def get(self, table_name, *keys, **kwargs):
num_uuids = self._table_names[table_name]
## A mapping between shards and columns to retrieve from each shard
shards = defaultdict(list)
## Track the keys that we find
found_keys = set()
## Determine all the shards that we need to contact
for key in keys:
assert len(key) == num_uuids
joined_key = join_uuids(*key)
row_names = [self._make_shard_name(table_name, joined_key)]
for row_name in row_names:
shards[row_name].append(joined_key)
## For each shard we contact, get all the keys it may hold
for row_name, columns in shards.iteritems():
try:
for key, val in self.tables[table_name].get(row_name, columns=columns,).iteritems():
key = split_uuids(key)
logger.critical('cassandra get(%r) yielding %r %d' % (table_name, key, len(val)))
## key is a list [(uuid, uuid)]
assert not any(k in found_keys for k in key)
found_keys.add(tuple(key))
yield tuple(key), val
except NotFoundException:
pass
## Raise an exception if we don't retrieve all the keys that were requested
missing_keys = set(keys) - found_keys
for k in missing_keys:
yield k, None
def scan(self, table_name, *key_ranges, **kwargs):
kwargs.pop('batch_size', 100)
if not key_ranges:
## get all columns
key_ranges = [['', '']]
# TODO: s/num_uuids/key_spec/g
num_uuids = self._table_names[table_name]
for start, finish in key_ranges:
specific_key_range = bool( start or finish )
if specific_key_range and start == finish and len(start) == num_uuids:
logger.warn('doing a scan on a single element, what?')
#logger.info('specific_key_range: %r %r' % (start, finish))
assert len(start) == num_uuids
joined_key = join_uuids(*start)
columns = [joined_key]
row_names = [self._make_shard_name(table_name, joined_key)]
start = None
finish = None
else:
columns = None
# TODO: require_uuid and num_uuids is obsolete. fix.
start = make_start_key(start, uuid_mode=self._require_uuid, num_uuids=num_uuids)
finish = make_end_key(finish, uuid_mode=self._require_uuid, num_uuids=num_uuids)
row_names = self._make_shard_names(table_name, start, finish)
total_count = 0
hit_empty = False
for row_name in row_names:
try:
for key, val in self._get_from_one_row(table_name, row_name, columns, start, finish, num_uuids):
assert len(key) == num_uuids
yield key, val
if start:
assert start <= join_uuids(*key)
if finish:
assert finish >= join_uuids(*key)
total_count += 1
#logger.critical('total_count: %d' % total_count)
except pycassa.NotFoundException:
hit_empty = True
def delete(self, table_name, *keys, **kwargs):
num_uuids = self._table_names[table_name]
batch_size = kwargs.pop('batch_size', 1000)
batch = self.tables[table_name].batch(queue_size=batch_size)
count = 0
for key in keys:
assert len(key) == num_uuids
joined_key = join_uuids(*key)
row_name = self._make_shard_name(table_name, joined_key)
columns = [joined_key]
#logger.critical('C* delete: table_name=%r columns=%r' % (table_name, columns))
batch.remove(row_name, columns=columns)
count += 1
batch.send()
logger.info('deleted %d tree_ids from %r' % (count, table_name))
def _get_from_one_row(self, table_name, row_name, columns, start, finish, num_uuids):
logger.debug('c* get: table_name=%r row_name=%r columns=%r start=%r finish=%r' % (
table_name, row_name, columns, start, finish))
if not columns:
assert start is not None and finish is not None
assert start <= finish
num_yielded = 0
while True:
## if we have
prev_start = start
logger.debug('cassandra get(%r...)' % row_name)
## if table_name == 'inbound':
## import ipdb
## ipdb.set_trace()
for key, val in self.tables[table_name].get(
row_name,
columns=columns,
column_start=start,
column_finish=finish,
column_count=1,
).iteritems():
key = split_uuids(key)
logger.critical('cassandra get(%r) yielding %r %d' % (table_name, key, len(val)))
yield key, val
num_yielded += 1
logger.debug('c* get: table_name=%r row_name=%r columns=%r start=%r finish=%r' % (
table_name, row_name, columns, start, finish))
## prepare to page ahead to next batch
if columns:
break
start = list(key)
start[-1] = uuid.UUID(int=key[-1].int+1)
assert len(start) == num_uuids
start = join_uuids(*start)
if start == prev_start or start > finish:
break
logger.debug('paging forward from %r to %r' % (prev_start, start))
## We need to raise a not found exception if the caller asked for
## a specific column and we didn't yield any results
if not columns and num_yielded == 0:
raise pycassa.NotFoundException
def put(self, table_name, *keys_and_values, **kwargs):
batch_size = kwargs.pop('batch_size', None)
tot_bytes = 0
cur_bytes = 0
tot_rows = 0
cur_rows = 0
num_uuids = self._table_names[table_name]
start = time.time()
logger.debug('starting save')
batch = self.tables[table_name].batch(queue_size=batch_size)
for key, blob in keys_and_values:
self.check_put_key_value(key, blob, table_name, num_uuids)
if len(blob) + cur_bytes >= self.thrift_framed_transport_size_in_mb * 2**19:
logger.critical('len(blob)=%d + cur_bytes=%d >= thrift_framed_transport_size_in_mb/2 = %d'
% (len(blob), cur_bytes, self.thrift_framed_transport_size_in_mb * 2**19))
## convert to MB and then cut in half
if cur_rows > 0:
logger.critical('pre-emptively sending only what has been batched, and will send this item in next batch.')
batch.send()
cur_bytes = 0
cur_rows = 0
cur_bytes += len(blob)
tot_bytes += len(blob)
cur_rows += 1
tot_rows += 1
if not isinstance(key, tuple):
## for consistency, always make things join_uuids,
## even if only one
key = (key,)
joined_key = join_uuids(*key)
row_name = self._make_shard_name(table_name, joined_key)
if len(blob) >= self.thrift_framed_transport_size_in_mb * 2**19:
logger.critical('len(blob)=%d >= thrift_framed_transport_size_in_mb / 2 = %d, so there is a risk that the total payload will exceed the full thrift_framed_transport_size_in_mb, and the only solution to this is to change Cassandra server-side config to allow larger frames...'
% (len(blob), self.thrift_framed_transport_size_in_mb * 2**19))
batch.insert(row_name, {joined_key: blob})
#logger.critical('saving %s %r %r' % (table_name, key, blob))
if tot_rows % 500 == 0:
logger.debug('num rows=%d, num MB=%d, thrift_framed_transport_size_in_mb=%d' % (
tot_rows, float(tot_bytes) / 2**20, self.thrift_framed_transport_size_in_mb))
batch.send()
elapsed = time.time() - start
row_rate = float(tot_rows) / elapsed
MB_rate = float(tot_bytes) / elapsed / 2**20
logger.info('%s.insert(%d rows, %d bytes in %.1f sec --> %.1f rows/sec %.3f MBps' % (
table_name, tot_rows, tot_bytes, elapsed, row_rate, MB_rate))
