bfile.py 20.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
# Copyright (c) 2020 VisualDL Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =======================================================================

import os
import tempfile
走神的阿圆's avatar
走神的阿圆 已提交
18 19
import hashlib
import base64
走神的阿圆's avatar
走神的阿圆 已提交
20
import time
走神的阿圆's avatar
走神的阿圆 已提交
21 22 23 24 25 26 27 28 29 30

try:
    import hdfs
    from hdfs.util import HdfsError
    HDFS_ENABLED = True
except ImportError:
    HDFS_ENABLED = False
try:
    from baidubce.services.bos.bos_client import BosClient
    from baidubce import exception
走神的阿圆's avatar
走神的阿圆 已提交
31 32
    from baidubce.bce_client_configuration import BceClientConfiguration
    from baidubce.auth.bce_credentials import BceCredentials
走神的阿圆's avatar
走神的阿圆 已提交
33 34 35
    BOS_ENABLED = True
except ImportError:
    BOS_ENABLED = False
36 37 38 39 40 41 42

# Note: Some codes here refer to TensorBoardX.
# A good default block size depends on the system in question.
# A somewhat conservative default chosen here.
_DEFAULT_BLOCK_SIZE = 16 * 1024 * 1024


走神的阿圆's avatar
走神的阿圆 已提交
43 44 45 46 47 48
def content_md5(buffer):
    md5 = hashlib.md5()
    md5.update(buffer)
    return base64.standard_b64encode(md5.digest())


49 50 51 52 53 54 55 56
class FileFactory(object):
    def __init__(self):
        self._register_factories = {}

    def register_filesystem(self, path, filesystem):
        self._register_factories.update({path: filesystem})

    def get_filesystem(self, path):
走神的阿圆's avatar
走神的阿圆 已提交
57 58
        if path.startswith(
                'hdfs://') and "hdfs" not in self._register_factories:
走神的阿圆's avatar
走神的阿圆 已提交
59 60
            if not HDFS_ENABLED:
                raise RuntimeError('Please install module named "hdfs".')
走神的阿圆's avatar
走神的阿圆 已提交
61 62 63 64 65
            try:
                default_file_factory.register_filesystem("hdfs", HDFileSystem())
            except hdfs.util.HdfsError:
                raise RuntimeError(
                    "Please initialize `~/.hdfscli.cfg` for HDFS.")
走神的阿圆's avatar
走神的阿圆 已提交
66 67 68 69 70 71 72
        elif path.startswith(
                'bos://') and "bos" not in self._register_factories:
            if not BOS_ENABLED:
                raise RuntimeError(
                    'Please install module named "bce-python-sdk".')
            default_file_factory.register_filesystem("bos", BosFileSystem())

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
        prefix = ""
        index = path.find("://")
        if index >= 0:
            prefix = path[:index]
        fs = self._register_factories.get(prefix, None)
        if fs is None:
            raise ValueError("No recognized filesystem for prefix %s" % prefix)
        return fs


default_file_factory = FileFactory()


class LocalFileSystem(object):
    def __init__(self):
        pass

    @staticmethod
    def exists(path):
        return os.path.exists(path)

    @staticmethod
    def makedirs(path):
        os.makedirs(path)

    @staticmethod
    def join(path, *paths):
        return os.path.join(path, *paths)

走神的阿圆's avatar
走神的阿圆 已提交
102 103 104 105 106 107 108 109 110
    def isfile(self, filename):
        return os.path.isfile(filename)

    def read_file(self, filename, binary_mode=True):
        mode = "rb" if binary_mode else "r"
        with open(filename, mode) as reader:
            data = reader.read()
        return data

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
    def read(self, filename, binary_mode=False, size=None, continue_from=None):
        mode = "rb" if binary_mode else "r"
        encoding = None if binary_mode else "utf-8"
        offset = None
        if continue_from is not None:
            offset = continue_from.get("last_offset", None)
        with open(filename, mode=mode, encoding=encoding) as fp:
            if offset is not None:
                fp.seek(offset)
            data = fp.read(size)
            continue_from_token = {"last_offset": fp.tell()}
            return data, continue_from_token

    def _write(self, filename, file_content, mode):
        encoding = None if "b" in mode else "utf-8"
        with open(filename, mode, encoding=encoding) as fp:
            fp.write(file_content)

    def append(self, filename, file_content, binary_mode=False):
        self._write(filename, file_content, "ab" if binary_mode else "a")

    def write(self, filename, file_content, binary_mode=False):
        self._write(filename, file_content, "wb" if binary_mode else "w")

    def walk(self, dir):
136 137
        if 'posix' == os.name:
            return os.walk(dir, followlinks=True)
138 139 140 141 142 143
        return os.walk(dir)


default_file_factory.register_filesystem("", LocalFileSystem())


144 145 146 147 148 149 150 151 152 153
class HDFileSystem(object):
    def __init__(self):
        self.cli = hdfs.config.Config().get_client('dev')

    def exists(self, path):
        if self.cli.status(hdfs_path=path[7:], strict=False) is None:
            return False
        else:
            return True

走神的阿圆's avatar
走神的阿圆 已提交
154 155 156 157 158 159 160 161
    def isfile(self, filename):
        return exists(filename)

    def read_file(self, filename, binary_mode=True):
        with self.cli.read(hdfs_path=filename[7:]) as reader:
            data = reader.read()
        return data

162 163 164 165 166
    def makedirs(self, path):
        self.cli.makedirs(hdfs_path=path[7:])

    @staticmethod
    def join(path, *paths):
走神的阿圆's avatar
走神的阿圆 已提交
167 168 169
        result = os.path.join(path, *paths)
        result.replace('\\', '/')
        return result
170 171 172 173 174 175 176

    def read(self, filename, binary_mode=False, size=0, continue_from=None):
        offset = 0
        if continue_from is not None:
            offset = continue_from.get("last_offset", 0)

        encoding = None if binary_mode else "utf-8"
走神的阿圆's avatar
走神的阿圆 已提交
177
        try:
走神的阿圆's avatar
走神的阿圆 已提交
178 179
            with self.cli.read(hdfs_path=filename[7:], offset=offset,
                               encoding=encoding) as reader:
走神的阿圆's avatar
走神的阿圆 已提交
180 181 182 183 184
                data = reader.read()
                continue_from_token = {"last_offset": offset + len(data)}
                return data, continue_from_token
        except HdfsError:
            raise EOFError('No more events to read on HDFS.')
185 186 187 188 189 190 191 192 193

    def append(self, filename, file_content, binary_mode=False):
        self.cli.write(hdfs_path=filename[7:], data=file_content, append=True)

    def write(self, filename, file_content, binary_mode=False):
        self.cli.write(hdfs_path=filename[7:], data=file_content)

    def walk(self, dir):
        walks = self.cli.walk(hdfs_path=dir[7:])
走神的阿圆's avatar
走神的阿圆 已提交
194
        return (['hdfs://' + root, dirs, files] for root, dirs, files in walks)
195 196


走神的阿圆's avatar
走神的阿圆 已提交
197 198
class BosFileSystem(object):
    def __init__(self):
走神的阿圆's avatar
走神的阿圆 已提交
199 200 201 202
        self.max_contents_count = 1
        self.max_contents_time = 1
        self.get_bos_config()
        self.bos_client = BosClient(self.config)
走神的阿圆's avatar
走神的阿圆 已提交
203 204
        self.file_length_map = {}

走神的阿圆's avatar
走神的阿圆 已提交
205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220
        self._file_contents_to_add = b''
        self._file_contents_count = 0
        self._start_append_time = time.time()

    def get_bos_config(self):
        bos_host = os.getenv("BOS_HOST")
        if not bos_host:
            raise KeyError('${BOS_HOST} is not found.')
        access_key_id = os.getenv("BOS_AK")
        if not access_key_id:
            raise KeyError('${BOS_AK} is not found.')
        secret_access_key = os.getenv("BOS_SK")
        if not secret_access_key:
            raise KeyError('${BOS_SK} is not found.')
        self.max_contents_count = int(os.getenv('BOS_CACHE_COUNT', 1))
        self.max_contents_time = int(os.getenv('BOS_CACHE_TIME', 1))
走神的阿圆's avatar
走神的阿圆 已提交
221
        bos_sts = os.getenv("BOS_STS")
走神的阿圆's avatar
走神的阿圆 已提交
222 223
        self.config = BceClientConfiguration(
            credentials=BceCredentials(access_key_id, secret_access_key),
走神的阿圆's avatar
走神的阿圆 已提交
224
            endpoint=bos_host, security_token=bos_sts)
走神的阿圆's avatar
走神的阿圆 已提交
225

走神的阿圆's avatar
走神的阿圆 已提交
226 227 228 229 230 231 232
    def isfile(self, filename):
        return exists(filename)

    def read_file(self, filename, binary=True):
        print('BosFileSystem not support --model yet.')
        return

走神的阿圆's avatar
走神的阿圆 已提交
233 234 235 236 237
    @staticmethod
    def _get_object_info(path):
        path = path[6:]
        index = path.index('/')
        bucket_name = path[0:index]
走神的阿圆's avatar
走神的阿圆 已提交
238
        object_key = path[index + 1:]
走神的阿圆's avatar
走神的阿圆 已提交
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
        return bucket_name, object_key

    def exists(self, path):
        bucket_name, object_key = BosFileSystem._get_object_info(path)
        try:
            self.bos_client.get_object_meta_data(bucket_name, object_key)
            return True
        except exception.BceError:
            return False

    def get_meta(self, bucket_name, object_key):
        return self.bos_client.get_object_meta_data(bucket_name, object_key)

    def makedirs(self, path):
        if not path.endswith('/'):
            path += '/'
        if self.exists(path):
            return
        bucket_name, object_key = BosFileSystem._get_object_info(path)
        if not object_key.endswith('/'):
            object_key += '/'
        init_data = b''
        self.bos_client.append_object(bucket_name=bucket_name,
                                      key=object_key,
                                      data=init_data,
                                      content_md5=content_md5(init_data),
                                      content_length=len(init_data))

    @staticmethod
    def join(path, *paths):
        result = os.path.join(path, *paths)
        result.replace('\\', '/')
        return result

    def read(self, filename, binary_mode=False, size=0, continue_from=None):
        bucket_name, object_key = BosFileSystem._get_object_info(filename)
        offset = 0
        if continue_from is not None:
            offset = continue_from.get("last_offset", 0)
走神的阿圆's avatar
走神的阿圆 已提交
278 279 280 281 282 283 284 285 286 287 288
        length = int(
            self.get_meta(bucket_name, object_key).metadata.content_length)
        if offset < length:
            data = self.bos_client.get_object_as_string(bucket_name=bucket_name,
                                                        key=object_key,
                                                        range=[offset,
                                                               length - 1])
        else:
            data = b''

        continue_from_token = {"last_offset": length}
走神的阿圆's avatar
走神的阿圆 已提交
289 290
        return data, continue_from_token

走神的阿圆's avatar
走神的阿圆 已提交
291 292 293 294 295 296 297 298 299 300 301 302 303 304
    def ready_to_append(self):
        if self._file_contents_count >= self.max_contents_count or \
                time.time() - self._start_append_time > self.max_contents_time:
            return True
        else:
            return False

    def append(self, filename, file_content, binary_mode=False, force=False):
        self._file_contents_to_add += file_content
        self._file_contents_count += 1

        if not force and not self.ready_to_append():
            return
        file_content = self._file_contents_to_add
走神的阿圆's avatar
走神的阿圆 已提交
305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322
        bucket_name, object_key = BosFileSystem._get_object_info(filename)
        if not self.exists(filename):
            init_data = b''
            self.bos_client.append_object(bucket_name=bucket_name,
                                          key=object_key,
                                          data=init_data,
                                          content_md5=content_md5(init_data),
                                          content_length=len(init_data),
                                          offset=0)
        content_length = len(file_content)

        offset = self.get_meta(bucket_name, object_key).metadata.content_length
        self.bos_client.append_object(bucket_name=bucket_name,
                                      key=object_key,
                                      data=file_content,
                                      content_md5=content_md5(file_content),
                                      content_length=content_length,
                                      offset=offset)
走神的阿圆's avatar
走神的阿圆 已提交
323 324 325
        self._file_contents_to_add = b''
        self._file_contents_count = 0
        self._start_append_time = time.time()
走神的阿圆's avatar
走神的阿圆 已提交
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360

    def write(self, filename, file_content, binary_mode=False):
        bucket_name, object_key = BosFileSystem._get_object_info(filename)

        self.bos_client.append_object(bucket_name=bucket_name,
                                      key=object_key,
                                      data=file_content,
                                      content_md5=content_md5(file_content),
                                      content_length=len(file_content))

    def walk(self, dir):
        class WalkGenerator():
            def __init__(self, bucket_name, contents):
                self.contents = None
                self.length = 0
                self.bucket = bucket_name
                self.handle_contents(contents)
                self.count = 0

            def handle_contents(self, contents):
                contents_map = {}
                for item in contents:
                    try:
                        rindex = item.rindex('/')
                        key = item[0:rindex]
                        value = item[rindex + 1:]
                    except ValueError:
                        key = '.'
                        value = item
                    if key in contents_map.keys():
                        contents_map[key].append(value)
                    else:
                        contents_map[key] = [value]
                temp_walk = []
                for key, value in contents_map.items():
走神的阿圆's avatar
走神的阿圆 已提交
361 362 363
                    temp_walk.append(
                        [BosFileSystem.join('bos://' + self.bucket, key), [],
                         value])
走神的阿圆's avatar
走神的阿圆 已提交
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
                self.length = len(temp_walk)
                self.contents = temp_walk

            def __iter__(self):
                return self

            def __next__(self):
                if self.count < self.length:
                    self.count += 1
                    return self.contents[self.count - 1]
                else:
                    raise StopIteration

        bucket_name, object_key = BosFileSystem._get_object_info(dir)

        if object_key in ['.', './']:
            prefix = None
        else:
            prefix = object_key if object_key.endswith(
                '/') else object_key + '/'
        response = self.bos_client.list_objects(bucket_name,
                                                prefix=prefix)
        contents = [content.key for content in response.contents]
        return WalkGenerator(bucket_name, contents)


390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411
class BFile(object):
    def __init__(self, filename, mode):
        if mode not in ('r', 'rb', 'br', 'w', 'wb', 'bw'):
            raise NotImplementedError("mode {} not supported by "
                                      "BFile.".format(mode))
        self._filename = filename
        self.fs = default_file_factory.get_filesystem(filename)
        self.fs_supports_append = hasattr(self.fs, 'append')
        self.buff = None
        self.buff_chunk_size = _DEFAULT_BLOCK_SIZE
        self.buff_offset = 0
        self.continuation_token = None
        self.write_temp = None
        self.write_started = False
        self.binary_mode = 'b' in mode
        self.write_mode = 'w' in mode
        self.closed = False

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
412 413 414 415
        self.close()
        self.buff = None
        self.buff_offset = 0
        self.continuation_token = None
416 417 418 419

    def __iter__(self):
        return self

走神的阿圆's avatar
走神的阿圆 已提交
420 421 422
    def isfile(self, filename):
        return self.fs.isfile(filename)

423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
    def _read_buffer_to_offset(self, new_buff_offset):
        """Read buffer from index self.buffer_offset to index new_buff_offset.

        self.buff_offset marks the last position of the last read,
        new_buff_offset indicates the last position of this read.
        self.buff_offset will be updated by new_buff_offset after this read.

        Returns:
            self.buff[i1: i2]: Content of self.buff.
        """
        old_buff_offset = self.buff_offset
        read_size = min(len(self.buff), new_buff_offset) - old_buff_offset
        self.buff_offset += read_size
        return self.buff[old_buff_offset:old_buff_offset + read_size]

走神的阿圆's avatar
走神的阿圆 已提交
438 439 440
    def read_file(self, filename, binnary=True):
        return self.fs.read_file(filename, binnary)

441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556
    def read(self, n=None):
        """Read `n` or all contents of self.buff or file.

        Returns:
            result: Data from self.buff or file.
        """
        result = None
        # If self.buff is not none and length of self.buff more than
        # self.buff_offset, means there are some content in self.buff have
        # not been read.
        if self.buff and len(self.buff) > self.buff_offset:
            if n is not None:
                chunk = self._read_buffer_to_offset(self.buff_offset + n)
                # If length of data in self.buff is more than `n`, then read `n`
                # data from local buffer.
                if len(chunk) == n:
                    return chunk
                result = chunk
                # The length of all data in self.buff may less than `n`,
                # so we should read other `n-length(self.buff)` data.
                n -= len(chunk)
            # If n is none, read all data in self.buff.
            else:
                # add all local buffer and update offsets
                result = self._read_buffer_to_offset(len(self.buff))

        # self.buff is empty if program is here.
        # Read from filesystem.
        # If n is not none, read max(n, self.buff_chunk_size) data from file,
        # otherwise read all data from file.
        # TODO(shenhuhan) N is limited to max_buff, but all-data is unlimited?
        read_size = max(self.buff_chunk_size, n) if n is not None else None
        self.buff, self.continuation_token = self.fs.read(
            self._filename, self.binary_mode, read_size,
            self.continuation_token)
        self.buff_offset = 0

        if n is not None:
            chunk = self._read_buffer_to_offset(n)
        else:
            # add all local buffer and update offsets
            chunk = self._read_buffer_to_offset(len(self.buff))
        result = result + chunk if result else chunk

        return result

    def write(self, file_content):
        """Write contents to file.

        Args:
            file_content: Contents waiting to be written to file.
        """
        if not self.write_mode:
            raise RuntimeError("File not opened in write mode")
        if self.closed:
            raise RuntimeError("File already closed")

        if self.fs_supports_append:
            if not self.write_started:
                self.fs.write(self._filename, file_content, self.binary_mode)
                self.write_started = True
            else:
                self.fs.append(self._filename, file_content, self.binary_mode)
        else:
            # add to temp file, but wait for flush to write to final filesystem
            if self.write_temp is None:
                mode = "w+b" if self.binary_mode else "w+"
                self.write_temp = tempfile.TemporaryFile(mode)
            self.write_temp.write(file_content)

    def __next__(self):
        line = None
        while True:
            if not self.buff:
                # read one unit into the buffer
                line = self.read(1)
                if line and (line[-1] == '\n' or not self.buff):
                    return line
                if not self.buff:
                    raise StopIteration()
            else:
                index = self.buff.find('\n', self.buff_offset)
                if index != -1:
                    # include line until now plus newline
                    chunk = self.read(index + 1 - self.buff_offset)
                    line = line + chunk if line else chunk
                    return line

                # read one unit past end of buffer
                chunk = self.read(len(self.buff) + 1 - self.buff_offset)
                line = line + chunk if line else chunk
                if line and (line[-1] == '\n' or not self.buff):
                    return line
                if not self.buff:
                    raise StopIteration()

    def next(self):
        return self.__next__()

    def flush(self):
        """Flush data to disk.
        """
        if self.closed:
            raise RuntimeError("File already closed")
        if not self.fs_supports_append:
            if self.write_temp is not None:
                # read temp file from the beginning
                self.write_temp.flush()
                self.write_temp.seek(0)
                chunk = self.write_temp.read()
                if chunk is not None:
                    # write full contents and keep in temp file
                    self.fs.write(self._filename, chunk, self.binary_mode)
                    self.write_temp.seek(len(chunk))

    def close(self):
走神的阿圆's avatar
走神的阿圆 已提交
557 558
        if isinstance(self.fs, BosFileSystem):
            self.fs.append(self._filename, b'', self.binary_mode, force=True)
559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
        self.flush()
        if self.write_temp is not None:
            self.write_temp.close()
            self.write_temp = None
            self.write_started = False
        self.closed = True


def exists(path):
    return default_file_factory.get_filesystem(path).exists(path)


def makedirs(path):
    return default_file_factory.get_filesystem(path).makedirs(path)


def join(path, *paths):
    return default_file_factory.get_filesystem(path).join(path, *paths)


def walk(dir):
    return default_file_factory.get_filesystem(dir).walk(dir)