| 1 |
#!/usr/bin/env python3 |
| 2 |
# -*- mode: python -*- |
| 3 |
# -*- coding: utf-8 -*- |
| 4 |
|
| 5 |
# Licensed to the Apache Software Foundation (ASF) under one |
| 6 |
# or more contributor license agreements. See the NOTICE file |
| 7 |
# distributed with this work for additional information |
| 8 |
# regarding copyright ownership. The ASF licenses this file |
| 9 |
# to you under the Apache License, Version 2.0 (the |
| 10 |
# "License"); you may not use this file except in compliance |
| 11 |
# with the License. You may obtain a copy of the License at |
| 12 |
# |
| 13 |
# http://www.apache.org/licenses/LICENSE-2.0 |
| 14 |
# |
| 15 |
# Unless required by applicable law or agreed to in writing, software |
| 16 |
# distributed under the License is distributed on an "AS IS" BASIS, |
| 17 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 18 |
# See the License for the specific language governing permissions and |
| 19 |
# limitations under the License. |
| 20 |
|
| 21 |
"""Read/Write Avro File Object Containers.""" |
| 22 |
|
| 23 |
import io |
| 24 |
import logging |
| 25 |
import os |
| 26 |
import zlib |
| 27 |
|
| 28 |
from avro import schema |
| 29 |
from avro import io as avro_io |
| 30 |
|
| 31 |
try: |
| 32 |
import snappy |
| 33 |
has_snappy = True |
| 34 |
except ImportError: |
| 35 |
has_snappy = False |
| 36 |
|
| 37 |
|
| 38 |
logger = logging.getLogger(__name__) |
| 39 |
|
| 40 |
# ------------------------------------------------------------------------------ |
| 41 |
# Constants |
| 42 |
|
| 43 |
# Version of the container file: |
| 44 |
VERSION = 1 |
| 45 |
|
| 46 |
# Magic code that starts a data container file: |
| 47 |
MAGIC = b'Obj' + bytes([VERSION]) |
| 48 |
|
| 49 |
# Size of the magic code, in number of bytes: |
| 50 |
MAGIC_SIZE = len(MAGIC) |
| 51 |
|
| 52 |
# Size of the synchronization marker, in number of bytes: |
| 53 |
SYNC_SIZE = 16 |
| 54 |
|
| 55 |
# Interval between synchronization markers, in number of bytes: |
| 56 |
# TODO: make configurable |
| 57 |
SYNC_INTERVAL = 1000 * SYNC_SIZE |
| 58 |
|
| 59 |
# Schema of the container header: |
| 60 |
META_SCHEMA = schema.Parse(""" |
| 61 |
{ |
| 62 |
"type": "record", "name": "org.apache.avro.file.Header", |
| 63 |
"fields": [{ |
| 64 |
"name": "magic", |
| 65 |
"type": {"type": "fixed", "name": "magic", "size": %(magic_size)d} |
| 66 |
}, { |
| 67 |
"name": "meta", |
| 68 |
"type": {"type": "map", "values": "bytes"} |
| 69 |
}, { |
| 70 |
"name": "sync", |
| 71 |
"type": {"type": "fixed", "name": "sync", "size": %(sync_size)d} |
| 72 |
}] |
| 73 |
} |
| 74 |
""" % { |
| 75 |
'magic_size': MAGIC_SIZE, |
| 76 |
'sync_size': SYNC_SIZE, |
| 77 |
}) |
| 78 |
|
| 79 |
# Codecs supported by container files: |
| 80 |
VALID_CODECS = frozenset(['null', 'deflate']) |
| 81 |
if has_snappy: |
| 82 |
VALID_CODECS = frozenset.union(VALID_CODECS, ['snappy']) |
| 83 |
|
| 84 |
# Not used yet |
| 85 |
VALID_ENCODINGS = frozenset(['binary']) |
| 86 |
|
| 87 |
# Metadata key associated to the codec: |
| 88 |
CODEC_KEY = "avro.codec" |
| 89 |
|
| 90 |
# Metadata key associated to the schema: |
| 91 |
SCHEMA_KEY = "avro.schema" |
| 92 |
|
| 93 |
|
| 94 |
# ------------------------------------------------------------------------------ |
| 95 |
# Exceptions |
| 96 |
|
| 97 |
|
| 98 |
class DataFileException(schema.AvroException): |
| 99 |
"""Problem reading or writing file object containers.""" |
| 100 |
|
| 101 |
def __init__(self, msg): |
| 102 |
super(DataFileException, self).__init__(msg) |
| 103 |
|
| 104 |
|
| 105 |
# ------------------------------------------------------------------------------ |
| 106 |
|
| 107 |
|
| 108 |
class DataFileWriter(object): |
| 109 |
"""Writes Avro data files.""" |
| 110 |
|
| 111 |
@staticmethod |
| 112 |
def GenerateSyncMarker(): |
| 113 |
"""Generates a random synchronization marker.""" |
| 114 |
return os.urandom(SYNC_SIZE) |
| 115 |
|
| 116 |
# TODO: make 'encoder' a metadata property |
| 117 |
def __init__( |
| 118 |
self, |
| 119 |
writer, |
| 120 |
datum_writer, |
| 121 |
writer_schema=None, |
| 122 |
codec='null', |
| 123 |
): |
| 124 |
"""Constructs a new DataFileWriter instance. |
| 125 |
|
| 126 |
If the schema is not present, presume we're appending. |
| 127 |
|
| 128 |
Args: |
| 129 |
writer: File-like object to write into. |
| 130 |
datum_writer: |
| 131 |
writer_schema: Schema |
| 132 |
codec: |
| 133 |
""" |
| 134 |
self._writer = writer |
| 135 |
self._encoder = avro_io.BinaryEncoder(writer) |
| 136 |
self._datum_writer = datum_writer |
| 137 |
self._buffer_writer = io.BytesIO() |
| 138 |
self._buffer_encoder = avro_io.BinaryEncoder(self._buffer_writer) |
| 139 |
self._block_count = 0 |
| 140 |
self._meta = {} |
| 141 |
|
| 142 |
# Ensure we have a writer that accepts bytes: |
| 143 |
self._writer.write(b'') |
| 144 |
|
| 145 |
# Whether the header has already been written: |
| 146 |
self._header_written = False |
| 147 |
|
| 148 |
if writer_schema is not None: |
| 149 |
if codec not in VALID_CODECS: |
| 150 |
raise DataFileException('Unknown codec: %r' % codec) |
| 151 |
self._sync_marker = DataFileWriter.GenerateSyncMarker() |
| 152 |
self.SetMeta('avro.codec', codec) |
| 153 |
self.SetMeta('avro.schema', str(writer_schema).encode('utf-8')) |
| 154 |
self.datum_writer.writer_schema = writer_schema |
| 155 |
else: |
| 156 |
# open writer for reading to collect metadata |
| 157 |
dfr = DataFileReader(writer, avro_io.DatumReader()) |
| 158 |
|
| 159 |
# TODO: collect arbitrary metadata |
| 160 |
# collect metadata |
| 161 |
self._sync_marker = dfr.sync_marker |
| 162 |
self.SetMeta('avro.codec', dfr.GetMeta('avro.codec')) |
| 163 |
|
| 164 |
# get schema used to write existing file |
| 165 |
schema_from_file = dfr.GetMeta('avro.schema').decode('utf-8') |
| 166 |
self.SetMeta('avro.schema', schema_from_file) |
| 167 |
self.datum_writer.writer_schema = schema.Parse(schema_from_file) |
| 168 |
|
| 169 |
# seek to the end of the file and prepare for writing |
| 170 |
writer.seek(0, 2) |
| 171 |
self._header_written = True |
| 172 |
|
| 173 |
# read-only properties |
| 174 |
|
| 175 |
@property |
| 176 |
def writer(self): |
| 177 |
return self._writer |
| 178 |
|
| 179 |
@property |
| 180 |
def encoder(self): |
| 181 |
return self._encoder |
| 182 |
|
| 183 |
@property |
| 184 |
def datum_writer(self): |
| 185 |
return self._datum_writer |
| 186 |
|
| 187 |
@property |
| 188 |
def buffer_encoder(self): |
| 189 |
return self._buffer_encoder |
| 190 |
|
| 191 |
@property |
| 192 |
def sync_marker(self): |
| 193 |
return self._sync_marker |
| 194 |
|
| 195 |
@property |
| 196 |
def meta(self): |
| 197 |
return self._meta |
| 198 |
|
| 199 |
def __enter__(self): |
| 200 |
return self |
| 201 |
|
| 202 |
def __exit__(self, type, value, traceback): |
| 203 |
# Perform a close if there's no exception |
| 204 |
if type is None: |
| 205 |
self.close() |
| 206 |
|
| 207 |
@property |
| 208 |
def block_count(self): |
| 209 |
return self._block_count |
| 210 |
|
| 211 |
def GetMeta(self, key): |
| 212 |
"""Reports the metadata associated to the given key. |
| 213 |
|
| 214 |
Args: |
| 215 |
key: Key of the metadata to report the value of. |
| 216 |
Returns: |
| 217 |
The metadata value, as bytes, or None if the key does not exist. |
| 218 |
""" |
| 219 |
return self._meta.get(key) |
| 220 |
|
| 221 |
def SetMeta(self, key, value): |
| 222 |
"""Sets the metadata value for the given key. |
| 223 |
|
| 224 |
Note: metadata is persisted and retrieved as bytes. |
| 225 |
|
| 226 |
Args: |
| 227 |
key: Key of the metadata to set. |
| 228 |
value: Value of the metadata, as bytes or str. |
| 229 |
Strings are automatically converted to bytes. |
| 230 |
""" |
| 231 |
if isinstance(value, str): |
| 232 |
value = value.encode('utf-8') |
| 233 |
assert isinstance(value, bytes), ( |
| 234 |
'Invalid metadata value for key %r: %r' % (key, value)) |
| 235 |
self._meta[key] = value |
| 236 |
|
| 237 |
def _WriteHeader(self): |
| 238 |
header = { |
| 239 |
'magic': MAGIC, |
| 240 |
'meta': self.meta, |
| 241 |
'sync': self.sync_marker, |
| 242 |
} |
| 243 |
logger.debug( |
| 244 |
'Writing Avro data file header:\n%s\nAvro header schema:\n%s', |
| 245 |
header, META_SCHEMA) |
| 246 |
self.datum_writer.write_data(META_SCHEMA, header, self.encoder) |
| 247 |
self._header_written = True |
| 248 |
|
| 249 |
# TODO: make a schema for blocks and use datum_writer |
| 250 |
def _WriteBlock(self): |
| 251 |
if not self._header_written: |
| 252 |
self._WriteHeader() |
| 253 |
|
| 254 |
if self.block_count <= 0: |
| 255 |
logger.info('Current block is empty, nothing to write.') |
| 256 |
return |
| 257 |
|
| 258 |
# write number of items in block |
| 259 |
self.encoder.write_long(self.block_count) |
| 260 |
|
| 261 |
# write block contents |
| 262 |
uncompressed_data = self._buffer_writer.getvalue() |
| 263 |
codec = self.GetMeta(CODEC_KEY).decode('utf-8') |
| 264 |
if codec == 'null': |
| 265 |
compressed_data = uncompressed_data |
| 266 |
compressed_data_length = len(compressed_data) |
| 267 |
elif codec == 'deflate': |
| 268 |
# The first two characters and last character are zlib |
| 269 |
# wrappers around deflate data. |
| 270 |
compressed_data = zlib.compress(uncompressed_data)[2:-1] |
| 271 |
compressed_data_length = len(compressed_data) |
| 272 |
elif codec == 'snappy': |
| 273 |
compressed_data = snappy.compress(uncompressed_data) |
| 274 |
compressed_data_length = len(compressed_data) + 4 # crc32 |
| 275 |
else: |
| 276 |
fail_msg = '"%s" codec is not supported.' % codec |
| 277 |
raise DataFileException(fail_msg) |
| 278 |
|
| 279 |
# Write length of block |
| 280 |
self.encoder.write_long(compressed_data_length) |
| 281 |
|
| 282 |
# Write block |
| 283 |
self.writer.write(compressed_data) |
| 284 |
|
| 285 |
# Write CRC32 checksum for Snappy |
| 286 |
if self.GetMeta(CODEC_KEY) == 'snappy': |
| 287 |
self.encoder.write_crc32(uncompressed_data) |
| 288 |
|
| 289 |
# write sync marker |
| 290 |
self.writer.write(self.sync_marker) |
| 291 |
|
| 292 |
logger.debug( |
| 293 |
'Writing block with count=%d nbytes=%d sync=%r', |
| 294 |
self.block_count, compressed_data_length, self.sync_marker) |
| 295 |
|
| 296 |
# reset buffer |
| 297 |
self._buffer_writer.seek(0) |
| 298 |
self._buffer_writer.truncate() |
| 299 |
self._block_count = 0 |
| 300 |
|
| 301 |
def append(self, datum): |
| 302 |
"""Append a datum to the file.""" |
| 303 |
self.datum_writer.write(datum, self.buffer_encoder) |
| 304 |
self._block_count += 1 |
| 305 |
|
| 306 |
# if the data to write is larger than the sync interval, write the block |
| 307 |
if self._buffer_writer.tell() >= SYNC_INTERVAL: |
| 308 |
self._WriteBlock() |
| 309 |
|
| 310 |
def sync(self): |
| 311 |
""" |
| 312 |
Return the current position as a value that may be passed to |
| 313 |
DataFileReader.seek(long). Forces the end of the current block, |
| 314 |
emitting a synchronization marker. |
| 315 |
""" |
| 316 |
self._WriteBlock() |
| 317 |
return self.writer.tell() |
| 318 |
|
| 319 |
def flush(self): |
| 320 |
"""Flush the current state of the file, including metadata.""" |
| 321 |
self._WriteBlock() |
| 322 |
self.writer.flush() |
| 323 |
|
| 324 |
def close(self): |
| 325 |
"""Close the file.""" |
| 326 |
self.flush() |
| 327 |
self.writer.close() |
| 328 |
|
| 329 |
|
| 330 |
# ------------------------------------------------------------------------------ |
| 331 |
|
| 332 |
|
| 333 |
class DataFileReader(object): |
| 334 |
"""Read files written by DataFileWriter.""" |
| 335 |
|
| 336 |
# TODO: allow user to specify expected schema? |
| 337 |
# TODO: allow user to specify the encoder |
| 338 |
def __init__(self, reader, datum_reader): |
| 339 |
"""Initializes a new data file reader. |
| 340 |
|
| 341 |
Args: |
| 342 |
reader: Open file to read from. |
| 343 |
datum_reader: Avro datum reader. |
| 344 |
""" |
| 345 |
self._reader = reader |
| 346 |
self._raw_decoder = avro_io.BinaryDecoder(reader) |
| 347 |
self._datum_decoder = None # Maybe reset at every block. |
| 348 |
self._datum_reader = datum_reader |
| 349 |
|
| 350 |
# read the header: magic, meta, sync |
| 351 |
self._read_header() |
| 352 |
|
| 353 |
# ensure codec is valid |
| 354 |
avro_codec_raw = self.GetMeta('avro.codec') |
| 355 |
if avro_codec_raw is None: |
| 356 |
self.codec = "null" |
| 357 |
else: |
| 358 |
self.codec = avro_codec_raw.decode('utf-8') |
| 359 |
if self.codec not in VALID_CODECS: |
| 360 |
raise DataFileException('Unknown codec: %s.' % self.codec) |
| 361 |
|
| 362 |
self._file_length = self._GetInputFileLength() |
| 363 |
|
| 364 |
# get ready to read |
| 365 |
self._block_count = 0 |
| 366 |
self.datum_reader.writer_schema = ( |
| 367 |
schema.Parse(self.GetMeta(SCHEMA_KEY).decode('utf-8'))) |
| 368 |
|
| 369 |
def __enter__(self): |
| 370 |
return self |
| 371 |
|
| 372 |
def __exit__(self, type, value, traceback): |
| 373 |
# Perform a close if there's no exception |
| 374 |
if type is None: |
| 375 |
self.close() |
| 376 |
|
| 377 |
def __iter__(self): |
| 378 |
return self |
| 379 |
|
| 380 |
def __next__(self): |
| 381 |
"""Implements the iterator interface.""" |
| 382 |
return next(self) |
| 383 |
|
| 384 |
# read-only properties |
| 385 |
@property |
| 386 |
def reader(self): |
| 387 |
return self._reader |
| 388 |
|
| 389 |
@property |
| 390 |
def raw_decoder(self): |
| 391 |
return self._raw_decoder |
| 392 |
|
| 393 |
@property |
| 394 |
def datum_decoder(self): |
| 395 |
return self._datum_decoder |
| 396 |
|
| 397 |
@property |
| 398 |
def datum_reader(self): |
| 399 |
return self._datum_reader |
| 400 |
|
| 401 |
@property |
| 402 |
def sync_marker(self): |
| 403 |
return self._sync_marker |
| 404 |
|
| 405 |
@property |
| 406 |
def meta(self): |
| 407 |
return self._meta |
| 408 |
|
| 409 |
@property |
| 410 |
def file_length(self): |
| 411 |
"""Length of the input file, in bytes.""" |
| 412 |
return self._file_length |
| 413 |
|
| 414 |
# read/write properties |
| 415 |
@property |
| 416 |
def block_count(self): |
| 417 |
return self._block_count |
| 418 |
|
| 419 |
def GetMeta(self, key): |
| 420 |
"""Reports the value of a given metadata key. |
| 421 |
|
| 422 |
Args: |
| 423 |
key: Metadata key (string) to report the value of. |
| 424 |
Returns: |
| 425 |
Value associated to the metadata key, as bytes. |
| 426 |
""" |
| 427 |
return self._meta.get(key) |
| 428 |
|
| 429 |
def SetMeta(self, key, value): |
| 430 |
"""Sets a metadata. |
| 431 |
|
| 432 |
Args: |
| 433 |
key: Metadata key (string) to set. |
| 434 |
value: Metadata value to set, as bytes. |
| 435 |
""" |
| 436 |
if isinstance(value, str): |
| 437 |
value = value.encode('utf-8') |
| 438 |
self._meta[key] = value |
| 439 |
|
| 440 |
def _GetInputFileLength(self): |
| 441 |
"""Reports the length of the input file, in bytes. |
| 442 |
|
| 443 |
Leaves the current position unmodified. |
| 444 |
|
| 445 |
Returns: |
| 446 |
The length of the input file, in bytes. |
| 447 |
""" |
| 448 |
current_pos = self.reader.tell() |
| 449 |
self.reader.seek(0, 2) |
| 450 |
file_length = self.reader.tell() |
| 451 |
self.reader.seek(current_pos) |
| 452 |
return file_length |
| 453 |
|
| 454 |
def is_EOF(self): |
| 455 |
return self.reader.tell() == self.file_length |
| 456 |
|
| 457 |
def _read_header(self): |
| 458 |
# seek to the beginning of the file to get magic block |
| 459 |
self.reader.seek(0, 0) |
| 460 |
|
| 461 |
# read header into a dict |
| 462 |
header = self.datum_reader.read_data( |
| 463 |
META_SCHEMA, META_SCHEMA, self.raw_decoder) |
| 464 |
|
| 465 |
# check magic number |
| 466 |
if header.get('magic') != MAGIC: |
| 467 |
fail_msg = "Not an Avro data file: %s doesn't match %s."\ |
| 468 |
% (header.get('magic'), MAGIC) |
| 469 |
raise schema.AvroException(fail_msg) |
| 470 |
|
| 471 |
# set metadata |
| 472 |
self._meta = header['meta'] |
| 473 |
|
| 474 |
# set sync marker |
| 475 |
self._sync_marker = header['sync'] |
| 476 |
|
| 477 |
def _read_block_header(self): |
| 478 |
self._block_count = self.raw_decoder.read_long() |
| 479 |
if self.codec == "null": |
| 480 |
# Skip a long; we don't need to use the length. |
| 481 |
self.raw_decoder.skip_long() |
| 482 |
self._datum_decoder = self._raw_decoder |
| 483 |
elif self.codec == 'deflate': |
| 484 |
# Compressed data is stored as (length, data), which |
| 485 |
# corresponds to how the "bytes" type is encoded. |
| 486 |
data = self.raw_decoder.read_bytes() |
| 487 |
# -15 is the log of the window size; negative indicates |
| 488 |
# "raw" (no zlib headers) decompression. See zlib.h. |
| 489 |
uncompressed = zlib.decompress(data, -15) |
| 490 |
self._datum_decoder = avro_io.BinaryDecoder(io.BytesIO(uncompressed)) |
| 491 |
elif self.codec == 'snappy': |
| 492 |
# Compressed data includes a 4-byte CRC32 checksum |
| 493 |
length = self.raw_decoder.read_long() |
| 494 |
data = self.raw_decoder.read(length - 4) |
| 495 |
uncompressed = snappy.decompress(data) |
| 496 |
self._datum_decoder = avro_io.BinaryDecoder(io.BytesIO(uncompressed)) |
| 497 |
self.raw_decoder.check_crc32(uncompressed); |
| 498 |
else: |
| 499 |
raise DataFileException("Unknown codec: %r" % self.codec) |
| 500 |
|
| 501 |
def _skip_sync(self): |
| 502 |
""" |
| 503 |
Read the length of the sync marker; if it matches the sync marker, |
| 504 |
return True. Otherwise, seek back to where we started and return False. |
| 505 |
""" |
| 506 |
proposed_sync_marker = self.reader.read(SYNC_SIZE) |
| 507 |
if proposed_sync_marker != self.sync_marker: |
| 508 |
self.reader.seek(-SYNC_SIZE, 1) |
| 509 |
return False |
| 510 |
else: |
| 511 |
return True |
| 512 |
|
| 513 |
# TODO: handle block of length zero |
| 514 |
# TODO: clean this up with recursion |
| 515 |
def __next__(self): |
| 516 |
"""Return the next datum in the file.""" |
| 517 |
if self.block_count == 0: |
| 518 |
if self.is_EOF(): |
| 519 |
raise StopIteration |
| 520 |
elif self._skip_sync(): |
| 521 |
if self.is_EOF(): raise StopIteration |
| 522 |
self._read_block_header() |
| 523 |
else: |
| 524 |
self._read_block_header() |
| 525 |
|
| 526 |
datum = self.datum_reader.read(self.datum_decoder) |
| 527 |
self._block_count -= 1 |
| 528 |
return datum |
| 529 |
|
| 530 |
def close(self): |
| 531 |
"""Close this reader.""" |
| 532 |
self.reader.close() |
| 533 |
|
| 534 |
|
| 535 |
if __name__ == '__main__': |
| 536 |
raise Exception('Not a standalone module') |