def mapPartitions(self, _table, func):
func_id, func_bytes = self.serialize_and_hash_func(func)
results = []
for partition in range(_table.partition):
operand = EggRoll.__get_storage_locator(_table, partition)
unary_p = processor_pb2.UnaryProcess(operand=operand,
info=processor_pb2.TaskInfo(task_id=self.job_id,
function_id=func_id,
function_bytes=func_bytes))
proc_id = partition % len(self.proc_list)
channel, stub = self.proc_list[proc_id]
results.append(stub.mapPartitions.future(unary_p))
for r in results:
result = r.result()
return _DTable(self, result.type, result.namespace, result.name, _table.partition)
def join(self, left, right, func):
func_id, func_bytes = self.serialize_and_hash_func(func)
results = []
res = None
for partition in range(left.partition):
l_op = EggRoll.__get_storage_locator(left, partition)
r_op = EggRoll.__get_storage_locator(right, partition)
binary_p = processor_pb2.BinaryProcess(left=l_op, right=r_op,
info=processor_pb2.TaskInfo(task_id=self.job_id,
function_id=func_id,
function_bytes=func_bytes))
proc_id = partition % len(self.proc_list)
channel, stub = self.proc_list[proc_id]
results.append(stub.join.future(binary_p))
for r in results:
res = r.result()
return _DTable(self, res.type, res.namespace, res.name, left.partition)
def reduce(self, _table: _DTable, func):
func_id, func_bytes = self.serialize_and_hash_func(func)
operand = storage_basic_pb2.StorageLocator(namespace=_table._namespace, type=_table._type, name=_table._name)
unary_p = processor_pb2.UnaryProcess(operand=operand,
info=processor_pb2.TaskInfo(task_id=self.job_id,
function_id=func_id,
function_bytes=func_bytes))
values = [_EggRoll._deserialize_operand(operand) for operand in self.proc_stub.reduce(unary_p)]
values = [v for v in filter(partial(is_not, None), values)]
if len(values) <= 0:
return None
if len(values) == 1:
return values[0]
else:
val, *remain = values
for _nv in remain:
val = func(val, _nv)
return val
def sample(self, _table, fraction, seed):
if fraction < 0 or fraction > 1:
raise ValueError("fraction must be in [0, 1]")
func_bytes = self._serdes.serialize((fraction, seed))
results = []
func_id = str(uuid.uuid1())
for p in range(_table.partition):
operand = EggRoll.__get_storage_locator(_table, p)
unary_p = processor_pb2.UnaryProcess(
operand=operand,
info=processor_pb2.TaskInfo(task_id=self.job_id,
function_id=func_id,
function_bytes=func_bytes))
proc_id = p % len(self.proc_list)
channel, stub = self.proc_list[proc_id]
results.append(stub.sample.future(unary_p))
for r in results:
result = r.result()
return _DTable(self, result.type, result.namespace, result.name,
_table.partition)
def reduce(self, _table, func):
func_id, func_bytes = self.serialize_and_hash_func(func)
rtn = None
results = []
for partition in range(_table.partition):
operand = EggRoll.__get_storage_locator(_table, partition)
proc_id = partition % len(self.proc_list)
channel, stub = self.proc_list[proc_id]
unary_p = processor_pb2.UnaryProcess(operand=operand, info=processor_pb2.TaskInfo(task_id=self.job_id,
function_id=func_id,
function_bytes=func_bytes))
results = results + list(stub.reduce(unary_p))
rs = []
for val in results:
if len(val.value) > 0:
rs.append(self._serdes.deserialize(val.value))
rs = [r for r in filter(partial(is_not, None), rs)]
if len(results) <= 0:
return rtn
rtn = rs[0]
for r in rs[1:]:
rtn = func(rtn, r)
return rtn