-
Notifications
You must be signed in to change notification settings - Fork 86
/
ak_from_parquet.py
266 lines (230 loc) · 9.37 KB
/
ak_from_parquet.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
# BSD 3-Clause License; see https://github.com/scikit-hep/awkward-1.0/blob/main/LICENSE
import awkward as ak
np = ak.nplike.NumpyMetadata.instance()
numpy = ak.nplike.Numpy.instance()
def from_parquet(
path,
columns=None,
row_groups=None,
storage_options=None,
max_gap=64_000,
max_block=256_000_000,
footer_sample_size=1_000_000,
conservative_optiontype=False,
highlevel=True,
behavior=None,
):
"""
Args:
path (str): Local filename or remote URL, passed to fsspec for resolution.
May contain glob patterns.
columns (None, str, or list of str): Glob pattern(s) with bash-like curly
brackets for matching column names. Nested records are separated by dots.
If a list of patterns, the logical-or is matched. If None, all columns
are read.
row_groups (None or set of int): Row groups to read; must be non-negative.
Order is ignored: the output array is presented in the order specified by
Parquet metadata. If None, all row groups/all rows are read.
storage_options: Passed to `fsspec.parquet.open_parquet_file`.
max_gap (int): Passed to `fsspec.parquet.open_parquet_file`.
max_block (int): Passed to `fsspec.parquet.open_parquet_file`.
footer_sample_size (int): Passed to `fsspec.parquet.open_parquet_file`.
conservative_optiontype (bool): Passed to `ak.from_arrow`.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
Reads data from a local or remote Parquet file or collection of files.
The data are eagerly (not lazily) read and must fit into memory. Use `columns`
and/or `row_groups` to select and filter manageable subsets of the data, and
use #ak.metadata_from_parquet to find column names and the range of row groups
that a dataset has.
See also #ak.to_parquet, #ak.metadata_from_parquet.
"""
with ak._v2._util.OperationErrorContext(
"ak._v2.from_parquet",
dict(
path=path,
columns=columns,
row_groups=row_groups,
storage_options=storage_options,
max_gap=max_gap,
max_block=max_block,
footer_sample_size=footer_sample_size,
conservative_optiontype=conservative_optiontype,
highlevel=highlevel,
behavior=behavior,
),
):
return _impl(
path,
columns,
row_groups,
storage_options,
max_gap,
max_block,
footer_sample_size,
conservative_optiontype,
highlevel,
behavior,
)
def _impl(
path,
columns,
row_groups,
storage_options,
max_gap,
max_block,
footer_sample_size,
conservative_optiontype,
highlevel,
behavior,
):
import awkward._v2._connect.pyarrow # noqa: F401
name = "ak._v2.from_parquet"
pyarrow_parquet = ak._v2._connect.pyarrow.import_pyarrow_parquet(name)
fsspec = ak._v2._connect.pyarrow.import_fsspec(name)
import fsspec.parquet
if row_groups is not None:
if not all(ak._v2._util.isint(x) and x >= 0 for x in row_groups):
raise ak._v2._util.error(
TypeError("row_groups must be a set of non-negative integers")
)
fs, _, paths = fsspec.get_fs_token_paths(
path, mode="rb", storage_options=storage_options
)
all_paths, path_for_metadata = _all_and_metadata_paths(path, fs, paths)
parquet_columns = None
subform = None
subrg = [None] * len(all_paths)
actual_paths = all_paths
if columns is not None or row_groups is not None:
with fsspec.parquet.open_parquet_file(
path_for_metadata,
fs=fs,
engine="pyarrow",
row_groups=[],
storage_options=storage_options,
max_gap=max_gap,
max_block=max_block,
footer_sample_size=footer_sample_size,
) as file_for_metadata:
parquetfile_for_metadata = pyarrow_parquet.ParquetFile(file_for_metadata)
if columns is not None:
# FIXME: get this from parquetfile_for_metadata
list_indicator = "list.item"
form = ak._v2._connect.pyarrow.form_handle_arrow(
parquetfile_for_metadata.schema_arrow, pass_empty_field=True
)
subform = form.select_columns(columns)
parquet_columns = subform.columns(list_indicator=list_indicator)
if row_groups is not None:
metadata = parquetfile_for_metadata.metadata
if any(not 0 <= rg < metadata.num_row_groups for rg in row_groups):
raise ak._v2._util.error(
ValueError(
f"one of the requested row_groups is out of range (must be less than {metadata.num_row_groups})"
)
)
split_paths = [p.split("/") for p in all_paths]
prev_index = None
prev_i = 0
actual_paths = []
subrg = []
for i in range(metadata.num_row_groups):
split_path = metadata.row_group(i).column(0).file_path.split("/")
index = None
for j, compare in enumerate(split_paths):
if split_path == compare[-len(split_path) :]:
index = j
break
if index is None:
eoln = "\n "
raise ak._v2._util.error(
LookupError(
f"""path {'/'.join(split_path)!r} from metadata not found in path matches:
{eoln.join(all_paths)}"""
)
)
if prev_index != index:
prev_index = index
prev_i = i
actual_paths.append(all_paths[index])
subrg.append([])
if i in row_groups:
subrg[-1].append(i - prev_i)
for k in range(len(subrg) - 1, -1, -1):
if len(subrg[k]) == 0:
del actual_paths[k]
del subrg[k]
arrays = []
for i, p in enumerate(actual_paths):
with fsspec.parquet.open_parquet_file(
p,
fs=fs,
engine="pyarrow",
columns=parquet_columns,
row_groups=subrg[i],
storage_options=storage_options,
max_gap=max_gap,
max_block=max_block,
footer_sample_size=footer_sample_size,
) as file:
parquetfile = pyarrow_parquet.ParquetFile(file)
if subform is None:
subform = ak._v2._connect.pyarrow.form_handle_arrow(
parquetfile.schema_arrow, pass_empty_field=True
)
if row_groups is None:
arrow_table = parquetfile.read(parquet_columns)
else:
arrow_table = parquetfile.read_row_groups(subrg[i], parquet_columns)
arrays.append(
ak._v2._connect.pyarrow.handle_arrow(
arrow_table,
conservative_optiontype=conservative_optiontype,
pass_empty_field=True,
)
)
if len(arrays) == 0:
return ak._v2.operations.convert.ak_from_buffers._impl(
subform, 0, _DictOfEmptyBuffers(), "", numpy, highlevel, behavior
)
else:
return ak._v2.operations.structure.ak_concatenate._impl(
arrays, 0, True, True, highlevel, behavior
)
class _DictOfEmptyBuffers:
def __getitem__(self, where):
return b"\x00\x00\x00\x00\x00\x00\x00\x00"
def _all_and_metadata_paths(path, fs, paths):
all_paths = []
for x in paths:
if fs.isfile(x):
is_meta = x.split("/")[-1] == "_metadata"
is_comm = x.split("/")[-1] == "_common_metadata"
all_paths.append((x, is_meta, is_comm))
elif fs.isdir(x):
for prefix, _, files in fs.walk(x):
for f in files:
is_meta = f == "_metadata"
is_comm = f == "_common_metadata"
if f.endswith((".parq", ".parquet")) or is_meta or is_comm:
if fs.isfile("/".join((prefix, f))):
all_paths.append(("/".join((prefix, f)), is_meta, is_comm))
path_for_metadata = [x for x, is_meta, is_comm in all_paths if is_meta]
if len(path_for_metadata) != 0:
path_for_metadata = path_for_metadata[0]
else:
path_for_metadata = [x for x, is_meta, is_comm in all_paths if is_comm]
if len(path_for_metadata) != 0:
path_for_metadata = path_for_metadata[0]
else:
if len(all_paths) != 0:
path_for_metadata = all_paths[0][0]
all_paths = [x for x, is_meta, is_comm in all_paths if not is_meta and not is_comm]
if len(all_paths) == 0:
raise ak._v2._util.error(
ValueError(f"no *.parquet or *.parq matches for path {path!r}")
)
return all_paths, path_for_metadata