def __init__(self,
connection,
space_id,
repository_path,
expected_size=None):
self._log = logging.getLogger("OutputValueFile")
self._connection = connection
assert space_id is not None
self._space_id = space_id
self._value_file_id = _get_next_value_file_id(connection)
self._value_file_path = compute_value_file_path(
repository_path, space_id, self._value_file_id)
self._expected_size = expected_size
self._log.debug("opening {0} expected size = {1}".format(
self._value_file_path, self._expected_size))
value_file_dir = os.path.dirname(self._value_file_path)
if not os.path.exists(value_file_dir):
os.makedirs(value_file_dir)
flags = os.O_WRONLY | os.O_CREAT
self._value_file_fd = os.open(self._value_file_path, flags)
self._creation_time = create_timestamp()
self._size = 0
self._md5 = hashlib.md5()
self._segment_sequence_count = 0
self._min_segment_id = None
self._max_segment_id = None
self._collection_ids = set()
def test_handoff_small_content(self):
"""test retrieving content that fits in a single message"""
file_size = 10 * 64 * 1024
file_content = random_string(file_size)
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 5
file_adler32 = zlib.adler32(file_content)
file_md5 = hashlib.md5(file_content)
message = {
"message-type" : "archive-key-entire",
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"file-size" : file_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"handoff-node-name" : None,
}
g = gevent.spawn(self._send_message_get_reply, message, file_content)
g.join(timeout=10.0)
self.assertEqual(g.ready(), True)
reply = g.value
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
print >> sys.stderr, "archive successful"
print >> sys.stderr, "press [Enter] to continue"
raw_input()
def close(self):
"""close the file and make it visible in the database"""
self.sync()
os.close(self._value_file_fd)
if self._segment_sequence_count == 0:
self._log.info("removing empty file %s" % (self._value_file_path,))
try:
os.unlink(self._value_file_path)
except Exception:
pass
return
self._log.info("closing %s size=%s segment_sequence_count=%s" % (
self._value_file_path, self._size, self._segment_sequence_count
))
value_file_row = value_file_template(
id=self._value_file_id,
space_id=self._space_id,
creation_time=self._creation_time,
close_time=create_timestamp(),
size=self._size,
hash=psycopg2.Binary(self._md5.digest()),
segment_sequence_count=self._segment_sequence_count,
min_segment_id=self._min_segment_id,
max_segment_id=self._max_segment_id,
distinct_collection_count=len(self._collection_ids),
collection_ids=sorted(list(self._collection_ids)),
garbage_size_estimate=0,
fragmentation_estimate=0,
last_cleanup_check_time=None,
last_integrity_check_time=None
)
_update_value_file_row(self._connection, value_file_row)
def close(self):
"""close the file and make it visible in the database"""
self._log.debug("closing %s size=%s segment_sequence_count=%s" % (
self._value_file_path, self._size, self._segment_sequence_count
))
os.fsync(self._value_file_fd)
os.close(self._value_file_fd)
value_file_row = value_file_template(
id=self._value_file_id,
space_id=self._space_id,
creation_time=self._creation_time,
close_time=create_timestamp(),
size=self._size,
hash=psycopg2.Binary(self._md5.digest()),
segment_sequence_count=self._segment_sequence_count,
min_segment_id=self._min_segment_id,
max_segment_id=self._max_segment_id,
distinct_collection_count=len(self._collection_ids),
collection_ids=sorted(list(self._collection_ids)),
garbage_size_estimate=0,
fragmentation_estimate=0,
last_cleanup_check_time=None,
last_integrity_check_time=None
)
_insert_value_file_row(self._connection, value_file_row)
def test_handoff_small_content(self):
"""test retrieving content that fits in a single message"""
file_size = 10 * 64 * 1024
file_content = random_string(file_size)
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 5
file_adler32 = zlib.adler32(file_content)
file_md5 = hashlib.md5(file_content)
message = {
"message-type": "archive-key-entire",
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"timestamp-repr": repr(timestamp),
"segment-num": segment_num,
"file-size": file_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"handoff-node-name": None,
}
g = gevent.spawn(self._send_message_get_reply, message, file_content)
g.join(timeout=10.0)
self.assertEqual(g.ready(), True)
reply = g.value
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
print >> sys.stderr, "archive successful"
print >> sys.stderr, "press [Enter] to continue"
raw_input()
def close(self):
"""close the file and make it visible in the database"""
self.sync()
os.close(self._value_file_fd)
if self._segment_sequence_count == 0:
self._log.info("removing empty file %s" %
(self._value_file_path, ))
try:
os.unlink(self._value_file_path)
except Exception:
pass
return
self._log.info(
"closing %s size=%s segment_sequence_count=%s" %
(self._value_file_path, self._size, self._segment_sequence_count))
value_file_row = value_file_template(
id=self._value_file_id,
space_id=self._space_id,
creation_time=self._creation_time,
close_time=create_timestamp(),
size=self._size,
hash=psycopg2.Binary(self._md5.digest()),
segment_sequence_count=self._segment_sequence_count,
min_segment_id=self._min_segment_id,
max_segment_id=self._max_segment_id,
distinct_collection_count=len(self._collection_ids),
collection_ids=sorted(list(self._collection_ids)),
garbage_size_estimate=0,
fragmentation_estimate=0,
last_cleanup_check_time=None,
last_integrity_check_time=None)
_update_value_file_row(self._connection, value_file_row)
def _handle_anti_entropy_audit_request(state, message, _data):
"""handle a requst to audit a specific collection, not some random one"""
log = logging.getLogger("_handle_anti_entropy_audit_request")
timestamp = create_timestamp()
state_key = (message["collection-id"], timestamp, )
database = AuditResultDatabase(state["central-database-connection"])
row_id = database.start_audit(message["collection-id"], timestamp)
database.close()
state["active-requests"][state_key] = _request_state_tuple(
client_tag=message["client-tag"],
timestamp=timestamp,
timeout=time.time()+_request_timeout,
retry_count=max_retry_count,
replies=dict(),
row_id=row_id,
)
request = {
"message-type" : "consistency-check",
"collection-id" : message["collection-id"],
"timestamp-repr": repr(timestamp),
}
for anti_entropy_client in state["anti-entropy-clients"]:
anti_entropy_client.queue_message_for_send(request)
def __init__(self,
connection,
space_id,
repository_path,
expected_size=None):
self._log = logging.getLogger("OutputValueFile")
self._connection = connection
assert space_id is not None
self._space_id = space_id
self._value_file_id = _get_next_value_file_id(connection)
self._value_file_path = compute_value_file_path(
repository_path, space_id, self._value_file_id)
self._expected_size = expected_size
self._log.debug("opening {0} expected size = {1}".format(
self._value_file_path, self._expected_size))
value_file_dir = os.path.dirname(self._value_file_path)
if not os.path.exists(value_file_dir):
os.makedirs(value_file_dir)
flags = os.O_WRONLY | os.O_CREAT
self._value_file_fd = os.open(self._value_file_path, flags)
self._creation_time = create_timestamp()
self._size = 0
self._md5 = hashlib.md5()
self._segment_sequence_count = 0
self._min_segment_id = None
self._max_segment_id = None
self._collection_ids = set()
def close(self):
"""close the file and make it visible in the database"""
self._log.debug(
"closing %s size=%s segment_sequence_count=%s" %
(self._value_file_path, self._size, self._segment_sequence_count))
os.fsync(self._value_file_fd)
os.close(self._value_file_fd)
value_file_row = value_file_template(
id=self._value_file_id,
space_id=self._space_id,
creation_time=self._creation_time,
close_time=create_timestamp(),
size=self._size,
hash=psycopg2.Binary(self._md5.digest()),
segment_sequence_count=self._segment_sequence_count,
min_segment_id=self._min_segment_id,
max_segment_id=self._max_segment_id,
distinct_collection_count=len(self._collection_ids),
collection_ids=sorted(list(self._collection_ids)),
garbage_size_estimate=0,
fragmentation_estimate=0,
last_cleanup_check_time=None,
last_integrity_check_time=None)
_insert_value_file_row(self._connection, value_file_row)
def run(self, halt_event):
"""
send 'request-handoffs' to all remote handoff servers
"""
if halt_event.is_set():
return
if len(self._state["pending-handoffs"]) > 0:
# run again later
return [(self.run, time.time() + _delay_interval, )]
self._log.debug("sending handoff requests")
message = {
"message-type" : "request-handoffs",
"request-timestamp-repr" : repr(create_timestamp()),
"node-name" : self._local_node_name,
"node-id" : self._local_node_id,
}
for handoff_server_client in self._state["handoff-server-clients"]:
handoff_server_client.queue_message_for_send(message)
return [(self.run, self.next_run(), )]
def _start_consistency_check(state, collection_id, row_id=None, retry_count=0):
log = logging.getLogger("_start_consistency_check")
timestamp = create_timestamp()
state_key = (collection_id, timestamp, )
database = AuditResultDatabase(state["central-database-connection"])
if row_id is None:
row_id = database.start_audit(collection_id, timestamp)
else:
database.restart_audit(row_id, timestamp)
database.close()
state["active-requests"][state_key] = _request_state_tuple(
client_tag=None,
timestamp=timestamp,
timeout=time.time()+_request_timeout,
retry_count=retry_count,
replies=dict(),
row_id=row_id,
)
request = {
"message-type" : "consistency-check",
"collection-id" : collection_id,
"timestamp-repr": repr(timestamp),
}
for anti_entropy_client in state["anti-entropy-clients"]:
anti_entropy_client.queue_message_for_send(request)
def test_purge_nonexistent_key(self):
"""test purgeing a key that does not exist, with no complicatons"""
collection_id = 1001
key = self._key_generator.next()
segment_num = 4
timestamp = create_timestamp()
reply = self._purge(collection_id, key, timestamp, segment_num)
self.assertEqual(reply["result"], "success", reply)
def test_simple_segment(self):
"""test writing an reading a simple segment of one sequence"""
collection_id = 1001
key = "aaa/bbb/ccc"
timestamp = create_timestamp()
segment_num = 42
sequence_num = 0
data_size = 1024
data = random_string(data_size)
data_adler32 = zlib.adler32(data)
data_md5 = hashlib.md5(data)
file_tombstone = False
writer = Writer(self._database_connection, _repository_path)
# clean out any segments that are laying around for this (test) keu
reader = Reader(self._database_connection, _repository_path)
writer.start_new_segment(collection_id, key, repr(timestamp), segment_num)
writer.store_sequence(
collection_id, key, repr(timestamp), segment_num, sequence_num, data
)
writer.finish_new_segment(
collection_id,
key,
repr(timestamp),
segment_num,
data_size,
data_adler32,
data_md5.digest(),
file_tombstone,
handoff_node_id=None,
)
writer.close()
file_info = most_recent_timestamp_for_key(
self._database_connection, collection_id, key
)
self.assertEqual(file_info.file_size, data_size)
self.assertEqual(file_info.file_adler32, data_adler32)
self.assertEqual(str(file_info.file_hash), data_md5.digest())
self.assertEqual(file_info.file_tombstone, file_tombstone)
reader = Reader(self._database_connection, _repository_path)
sequence_generator = reader.generate_all_sequence_rows_for_segment(
collection_id, key, file_info.timestamp, file_info.segment_num
)
# first yield should be a count
sequence_count = sequence_generator.next()
self.assertEqual(sequence_count, 1)
sequence_data = sequence_generator.next()
self.assertEqual(len(sequence_data), len(data))
self.assertEqual(sequence_data, data)
def xxxtest_archive_key_entire_with_meta(self):
"""
test archiving a key in a single message, including meta data
"""
file_size = 10 * 64 * 1024
content_item = random_string(file_size)
user_request_id = uuid.uuid1().hex
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 2
meta_key = "".join([nimbus_meta_prefix, "test_key"])
meta_value = "pork"
file_adler32 = zlib.adler32(content_item)
file_md5 = hashlib.md5(content_item)
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
message = {
"message-type" : "archive-key-entire",
"priority" : archive_priority,
"user-request-id" : user_request_id,
"collection-id" : collection_id,
"key" : key,
"unified-id" : unified_id,
"timestamp-repr" : repr(timestamp),
"conjoined-part" : 0,
"segment-num" : segment_num,
"segment-size" : file_size,
"zfec-padding-size" : 4,
"segment-adler32" : file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size" : file_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"source-node-name" : _local_node_name,
"handoff-node-name" : None,
meta_key : meta_value
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=content_item
)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
def run(self, halt_event):
if halt_event.is_set():
return
# we want to dump everything for the previous hour
current_time = create_timestamp()
current_hour = floor_hour(current_time)
prev_hour = current_hour - datetime.timedelta(hours=1)
self._flush_to_database(prev_hour)
return [(self.run, self.next_run(), )]
def xxxtest_destroy_nonexistent_key(self):
"""test destroying a key that does not exist, with no complications"""
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
collection_id = 1001
key = self._key_generator.next()
segment_num = 4
timestamp = create_timestamp()
reply = self._destroy(collection_id, key, unified_id, timestamp,
segment_num)
self.assertEqual(reply["result"], "success", reply["error-message"])
def xxxtest_destroy_tombstone(self):
"""test destroying a key that has already been destroyed"""
file_size = 10 * 64 * 1024
content_item = random_string(file_size)
message_id = uuid.uuid1().hex
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
archive_timestamp = create_timestamp()
destroy_1_timestamp = archive_timestamp + timedelta(seconds=1)
destroy_2_timestamp = destroy_1_timestamp + timedelta(seconds=1)
segment_num = 2
file_adler32 = zlib.adler32(content_item)
file_md5 = hashlib.md5(content_item)
message = {
"message-type" : "archive-key-entire",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(archive_timestamp),
"segment-num" : segment_num,
"segment-size" : file_size,
"segment-adler32" : file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size" : file_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=content_item
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
reply = self._destroy(
collection_id, key, destroy_1_timestamp, segment_num
)
self.assertEqual(reply["result"], "success", reply["error-message"])
reply = self._destroy(
collection_id, key, destroy_2_timestamp, segment_num
)
self.assertEqual(reply["result"], "success", reply["error-message"])
def xxxtest_destroy_nonexistent_key(self):
"""test destroying a key that does not exist, with no complications"""
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
collection_id = 1001
key = self._key_generator.next()
segment_num = 4
timestamp = create_timestamp()
reply = self._destroy(collection_id, key, unified_id, timestamp,
segment_num)
self.assertEqual(reply["result"], "success", reply["error-message"])
def xxxtest_archive_key_entire_with_meta(self):
"""
test archiving a key in a single message, including meta data
"""
file_size = 10 * 64 * 1024
content_item = random_string(file_size)
user_request_id = uuid.uuid1().hex
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 2
meta_key = "".join([nimbus_meta_prefix, "test_key"])
meta_value = "pork"
file_adler32 = zlib.adler32(content_item)
file_md5 = hashlib.md5(content_item)
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
message = {
"message-type": "archive-key-entire",
"priority": archive_priority,
"user-request-id": user_request_id,
"collection-id": collection_id,
"key": key,
"unified-id": unified_id,
"timestamp-repr": repr(timestamp),
"conjoined-part": 0,
"segment-num": segment_num,
"segment-size": file_size,
"zfec-padding-size": 4,
"segment-adler32": file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size": file_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"source-node-name": _local_node_name,
"handoff-node-name": None,
meta_key: meta_value
}
reply = send_request_and_get_reply(_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=content_item)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
def retrieve(self, response, timeout):
try:
return self._retrieve(response, timeout)
except Exception, instance:
self._log.error("request {0} _retrieve exception".format(self.user_request_id))
self._log.exception("request {0}".format(self.user_request_id))
queue_entry = redis_queue_entry_tuple(
timestamp=create_timestamp(), collection_id=self._collection_id, value=1
)
self._redis_queue.put(("retrieve_error", queue_entry))
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
raise RetrieveFailedError(instance)
def xxxtest_destroy_tombstone(self):
"""test destroying a key that has already been destroyed"""
file_size = 10 * 64 * 1024
content_item = random_string(file_size)
message_id = uuid.uuid1().hex
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
archive_timestamp = create_timestamp()
destroy_1_timestamp = archive_timestamp + timedelta(seconds=1)
destroy_2_timestamp = destroy_1_timestamp + timedelta(seconds=1)
segment_num = 2
file_adler32 = zlib.adler32(content_item)
file_md5 = hashlib.md5(content_item)
message = {
"message-type": "archive-key-entire",
"message-id": message_id,
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"timestamp-repr": repr(archive_timestamp),
"segment-num": segment_num,
"segment-size": file_size,
"segment-adler32": file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size": file_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"handoff-node-name": None,
}
reply = send_request_and_get_reply(_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=content_item)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
reply = self._destroy(collection_id, key, destroy_1_timestamp,
segment_num)
self.assertEqual(reply["result"], "success", reply["error-message"])
reply = self._destroy(collection_id, key, destroy_2_timestamp,
segment_num)
self.assertEqual(reply["result"], "success", reply["error-message"])
def next_run(cls):
log = logging.getLogger("StateCleaner")
current_time = create_timestamp()
next_time = datetime.datetime(
year = current_time.year,
month = current_time.month,
day = current_time.day,
hour = current_time.hour,
minute = 5,
second = 0,
microsecond = 0
)
if current_time.minute >= 5:
next_time += datetime.timedelta(hours=1)
log.info("next dump time = %s", (next_time, ))
return time.mktime(next_time.timetuple())
def retrieve(self, response, timeout):
try:
return self._retrieve(response, timeout)
except Exception, instance:
self._log.error("request {0} _retrieve exception".format(
self.user_request_id))
self._log.exception("request {0}".format(self.user_request_id))
queue_entry = \
redis_queue_entry_tuple(timestamp=create_timestamp(),
collection_id=self._collection_id,
value=1)
self._redis_queue.put((
"retrieve_error",
queue_entry,
))
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
raise RetrieveFailedError(instance)
def __init__(self, connection, space_id, repository_path):
self._space_id = space_id
self._value_file_id = _insert_value_file_default_row(
connection, space_id)
self._log = logging.getLogger("VF%08d" % (self._value_file_id, ))
self._connection = connection
self._value_file_path = compute_value_file_path(
repository_path, space_id, self._value_file_id)
self._log.info("opening %s" % (self._value_file_path, ))
self._value_file_fd = _open_value_file(self._value_file_path)
self._creation_time = create_timestamp()
self._size = 0
self._md5 = hashlib.md5()
self._segment_sequence_count = 0
self._min_segment_id = None
self._max_segment_id = None
self._collection_ids = set()
self._synced = True # treat as synced until we write
def __init__(self, connection, space_id, repository_path):
self._space_id = space_id
self._value_file_id = _insert_value_file_default_row(connection,
space_id)
self._log = logging.getLogger("VF%08d" % (self._value_file_id, ))
self._connection = connection
self._value_file_path = compute_value_file_path(
repository_path, space_id, self._value_file_id
)
self._log.info("opening %s" % (self._value_file_path, ))
self._value_file_fd = _open_value_file(self._value_file_path)
self._creation_time = create_timestamp()
self._size = 0
self._md5 = hashlib.md5()
self._segment_sequence_count = 0
self._min_segment_id = None
self._max_segment_id = None
self._collection_ids = set()
self._synced = True # treat as synced until we write
def test_archive_key_entire(self):
"""test archiving all data for a key in a single message"""
file_size = 10 * 64 * 1024
content_item = random_string(file_size)
message_id = uuid.uuid1().hex
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 2
file_adler32 = zlib.adler32(content_item)
file_md5 = hashlib.md5(content_item)
message = {
"message-type" : "archive-key-entire",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"segment-size" : file_size,
"segment-adler32" : file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size" : file_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=content_item
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
def _detail_generator(
total_bytes_added, total_bytes_removed, total_bytes_retrieved
):
current_time = create_timestamp()
for i in xrange(1000):
message = {
"message-type" : "space-accounting-detail",
"collection-id" : _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event" : "bytes_added",
"value" : total_bytes_added / 1000,
}
# FIXME: losing messages if we pump them all in w/o delay
time.sleep(.01)
yield message, None
for i in xrange(50):
message = {
"message-type" : "space-accounting-detail",
"collection-id" : _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event" : "bytes_removed",
"value" : total_bytes_removed / 50,
}
yield message, None
for i in xrange(25):
message = {
"message-type" : "space-accounting-detail",
"collection-id" : _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event" : "bytes_retrieved",
"value" : total_bytes_retrieved / 25,
}
yield message, None
def _detail_generator(total_bytes_added, total_bytes_removed,
total_bytes_retrieved):
current_time = create_timestamp()
for i in xrange(1000):
message = {
"message-type": "space-accounting-detail",
"collection-id": _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event": "bytes_added",
"value": total_bytes_added / 1000,
}
# FIXME: losing messages if we pump them all in w/o delay
time.sleep(.01)
yield message, None
for i in xrange(50):
message = {
"message-type": "space-accounting-detail",
"collection-id": _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event": "bytes_removed",
"value": total_bytes_removed / 50,
}
yield message, None
for i in xrange(25):
message = {
"message-type": "space-accounting-detail",
"collection-id": _collection_id,
"timestamp-repr": repr(current_time + timedelta(seconds=i)),
"event": "bytes_retrieved",
"value": total_bytes_retrieved / 25,
}
yield message, None
variable_value = urllib.unquote_plus(variable_value)
variable_value = variable_value.decode("utf-8")
kwargs[variable_name] = variable_value
# translate version id to the form we use internally
if "version_id_marker" in kwargs:
kwargs["version_id_marker"] = self._id_translator.internal_id(kwargs["version_id_marker"])
self._log.info(
"request {0}: "
"_list_versions: collection = ({1}) {2} {3} {4}".format(
user_request_id, collection_row["id"], collection_row["name"], collection_row["versioning"], kwargs
)
)
queue_entry = redis_queue_entry_tuple(timestamp=create_timestamp(), collection_id=collection_row["id"], value=1)
self._redis_queue.put(("listmatch_request", queue_entry))
try:
result_dict = list_versions(
self._interaction_pool, collection_row["id"], collection_row["versioning"], **kwargs
)
# segment_visibility raises ValueError if it is unhappy
except ValueError, instance:
self._log.error("request {0}: {1}".format(user_request_id, instance))
raise exc.HTTPBadRequest(instance)
except Exception:
self._log.exception("request {0}".format(user_request_id))
queue_entry = redis_queue_entry_tuple(
timestamp=create_timestamp(), collection_id=collection_row["id"], value=1
)
break
if prev_data is None:
if first_block:
assert len(data) > offset_into_first_block, (len(data), offset_into_first_block)
prev_data = data[offset_into_first_block:]
else:
prev_data = data
continue
self._log.debug("request {0} yielding {1} bytes".format(self.user_request_id, len(prev_data)))
yield prev_data
retrieve_bytes += len(prev_data)
prev_data = data
first_block = False
urllib_response.close()
self._log.debug("request {0} internal request complete".format(self.user_request_id))
# end - for entry in self._generate_key_rows(self._key_rows):
if response.status_int in [httplib.OK, httplib.PARTIAL_CONTENT]:
redis_entries = [("retrieve_success", 1), ("success_bytes_out", retrieve_bytes)]
else:
redis_entries = [("retrieve_error", 1), ("error_bytes_out", retrieve_bytes)]
timestamp = create_timestamp()
for key, value in redis_entries:
queue_entry = redis_queue_entry_tuple(timestamp=timestamp, collection_id=self._collection_id, value=value)
self._redis_queue.put((key, queue_entry))
def _handle_consistency_check_reply(state, message, _data):
log = logging.getLogger("_handle_consistency_check_reply")
timestamp = parse_timestamp_repr(message["timestamp-repr"])
state_key = (message["collection-id"], timestamp, )
try:
request_state = state["active-requests"][state_key]
except KeyError:
log.warn("Unknown state_key %s from %s" % (
state_key, message["node-name"]
))
return
if message["node-name"] in request_state.replies:
error_message = "duplicate reply from %s %s" % (
message["node-name"],
state_key,
)
log.error(error_message)
return
if message["result"] != "success":
log.error("%s (%s) %s from %s" % (
state_key,
message["result"],
message["error-message"],
message["node-name"],
))
reply_value = _error_reply
else:
reply_value = (message["count"], message["encoded-md5-digest"], )
request_state.replies[message["node-name"]] = reply_value
# not done yet, wait for more replies
if len(request_state.replies) < len(state["anti-entropy-clients"]):
return
# at this point we should have a reply from every node, so
# we don't want to preserve state anymore
del state["active-requests"][state_key]
database = AuditResultDatabase(state["central-database-connection"])
timestamp = create_timestamp()
# push the results into a dict to see how many unique entries there are
md5_digest_dict = dict()
md5_digest_dict[_error_reply] = list()
for node_name in request_state.replies.keys():
node_reply = request_state.replies[node_name]
if node_reply == _error_reply:
md5_digest_dict[_error_reply].append(node_name)
continue
_count, encoded_md5_digest = node_reply
if not encoded_md5_digest in md5_digest_dict:
md5_digest_dict[encoded_md5_digest] = list()
md5_digest_dict[encoded_md5_digest].append(node_name)
# if this audit was started by an anti-entropy-audit-request message,
# we want to send a reply
if request_state.client_tag is not None:
reply = {
"message-type" : "anti-entropy-audit-reply",
"client-tag" : request_state.client_tag,
"collection-id" : message["collection-id"],
"result" : None,
"error-message" : None,
}
else:
reply = None
error_reply_list = md5_digest_dict.pop(_error_reply)
if reply is not None:
reply["error-reply-nodes"] = error_reply_list
if len(md5_digest_dict) > 1:
log.error("found %s different hashes for (%s)" % (
len(md5_digest_dict),
message["collection-id"],
))
for index, value in enumerate(md5_digest_dict.values()):
log.info(str(value))
if reply is not None:
reply["mistmatch-nodes-%s" % (index+1, )] = value
# ok = no errors and all nodes have the same hash for every collection
if len(error_reply_list) == 0 and len(md5_digest_dict) == 1:
description = "collection %s compares ok" % (
message["collection-id"],
)
log.info(description)
state["event-push-client"].info(
"audit-ok", description, collection_id=message["collection-id"]
)
database.successful_audit(request_state.row_id, timestamp)
if reply is not None:
reply["result"] = "success"
state["resilient-server"].send_reply(reply)
return
# we have error(s), but the non-errors compare ok
if len(error_reply_list) > 0 and len(md5_digest_dict) == 1:
# if we come from anti-entropy-audit-request, don't retry
if reply is not None:
database.audit_error(request_state.row_id, timestamp)
database.close()
description = "There were error replies from %s nodes" % (
len(error_reply_list) ,
)
log.error(description)
state["event-push-client"].error(
"consistency-check-errors-replies",
description,
collection_id=message["collection-id"],
error_reply_nodes=error_reply_list
)
reply["result"] = "error"
reply["error-message"] = description
state["resilient-server"].send_reply(reply)
return
if request_state.retry_count >= max_retry_count:
description = "collection %s %s errors, too many retries" % (
message["collection-id"],
len(error_reply_list)
)
log.error(description)
state["event-push-client"].error(
"audit-errors",
description,
collection_id=message["collection-id"]
)
database.audit_error(request_state.row_id, timestamp)
# TODO: needto do something here
else:
description = "%s Error replies from %s nodes, will retry" % (
message["collection-id"],
len(error_reply_list)
)
log.warn(description)
state["event-push-client"].warn(
"audit-retry",
description,
collection_id=message["collection-id"]
)
state["retry-list"].append(
retry_entry_tuple(
retry_time=retry_time(),
collection_id=message["collection-id"],
row_id=request_state.row_id,
retry_count=request_state.retry_count,
)
)
database.wait_for_retry(request_state.row_id)
database.close()
return
# if we make it here, we have some form of mismatch, possibly mixed with
# errors
description = "%s error replies from %s nodes; hash mismatch(es) = %r" % (
message["collection-id"],
len(error_reply_list),
md5_digest_dict.values()
)
log.error(description)
state["event-push-client"].warn(
"audit-retry",
description,
collection_id=message["collection-id"]
)
# if we come from anti-entropy-audit-request, don't retry
if reply is not None:
database.audit_error(request_state.row_id, timestamp)
database.close()
reply["result"] = "audit-error"
reply["error-message"] = description
state["resilient-server"].send_reply(reply)
return
if request_state.retry_count >= max_retry_count:
log.error("%s too many retries" % (message["collection-id"], ))
database.audit_error(request_state.row_id, timestamp)
# TODO: need to do something here
else:
state["retry-list"].append(
retry_entry_tuple(
retry_time=retry_time(),
collection_id=message["collection-id"],
row_id=request_state.row_id,
retry_count=request_state.retry_count,
)
)
database.wait_for_retry(request_state.row_id)
database.close()
def test_retrieve_large_content(self):
"""test retrieving content that fits in a multiple messages"""
slice_size = 1024 * 1024
slice_count = 10
total_size = slice_size * slice_count
test_data = random_string(total_size)
collection_id = 1001
archive_priority = create_priority()
timestamp = create_timestamp()
key = self._key_generator.next()
segment_num = 4
sequence_num = 0
file_adler32 = zlib.adler32(test_data)
file_md5 = hashlib.md5(test_data)
slice_start = 0
slice_end = slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type": "archive-key-start",
"message-id": message_id,
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"conjoined-unified-id": None,
"conjoined-part": 0,
"timestamp-repr": repr(timestamp),
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"segment-adler32": segment_adler32,
"segment-md5-digest": b64encode(segment_md5.digest()),
"sequence-num": sequence_num,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-start-reply")
self.assertEqual(reply["result"], "success")
for _ in range(slice_count - 2):
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type": "archive-key-next",
"message-id": message_id,
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"conjoined-unified-id": None,
"conjoined-part": 0,
"timestamp-repr": repr(timestamp),
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"segment-adler32": segment_adler32,
"segment-md5-digest": b64encode(segment_md5.digest()),
"sequence-num": sequence_num,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-next-reply")
self.assertEqual(reply["result"], "success")
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
self.assertEqual(slice_end, total_size)
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type": "archive-key-final",
"message-id": message_id,
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"conjoined-unified-id": None,
"conjoined-part": 0,
"timestamp-repr": repr(timestamp),
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"segment-adler32": segment_adler32,
"segment-md5-digest": b64encode(segment_md5.digest()),
"sequence-num": sequence_num,
"file-size": total_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"handoff-node-name": None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
# get file info from the local database
_conjoined_row, segment_rows = current_status_of_key(
self._database_connection, collection_id, key)
self.assertEqual(len(segment_rows), 1)
retrieved_data_list = list()
message_id = uuid.uuid1().hex
message = {
"message-type": "retrieve-key-start",
"message-id": message_id,
"collection-id": collection_id,
"key": key,
"timestamp-repr": repr(timestamp),
"conjoined-unified-id": None,
"conjoined-part": 0,
"segment-num": segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name, _data_reader_address, _local_node_name,
_client_address, message)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
self.assertEqual(reply["completed"], False)
print "sequence-num =", reply["sequence-num"]
retrieved_data_list.append(data)
while True:
message_id = uuid.uuid1().hex
message = {
"message-type": "retrieve-key-next",
"message-id": message_id,
"collection-id": collection_id,
"key": key,
"timestamp-repr": repr(timestamp),
"conjoined-unified-id": None,
"conjoined-part": 0,
"segment-num": segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name, _data_reader_address, _local_node_name,
_client_address, message)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
retrieved_data_list.append(data)
print "sequence-num =", reply["sequence-num"]
if reply["completed"]:
break
retrieved_data = "".join(retrieved_data_list)
self.assertEqual(len(retrieved_data), len(test_data))
self.assertEqual(retrieved_data, test_data)
def xxxtest_large_archive(self):
"""
test archiving a file that needs more than one message.
For example, a 10 Mb file: each node would get 10 120kb
zefec shares.
"""
slice_size = 1024 * 1024
slice_count = 10
total_size = slice_size * slice_count
test_data = random_string(total_size)
user_request_id = uuid.uuid1().hex
collection_id = 1001
archive_priority = create_priority()
timestamp = create_timestamp()
key = self._key_generator.next()
segment_num = 4
sequence_num = 0
file_adler32 = zlib.adler32(test_data)
file_md5 = hashlib.md5(test_data)
slice_start = 0
slice_end = slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
message = {
"message-type" : "archive-key-start",
"priority" : archive_priority,
"user-request-id" : user_request_id,
"collection-id" : collection_id,
"key" : key,
"unified-id" : unified_id,
"timestamp-repr" : repr(timestamp),
"conjoined-part" : 0,
"segment-num" : segment_num,
"segment-size" : len(test_data[slice_start:slice_end]),
"zfec-padding-size" : 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32" : segment_adler32,
"sequence-num" : sequence_num,
"source-node-name" : _local_node_name,
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-type"], "archive-key-start-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
for _ in range(slice_count-2):
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type" : "archive-key-next",
"priority" : archive_priority,
"user-request-id" : user_request_id,
"collection-id" : collection_id,
"key" : key,
"unified-id" : unified_id,
"timestamp-repr" : repr(timestamp),
"conjoined-part" : 0,
"segment-num" : segment_num,
"segment-size" : len(test_data[slice_start:slice_end]),
"zfec-padding-size" : 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32" : segment_adler32,
"sequence-num" : sequence_num,
"source-node-name" : _local_node_name,
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-type"], "archive-key-next-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message = {
"message-type" : "archive-key-final",
"priority" : archive_priority,
"user-request-id" : user_request_id,
"collection-id" : collection_id,
"key" : key,
"unified-id" : unified_id,
"timestamp-repr" : repr(timestamp),
"conjoined-part" : 0,
"segment-num" : segment_num,
"segment-size" : len(test_data[slice_start:slice_end]),
"zfec-padding-size" : 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32" : segment_adler32,
"sequence-num" : sequence_num,
"file-size" : total_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"source-node-name" : _local_node_name,
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
def audit_segments(halt_event, work_dir):
log = logging.getLogger("audit_segments")
if not os.path.exists(anti_entropy_dir):
log.info("creating {0}".format(anti_entropy_dir))
os.mkdir(anti_entropy_dir)
meta_repair_file_path = compute_meta_repair_file_path()
meta_repair_file = \
gzip.GzipFile(filename=meta_repair_file_path, mode="wb")
data_repair_file_path = compute_data_repair_file_path()
data_repair_file = \
gzip.GzipFile(filename=data_repair_file_path, mode="wb")
counts = {
"total" : 0,
anti_entropy_missing_replicas : 0,
anti_entropy_incomplete_finalization : 0,
anti_entropy_damaged_records : 0,
anti_entropy_missing_tombstones : 0,
anti_entropy_database_inconsistancy : 0,
}
current_time = create_timestamp()
min_segment_age = parse_timedelta_str(_min_segment_age)
newest_allowable_timestamp = current_time - min_segment_age
log.info("newest allowable timestamp = {0}".format(
newest_allowable_timestamp.isoformat()))
for row_key, segment_status, segment_data in generate_work(work_dir):
if halt_event.is_set():
log.info("halt_event is set: exiting")
return
assert segment_status == anti_entropy_pre_audit
counts["total"] += 1
# missing replicas needs to run first, because the other tests
# assume there are no missing replicas
if _missing_replicas(segment_data, newest_allowable_timestamp):
log.debug("missing_replicas {0}".format(row_key))
counts[anti_entropy_missing_replicas] += 1
store_sized_pickle(
(row_key, anti_entropy_missing_replicas, segment_data, ),
data_repair_file)
continue
# _missing_tombstones needs to run ahead of _incomplete_finalization
if _missing_tombstones(segment_data, newest_allowable_timestamp):
log.debug("missing_tombstones {0}".format(row_key))
counts[anti_entropy_missing_tombstones] += 1
store_sized_pickle(
(row_key, anti_entropy_missing_tombstones, segment_data, ),
meta_repair_file)
continue
if _incomplete_finalization(segment_data, newest_allowable_timestamp):
log.debug("incomplete_finalization {0}".format(row_key))
counts[anti_entropy_incomplete_finalization] += 1
store_sized_pickle(
(row_key, anti_entropy_incomplete_finalization, segment_data,),
data_repair_file)
continue
if _damaged_records(segment_data):
log.debug("damaged_records {0}".format(row_key))
counts[anti_entropy_damaged_records] += 1
store_sized_pickle(
(row_key, anti_entropy_damaged_records, segment_data,),
data_repair_file)
continue
if _database_inconsistancy(row_key, segment_data):
log.debug("database_inconsistancy {0}".format(row_key))
counts[anti_entropy_database_inconsistancy] += 1
store_sized_pickle(
(row_key, anti_entropy_database_inconsistency, segment_data,),
data_repair_file)
continue
meta_repair_file.close()
data_repair_file.close()
keys = ["total",
anti_entropy_missing_replicas,
anti_entropy_incomplete_finalization,
anti_entropy_damaged_records,
anti_entropy_missing_tombstones,
anti_entropy_database_inconsistancy]
for key in keys:
log.info("{0} {1:,}".format(key, counts[key]))
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 test_retrieve_small_content(self):
"""test retrieving content that fits in a single message"""
file_size = 10 * 64 * 1024
file_content = random_string(file_size)
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 2
file_adler32 = zlib.adler32(file_content)
file_md5 = hashlib.md5(file_content)
message_id = uuid.uuid1().hex
message = {
"message-type" : "archive-key-entire",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"segment-size" : file_size,
"segment-adler32" : file_adler32,
"segment-md5-digest" : b64encode(file_md5.digest()),
"file-size" : file_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=file_content
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
# get file info from the local database
_conjoined_row, segment_rows = current_status_of_key(
self._database_connection, collection_id, key
)
self.assertEqual(len(segment_rows), 1)
message_id = uuid.uuid1().hex
message = {
"message-type" : "retrieve-key-start",
"message-id" : message_id,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(timestamp),
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"segment-num" : segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name,
_data_reader_address,
_local_node_name,
_client_address,
message
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
self.assertEqual(reply["completed"], True)
self.assertEqual(len(data), len(file_content))
self.assertEqual(data, file_content)
def xxxtest_large_archive(self):
"""
test archiving a file that needs more than one message.
For example, a 10 Mb file: each node would get 10 120kb
zefec shares.
"""
slice_size = 1024 * 1024
slice_count = 10
total_size = slice_size * slice_count
test_data = random_string(total_size)
user_request_id = uuid.uuid1().hex
collection_id = 1001
archive_priority = create_priority()
timestamp = create_timestamp()
key = self._key_generator.next()
segment_num = 4
sequence_num = 0
file_adler32 = zlib.adler32(test_data)
file_md5 = hashlib.md5(test_data)
slice_start = 0
slice_end = slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
unified_id_factory = UnifiedIDFactory(1)
unified_id = unified_id_factory.next()
message = {
"message-type": "archive-key-start",
"priority": archive_priority,
"user-request-id": user_request_id,
"collection-id": collection_id,
"key": key,
"unified-id": unified_id,
"timestamp-repr": repr(timestamp),
"conjoined-part": 0,
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"zfec-padding-size": 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32": segment_adler32,
"sequence-num": sequence_num,
"source-node-name": _local_node_name,
"handoff-node-name": None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-type"], "archive-key-start-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
for _ in range(slice_count - 2):
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type": "archive-key-next",
"priority": archive_priority,
"user-request-id": user_request_id,
"collection-id": collection_id,
"key": key,
"unified-id": unified_id,
"timestamp-repr": repr(timestamp),
"conjoined-part": 0,
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"zfec-padding-size": 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32": segment_adler32,
"sequence-num": sequence_num,
"source-node-name": _local_node_name,
"handoff-node-name": None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-type"], "archive-key-next-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success",
reply["error-message"])
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message = {
"message-type": "archive-key-final",
"priority": archive_priority,
"user-request-id": user_request_id,
"collection-id": collection_id,
"key": key,
"unified-id": unified_id,
"timestamp-repr": repr(timestamp),
"conjoined-part": 0,
"segment-num": segment_num,
"segment-size": len(test_data[slice_start:slice_end]),
"zfec-padding-size": 4,
"segment-md5-digest": b64encode(segment_md5.digest()),
"segment-adler32": segment_adler32,
"sequence-num": sequence_num,
"file-size": total_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"source-node-name": _local_node_name,
"handoff-node-name": None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end])
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["user-request-id"], user_request_id)
self.assertEqual(reply["result"], "success", reply["error-message"])
def test_retrieve_large_content(self):
"""test retrieving content that fits in a multiple messages"""
slice_size = 1024 * 1024
slice_count = 10
total_size = slice_size * slice_count
test_data = random_string(total_size)
collection_id = 1001
archive_priority = create_priority()
timestamp = create_timestamp()
key = self._key_generator.next()
segment_num = 4
sequence_num = 0
file_adler32 = zlib.adler32(test_data)
file_md5 = hashlib.md5(test_data)
slice_start = 0
slice_end = slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type" : "archive-key-start",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"segment-size" : len(test_data[slice_start:slice_end]),
"segment-adler32" : segment_adler32,
"segment-md5-digest" : b64encode(segment_md5.digest()),
"sequence-num" : sequence_num,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-start-reply")
self.assertEqual(reply["result"], "success")
for _ in range(slice_count-2):
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type" : "archive-key-next",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"segment-size" : len(
test_data[slice_start:slice_end]
),
"segment-adler32" : segment_adler32,
"segment-md5-digest" : b64encode(segment_md5.digest()),
"sequence-num" : sequence_num,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-next-reply")
self.assertEqual(reply["result"], "success")
sequence_num += 1
slice_start += slice_size
slice_end += slice_size
self.assertEqual(slice_end, total_size)
segment_adler32 = zlib.adler32(test_data[slice_start:slice_end])
segment_md5 = hashlib.md5(test_data[slice_start:slice_end])
message_id = uuid.uuid1().hex
message = {
"message-type" : "archive-key-final",
"message-id" : message_id,
"priority" : archive_priority,
"collection-id" : collection_id,
"key" : key,
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"timestamp-repr" : repr(timestamp),
"segment-num" : segment_num,
"segment-size" : len(test_data[slice_start:slice_end]),
"segment-adler32" : segment_adler32,
"segment-md5-digest" : b64encode(segment_md5.digest()),
"sequence-num" : sequence_num,
"file-size" : total_size,
"file-adler32" : file_adler32,
"file-hash" : b64encode(file_md5.digest()),
"handoff-node-name" : None,
}
reply = send_request_and_get_reply(
_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=test_data[slice_start:slice_end]
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
# get file info from the local database
_conjoined_row, segment_rows = current_status_of_key(
self._database_connection, collection_id, key
)
self.assertEqual(len(segment_rows), 1)
retrieved_data_list = list()
message_id = uuid.uuid1().hex
message = {
"message-type" : "retrieve-key-start",
"message-id" : message_id,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(timestamp),
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"segment-num" : segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name,
_data_reader_address,
_local_node_name,
_client_address,
message
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
self.assertEqual(reply["completed"], False)
print "sequence-num =", reply["sequence-num"]
retrieved_data_list.append(data)
while True:
message_id = uuid.uuid1().hex
message = {
"message-type" : "retrieve-key-next",
"message-id" : message_id,
"collection-id" : collection_id,
"key" : key,
"timestamp-repr" : repr(timestamp),
"conjoined-unified-id" : None,
"conjoined-part" : 0,
"segment-num" : segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name,
_data_reader_address,
_local_node_name,
_client_address,
message
)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
retrieved_data_list.append(data)
print "sequence-num =", reply["sequence-num"]
if reply["completed"]:
break
retrieved_data = "".join(retrieved_data_list)
self.assertEqual(len(retrieved_data), len(test_data))
self.assertEqual(retrieved_data, test_data)
def __init__(self, halt_event):
self._log = logging.getLogger("WebWriter")
memcached_client = memcache.Client(_memcached_nodes)
self._interaction_pool = gdbpool.interaction_pool.DBInteractionPool(
get_central_database_dsn(),
pool_name=_central_pool_name,
pool_size=_database_pool_size,
do_log=True)
authenticator = InteractionPoolAuthenticator(memcached_client,
self._interaction_pool)
# Ticket #25: must run database operation in a greenlet
greenlet = gevent.Greenlet.spawn(_get_cluster_row_and_node_row,
self._interaction_pool)
greenlet.join()
self._cluster_row, node_row = greenlet.get()
self._unified_id_factory = UnifiedIDFactory(node_row.id)
self._deliverator = Deliverator()
self._zeromq_context = zmq.Context()
self._pull_server = GreenletPULLServer(self._zeromq_context,
_web_writer_pipeliner_address,
self._deliverator)
self._pull_server.link_exception(self._unhandled_greenlet_exception)
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_writer_pipeliner_address,
self._deliverator,
connect_messages=[])
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_writer_clients.append(resilient_client)
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")
# message sent to data writers telling them the server
# is (re)starting, thereby invalidating any archives
# that are in progress for this node
unified_id = self._unified_id_factory.next()
timestamp = create_timestamp()
self._event_push_client.info("web-writer-start",
"web writer (re)start",
unified_id=unified_id,
timestamp_repr=repr(timestamp),
source_node_name=_local_node_name)
id_translator_keys_path = os.environ.get(
"NIMBUS_IO_ID_TRANSLATION_KEYS",
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"])
redis_queue = gevent.queue.Queue()
self._redis_sink = OperationalStatsRedisSink(halt_event, redis_queue,
_local_node_name)
self._redis_sink.link_exception(self._unhandled_greenlet_exception)
self.application = Application(self._cluster_row,
self._unified_id_factory,
self._id_translator,
self._data_writer_clients,
authenticator, self._accounting_client,
self._event_push_client, redis_queue)
self.wsgi_server = WSGIServer((_web_writer_host, _web_writer_port),
application=self.application,
backlog=_wsgi_backlog)
def _retrieve(self, response, timeout):
self._log.debug("request {0}: start _retrieve".format((self.user_request_id)))
self._cache_key_rows_in_memcached(self._key_rows)
self.total_file_size = sum([row["file_size"] for row in self._key_rows])
self._log.debug("total_file_size = {0}".format(self.total_file_size))
queue_entry = redis_queue_entry_tuple(timestamp=create_timestamp(), collection_id=self._collection_id, value=1)
self._redis_queue.put(("retrieve_request", queue_entry))
retrieve_bytes = 0L
self._log.debug("start key_rows loop")
first_block = True
for entry in self._generate_key_rows(self._key_rows):
key_row, block_offset, block_count, offset_into_first_block, offset_into_last_block = entry
self._log.debug(
"request {0}: {1} {2}".format(self.user_request_id, key_row["unified_id"], key_row["conjoined_part"])
)
# if a cache port is defined, and this response isn't larger than
# the configured maximum, send the request through the cache.
target_port = _web_internal_reader_port
if (
_web_internal_reader_cache_port is not None
and key_row["file_size"] <= _web_internal_reader_max_cache_size
):
target_port = _web_internal_reader_cache_port
uri = "http://{0}:{1}/data/{2}/{3}".format(
_web_internal_reader_host, target_port, key_row["unified_id"], key_row["conjoined_part"]
)
self._log.info("request {0} internally requesting {1}".format(self.user_request_id, uri))
headers = {"x-nimbus-io-user-request-id": self.user_request_id}
if block_offset > 0 and block_count is None:
headers["range"] = "bytes={0}-".format(block_offset * block_size)
headers["x-nimbus-io-expected-content-length"] = str(key_row["file_size"] - (block_offset * block_size))
expected_status = httplib.PARTIAL_CONTENT
elif block_count is not None:
headers["range"] = "bytes={0}-{1}".format(
block_offset * block_size, (block_offset + block_count) * block_size - 1
)
headers["x-nimbus-io-expected-content-length"] = str(block_count * block_size)
expected_status = httplib.PARTIAL_CONTENT
else:
headers["x-nimbus-io-expected-content-length"] = (str(key_row["file_size"]),)
expected_status = httplib.OK
request = urllib2.Request(uri, headers=headers)
self._log.debug(
"request {0} start internal; expected={1}; headers={2}".format(
self.user_request_id, repr(expected_status), headers
)
)
try:
urllib_response = urllib2.urlopen(request, timeout=timeout)
except urllib2.HTTPError, instance:
if instance.code == httplib.NOT_FOUND:
self._log.error("request {0}: got 404".format(self.user_request_id))
response.status_int = httplib.NOT_FOUND
break
if instance.code == httplib.PARTIAL_CONTENT and expected_status == httplib.PARTIAL_CONTENT:
urllib_response = instance
else:
message = "urllib2.HTTPError '{0}' '{1}'".format(instance.code, instance)
self._log.error("request {0}: exception {1}".format(self.user_request_id, message))
self._log.exception(message)
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
break
except (httplib.HTTPException, socket.error) as instance:
message = "{0}, '{1}'".format(instance.__class__.__name__, instance.message)
self._log.error("request {0}: exception {1}".format(self.user_request_id, message))
self._log.exception(message)
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
break
# translate version id to the form we use internally
if "version_id_marker" in kwargs:
kwargs["version_id_marker"] = self._id_translator.internal_id(
kwargs["version_id_marker"])
self._log.info("request {0}: " \
"_list_versions: collection = ({1}) {2} {3} {4}".format(
user_request_id,
collection_row["id"],
collection_row["name"],
collection_row["versioning"],
kwargs))
queue_entry = \
redis_queue_entry_tuple(timestamp=create_timestamp(),
collection_id=collection_row["id"],
value=1)
self._redis_queue.put((
"listmatch_request",
queue_entry,
))
try:
result_dict = list_versions(self._interaction_pool,
collection_row["id"],
collection_row["versioning"], **kwargs)
# segment_visibility raises ValueError if it is unhappy
except ValueError, instance:
self._log.error("request {0}: {1}".format(user_request_id,
instance))
first_block = False
urllib_response.close()
self._log.debug("request {0} internal request complete".format(
self.user_request_id))
# end - for entry in self._generate_key_rows(self._key_rows):
if response.status_int in [
httplib.OK,
httplib.PARTIAL_CONTENT,
]:
redis_entries = [("retrieve_success", 1),
("success_bytes_out", retrieve_bytes)]
else:
redis_entries = [("retrieve_error", 1),
("error_bytes_out", retrieve_bytes)]
timestamp = create_timestamp()
for key, value in redis_entries:
queue_entry = \
redis_queue_entry_tuple(timestamp=timestamp,
collection_id=self._collection_id,
value=value)
self._redis_queue.put((
key,
queue_entry,
))
for data in data_list:
yield data
sent += len(data)
except RetrieveFailedError, instance:
self._log.error('retrieve failed: {0} {1}'.format(
description, instance))
self._stats["retrieves"] -= 1
response.status_int = 503
return
end_time = time.time()
self._stats["retrieves"] -= 1
self.accounting_client.retrieved(collection_id, create_timestamp(),
sent)
self._log.info(
"request {0} successful retrieve".format(user_request_id))
response_headers = dict()
if "range" in req.headers:
status_int = httplib.PARTIAL_CONTENT
response_headers["Content-Range"] = \
_content_range_header(lower_bound,
upper_bound,
total_file_size)
response_headers["Content-Length"] = slice_size
else:
status_int = httplib.OK
def audit_segments(halt_event, work_dir):
log = logging.getLogger("audit_segments")
if not os.path.exists(anti_entropy_dir):
log.info("creating {0}".format(anti_entropy_dir))
os.mkdir(anti_entropy_dir)
meta_repair_file_path = compute_meta_repair_file_path()
meta_repair_file = \
gzip.GzipFile(filename=meta_repair_file_path, mode="wb")
data_repair_file_path = compute_data_repair_file_path()
data_repair_file = \
gzip.GzipFile(filename=data_repair_file_path, mode="wb")
counts = {
"total": 0,
anti_entropy_missing_replicas: 0,
anti_entropy_incomplete_finalization: 0,
anti_entropy_damaged_records: 0,
anti_entropy_missing_tombstones: 0,
anti_entropy_database_inconsistancy: 0,
}
current_time = create_timestamp()
min_segment_age = parse_timedelta_str(_min_segment_age)
newest_allowable_timestamp = current_time - min_segment_age
log.info("newest allowable timestamp = {0}".format(
newest_allowable_timestamp.isoformat()))
for row_key, segment_status, segment_data in generate_work(work_dir):
if halt_event.is_set():
log.info("halt_event is set: exiting")
return
assert segment_status == anti_entropy_pre_audit
counts["total"] += 1
# missing replicas needs to run first, because the other tests
# assume there are no missing replicas
if _missing_replicas(segment_data, newest_allowable_timestamp):
log.debug("missing_replicas {0}".format(row_key))
counts[anti_entropy_missing_replicas] += 1
store_sized_pickle((
row_key,
anti_entropy_missing_replicas,
segment_data,
), data_repair_file)
continue
# _missing_tombstones needs to run ahead of _incomplete_finalization
if _missing_tombstones(segment_data, newest_allowable_timestamp):
log.debug("missing_tombstones {0}".format(row_key))
counts[anti_entropy_missing_tombstones] += 1
store_sized_pickle((
row_key,
anti_entropy_missing_tombstones,
segment_data,
), meta_repair_file)
continue
if _incomplete_finalization(segment_data, newest_allowable_timestamp):
log.debug("incomplete_finalization {0}".format(row_key))
counts[anti_entropy_incomplete_finalization] += 1
store_sized_pickle((
row_key,
anti_entropy_incomplete_finalization,
segment_data,
), data_repair_file)
continue
if _damaged_records(segment_data):
log.debug("damaged_records {0}".format(row_key))
counts[anti_entropy_damaged_records] += 1
store_sized_pickle((
row_key,
anti_entropy_damaged_records,
segment_data,
), data_repair_file)
continue
if _database_inconsistancy(row_key, segment_data):
log.debug("database_inconsistancy {0}".format(row_key))
counts[anti_entropy_database_inconsistancy] += 1
store_sized_pickle((
row_key,
anti_entropy_database_inconsistency,
segment_data,
), data_repair_file)
continue
meta_repair_file.close()
data_repair_file.close()
keys = [
"total", anti_entropy_missing_replicas,
anti_entropy_incomplete_finalization, anti_entropy_damaged_records,
anti_entropy_missing_tombstones, anti_entropy_database_inconsistancy
]
for key in keys:
log.info("{0} {1:,}".format(key, counts[key]))
def test_retrieve_small_content(self):
"""test retrieving content that fits in a single message"""
file_size = 10 * 64 * 1024
file_content = random_string(file_size)
collection_id = 1001
key = self._key_generator.next()
archive_priority = create_priority()
timestamp = create_timestamp()
segment_num = 2
file_adler32 = zlib.adler32(file_content)
file_md5 = hashlib.md5(file_content)
message_id = uuid.uuid1().hex
message = {
"message-type": "archive-key-entire",
"message-id": message_id,
"priority": archive_priority,
"collection-id": collection_id,
"key": key,
"conjoined-unified-id": None,
"conjoined-part": 0,
"timestamp-repr": repr(timestamp),
"segment-num": segment_num,
"segment-size": file_size,
"segment-adler32": file_adler32,
"segment-md5-digest": b64encode(file_md5.digest()),
"file-size": file_size,
"file-adler32": file_adler32,
"file-hash": b64encode(file_md5.digest()),
"handoff-node-name": None,
}
reply = send_request_and_get_reply(_local_node_name,
_data_writer_address,
_local_node_name,
_client_address,
message,
data=file_content)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "archive-key-final-reply")
self.assertEqual(reply["result"], "success")
# get file info from the local database
_conjoined_row, segment_rows = current_status_of_key(
self._database_connection, collection_id, key)
self.assertEqual(len(segment_rows), 1)
message_id = uuid.uuid1().hex
message = {
"message-type": "retrieve-key-start",
"message-id": message_id,
"collection-id": collection_id,
"key": key,
"timestamp-repr": repr(timestamp),
"conjoined-unified-id": None,
"conjoined-part": 0,
"segment-num": segment_num
}
reply, data = send_request_and_get_reply_and_data(
_local_node_name, _data_reader_address, _local_node_name,
_client_address, message)
self.assertEqual(reply["message-id"], message_id)
self.assertEqual(reply["message-type"], "retrieve-key-reply")
self.assertEqual(reply["completed"], True)
self.assertEqual(len(data), len(file_content))
self.assertEqual(data, file_content)
def _retrieve(self, response, timeout):
self._log.debug("request {0}: start _retrieve".format(
(self.user_request_id)))
self._cache_key_rows_in_memcached(self._key_rows)
self.total_file_size = sum(
[row["file_size"] for row in self._key_rows])
self._log.debug("total_file_size = {0}".format(self.total_file_size))
queue_entry = \
redis_queue_entry_tuple(timestamp=create_timestamp(),
collection_id=self._collection_id,
value=1)
self._redis_queue.put((
"retrieve_request",
queue_entry,
))
retrieve_bytes = 0L
self._log.debug("start key_rows loop")
first_block = True
for entry in self._generate_key_rows(self._key_rows):
key_row, \
block_offset, \
block_count, \
offset_into_first_block, \
offset_into_last_block = entry
self._log.debug("request {0}: {1} {2}".format(
self.user_request_id, key_row["unified_id"],
key_row["conjoined_part"]))
# if a cache port is defined, and this response isn't larger than
# the configured maximum, send the request through the cache.
target_port = _web_internal_reader_port
if (_web_internal_reader_cache_port is not None
and key_row["file_size"] <=
_web_internal_reader_max_cache_size):
target_port = _web_internal_reader_cache_port
uri = "http://{0}:{1}/data/{2}/{3}".format(
_web_internal_reader_host, target_port, key_row["unified_id"],
key_row["conjoined_part"])
self._log.info("request {0} internally requesting {1}".format(
self.user_request_id, uri))
headers = {"x-nimbus-io-user-request-id": self.user_request_id}
if block_offset > 0 and block_count is None:
headers["range"] = \
"bytes={0}-".format(block_offset * block_size)
headers["x-nimbus-io-expected-content-length"] = \
str(key_row["file_size"] - (block_offset * block_size))
expected_status = httplib.PARTIAL_CONTENT
elif block_count is not None:
headers["range"] = \
"bytes={0}-{1}".format(
block_offset * block_size,
(block_offset + block_count) * block_size - 1)
headers["x-nimbus-io-expected-content-length"] = \
str(block_count * block_size)
expected_status = httplib.PARTIAL_CONTENT
else:
headers["x-nimbus-io-expected-content-length"] = \
str(key_row["file_size"]),
expected_status = httplib.OK
request = urllib2.Request(uri, headers=headers)
self._log.debug(
"request {0} start internal; expected={1}; headers={2}".format(
self.user_request_id, repr(expected_status), headers))
try:
urllib_response = urllib2.urlopen(request, timeout=timeout)
except urllib2.HTTPError, instance:
if instance.code == httplib.NOT_FOUND:
self._log.error("request {0}: got 404".format(
self.user_request_id))
response.status_int = httplib.NOT_FOUND
break
if instance.code == httplib.PARTIAL_CONTENT and \
expected_status == httplib.PARTIAL_CONTENT:
urllib_response = instance
else:
message = "urllib2.HTTPError '{0}' '{1}'".format(
instance.code, instance)
self._log.error("request {0}: exception {1}".format(
self.user_request_id, message))
self._log.exception(message)
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
break
except (
httplib.HTTPException,
socket.error,
) as instance:
message = "{0}, '{1}'".format(instance.__class__.__name__,
instance.message)
self._log.error("request {0}: exception {1}".format(
self.user_request_id, message))
self._log.exception(message)
response.status_int = httplib.SERVICE_UNAVAILABLE
response.retry_after = _retrieve_retry_interval
break
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("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 __init__(self, halt_event):
self._log = logging.getLogger("WebWriter")
memcached_client = memcache.Client(_memcached_nodes)
self._interaction_pool = gdbpool.interaction_pool.DBInteractionPool(
get_central_database_dsn(),
pool_name=_central_pool_name,
pool_size=_database_pool_size,
do_log=True)
authenticator = InteractionPoolAuthenticator(memcached_client,
self._interaction_pool)
# Ticket #25: must run database operation in a greenlet
greenlet = gevent.Greenlet.spawn(_get_cluster_row_and_node_row,
self._interaction_pool)
greenlet.join()
self._cluster_row, node_row = greenlet.get()
self._unified_id_factory = UnifiedIDFactory(node_row.id)
self._deliverator = Deliverator()
self._zeromq_context = zmq.Context()
self._pull_server = GreenletPULLServer(
self._zeromq_context,
_web_writer_pipeliner_address,
self._deliverator
)
self._pull_server.link_exception(self._unhandled_greenlet_exception)
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_writer_pipeliner_address,
self._deliverator,
connect_messages=[]
)
resilient_client.link_exception(self._unhandled_greenlet_exception)
self._data_writer_clients.append(resilient_client)
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"
)
# message sent to data writers telling them the server
# is (re)starting, thereby invalidating any archives
# that are in progress for this node
unified_id = self._unified_id_factory.next()
timestamp = create_timestamp()
self._event_push_client.info("web-writer-start",
"web writer (re)start",
unified_id=unified_id,
timestamp_repr=repr(timestamp),
source_node_name=_local_node_name)
id_translator_keys_path = os.environ.get(
"NIMBUS_IO_ID_TRANSLATION_KEYS",
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"]
)
redis_queue = gevent.queue.Queue()
self._redis_sink = OperationalStatsRedisSink(halt_event,
redis_queue,
_local_node_name)
self._redis_sink.link_exception(self._unhandled_greenlet_exception)
self.application = Application(
self._cluster_row,
self._unified_id_factory,
self._id_translator,
self._data_writer_clients,
authenticator,
self._accounting_client,
self._event_push_client,
redis_queue
)
self.wsgi_server = WSGIServer((_web_writer_host, _web_writer_port),
application=self.application,
backlog=_wsgi_backlog
)
def _value_file_status(connection, entry):
log = logging.getLogger("_value_file_status")
batch_key = make_batch_key(entry)
value_file_path = compute_value_file_path(_repository_path,
entry.space_id,
entry.value_file_id)
# Always do a stat on the value file.
try:
stat_result = os.stat(value_file_path)
except OSError as instance:
# If the value file is missing, consider all of the segment_sequences
# to be missing, and handle it as such.
if instance.errno == errno.ENOENT:
log.error("value file missing {0} {1}".format(batch_key,
value_file_path))
return _value_file_missing
log.error("Error stat'ing value file {0} {1} {2}".format(
str(instance), batch_key, value_file_path))
raise
# If the value file is still open, consider all data in it undammaged.
if entry.value_file_close_time is None:
return _value_file_valid
# If the value file exists, is closed, and has an md5 in the database,
# has a size in the database, and the size in the stat matches the size
# in the database, and has a close_time or a last_integrity_check_time
# that is younger than (MAX_TIME_BETWEEN_VALUE_FILE_INTEGRITY_CHECK)
# consider all records in the file undammaged. (This is the common case.)
if entry.value_file_hash is None:
log.info("Value file row has no md5 hash {0} {1}".format(batch_key,
entry))
return _value_file_questionable
if entry.value_file_size is None:
log.info("Value file row has no size {0} {1}".format(batch_key,
entry))
return _value_file_questionable
if entry.value_file_size != stat_result.st_size:
log.info("Value file row size {0} != stat size {1} {2}".format(
entry.value_file_size, stat_result.st_size, batch_key))
return _value_file_questionable
current_time = create_timestamp()
value_file_row_age = current_time - entry.value_file_close_time
if entry.value_file_last_integrity_check_time is not None:
value_file_row_age = \
current_time - entry.value_file_last_integrity_check_time
if value_file_row_age < _max_value_file_time:
return _value_file_valid
value_file_result = _value_file_valid
# If the value matches all the previous criteria EXCEPT the
# MAX_TIME_BETWEEN_VALUE_FILE_INTEGRITY_CHECK, then read the whole file,
# and calculate the md5. If it matches, consider the whole file good as
# above. Update last_integrity_check_time regardless.
md5_sum = hashlib.md5()
try:
with open(value_file_path, "rb") as input_file:
while True:
data = input_file.read(_read_buffer_size)
if len(data) == 0:
break
md5_sum.update(data)
except (OSError, IOError) as instance:
log.error("Error reading {0} {1}".format(value_file_path,
instance))
value_file_result = _value_file_questionable
if value_file_result == _value_file_valid and \
md5_sum.digest() != bytes(entry.value_file_hash):
log.error(
"md5 mismatch {0} {1} {2} {3}".format(md5_sum.digest(),
bytes(entry.value_file_hash),
batch_key,
value_file_path))
value_file_result = _value_file_questionable
# we're only supposed to do this after we've also read the file
# and inserted any damage. not before. otherwise it's a race condition --
# we may crash before finishing checking the file, and then the file
# doesn't get checked, but it's marked as checked.
_update_value_file_last_integrity_check_time(connection,
entry.value_file_id,
create_timestamp())
return value_file_result
