def run_coder_benchmarks(num_runs, input_size, seed, verbose):
random.seed(seed)
# TODO(BEAM-4441): Pick coders using type hints, for example:
# tuple_coder = typecoders.registry.get_coder(typehints.Tuple[int, ...])
benchmarks = [
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_int),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_int),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_string),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_string),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_list),
coder_benchmark_factory(
coders.IterableCoder(coders.FastPrimitivesCoder()), small_list),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_list),
coder_benchmark_factory(
coders.IterableCoder(coders.FastPrimitivesCoder()), large_list),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_tuple),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_tuple),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_dict),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_dict),
coder_benchmark_factory(
coders.WindowedValueCoder(coders.FastPrimitivesCoder()),
wv_with_one_window),
coder_benchmark_factory(
coders.WindowedValueCoder(coders.FastPrimitivesCoder()),
wv_with_multiple_windows),
]
suite = [
utils.BenchmarkConfig(b, input_size, num_runs) for b in benchmarks
]
utils.run_benchmarks(suite, verbose=verbose)
def run_coder_benchmarks(num_runs,
input_size,
seed,
verbose,
filter_regex='.*'):
random.seed(seed)
# TODO(BEAM-4441): Pick coders using type hints, for example:
# tuple_coder = typecoders.registry.get_coder(typing.Tuple[int, ...])
benchmarks = [
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_int),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_int),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_string),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_string),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_list),
coder_benchmark_factory(
coders.IterableCoder(coders.FastPrimitivesCoder()), small_list),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_list),
coder_benchmark_factory(
coders.IterableCoder(coders.FastPrimitivesCoder()), large_list),
coder_benchmark_factory(
coders.IterableCoder(coders.FastPrimitivesCoder()),
large_iterable),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_tuple),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_tuple),
coder_benchmark_factory(coders.FastPrimitivesCoder(), small_dict),
coder_benchmark_factory(coders.FastPrimitivesCoder(), large_dict),
coder_benchmark_factory(coders.ProtoCoder(test_message.MessageWithMap),
small_message_with_map),
coder_benchmark_factory(coders.ProtoCoder(test_message.MessageWithMap),
large_message_with_map),
coder_benchmark_factory(
coders.DeterministicProtoCoder(test_message.MessageWithMap),
small_message_with_map),
coder_benchmark_factory(
coders.DeterministicProtoCoder(test_message.MessageWithMap),
large_message_with_map),
coder_benchmark_factory(
coders.WindowedValueCoder(coders.FastPrimitivesCoder()),
wv_with_one_window),
coder_benchmark_factory(
coders.WindowedValueCoder(coders.FastPrimitivesCoder(),
coders.IntervalWindowCoder()),
wv_with_multiple_windows),
coder_benchmark_factory(
coders.WindowedValueCoder(coders.FastPrimitivesCoder(),
coders.GlobalWindowCoder()),
globally_windowed_value),
coder_benchmark_factory(
coders.LengthPrefixCoder(coders.FastPrimitivesCoder()), small_int)
]
suite = [
utils.BenchmarkConfig(b, input_size, num_runs) for b in benchmarks
if re.search(filter_regex, b.__name__, flags=re.I)
]
utils.run_benchmarks(suite, verbose=verbose)
def test_windowedvalue_coder_paneinfo(self):
coder = coders.WindowedValueCoder(coders.VarIntCoder(),
coders.GlobalWindowCoder())
test_paneinfo_values = [
windowed_value.PANE_INFO_UNKNOWN,
windowed_value.PaneInfo(
True, True, windowed_value.PaneInfoTiming.EARLY, 0, -1),
windowed_value.PaneInfo(
True, False, windowed_value.PaneInfoTiming.ON_TIME, 0, 0),
windowed_value.PaneInfo(
True, False, windowed_value.PaneInfoTiming.ON_TIME, 10, 0),
windowed_value.PaneInfo(
False, True, windowed_value.PaneInfoTiming.ON_TIME, 0, 23),
windowed_value.PaneInfo(
False, True, windowed_value.PaneInfoTiming.ON_TIME, 12, 23),
windowed_value.PaneInfo(
False, False, windowed_value.PaneInfoTiming.LATE, 0, 123),]
test_values = [windowed_value.WindowedValue(123, 234, (GlobalWindow(),), p)
for p in test_paneinfo_values]
# Test unnested.
self.check_coder(coder, windowed_value.WindowedValue(
123, 234, (GlobalWindow(),), windowed_value.PANE_INFO_UNKNOWN))
for value in test_values:
self.check_coder(coder, value)
# Test nested.
for value1 in test_values:
for value2 in test_values:
self.check_coder(coders.TupleCoder((coder, coder)), (value1, value2))
def test_windowed_value_coder(self):
coder = coders.WindowedValueCoder(
coders.VarIntCoder(), coders.GlobalWindowCoder())
# Verify cloud object representation
self.assertEqual({
'@type': 'kind:windowed_value',
'is_wrapper': True,
'component_encodings': [
coders.VarIntCoder().as_cloud_object(),
coders.GlobalWindowCoder().as_cloud_object(),
],
},
coder.as_cloud_object())
# Test binary representation
self.assertEqual(
b'\x7f\xdf;dZ\x1c\xac\t\x00\x00\x00\x01\x0f\x01',
coder.encode(window.GlobalWindows.windowed_value(1)))
# Test decoding large timestamp
self.assertEqual(
coder.decode(b'\x7f\xdf;dZ\x1c\xac\x08\x00\x00\x00\x01\x0f\x00'),
windowed_value.create(0, MIN_TIMESTAMP.micros, (GlobalWindow(), )))
# Test unnested
self.check_coder(
coders.WindowedValueCoder(coders.VarIntCoder()),
windowed_value.WindowedValue(3, -100, ()),
windowed_value.WindowedValue(-1, 100, (1, 2, 3)))
# Test Global Window
self.check_coder(
coders.WindowedValueCoder(
coders.VarIntCoder(), coders.GlobalWindowCoder()),
window.GlobalWindows.windowed_value(1))
# Test nested
self.check_coder(
coders.TupleCoder((
coders.WindowedValueCoder(coders.FloatCoder()),
coders.WindowedValueCoder(coders.StrUtf8Coder()))),
(
windowed_value.WindowedValue(1.5, 0, ()),
windowed_value.WindowedValue("abc", 10, ('window', ))))
def test_param_windowed_value_coder(self):
from apache_beam.transforms.window import IntervalWindow
from apache_beam.utils.windowed_value import PaneInfo
wv = windowed_value.create(
b'',
# Milliseconds to microseconds
1000 * 1000,
(IntervalWindow(11, 21),),
PaneInfo(True, False, 1, 2, 3))
windowed_value_coder = coders.WindowedValueCoder(
coders.BytesCoder(), coders.IntervalWindowCoder())
payload = windowed_value_coder.encode(wv)
coder = coders.ParamWindowedValueCoder(
payload, [coders.VarIntCoder(), coders.IntervalWindowCoder()])
# Test binary representation
self.assertEqual(b'\x01',
coder.encode(window.GlobalWindows.windowed_value(1)))
# Test unnested
self.check_coder(
coders.ParamWindowedValueCoder(
payload, [coders.VarIntCoder(), coders.IntervalWindowCoder()]),
windowed_value.WindowedValue(
3,
1,
(window.IntervalWindow(11, 21),),
PaneInfo(True, False, 1, 2, 3)),
windowed_value.WindowedValue(
1,
1,
(window.IntervalWindow(11, 21),),
PaneInfo(True, False, 1, 2, 3)))
# Test nested
self.check_coder(
coders.TupleCoder((
coders.ParamWindowedValueCoder(
payload, [
coders.FloatCoder(),
coders.IntervalWindowCoder()]),
coders.ParamWindowedValueCoder(
payload, [
coders.StrUtf8Coder(),
coders.IntervalWindowCoder()]))),
(windowed_value.WindowedValue(
1.5,
1,
(window.IntervalWindow(11, 21),),
PaneInfo(True, False, 1, 2, 3)),
windowed_value.WindowedValue(
"abc",
1,
(window.IntervalWindow(11, 21),),
PaneInfo(True, False, 1, 2, 3))))
def test_nested_observables(self):
class FakeObservableIterator(observable.ObservableMixin):
def __iter__(self):
return iter([1, 2, 3])
# Coder for elements from the observable iterator.
elem_coder = coders.VarIntCoder()
iter_coder = coders.TupleSequenceCoder(elem_coder)
# Test nested WindowedValue observable.
coder = coders.WindowedValueCoder(iter_coder)
observ = FakeObservableIterator()
value = windowed_value.WindowedValue(observ, 0, ())
self.assertEqual(
coder.get_impl().get_estimated_size_and_observables(value)[1],
[(observ, elem_coder.get_impl())])
# Test nested tuple observable.
coder = coders.TupleCoder((coders.StrUtf8Coder(), iter_coder))
value = (u'123', observ)
self.assertEqual(
coder.get_impl().get_estimated_size_and_observables(value)[1],
[(observ, elem_coder.get_impl())])
class StandardCodersTest(unittest.TestCase):
_urn_to_coder_class = {
'urn:beam:coders:bytes:0.1': coders.BytesCoder,
'urn:beam:coders:varint:0.1': coders.VarIntCoder,
'urn:beam:coders:kv:0.1': lambda k, v: coders.TupleCoder((k, v)),
'urn:beam:coders:interval_window:0.1': coders.IntervalWindowCoder,
'urn:beam:coders:stream:0.1': lambda t: coders.IterableCoder(t),
'urn:beam:coders:global_window:0.1': coders.GlobalWindowCoder,
'urn:beam:coders:windowed_value:0.1':
lambda v, w: coders.WindowedValueCoder(v, w)
}
_urn_to_json_value_parser = {
'urn:beam:coders:bytes:0.1': lambda x: x,
'urn:beam:coders:varint:0.1': lambda x: x,
'urn:beam:coders:kv:0.1':
lambda x, key_parser, value_parser: (key_parser(x['key']),
value_parser(x['value'])),
'urn:beam:coders:interval_window:0.1':
lambda x: IntervalWindow(
start=Timestamp(micros=(x['end'] - x['span']) * 1000),
end=Timestamp(micros=x['end'] * 1000)),
'urn:beam:coders:stream:0.1': lambda x, parser: map(parser, x),
'urn:beam:coders:global_window:0.1': lambda x: window.GlobalWindow(),
'urn:beam:coders:windowed_value:0.1':
lambda x, value_parser, window_parser: windowed_value.create(
value_parser(x['value']), x['timestamp'] * 1000,
tuple([window_parser(w) for w in x['windows']]))
}
def test_standard_coders(self):
for name, spec in _load_test_cases(STANDARD_CODERS_YAML):
logging.info('Executing %s test.', name)
self._run_standard_coder(name, spec)
def _run_standard_coder(self, name, spec):
coder = self.parse_coder(spec['coder'])
parse_value = self.json_value_parser(spec['coder'])
nested_list = [spec['nested']] if 'nested' in spec else [True, False]
for nested in nested_list:
for expected_encoded, json_value in spec['examples'].items():
value = parse_value(json_value)
expected_encoded = expected_encoded.encode('latin1')
if not spec['coder'].get('non_deterministic', False):
actual_encoded = encode_nested(coder, value, nested)
if self.fix and actual_encoded != expected_encoded:
self.to_fix[spec['index'], expected_encoded] = actual_encoded
else:
self.assertEqual(expected_encoded, actual_encoded)
self.assertEqual(decode_nested(coder, expected_encoded, nested),
value)
else:
# Only verify decoding for a non-deterministic coder
self.assertEqual(decode_nested(coder, expected_encoded, nested),
value)
def parse_coder(self, spec):
return self._urn_to_coder_class[spec['urn']](
*[self.parse_coder(c) for c in spec.get('components', ())])
def json_value_parser(self, coder_spec):
component_parsers = [
self.json_value_parser(c) for c in coder_spec.get('components', ())]
return lambda x: self._urn_to_json_value_parser[coder_spec['urn']](
x, *component_parsers)
# Used when --fix is passed.
fix = False
to_fix = {}
@classmethod
def tearDownClass(cls):
if cls.fix and cls.to_fix:
print "FIXING", len(cls.to_fix), "TESTS"
doc_sep = '\n---\n'
docs = open(STANDARD_CODERS_YAML).read().split(doc_sep)
def quote(s):
return json.dumps(s.decode('latin1')).replace(r'\u0000', r'\0')
for (doc_ix, expected_encoded), actual_encoded in cls.to_fix.items():
print quote(expected_encoded), "->", quote(actual_encoded)
docs[doc_ix] = docs[doc_ix].replace(
quote(expected_encoded) + ':', quote(actual_encoded) + ':')
open(STANDARD_CODERS_YAML, 'w').write(doc_sep.join(docs))
