_conn.begin_transaction()
local_conn.execute(
"create temp table tmp_motoboto_collection_ids (id int4 not null)", [])
for row in collection_id_rows:
local_conn.execute(
"insert into tmp_motoboto_collection_ids values (%s)", row)
for query in _delete_test_collections_data:
rowcount = local_conn.execute(query, [])
if rowcount:
print "Deleted %s via %s" % (
rowcount,
query.split("\n", 1)[0],
)
local_conn.commit()
def _setup(_halt_event, state):
log = logging.getLogger("_setup")
# do the event push client first, because we may need to
# push an execption event from setup
state["event-push-client"] = EventPushClient(state["zmq-context"], "data_writer")
log.info("binding resilient-server to %s" % (_data_writer_address,))
state["resilient-server"] = ResilientServer(state["zmq-context"], _data_writer_address, state["receive-queue"])
state["resilient-server"].register(state["pollster"])
state["queue-dispatcher"] = DequeDispatcher(state, state["receive-queue"], _dispatch_table)
central_connection = get_central_connection()
state["cluster-row"] = get_cluster_row(central_connection)
state["node-rows"] = get_node_rows(central_connection, state["cluster-row"].id)
central_connection.close()
state["node-id-dict"] = dict([(node_row.name, node_row.id) for node_row in state["node-rows"]])
state["database-connection"] = get_node_local_connection()
# Ticket #1646 mark output value files as closed at startup
mark_value_files_as_closed(state["database-connection"])
state["writer"] = Writer(state["database-connection"], _repository_path)
state["stats-reporter"] = StatsReporter(state)
state["event-push-client"].info("program-start", "data_writer starts")
return [
(state["pollster"].run, time.time()),
(state["queue-dispatcher"].run, time.time()),
(state["stats-reporter"].run, state["stats-reporter"].next_run()),
]
def __init__(self, transaction=True):
self._log = logging.getLogger("SpaceAccountingDatabase")
self._connection = get_central_connection()
if transaction:
self._connection.execute("BEGIN;")
def main():
"""
main entry point
return 0 for success (exit code)
"""
initialize_logging(_log_path)
log = logging.getLogger("main")
log.info("program starts")
halt_event = Event()
set_signal_handler(halt_event)
zeromq_context = zmq.Context()
event_push_client = EventPushClient(zeromq_context,
"redis_stats_collector")
event_push_client.info("program-start", "flush_stats_from_redis starts")
# don't flush anything newer than 1 minute ago
current_time = datetime.utcnow()
timestamp_cutoff = current_time - timedelta(minutes=1)
return_code = 0
central_db_connection = None
collection_ops_accounting_rows = list()
# values to be added to the dedupe table
new_dedupes = list()
# keys to be deleted (a list for each node
node_keys_processed = [list() for _ in _node_names]
try:
central_db_connection = get_central_connection()
# On startup, the program connects to the central database and tries
# to acquire a pg_advisory_lock appropriate for this program and the
# data center it is running in using the pg_try_advisory_lock function.
# If it cannot acquire the lock, it notes the status of the lock
# and exits. This central locking mechanism lets us avoid single points
# of failure by configuring the program to run on multiple nodes.
with advisory_lock(central_db_connection, "redis_stats_collector"):
node_dict = _retrieve_node_dict(central_db_connection)
for node_name, keys_processed in \
zip(_node_names, node_keys_processed):
node_id = node_dict[node_name]
log.debug("processing node {0} node_id={1}".format(node_name,
node_id))
# The program then selects into memory all recently collected
# keys from the central database table
# collection_ops_accounting_flush_dedupe and stores them in a
# dedupe set. This set allows runs of the collection/flush
# program to be idempotent across some time period (
# but we won't keep the list of old keys forever.)
dedupe_set = _retrieve_dedupe_set(central_db_connection,
node_id)
# The program then visits the Redis instance on every storage
# node in the local data center, collecting the data from all
# past stats keys -- aggregating it into the program's memory.
# The aggregation should involve buckets for each
# storage_node_id and redis key, corresponding to the columns
# in the database.
_process_one_node(node_name,
node_dict[node_name],
timestamp_cutoff,
dedupe_set,
collection_ops_accounting_rows,
new_dedupes,
keys_processed)
# After collecting past keys from every storage node,
# inside a central database transaction:
# 1. Insert the collected stats into the central database
# collection_ops_accounting
# 2. Insert collected keys into recently collected keys
# collection_ops_accounting_flush_dedupe.
# 3. commit transaction
log.debug("updating central database")
central_db_connection.begin_transaction()
try:
_insert_accounting_rows(central_db_connection,
collection_ops_accounting_rows)
_insert_dedupe_rows(central_db_connection,
timestamp_cutoff,
new_dedupes)
except Exception:
central_db_connection.rollback()
raise
else:
central_db_connection.commit()
# Then revisit the Redis nodes, and delete the keys we flushed
# into the database, and any keys we skipped because they were
# found in the dedupe set.
for node_name, keys_processed in zip(_node_names,
node_keys_processed):
_remove_processed_keys(node_name, keys_processed)
except Exception as instance:
log.exception("Uhandled exception {0}".format(instance))
event_push_client.exception(
unhandled_exception_topic,
str(instance),
exctype=instance.__class__.__name__
)
return_code = 1
if central_db_connection is not None:
central_db_connection.close()
event_push_client.close()
zeromq_context.term()
log.info("program terminates return_code = {0}".format(return_code))
return return_code
def __init__(self, transaction=True):
self._log = logging.getLogger("SpaceAccountingDatabase")
self._connection = get_central_connection()
if transaction:
self._connection.execute("BEGIN;")
def __init__(self):
self._log = logging.getLogger("WebInternalReader")
memcached_client = memcache.Client(_memcached_nodes)
self._central_connection = get_central_connection()
self._cluster_row = get_cluster_row(self._central_connection)
self._node_local_connection = get_node_local_connection()
self._deliverator = Deliverator()
self._zeromq_context = zmq.Context()
self._pull_server = GreenletPULLServer(
self._zeromq_context,
_web_internal_reader_pipeline_address,
self._deliverator
)
self._pull_server.link_exception(self._unhandled_greenlet_exception)
self._data_reader_clients = list()
self._data_readers = list()
for node_name, address in zip(_node_names, _data_reader_addresses):
resilient_client = GreenletResilientClient(
self._zeromq_context,
node_name,
address,
_client_tag,
_web_internal_reader_pipeline_address,
self._deliverator,
connect_messages=[]
)
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_reader_clients.append(resilient_client)
data_reader = DataReader(
node_name, resilient_client
)
self._data_readers.append(data_reader)
self._space_accounting_dealer_client = GreenletDealerClient(
self._zeromq_context,
_local_node_name,
_space_accounting_server_address
)
self._space_accounting_dealer_client.link_exception(
self._unhandled_greenlet_exception
)
push_client = GreenletPUSHClient(
self._zeromq_context,
_local_node_name,
_space_accounting_pipeline_address,
)
self._accounting_client = SpaceAccountingClient(
_local_node_name,
self._space_accounting_dealer_client,
push_client
)
self._event_push_client = EventPushClient(
self._zeromq_context,
"web-internal-reader"
)
# message sent to data readers telling them the server
# is (re)starting, thereby invalidating any archvies or retrieved
# that are in progress for this node
timestamp = create_timestamp()
self._event_push_client.info("web-reader-start",
"web reader (re)start",
timestamp_repr=repr(timestamp),
source_node_name=_local_node_name)
self._watcher = Watcher(
_stats,
self._data_reader_clients,
self._event_push_client
)
self.application = Application(
memcached_client,
self._central_connection,
self._node_local_connection,
self._cluster_row,
self._data_readers,
self._accounting_client,
self._event_push_client,
_stats
)
self.wsgi_server = WSGIServer(
(_web_internal_reader_host, _web_internal_reader_port),
application=self.application,
backlog=_wsgi_backlog
)
def __init__(self):
self._log = logging.getLogger("WebInternalReader")
memcached_client = memcache.Client(_memcached_nodes)
self._central_connection = get_central_connection()
self._cluster_row = get_cluster_row(self._central_connection)
self._node_local_connection = get_node_local_connection()
self._deliverator = Deliverator()
self._zeromq_context = zmq.Context()
self._pull_server = GreenletPULLServer(
self._zeromq_context, _web_internal_reader_pipeline_address,
self._deliverator)
self._pull_server.link_exception(self._unhandled_greenlet_exception)
self._data_reader_clients = list()
self._data_readers = list()
for node_name, address in zip(_node_names, _data_reader_addresses):
resilient_client = GreenletResilientClient(
self._zeromq_context,
node_name,
address,
_client_tag,
_web_internal_reader_pipeline_address,
self._deliverator,
connect_messages=[])
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_reader_clients.append(resilient_client)
data_reader = DataReader(node_name, resilient_client)
self._data_readers.append(data_reader)
self._space_accounting_dealer_client = GreenletDealerClient(
self._zeromq_context, _local_node_name,
_space_accounting_server_address)
self._space_accounting_dealer_client.link_exception(
self._unhandled_greenlet_exception)
push_client = GreenletPUSHClient(
self._zeromq_context,
_local_node_name,
_space_accounting_pipeline_address,
)
self._accounting_client = SpaceAccountingClient(
_local_node_name, self._space_accounting_dealer_client,
push_client)
self._event_push_client = EventPushClient(self._zeromq_context,
"web-internal-reader")
# message sent to data readers telling them the server
# is (re)starting, thereby invalidating any archvies or retrieved
# that are in progress for this node
timestamp = create_timestamp()
self._event_push_client.info("web-reader-start",
"web reader (re)start",
timestamp_repr=repr(timestamp),
source_node_name=_local_node_name)
self._watcher = Watcher(_stats, self._data_reader_clients,
self._event_push_client)
self.application = Application(memcached_client,
self._central_connection,
self._node_local_connection,
self._cluster_row, self._data_readers,
self._accounting_client,
self._event_push_client, _stats)
self.wsgi_server = WSGIServer(
(_web_internal_reader_host, _web_internal_reader_port),
application=self.application,
backlog=_wsgi_backlog)
def _setup(_halt_event, state):
log = logging.getLogger("_setup")
status_checkers = list()
# do the event push client first, because we may need to
# push an execption event from setup
state["event-push-client"] = EventPushClient(
state["zmq-context"],
"handoff_server"
)
central_connection = get_central_connection()
state["cluster-row"] = get_cluster_row(central_connection)
state["node-rows"] = get_node_rows(
central_connection, state["cluster-row"].id
)
central_connection.close()
state["node-id-dict"] = dict(
[(node_row.name, node_row.id, ) for node_row in state["node-rows"]]
)
state["node-name-dict"] = dict(
[(node_row.id, node_row.name, ) for node_row in state["node-rows"]]
)
state["database-connection"] = get_node_local_connection()
for node_row, handoff_server_address in zip(
state["node-rows"], _handoff_server_addresses
):
if node_row.name == _local_node_name:
log.info("binding resilient-server to %s" % (
handoff_server_address,
))
state["resilient-server"] = ResilientServer(
state["zmq-context"],
handoff_server_address,
state["receive-queue"]
)
state["resilient-server"].register(state["pollster"])
else:
handoff_server_client = ResilientClient(
state["zmq-context"],
state["pollster"],
node_row.name,
handoff_server_address,
_client_tag,
_handoff_server_pipeline_address
)
state["handoff-server-clients"].append(handoff_server_client)
# don't run all the status checkers at the same time
status_checkers.append(
(handoff_server_client.run,
time.time() + random.random() * 60.0, )
)
log.info("binding pull-server to %s" % (_handoff_server_pipeline_address, ))
state["pull-server"] = PULLServer(
state["zmq-context"],
_handoff_server_pipeline_address,
state["receive-queue"]
)
state["pull-server"].register(state["pollster"])
for node_row, data_reader_address in zip(
state["node-rows"], _data_reader_addresses
):
data_reader_client = ResilientClient(
state["zmq-context"],
state["pollster"],
node_row.name,
data_reader_address,
_client_tag,
_handoff_server_pipeline_address
)
state["reader-client-dict"][data_reader_client.server_node_name] = \
data_reader_client
# don't run all the status checkers at the same time
status_checkers.append(
(data_reader_client.run, time.time() + random.random() * 60.0, )
)
for node_row, data_writer_address in zip(
state["node-rows"], _data_writer_addresses
):
data_writer_client = ResilientClient(
state["zmq-context"],
state["pollster"],
node_row.name,
data_writer_address,
_client_tag,
_handoff_server_pipeline_address
)
state["writer-client-dict"][data_writer_client.server_node_name] = \
data_writer_client
# don't run all the status checkers at the same time
status_checkers.append(
(data_writer_client.run, time.time() + random.random() * 60.0, )
)
state["queue-dispatcher"] = DequeDispatcher(
state,
state["receive-queue"],
_dispatch_table
)
state["handoff-requestor"] = HandoffRequestor(state, _local_node_name)
state["handoff-starter"] = HandoffStarter(
state, _local_node_name, state["event-push-client"]
)
state["event-push-client"].info("program-start", "handoff_server starts")
timer_driven_callbacks = [
(state["handoff-starter"].run, state["handoff-starter"].next_run(), ),
(state["pollster"].run, time.time(), ),
(state["queue-dispatcher"].run, time.time(), ),
# try to spread out handoff polling, if all nodes start together
(state["handoff-requestor"].run,
time.time() + random.random() * handoff_polling_interval)
]
timer_driven_callbacks.extend(status_checkers)
return timer_driven_callbacks
def __init__(self):
self._log = logging.getLogger("WebServer")
authenticator = SqlAuthenticator()
self._central_connection = get_central_connection()
self._cluster_row = get_cluster_row(self._central_connection)
self._node_local_connection = get_node_local_connection()
self._unified_id_factory = UnifiedIDFactory(
self._central_connection,
_get_shard_id(self._central_connection, self._cluster_row.id)
)
self._deliverator = Deliverator()
self._zeromq_context = zmq.Context()
self._pull_server = GreenletPULLServer(
self._zeromq_context,
_web_server_pipeline_address,
self._deliverator
)
self._pull_server.link_exception(self._unhandled_greenlet_exception)
# message sent to data readers and writers telling them the server
# is (re)starting, thereby invalidating any archvies or retrieved
# that are in progress for this node
timestamp = create_timestamp()
start_message = {
"message-type" : "web-server-start",
"priority" : create_priority(),
"unified-id" : self._unified_id_factory.next(),
"timestamp-repr" : repr(timestamp),
"source-node-name" : _local_node_name,
}
self._data_writer_clients = list()
for node_name, address in zip(_node_names, _data_writer_addresses):
resilient_client = GreenletResilientClient(
self._zeromq_context,
node_name,
address,
_client_tag,
_web_server_pipeline_address,
self._deliverator,
connect_messages=[start_message, ]
)
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_writer_clients.append(resilient_client)
self._data_reader_clients = list()
self._data_readers = list()
for node_name, address in zip(_node_names, _data_reader_addresses):
resilient_client = GreenletResilientClient(
self._zeromq_context,
node_name,
address,
_client_tag,
_web_server_pipeline_address,
self._deliverator,
connect_messages=[start_message, ]
)
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_reader_clients.append(resilient_client)
data_reader = DataReader(
node_name, resilient_client
)
self._data_readers.append(data_reader)
self._space_accounting_dealer_client = GreenletDealerClient(
self._zeromq_context,
_local_node_name,
_space_accounting_server_address
)
self._space_accounting_dealer_client.link_exception(
self._unhandled_greenlet_exception
)
push_client = GreenletPUSHClient(
self._zeromq_context,
_local_node_name,
_space_accounting_pipeline_address,
)
self._accounting_client = SpaceAccountingClient(
_local_node_name,
self._space_accounting_dealer_client,
push_client
)
self._event_push_client = EventPushClient(
self._zeromq_context,
"web-server"
)
self._watcher = Watcher(
_stats,
self._data_reader_clients,
self._data_writer_clients,
self._event_push_client
)
id_translator_keys_path = os.path.join(
_repository_path, "id_translator_keys.pkl"
)
with open(id_translator_keys_path, "r") as input_file:
id_translator_keys = pickle.load(input_file)
self._id_translator = InternalIDTranslator(
id_translator_keys["key"],
id_translator_keys["hmac_key"],
id_translator_keys["iv_key"],
id_translator_keys["hmac_size"]
)
self.application = Application(
self._central_connection,
self._node_local_connection,
self._cluster_row,
self._unified_id_factory,
self._id_translator,
self._data_writer_clients,
self._data_readers,
authenticator,
self._accounting_client,
self._event_push_client,
_stats
)
self.wsgi_server = WSGIServer(
(_web_server_host, _web_server_port),
application=self.application,
backlog=_wsgi_backlog
)
def setUp(self):
if not hasattr(self, "_log"):
self._log = logging.getLogger("TestHandoffServer")
self.tearDown()
database_connection = get_central_connection()
cluster_row = get_cluster_row(database_connection)
node_rows = get_node_rows(database_connection, cluster_row.id)
database_connection.close()
self._key_generator = generate_key()
self._event_publisher_processes = list()
self._data_writer_processes = list()
self._data_reader_processes = list()
self._handoff_server_processes = list()
for i in xrange(_node_count):
node_name = _generate_node_name(i)
repository_path = _repository_path(node_name)
os.makedirs(repository_path)
process = start_event_publisher(node_name,
_event_publisher_pull_addresses[i],
_event_publisher_pub_addresses[i])
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._event_publisher_processes.append(process)
time.sleep(1.0)
process = start_data_writer(_cluster_name, node_name,
_data_writer_addresses[i],
_event_publisher_pull_addresses[i],
repository_path)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._data_writer_processes.append(process)
time.sleep(1.0)
process = start_data_reader(node_name, _data_reader_addresses[i],
_event_publisher_pull_addresses[i],
repository_path)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._data_reader_processes.append(process)
time.sleep(1.0)
process = start_handoff_server(
_cluster_name, node_name, _handoff_server_addresses,
_handoff_server_pipeline_addresses[i], _data_reader_addresses,
_data_writer_addresses, _event_publisher_pull_addresses[i],
_repository_path(node_name))
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._handoff_server_processes.append(process)
time.sleep(1.0)
self._context = zmq.context.Context()
self._pollster = GreenletZeroMQPollster()
self._deliverator = Deliverator()
self._pull_server = GreenletPULLServer(self._context, _client_address,
self._deliverator)
self._pull_server.register(self._pollster)
backup_nodes = random.sample(node_rows[1:], 2)
self._log.debug("backup nodes = %s" % ([n.name
for n in backup_nodes], ))
self._resilient_clients = list()
for node_row, address in zip(node_rows, _data_writer_addresses):
if not node_row in backup_nodes:
continue
resilient_client = GreenletResilientClient(
self._context,
self._pollster,
node_row.name,
address,
_local_node_name,
_client_address,
self._deliverator,
)
self._resilient_clients.append(resilient_client)
self._log.debug("%s resilient clients" %
(len(self._resilient_clients), ))
self._data_writer_handoff_client = DataWriterHandoffClient(
node_rows[0].name, self._resilient_clients)
self._pollster.start()
def setUp(self):
if not hasattr(self, "_log"):
self._log = logging.getLogger("TestHandoffServer")
self.tearDown()
database_connection = get_central_connection()
cluster_row = get_cluster_row(database_connection)
node_rows = get_node_rows(database_connection, cluster_row.id)
database_connection.close()
self._key_generator = generate_key()
self._event_publisher_processes = list()
self._data_writer_processes = list()
self._data_reader_processes = list()
self._handoff_server_processes = list()
for i in xrange(_node_count):
node_name = _generate_node_name(i)
repository_path = _repository_path(node_name)
os.makedirs(repository_path)
process = start_event_publisher(
node_name,
_event_publisher_pull_addresses[i],
_event_publisher_pub_addresses[i]
)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._event_publisher_processes.append(process)
time.sleep(1.0)
process = start_data_writer(
_cluster_name,
node_name,
_data_writer_addresses[i],
_event_publisher_pull_addresses[i],
repository_path
)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._data_writer_processes.append(process)
time.sleep(1.0)
process = start_data_reader(
node_name,
_data_reader_addresses[i],
_event_publisher_pull_addresses[i],
repository_path
)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._data_reader_processes.append(process)
time.sleep(1.0)
process = start_handoff_server(
_cluster_name,
node_name,
_handoff_server_addresses,
_handoff_server_pipeline_addresses[i],
_data_reader_addresses,
_data_writer_addresses,
_event_publisher_pull_addresses[i],
_repository_path(node_name)
)
poll_result = poll_process(process)
self.assertEqual(poll_result, None)
self._handoff_server_processes.append(process)
time.sleep(1.0)
self._context = zmq.context.Context()
self._pollster = GreenletZeroMQPollster()
self._deliverator = Deliverator()
self._pull_server = GreenletPULLServer(
self._context,
_client_address,
self._deliverator
)
self._pull_server.register(self._pollster)
backup_nodes = random.sample(node_rows[1:], 2)
self._log.debug("backup nodes = %s" % (
[n.name for n in backup_nodes],
))
self._resilient_clients = list()
for node_row, address in zip(node_rows, _data_writer_addresses):
if not node_row in backup_nodes:
continue
resilient_client = GreenletResilientClient(
self._context,
self._pollster,
node_row.name,
address,
_local_node_name,
_client_address,
self._deliverator,
)
self._resilient_clients.append(resilient_client)
self._log.debug("%s resilient clients" % (
len(self._resilient_clients),
))
self._data_writer_handoff_client = DataWriterHandoffClient(
node_rows[0].name,
self._resilient_clients
)
self._pollster.start()
def _setup(_halt_event, state):
log = logging.getLogger("_setup")
status_checkers = list()
# do the event push client first, because we may need to
# push an execption event from setup
state["event-push-client"] = EventPushClient(
state["zmq-context"],
"anti_entropy_server"
)
state["central-database-connection"] = get_central_connection()
state["local-database-connection"] = get_node_local_connection()
state["cluster-row"] = get_cluster_row(
state["central-database-connection"]
)
local_anti_entropy_server_address = None
for node_name, address in zip(_node_names, _anti_entropy_server_addresses):
if node_name == _local_node_name:
local_anti_entropy_server_address = address
break
assert local_anti_entropy_server_address is not None
log.info("binding resilient-server to %s" % (
local_anti_entropy_server_address,
))
state["resilient-server"] = ResilientServer(
state["zmq-context"],
local_anti_entropy_server_address,
state["receive-queue"]
)
state["resilient-server"].register(state["pollster"])
log.info("binding pull-server to %s" % (
_anti_entropy_server_pipeline_address,
))
state["pull-server"] = PULLServer(
state["zmq-context"],
_anti_entropy_server_pipeline_address,
state["receive-queue"]
)
state["pull-server"].register(state["pollster"])
state["anti-entropy-clients"] = list()
for node_name, anti_entropy_server_address in zip(
_node_names, _anti_entropy_server_addresses
):
resilient_client = ResilientClient(
state["zmq-context"],
state["pollster"],
node_name,
anti_entropy_server_address,
_client_tag,
_anti_entropy_server_pipeline_address
)
state["anti-entropy-clients"].append(resilient_client)
status_checkers.append(
(resilient_client.run, time.time() + random.random() * 60.0, )
)
state["queue-dispatcher"] = DequeDispatcher(
state,
state["receive-queue"],
_dispatch_table
)
state["collection-list-requestor"] = CollectionListRequestor(state)
state["consistency-check-starter"] = ConsistencyCheckStarter(
state, _start_consistency_check
)
state["retry-manager"] = RetryManager(
state, _start_consistency_check
)
state["state-cleaner"] = StateCleaner(state)
state["event-push-client"].info(
"program-start", "anti_entropy_server starts"
)
# start the collection list requestor right away
# start the consistency check starter a little later, when
# we presumably have some collection ids
timer_driven_callbacks = [
(state["pollster"].run, time.time(), ),
(state["queue-dispatcher"].run, time.time(), ),
(state["collection-list-requestor"].run, time.time(), ),
(state["consistency-check-starter"].run, time.time()+60.0, ),
(state["retry-manager"].run, state["retry-manager"].next_run(), ),
(state["state-cleaner"].run, state["state-cleaner"].next_run(), ),
]
timer_driven_callbacks.extend(status_checkers)
return timer_driven_callbacks
def _setup(state):
log = logging.getLogger("_setup")
# do the event push client first, because we may need to
# push an execption event from setup
state["event-push-client"] = EventPushClient(
state["zmq-context"],
"data_writer"
)
log.info("binding resilient-server to {0}".format(_data_writer_address))
state["resilient-server"] = ResilientServer(
state["zmq-context"],
_data_writer_address,
state["message-queue"]
)
state["resilient-server"].register(state["pollster"])
log.info("binding reply-pull-server to {0}".format(
_writer_thread_reply_address))
state["reply-pull-server"] = ReplyPULLServer(
state["zmq-context"],
_writer_thread_reply_address,
state["resilient-server"].send_reply
)
state["reply-pull-server"].register(state["pollster"])
log.info("binding anti-entropy-server to {0}".format(
_data_writer_anti_entropy_address))
state["anti-entropy-server"] = REPServer(
state["zmq-context"],
_data_writer_anti_entropy_address,
state["message-queue"]
)
state["anti-entropy-server"].register(state["pollster"])
topics = ["web-writer-start", ]
log.info("connecting sub-client to {0} subscribing to {1}".format(
_event_aggregator_pub_address,
topics))
state["sub-client"] = SUBClient(
state["zmq-context"],
_event_aggregator_pub_address,
topics,
state["message-queue"]
)
state["sub-client"].register(state["pollster"])
central_connection = get_central_connection()
state["cluster-row"] = get_cluster_row(central_connection)
state["node-rows"] = get_node_rows(
central_connection, state["cluster-row"].id
)
central_connection.close()
state["node-id-dict"] = dict(
[(node_row.name, node_row.id, ) for node_row in state["node-rows"]]
)
state["event-push-client"].info("program-start", "data_writer starts")
state["reply-push-client"] = PUSHClient(state["zmq-context"],
_writer_thread_reply_address)
state["writer-thread"] = WriterThread(state["halt-event"],
state["node-id-dict"],
state["message-queue"],
state["reply-push-client"])
state["writer-thread"].start()
state["sync-thread"] = SyncThread(state["halt-event"],
state["message-queue"])
state["sync-thread"].start()
