Skip to content

[python] Support DataSplit v9 on release-2.0 #10149

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
XiaoHongbo-Hope wants to merge 1 commit into from
Closed

[python] Support DataSplit v9 on release-2.0 #10149

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 12 additions & 12 deletions paimon-python/pypaimon/read/split_serializer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@
"""Deserialize the cross-language ``SplitSerializer`` v1 binary into a pypaimon
:class:`DataSplit`.

Mirror of the Java ``DataSplit#serialize`` (VERSION 8) frame wrapped in the
Mirror of the Java ``DataSplit#serialize`` (versions 8 and 9) frame wrapped in the
``SplitSerializer`` v1 header, as produced by ``pypaimon_rust``'s
``Split.serialize()``. Extracts the fields the reader needs, plus per-file
min/max keys for PK merge-on-read; key/value stats (planning-only) stay empty.
Expand Down Expand Up @@ -46,16 +46,14 @@
_INDEXED_SPLIT_MAGIC = -938472394838495695
_INDEXED_SPLIT_VERSION = 1
_SPLIT_MAGIC = -2394839472490812314
_SPLIT_VERSION = 8

_DFM_ARITY = 20
_DFM_ARITY_BY_VERSION = {8: 20, 9: 21}


def _f(idx, name, dtype):
return DataField(idx, name, dtype)


# DataFileMeta 20-field layout (order/types mirror DataFileMetaSerializer#toRow).
# DataFileMeta layout (order/types mirror DataFileMetaSerializer#toRow).
# Fields 3/4 (min/max key) are decoded for PK tables; 5/6 (stats) stay unread.
_DFM_FIELDS: List[DataField] = [
_f(0, '_FILE_NAME', AtomicType('STRING')),
Expand All @@ -78,10 +76,11 @@ def _f(idx, name, dtype):
_f(17, '_EXTERNAL_PATH', AtomicType('STRING')),
_f(18, '_FIRST_ROW_ID', AtomicType('BIGINT')),
_f(19, '_WRITE_COLS', AtomicType('BYTES')),
# Added in v9 for compaction/index refresh; unused by the Python reader.
_f(20, '_WRITE_COLS_SEQUENCES', AtomicType('BYTES')),
]

# Arity is fixed by DataSplit VERSION 8; keep the field list and arity in lockstep.
assert len(_DFM_FIELDS) == _DFM_ARITY
assert len(_DFM_FIELDS) == max(_DFM_ARITY_BY_VERSION.values())


def _decode_str_array(b: Optional[bytes]) -> Optional[List[str]]:
Expand Down Expand Up @@ -214,9 +213,9 @@ def _read_datasplit_body(r: _Reader, partition_fields: List[DataField],
if r.i64() != _SPLIT_MAGIC:
raise ValueError("bad DataSplit magic")
version = r.i32()
if version != _SPLIT_VERSION:
if version not in _DFM_ARITY_BY_VERSION:
raise ValueError(
"unsupported DataSplit version %d (expected %d)" % (version, _SPLIT_VERSION))
"unsupported DataSplit version %d (expected 8 or 9)" % version)
snapshot_id = r.i64() # scanned snapshot; row-id conflict detection needs it
partition = GenericRowDeserializer.from_bytes(r.take(r.i32()), partition_fields)
bucket = r.i32()
Expand All @@ -228,7 +227,8 @@ def _read_datasplit_body(r: _Reader, partition_fields: List[DataField],
if r.u8() != 0: # beforeDeletionFiles must be null
raise ValueError("cannot deserialize a split with before deletion files")
file_count = r.i32()
files = [_datafilemeta_from_row(r.take(r.i32()), bucket_path, key_fields)
arity = _DFM_ARITY_BY_VERSION[version]
files = [_datafilemeta_from_row(r.take(r.i32()), bucket_path, arity, key_fields)
for _ in range(file_count)]
data_deletion_files = _read_deletion_list(r)
r.u8() # isStreaming
Expand All @@ -250,9 +250,9 @@ def _decode_key(b: Optional[bytes], key_fields: Optional[List[DataField]]) -> Ge
return GenericRowDeserializer.from_bytes(b, key_fields)


def _datafilemeta_from_row(row_bytes: bytes, bucket_path: str,
def _datafilemeta_from_row(row_bytes: bytes, bucket_path: str, arity: int,
key_fields: Optional[List[DataField]] = None) -> DataFileMeta:
row = BinaryRow(struct.pack('>i', _DFM_ARITY) + row_bytes, _DFM_FIELDS)
row = BinaryRow(struct.pack('>i', arity) + row_bytes, _DFM_FIELDS[:arity])
g = row.get_field
file_name = g(0)
external_path = g(17)
Expand Down
68 changes: 67 additions & 1 deletion paimon-python/pypaimon/tests/split_serializer_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@
# under the License.

import base64
import struct
import unittest

from pypaimon.globalindex.indexed_split import IndexedSplit
Expand Down Expand Up @@ -56,7 +57,7 @@
)

# Same split serialized as an IndexedSplit (type id 3): identical DataSplit body
# with a trailing row-ranges/scores section that the reader skips.
# with a trailing row-ranges/scores section that must survive decoding.
_GOLDEN_INDEXED_SPLIT_V1 = base64.b64decode(
"U1BMSVRfVjEAAAABAAAAA/L54FRCJC4xAAAAAd7D0jAsGexmAAAACAAAAAAAAAAqAAAAHAAAAAIA"
"AAAAAAAAAOoHAAAAAAAABwAAAAAAAAAAAAADABRkdD0yMDI2MDcwNi9idWNrZXQtMwEAAAAIAAAA"
Expand All @@ -78,6 +79,26 @@
)


# Java DataSplitCompatibleTest's v9 golden, also used by paimon-rust's
# serialize_matches_datasplit_v9 test. Includes the new non-null per-column
# sequence numbers [15, 100, 150, 200] after write_cols.
_GOLDEN_DATA_SPLIT_V9 = base64.b64decode(
"3sPSMCwZ7GYAAAAJAAAAAAAAABIAAAAUAAAAAQAAAAAAAAAAYWFhYWEAAIUAAAAUAAdteSBwYXRo"
"AQAAACAAAAAAAAAAAAEAAAJoAAAAAAAAAABteV9maWxlhwAAEAAAAAAAAAQAAAAAAAAUAAAAsAAA"
"ABQAAADIAAAAYAAAAOAAAACAAAAAQAEAAA8AAAAAAAAAyAAAAAAAAAAFAAAAAAAAAAMAAAAAAAAA"
"GAAAAMABAABgTEpMfwEAAAsAAAAAAAAAAQIEAAAAAIMBAAAAAAAAACAAAADYAQAAGQAAAPgBAAAM"
"AAAAAAAAACgAAAAYAgAAKAAAAEACAAAAAAABAAAAAAAAAABtaW5fa2V5hwAAAAAAAAABAAAAAAAA"
"AABtYXhfa2V5hwAAAAAAAAAAAAAAABQAAAAgAAAAFAAAADgAAAAQAAAAUAAAAAAAAAEAAAAAAAAA"
"AG1pbl9rZXmHAAAAAAAAAAEAAAAAAAAAAG1heF9rZXmHAAAAAAEAAAAAAAAAAAAAAAAAAAAAAAAA"
"AAAAACQAAAAgAAAAJAAAAEgAAAAQAAAAcAAAAAAAAAEAAAAAAAAAAAkAAAAQAAAAbWluX3ZhbHVl"
"AAAAAAAAAAAAAAAAAAABAAAAAAAAAAAJAAAAEAAAAG1heF92YWx1ZQAAAAAAAAAAAAAAAQAAAAAA"
"AAAAAAAAAAAAAAIAAAAAAAAAZXh0cmExAIZleHRyYTIAhgMAAAAAAAAAZmllbGQxAIZmaWVsZDIA"
"hmZpZWxkMwCGaGRmczovLy9wYXRoL3RvL3dhcmVob3VzZQAAAAAAAAAEAAAAAAAAAGEAAAAAAACB"
"YgAAAAAAAIFjAAAAAAAAgWYAAAAAAACBBAAAAAAAAAAPAAAAAAAAAGQAAAAAAAAAlgAAAAAAAADI"
"AAAAAAAAAAEAAAABAQANZGVsZXRpb25fZmlsZQAAAAAAAABkAAAAAAAAABYAAAAAAAAAIQAA"
)


class SplitSerializerTest(unittest.TestCase):

def _partition_fields(self):
Expand Down Expand Up @@ -139,6 +160,51 @@ def test_decode_modified_utf8_supplementary_char(self):
_decode_modified_utf8(bytes([0xED, 0xA0, 0xBD, 0xED, 0xB8, 0x80])),
'\U0001F600')

def test_deserialize_v9_data_and_indexed_splits(self):
fields = [DataField(0, 's', AtomicType('STRING'))]
for indexed in (False, True):
with self.subTest(indexed=indexed):
if indexed:
data = (_GOLDEN_INDEXED_SPLIT_V1[:28] + _GOLDEN_DATA_SPLIT_V9
+ struct.pack('>iqqBif', 1, 13, 14, 1, 1, 0.75))
else:
data = _GOLDEN_DATA_SPLIT_V1[:16] + _GOLDEN_DATA_SPLIT_V9
split = deserialize_split_v1(data, fields, fields)
self.assertEqual(split.snapshot_id, 18)
self.assertEqual(list(split.partition.values), ['aaaaa'])
self.assertEqual(split.bucket, 20)
self.assertFalse(split.raw_convertible)
self.assertEqual(len(split.files), 1)
file = split.files[0]
self.assertEqual(file.file_name, 'my_file')
self.assertEqual(file.file_path, 'hdfs:///path/to/warehouse')
self.assertEqual((file.file_size, file.row_count), (1024 * 1024, 1024))
self.assertEqual(list(file.min_key.values), ['min_key'])
self.assertEqual(list(file.max_key.values), ['max_key'])
self.assertEqual((file.min_sequence_number, file.max_sequence_number), (15, 200))
self.assertEqual((file.schema_id, file.level), (5, 3))
self.assertEqual(file.extra_files, ['extra1', 'extra2'])
self.assertEqual(file.embedded_index, bytes([1, 2, 4]))
self.assertEqual(file.value_stats_cols, ['field1', 'field2', 'field3'])
self.assertEqual(file.first_row_id, 12)
self.assertEqual(file.write_cols, ['a', 'b', 'c', 'f'])
self.assertEqual(len(split.data_deletion_files), 1)
dv = split.data_deletion_files[0]
self.assertEqual(
(dv.dv_index_path, dv.offset, dv.length, dv.cardinality),
('deletion_file', 100, 22, 33))
if indexed:
self.assertEqual([(r.from_, r.to) for r in split.row_ranges()], [(13, 14)])
self.assertEqual(split.scores(), [0.75])

def test_unsupported_data_split_version_raises(self):
for version in (7, 10):
with self.subTest(version=version):
data = (_GOLDEN_DATA_SPLIT_V1[:24] + struct.pack('>i', version)
+ _GOLDEN_DATA_SPLIT_V1[28:])
with self.assertRaisesRegex(ValueError, 'unsupported DataSplit version'):
deserialize_split_v1(data, self._partition_fields())

def test_decode_str_array_inline_and_pointer(self):
# Covers both element encodings: inline (<=7 bytes) and var pointer (>7).
self.assertEqual(_decode_str_array(_BINARY_ARRAY_STR), ['id', 'longcolumn12'])
Expand Down
Loading